Publications

Fine-grained wildlife classification in aerial imagery is limited not only by model performance, but also by unreliable labels: animals occupy few pixels, key visual cues vary seasonally, and modality-specific evidence can be ambiguous. We study adult-male identification in red deer (Cervus elaphus), where the antler cycle defines predictable windows of reliable evidence for both annotation and prediction. Using 7,295 RGB-only, thermal-only, and matched RGB+thermal crop sets from low-altitude UAV surveys, labeled by three annotators, we show that seasonal structure links (I) annotation quality, (II) downstream classification, and (III) selective prediction. Matched RGB+thermal review resolves more samples than either single modality, recovering majority-male labels otherwise missed by RGB or thermal alone, in human-based as well as model-based classification. Months with high annotator abstention also show lower classifier confidence, and soft seasonal priors mainly benefit the season-limited thermal view. Uncertainty-band abstention further raises covered accuracy to 98.9%, though at reduced coverage and with deferral that falls disproportionately on males. Overall, a biologically grounded seasonal calendar predicts where annotation and prediction are unreliable, and can guide both annotation protocol design and modality weighting.

This poster investigates the use of synthetic, non-radiometric thermal UAV imagery for training animal-detection models, presented as Article 68 in the SIGGRAPH 2026 Posters programme. By combining computer-graphics techniques with computer-vision methods, the work explores how simulated thermal drone footage of forests can support wildlife detection where real annotated thermal data is scarce.

The BAMBI dataset is a large-scale airborne multispectral wildlife dataset comprising 389 paired RGB and thermal aerial video sequences recorded across diverse forest and forest-adjacent habitats in Austria. Each frame is geo-referenced with precise global coordinates (longitude, latitude, and altitude), enabling learning and evaluation in both image space and geographic space. The dataset contains 5,305 matched RGB-thermal labelled image pairs with 13,057 co-registered bounding box annotations, enabling direct comparison between both sensing modalities and supporting research on multimodal wildlife detection, cross-modal perception, and RGB-thermal fusion methods.

Hospitals, due to their complex and dynamic nature, face significant challenges in aligning clinical workflows with established treatment protocols. In this study, we analyzed HL7 FHIR AuditEvent sequences from 10 randomly selected chest pain cases to model clinical process flows and assess adherence to a standard operating procedure (SOP) defined by Medical University of Vienna. Despite the limited dataset, our findings reveal several deviations from the defined SOP, illustrating the fragmented nature of real-world clinical workflows. This work demonstrates the potential of combining medical data standards with process analytics to turn clinical data into actionable insights for improving patient care.

Achieving interoperability with HL7 FHIR in practice requires creating a FHIR Implementation Guide (IG). Between the many IGs at international, national, and local levels, incompatibilities are inevitable — for example, one IG requiring a field to be filled in while another explicitly forbids it. The same application, such as a FHIR server or FHIR facade, must then handle inherently incompatible resource structures. We use Germany as a real-world failure case: three competing IGs (basisprofile-de, KBV-Basis-Profile, and ISiK), which were not aligned during development, have become mandatory; many organizations must implement all three for the same patient. We systematically compare KBV Base 1.x and ISiK Basismodul v2, identifying hard structural conflicts including closed versus open slicing on Condition.onset[x] and disjoint address and name type constraints on Patient, Practitioner, and RelatedPerson. We show one possible technical response: we cannot resolve the fact that FHIR disallows one resource from conforming to all three IGs at once, but our namespace-based data management in Neo4j is designed to allow storing data without redundancy and delivering the same patient in three different formats depending on which profile the client requests. Future work includes integrating this model with FHIR profile comparison tooling, evaluating performance at scale, and aligning with HL7 Europe EHDS implementation guides as they mature.

Recent advancements in neural scene representations, specifically Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS), have revolutionized novel-view synthesis. However, these methods are predominantly optimized for and evaluated on ground-based, visible-spectrum (RGB) data, leaving a significant gap in their application to aerial, radiometric thermal imagery, which is critical for industrial inspection, search-and-rescue, and environmental monitoring. In this work, we systematically investigate the applicability of these state-of-the-art paradigms to radiometric thermal imagery acquired from airborne drone platforms. We introduce a novel, publicly available multimodal dataset captured using a DJI M30T system, comprising synchronized RGB and radiometric thermal frames of a building. We conduct a comprehensive evaluation comparing specialized thermal approaches (ThermalNeRF, ThermoNeRF, Thermal3DGS) against general-purpose methods (nerfacto, gsplat). Our assessment utilizes a suite of quantitative metrics (PSNR, SSIM, MAE, LPIPS, and DISTS) complemented by qualitative visual analysis. Results indicate that Thermal3DGS achieves state-of-the-art performance in the thermal domain (PSNR 22.99, SSIM 0.845), effectively mitigating artifacts common in low-texture thermal data. Conversely, gsplat demonstrates superior RGB synthesis and competitive thermal performance, suggesting that general-purpose splatting representations are robust enough for cross-spectral applications. This work bridges the gap between aerial radiometric sensing and neural rendering, demonstrating that off-the-shelf drone thermography can be utilized for high-fidelity 3D thermal reconstruction with minimal adaptation.

We employ an iterative LLM-assisted code-generation workflow coupled with static analysis. For each prompt, the LLM produces an initial code candidate, which is evaluated using SonarQube. Analysis findings (security, quality, maintainability) are fed back to the model to guide regeneration. Iteration continues until issue thresholds are satisfied or a maximum iteration count is reached. We evaluated our approach on 50 code generation prompts covering security vulnerabilities, code quality, algorithmic tasks, and complexity challenges. Each prompt underwent up to five iterative refinement cycles, incorporating SonarQube feedback into the LLM’s next generation. For each run, we recorded initial and final issue counts, iteration numbers, and categorized issues by type and severity. 94% of all runs successfully completed, resolving all detected issues after an average of 1.34 refinement iterations, with the majority (74%) requiring only a single iteration.

This work studies multimodal sex and life-stage classification of red deer from aerial RGB-thermal drone video. Sex classification is challenging in thermal-only footage because antlers, a key visual cue, become nearly invisible in thermal imagery, whereas they remain clearly visible in RGB. We show that fusing RGB and thermal modalities yields the best classification results, demonstrating that relying on a single sensing modality is insufficient for reliable sex and life-stage classification of wildlife from aerial video.

Drone-based thermal imaging is increasingly used for wildlife monitoring, but tracking individual animals in aerial footage remains difficult in forested habitats due to low texture, occlusions, and dynamic viewpoints. Traditional image-space tracking methods often fail because motion and appearance cues degrade in thermal data. This work investigates geo-referenced, rather than purely image-space, bounding boxes for tracking wildlife in thermal drone videos, comparing local (image-space) and global (geo-referenced) tracking strategies.

Software systems that process structured inputs often lack complete and up-to-date specifications, which specify the input syntax and the semantics of input processing. While grammar mining techniques have focused on recovering syntactic structures, the semantics of input processing remains largely unexplored. In this work, we introduce a novel approach for inferring attributed grammars from parser implementations. Given an input grammar, our technique dynamically analyzes the implementation of recursive descent parsers to reconstruct the semantic aspects of input handling, resulting in specifications in the form of attributed grammars. By observing program executions and mapping the program’s runtime behavior to the grammar, we systematically extract and embed semantic actions into the grammar rules. This enables comprehensive specification recovery. We demonstrate the feasibility of our approach using an initial set of programs, showing that it can accurately reproduce program behavior through the generated attributed grammars.

A vast number of software systems include components that parse and process structured input. In addition to programming languages, which are analyzed by compilers or interpreters, there are numerous components that process standardized or proprietary data formats of varying complexity. Even if such components were initially developed and tested based on a specification, such as a grammar, numerous modifications and adaptations over the course of software evolution can make it impossible to precisely determine which inputs they actually accept.

In this situation, grammar mining can be used to reconstruct the specification in the form of a grammar. Established approaches already produce useful results, provided that sufficient input data is available to fully cover the input language. However, achieving this completeness is a major challenge. In practice, only input data recorded during the operation of the software systems is available. If this data is used for grammar mining, the resulting grammar reflects only the actual processed inputs but not the complete grammar of the input language accepted by the software component. As a result, edge cases or previously supported features that no longer appear in the available input data are missing from the generated grammar.

This work addresses this challenge by introducing a novel approach for the automatic generation of inputs for grammar mining. Although input generators have already been used for fuzz testing, it remains unclear whether they are also suitable for grammar miners. Building on the grammar miner Mimid, this work presents a fully automated approach to input generation. The approach leverages Dynamic Symbolic Execution (DSE) and extends it with two mechanisms to overcome the limitations of DSE regarding structured input parsers. First, the search for new inputs is guided by an iterative expansion that starts with a single-character input and gradually extends it. Second, input generation is structured into a novel three-phase approach, which separates the generation of inputs for parser functions.

The proposed method was evaluated against a diverse set of eleven benchmark applications from the existing literature. Results demonstrate that the approach achieves precision and recall for extracted grammars close to those derived from state-of-the-art grammar miners such as Mimid. Notably, it successfully uncovers subtle features and edge cases in parsers that are typically missed by such grammar miners. The effectiveness of the method is supported by empirical evidence, showing that it can achieve high performance in various domains without requiring prior input samples.

This contribution is significant for researchers and practitioners in software engineering, offering an automated, scalable, and precise solution for grammar mining. By eliminating the need for manual input generation, the approach not only reduces workload but also enhances the robustness and comprehensiveness of the extracted grammars. Following this approach, software engineers can reconstruct specification from existing (legacy) parsers.

We use unmanned aerial drones to estimate wildlife density in southeastern Austria and compare these estimates to camera trap data. Using thermal (IR) and RGB imagery, drones enable efficient, non-intrusive animal counting. Surveys were conducted during the leafless period on single days in October and November 2024 in three areas of a sub-Illyrian hill and terrace landscape. Animals were manually identified in the recorded footage, and three density extrapolation approaches of increasing sophistication were applied: basic area-based methods, bootstrapping, and zero-inflated negative binomial modeling. Results were compared with Random Encounter Model estimates derived from camera trap data. Drone-based approaches generally produced higher density estimates than the comparison method, with one exception in October. Both methodologies provide valuable but distinct perspectives on animal populations; drones offer a promising, scalable method for wildlife density estimation.

Traffic accidents involving wildlife pose a widespread problem globally, harming both humans and nature. These incidents often result in heavy vehicle damage, leading to expensive repairs and insurance claims. To mitigate these accidents, efforts are underway to better understand wildlife populations near high-risk roads and implement preventive measures such as visual or audible wildlife warning devices. To prevent wildlife accidents, high-risk areas must be identified first. In this work, we propose a model that predicts dangerous areas based on animal sightings and apply it to two road segments in Austria.

The provision of customized orthopedic aids, such as prostheses and orthoses, is essential for treating congenital malformations, chronic diseases, and musculoskeletal injuries. Traditionally, these aids have been crafted manually by skilled technicians; however, additive manufacturing techniques and digital methodologies in the context of Industry 4.0 are revolutionizing this field. This work explores the automated customization of orthopedic aids using a two-step pipeline that leverages digital scans and structural analysis: the first step adapts orthopedic aids based on 3D scans of the patient’s body, and the second step employs the Finite Element Method (FEM) to analyze structural integrity.

This paper targets the detection of misinformation in GPT3 generated texts and the FEVER[7] dataset, using Large Language Models (LLM) and Google search. Given the uncertainty associated with LLM-produced text, a requisite arises for a fact-checking system geared towards non-human-written content under varied contexts. In general, there are two different approaches to perform fact checking: manual checking, favored by entities like politifact, and automated or semi-automated checking, like ProoFVer. Our research delvesinto fully automated checking of LLM-produced texts.

In an era characterized by rapid technological advancements and increasingly complex socio-technical systems, especially regarding Artificial Intelligence (AI) based E-Commerce Optimization, the demand for a comprehensive method to resolve problems, which could result from the lack of trust in AI-Software-Systems, has never been more significant. Systems thinking offers a robust framework for understanding the intricate relationships between various components of a system, making it particularly relevant in today’s digital and AI-based software landscape. Concurrently, the E-Commerce sector is undergoing a transformation, driven by advancements in AI and data analytics grounded in legal changes regarding the General Data Protection Regulation (GDPR) and the upcoming EU AI Act. We explore the synergies between AI-based E-Commerce and Systems Thinking by introducing the AI-DATA model, a human-centered Customer Journey Optimization approach in the context of E-Commerce marketing strategy based on Trustworthy AI. AI-DATA is a phase model consisting of Awareness, Interest, Desire, Action, Trust and Again, and is an extension of the AIDA model, which also considers the customer path to contain activities after a transaction has finished, and a cycle of repeat customers.

For some years now, radiation fields and in particular neural radiation fields (NeRF) have represented a ground-breaking advance in computer graphics. They make it possible to generate high-quality new views for scenes that have been captured from different angles using multiple photos or videos. Instead of conventional methods that rely on geometric representations or explicit scene networks, NeRF uses neural networks to directly model the volumetric scene function. In this way, the approach has dramatically changed the landscape of novel-view synthesis, offering unprecedented realism and flexibility in the representation of complex scenes. However, the training of NeRF models is typically based on computationally intensive image-based reconstructions of camera positions and visual features of the targeted scenes using Structure from Motion (SfM). In aerial imaging, camera positions are already explicitly available through accurate global navigation satellite systems (e.g. GPS) and internal sensors of aircraft. In this work, we present a novel processing pipeline developed to effectively utilise image and sensor data captured by unmanned aerial vehicles (UAVs) to train NeRF-like models without the need for SfM.

Over the last 20 years, process mining (PM) has established itself as an independent discipline that aims to analyse and optimise real processes using data from information systems. Our work develops a system for standardised recording and evaluation of healthcare processes by capturing HL7 FHIR audit events and converting them into PM-compatible formats such as XES or OCEL. A simulation of the breast cancer screening process showed that the recorded process steps could be successfully reconstructed and analysed. This system enables the verification of process conformity, early problem identification and contributes to the improvement of health quality.

While adherence to clinical guidelines improves the quality and consistency of care, personalised healthcare also requires a deep understanding of individual disease models and treatment plans. The structured preparation of routine medical data in a specific clinical context, e.g. a treatment pathway outlined in a medical guideline, is currently a challenging task. Medical data is often stored in different formats and systems, and the relevant clinical knowledge that defines the context is not available in machine-readable formats. We present an approach to extracting information from free-text medical documentation using structured clinical knowledge to guide information extraction into a structured and encoded format, overcoming the known challenges for natural language processing algorithms. The preliminary results are encouraging, as one of our methods was able to extract 100% of all data points with an accuracy of 85% in detail. This progress demonstrates the potential of our approach to effectively utilise unstructured clinical data to improve the quality of patient care and reduce the workload of medical staff.

Internet of Things (IoT) devices are ubiquitous, but often cannot be permanently connected to the power grid and therefore have to rely on batteries and other energy sources like solar power. These mobile devices should run over a long period without the need for recharging, but still be able to perform comparatively complex tasks from the realm of machine learning like a classification task on live sensor data. The state-of-the-art as far as accuracy is concerned are Neural Network models like LSTM (Long short-term memory) or CNN (Convolutional neural network). However, these models usually take up more Flash memory, more RAM of the devices and more CPU time than traditional machine learning approaches, which in turn causes a higher energy demand. Another possibility is to transfer the time series data to an external server to do the calculations there and send back the result. For a continuous classification of sensor data, this approach is often not feasible, since the constant transmission of a large quantity of data over a wireless network is required. We discuss a pipeline for efficient time series classification of live sensor data on low-energy IoT devices with on-device processing. We present our case study of detecting boats docking at an IOT-enabled buoy, which is required to be energy-self-sufficient for several months at a time.

In the quest for improving systems, understanding user behaviour is a crucial step towards achieving optimal results. One technique that aids in comprehending user behaviour is process mining, which involves analysing event logs of a system. However, the accuracy of the process mining results can be impeded by large amounts of data or very unstructured processes, leading to confusion and difficulties in interpretation. To overcome this challenge, preprocessing techniques exist to enhance the resulting process maps. This work proposes a preprocessing method that utilizes an event hierarchy to produce a sanitized process map, that is easier to interpret. The proposed method involves analysing the event logs and creating a hierarchical structure of the events. This hierarchy allows the identification of relevant events and their corresponding dependencies, which in turn enables the extraction of useful process information. This preprocessing method has been tested on event data from an online newspaper, and it has shown promising results in improving readability of the resulting process maps. By utilizing an event hierarchy, relevant information can be extracted from large amounts of data and unstructured processes, leading to cleaner process maps that aid in understanding user behaviour.

Identification and quantitative segmentation of individual blood vessels in mice visualized with preclinical imaging techniques is a tedious, manual or semiautomated task that can require weeks of reviewing hundreds of levels of individual data sets. Preclinical imaging, such as micro-magnetic resonance imaging (µMRI) can produce tomographic datasets of murine vasculature across length scales and organs, which is of outmost importance to study tumor progression, angiogenesis, or vascular risk factors for diseases such as Alzheimer’s. Training a neural network capable of accurate segmentation results requires a sufficiently large amount of labelled data, which takes a long time to compile. Recently, several reasonably automated approaches have emerged in the preclinical context but still require significant manual input and are less accurate than the deep learning approach presented in this paper – quantified by the Dice score. In this work, the implementation of a shallow, three-dimensional U-Net architecture for the segmentation of vessels in murine brains is presented, which is (1) open-source, (2) can be achieved with a small dataset (in this work only 8 μMRI imaging stacks of mouse brains were available), and (3) requires only a small subset of labelled training data. The presented model is evaluated together with two post-processing methodologies using a cross-validation, which results in an average Dice score of 61.34% in its best setup. The results show, that the methodology is able to detect blood vessels faster and more reliably compared to state-of-the-art vesselness filters with an average Dice score of 43.88% for the used dataset.

This paper presents an ontology and process mining-based system that employs semantic technologies for targeted e-commerce marketing strategy suggestions. The proposed approach utilizes ontologies as a knowledge representation formalism to capture the domain knowledge, product descriptions, and website texts in the e-commerce domain. The ontology is populated with relevant attributes and corresponding values, allowing for a structured and standardized representation of users and content.

To suggest targeted marketing strategies, the system employs interest and intensity calculation techniques that utilize the concept of scroll depth and others to measure the interest in different contents and customer preferences.

Process mining techniques identify moments of truth along the customer journey and happy paths a customer usually goes through. Once the ontology is combined with the process mining results, it can be identified in which phase of the customer journey a user is, and it can be determined what marketing actions to employ to get them to enter the next phase of a marketing model.

Overall, the proposed approach is a promising approach to address the challenges of personalized e-commerce marketing strategy suggestions by utilizing ontology-based knowledge representation and intensity calculation techniques. The system has the potential to provide personalized and targeted marketing strategies for e-commerce businesses, leading to increased customer engagement, loyalty, and sales.

Although orthopedics is becoming increasingly important as a medical domain, especially in emerging countries, the level of automation is still marginal and hardly any Industry 4.0 paradigms have been implemented. In this scientific work, solution concepts for holistic process automation in orthopedics are introduced so that prosthetic covers and orthoses for different body regions can be automated by using AI and evaluated with sensor networks. In this process, body scan models are adapted to the conditions of the anatomy or prosthesis models, so that stability as well as fitting accuracy are given in comparison with the other half of the body. Automation in the field of orthopedics leads not only to a significant reduction in costs but can also help to close the research gap regarding objectifiability of results. The first partial aspects have already been successfully implemented for leg prostheses, arm prostheses and shoe insoles with the aid of machine learning processes and physical models for elastic form fitting. As soon as the overall process has been realized, the applicability will be validated in the following year of the project by means of clinical studies and evaluated by utilizing sensor networks for pressure and temperature measurements.

Climate change poses a grave threat to habitats such as forests, endangering the integrity and biodiversity of the global flora and fauna. Accurate surveying techniques are crucial to model populations, detect over and under populations, and address them accordingly. This work proposes a process for creating a biodiversity model of a forest’s fauna using uncrewed aerial vehicles equipped with RGB and thermal cameras. Real-world data, combined with computer-generated imagery and artificial intelligence models, will allow training suitable computer vision models. These models will serve as a reliable and objective data source, enabling the creation of statistical models to describe the monitored forests’ conditions and the biodiversity of its fauna. The proposed methodology is expected to have significant implications for conservation efforts. It should represent a reliable and efficient way to monitor and evaluate forest ecosystems, identifying areas of concern and prioritizing conservation efforts. By providing a comprehensive understanding of the biodiversity within a forest, it could help policymakers make informed decisions and develop effective conservation strategies. Ultimately, this work aims to contribute to the preservation of our planet’s biodiversity and the protection of its habitats.

Time series data is generated in various application areas, such as IoT devices or sensors in vehicles. This type of data is often characterized by a high resource demand due to the interval at which information is measured ranges from daily down to milliseconds. Next to the frequency, the number of data sources, for example hundreds of sensors in modern airplanes generating time series concurrently, is typical for such big data scenarios. Such scenarios require the persistence of the measurements for further evaluations. In this work, we introduce an artificial data benchmark for relational, NoSQL, and NewSQL database management systems in the context of time series. We compare these databases by having multiple read and write data sources accessing the database management systems simultaneously. The evaluation shows that no tested system outperforms all other systems. While DolphinDB shows the highest read performance in single-user scenarios, CrateDB is able to show its advantages regarding when multiple users access the data simultaneously.

Background:

Medical information systems frequently use event logging, but these logs are not suitable for process mining as they are not logged in a standardized format.

Objectives:

Our goal is to enrich medical event logs for use in process mining.

Method:

We present an approach to convert events from standards- based repositories into the XES and OCEL formats commonly used in process mining.

Results:

We tested this approach using simulated data from the Austrian breast cancer screening program.

Conclusion:

We aim to apply it to analyze care guidelines and improve hospital processes in the future.

Undetected errors in software systems are a common cause of vulnerabilities and security holes. Grammar Fuzzing is an effective method for testing these systems, but it has limitations such as lack of knowledge about the semantics of the program and difficulty obtaining grammar for these systems. To address these limitations, we propose an approach to automatically mine grammars, and enhance it with semantic rules and contextual constraints to create attribute grammars. These attribute grammars can then be used for fuzzing. Our preliminary results show that this automated extraction process is feasible, as we successfully applied it to an expression parser and were able to extract an attribute grammar representing the parser’s functionality.

Energy efficiency is an important topic in the area of mobile computing. Developers are often unaware of the impact their choices ondata type use and algorithm design have on this non-functional property. Software energy consumption profiling can be utilized toidentify the energy behaviour of implemented methods, while pattern mining can be utilized to identify recurring patterns in themethods being run. We present a methodology to combine energy consumption profiling and discriminative pattern mining to identifyenergy efficiency patterns. In a study of eight sorting algorithms implemented in Java with the data types int, double and Comparable,profiled on the Android platform, we manage to identify significant patterns in the source code of these 24 implementations. Theresults show that patterns can be identified for both, the data type in use, and for the energy behaviour of efficient or inefficient sortingalgorithms, that explain the observed energy profiles.

Deep learning approaches are highly influenced by two factors, namely the complexity of the task and the size of the training data set. In terms of both, the extraction of features of low-stature alpine plants represents a challenging domain due to their fuzzy appearance, a great structural variety in plant organs and the high effort associated with acquiring high-quality training data for such plants. For this reason, this study proposes an approach for training deep learning models in the context of alpine vegetation based on a combination of real-world and artificial data synthesised using Generative Adversarial Networks. The evaluation of this approach indicates that synthetic data can be used to increase the size of training data sets. With this at hand, the results and robustness of deep learning models are demonstrated with a U-Net segmentation model. The evaluation is carried out using a cross-validation for three alpine plants, namely Soldanella pusilla, Gnaphalium supinum, and Euphrasia minima. Improved segmentation accuracy was achieved for the latter two species. Dice Scores of 24.16% vs 26.18% were quantified for Gnaphalium with 100 real-world training images. In the case of Euphrasia, Dice Scores improved from 33.56% to 42.96% using only 20 real-world training images.

Nursing care is a crucial part of health care, especially in our ageing society. Digitalization in this area is lacking when attempting to conduct a machine-readable, standardized, healthcare data exchange between different services. To tackle this issue two standards, International Classification for Nursing Practice (ICNP) and Systematized Nomenclature of Medicine Clinical Terms (SNOMED CT), identified as highly relevant code systems in this sector via expert interviews, were analysed. A competency catalogue of 241 terms was then mapped to both of these systems, in some case extending them with new concepts and terms. The results show that neither code system was able to translate many of the terms without extension, with ICMP covering 12,03% and SNOMED CT covering 26,14%. However, both systems can be extended, with SNOMED CT showing a better capability to support data exchange in nursing care.

Background:

Many components must work together to continuously improve processes in healthcare organizations. Process mining has recently developed into a discipline that can make a significant contribution here.

Objectives:

We want to extend an existing management tool to assess and improve the capability of organizations in this area.

Method:

We add a dimension to the adoption readiness assessment and maturity model for sharable clinical pathways to assess and improve event data quality.

Results:

We present different approaches for formal and checkpoint assessments and an embedding of the improvement strategy with examples.

Conclusion:

The additional dimension from the process mining domain integrates with the existing model. At all levels, links can be established between the various aspects of event data quality with existing dimensions. The model has yet to be tested in a real-world use case.

This study addresses the representation and exchange of geospatial geometric 3D models, which is a common requirement in various applications like outdoor mixed reality, urban planning, and disaster risk management. Over the years, multiple file formats have been developed to cater to diverse needs, offering a wide range of supported features and target areas of application. However, classic exchange formats like the JavaScript Object Notation and the Extensible Markup Language have been predominantly favored as a basis for exchanging geospatial information, leaving out common geometric information exchange formats such as Wavefront’s OBJ, Stanford’s PLY, and OFF. To bridge this gap, our research proposes three novel extensions for the mentioned geometric file formats, with a primary focus on minimizing storage requirements while effectively representing geospatial data and also allowing to store semantic meta-information. The extensions, named GeoOBJ, GeoOFF, and GeoPLY, offer significant reductions in storage needs, ranging from 14 to 823% less compared to standard file formats, while retaining support for an adequate number of semantic features. Through extensive evaluations, we demonstrate the suitability of these proposed extensions for geospatial information representation, showcasing their efficacy in delivering low storage overheads and seamless incorporation of critical semantic features. The findings underscore the potential of GeoOBJ, GeoOFF, and GeoPLY as viable solutions for efficient geospatial data representation, empowering various applications to operate optimally with minimal storage constraints.

The adoption of the FHIR (Fast Healthcare Interoperability Resources) standard has led to an exponential growth of modular healthcare data that needs to be managed efficiently. Graph databases such as Neo4j offer an effective way to store and query this data, but can become complex when dealing with FHIR resources that contain numerous extensions. We explore the use of namespaces in Neo4j graph databases to manage FHIR resources and compare it with the existing tool, CyFHIR. We demonstrate that by embedding extensions using the namespace concept, the complexity of the graph can be significantly reduced. Furthermore, we evaluate our approach on a generated dataset and show that the use of namespaces in Neo4j outperforms CyFHIR conventional methods for storing FHIR resources in graph databases. Our findings suggest that the use of namespaces can be a valuable addition to Neo4j graph databases for managing complex FHIR resources.

In this work, a domain-driven design process is proposed, allowing to define a contract regarding the structure of the used data within a multi-language full-stack application using a model-to-text transformation approach. This process is based on a metamodel, which allows for the implementation of language-specific transformers to generate representations of the data as domain models for arbitrary programming languages. It is evaluated in the context of a modern full-stack microservice architecture including a C# and Python based backend with access to a database using object-relational-mapping principles and a TypeScript based frontend, connected by a GraphQL interface. The proposed methodology is intended for an agile modelling process including automatic adaptions, but also the preservation of user-defined extensions within the generated source-code artifacts.

Foot disorders, a commonly overlooked issue, are prevalent in developed societies. These disorders can have a significant impact on a person’s quality of life and can even be debilitating, regardless of age. Non-invasive pedobarographic examinations, which evaluate the pressure fields of the plantar surface of the foot and a supporting surface, allow for analysis of a patient’s gait and posture using 2D footprints or scans. This data can then be used by orthopaedists to create customized shoes or insoles as correctives. However, the lack of standardized protocols and guidelines and the scarcity of evidence-based information can lead to a subjective evaluation and selection of correctives by the orthopaedist. This study proposes an objective and quantifiable method for the classification of footprints using computer vision paradigms, in order to create more appropriate correctives. The results show that the proposed machine learning models are able to correctly identify the required one of three correctives with an accuracy of 70% for RGB scans and 49% for blueprints. Based on the current results, future work should focus not only on the classification of suitable correctives, but also on the determination of the corrective’s position to ensure the best possible patient outcomes.

This study addresses the pressing issue of wildlife-related road accidents in Austria by developing an innovative approach using georeferenced drone data and a sophisticated statistical risk model. Despite technological advancements, accidents involving wild animals continue to pose a significant threat to road safety. Our research focuses on comprehensive data collection using camera drones equipped with visual and thermal sensors. Through extensive testing in high-risk areas of Upper and Lower Austria, we identified and counted various animals, including roe deer, rabbits, and pheasants, during different times of the day. The collected data were analyzed to create a robust statistical risk model, providing insights into possible animal effects on roads.

Interconnected data or, in particular, graph structures are a valuable source of information. Gaining insights and knowledge from graph structures is applied throughout a wide range of application areas, for which efficient tools are desired. In this work we present an open source Java graph transformation framework. The framework provides a simple fluent Application Programming Interface (API) to transform a provided graph structure to a desired target format and, in turn, allow further analysis. First, we provide an overview on the architecture of the framework and its core components. Second, we provide an illustrative example which shows how to use the framework’s core API for transforming and verifying graph structures. Next to that, we present an instantiation of the framework in the context of analyzing the third-party dependencies amongst open source libraries on the Android platform. The example scenario provides insights on a typical scenario in which the graph transformation framework is applied to efficiently process complex graph structures. The framework is open-source and actively developed, and we further provide information on how to obtain it from its official GitHub page.

Managing the member data of social clubs can be a tedious task. However, there are software solutions available that can help streamline this process. This although means, that existing member data, that is often in the form of text-based data formats like CSV, or semi-structured formats like XML, or Excel needs to be imported in those tools. Unfortunately, the data in these formats may contain errors, inconsistencies, and missing values, which can compromise the usability of this data. In this work, a rule-based data cleansing pipeline designed to clean, enrich, and transform social club member data into a suitable format for import into software solutions is presented. The approach is evaluated on a small data sample and shows promising results for such an application scenario.

Due to the apparent effects of climate change on the Earth’s ecosystems, it is more important than ever to monitor flora and fauna in affected regions, e.g. mountain areas above the tree line. In the alpine ecosystem, and not just there, Vegetation plays a fundamental role and is the subject of this study. The work aims to develop algorithms for recognising small stature alpine plants from close range top view images. Ideally, automated assessment algorithms of the plant cover should objectively help scientists observe and interpret the state of the plant ecosystem over a long time series. Therefore, the aim in this respect was to derive visualisations that accurately describe plant growth and displacement (translocation). Additionally, recording changes in biodiversity was an intent. This work uses multi-temporal data comprising RGB images and multi-label masks to accomplish the aforementioned task. The evaluated methods involve mask comparison, optical flow estimation, detection of individual plants, and descriptive statistical analysis of image feature properties. Tests on the given data set show that all methods but the optical flow estimation have great potential. The mask comparison method captured plant growth and translocation most satisfactory. Individual plant detection and statistical analysis further helped to evaluate changes in biodiversity. When combined, the proposed methods give an immediate overview about relevant changes in the multi-temporal transects, which has not been done before for close-distance images of alpine plants.

Emergency responder training can benefit from outdoor use of Mixed Reality (MR) devices to make trainings more realistic and allow simulations that would otherwise not be possible due to safety risks or cost-effectiveness. But outdoor use of MR requires knowledge of the topography and objects in the area to enable accurate interaction of the real world trainees experience and the virtual elements that are placed in them. An approach utilizing elevation data and geographic information systems to create effective occlusion models is shown, that can be used in such outdoor training simulations. The initial results show that this approach enables accurate occlusion and placement of virtual objects within an urban environment. This improves immersion and spatial perception for trainees. In the future, improvements of the approach are planned with on the fly updates to outdated information in the occlusion models.

Recommendation systems can be used in many situations in daily life. Recommending people on social media networks, products in various online shops, music, or movies are only a few use cases of these systems. The cold start problem, when no information about a new or infrequent user is available, is challenging for recommendation systems. We deal with creating restaurant and category recommendations for restaurant visitors. Recommendations are generated with different metrics and technologies based on user profiles to make recommendations as individual as possible. We use kMeans and Mean-Shift for clustering users to build a base for recommendations generated using user-based and contentbased collaborative filtering methods. These suggestions consider the location of restaurants, the similarity between users and restaurants, and the ratings users give. We mitigate the cold-start problem by using matrix factorization and spatial information for users with few restaurant visits in the past. Recommendations are evaluated and adapted as a result of other user behavior to obtain better results. As a result, we can query recommendations via an Application Programming Interface (API), which consist of a mixture of location and user-based recommendation to please the users’ needs by combining exploration and exploitation.

We present a novel approach to store data with different contexts inside a property graph model. We introduce namespaces, similar to namespaces in XML, and extend nodes and relationships with labels to assign them to a specific context, i.e. namespace. Individual properties of a node or relationship can also be put in a namespace. This work is specifically targeting the utilization in graph databases, with a reference implementation provided via the Neo4j database. In addition to the theoretical approach, an object to graph mapper for the programming language Java is implemented and used to evaluate the approach. As an evaluation example, a university organization is used, which is split into two domains. The experiments show, that information of different domains can be stored within the same model using namespaces. Thus, it is possible to reuse shared information over multiple contexts, which reduces data duplication in the graph database, as otherwise multiple nodes would be required.

Aerial photographs of buildings are often used as memorabilia sold by trading companies. Such photographs come with an issue regarding the address of the shown buildings, since the recording location of the camera may be known, but shows a spatial distance to the actual subject of the image. In addition to that, also this recording location is often not known in detail but only roughly in the form of the flight route/area. To address this problem, a methodology for reverse geocoding is proposed, allowing to identify the position of buildings that are photographed from aerial vehicles. This is done using a process for extending recording locations and a second process based on the registration of invariant features within aerial shots compared to maps.

The integration of renewable local energy generation in single households – turning the household into a “prosumer” – is an important way to support an ecological transition of the electric power system. However, due to the volatile and distributed nature of most renewable energy sources, the power system may face stability problems when integrating a large number of renewables. The paper at hand describes an approach to overcome these shortages in a two-fold manner: First, the effects of the installed renewables shall be limited locally to a group of households – a so-called “energy community”. To do so, all the participating households are using existing self-consumption optimization tools. However, when a household has excess energy which can not be consumed locally, this energy is shared among the other participating households by using a cluster storage device, thus enabling a community self-consumption before feeding into the low-voltage distribution grid. Second, the connected operator may request flexibility from the participating households. For that, additional loads or load sheds are triggered by the requesting grid operator, depending on the current situation in the grid. The households decide autonomously about the amount of granted flexibility, receiving respective financial incentives. This work introduces an energy management concept and a prototypical control infrastructure used for the aforementioned functionalities. In a number of simulations and field tests, the proposed approach was successfully evaluated. The article provides a comprehensive overview of the gained results and the conclusions derived from them.

The inspection of products and the assessment of quality is connected with high costs and time effort in many industrial domains. This also applies to the forestry industry. Utilizing state-of-the-art deep learning models allows the analysis automation of wooden piles in a vision-based manner. In this work a three-step approach is presented for the localization, segmentation and multi-facet classification of individual logs based on a client/server architecture allowing to determine the quality, volume and like this the value of a wooden pile based on a smartphone application. Using multiple YOLOv4 and U-NET models leads to a client-side log localization accuracy of 82.9% with low storage requirements of 23 MB and a server-side log detection accuracy of 94.1%, together with a log type classification accuracy of 95% and 96% according to the quality assessment of spruce logs. In addition, the trained segmentation model reaches an accuracy of 89%.

Historical and current aerial photographs are only of great value if the geolocation or address of the photographed areas is also available. In Western Europe, especially Austria, Germany and Czech Republic, there is a market for the sale of aerial photographs of one’s own private residential building. Automated geolocation is a feasible way to enable the sales agents to assign the addresses for the sale more quickly. In the course of this research work, a process chain is modeled that allows the assignment of aerial photographs to residential addresses using machine vision. After model-based rectifying the aerial images to compensate for perspective distortions, larger image blocks get assembled using image stitching. The assignment to a 2D reference map, such as satellite imagery via Google Maps, is done by applying a U-Net CNN after extracting durable image features such as roads or buildings. The mapping of aerial imagery to two-dimensional cartography is either automated via registration approaches or based on manually placed corresponding landmarks and homography. Test runs on imagery between the years 1969 and 2020 show that the labor-intensive process of geolocation of aerial imagery can be solved by the proposed process model in a hybrid way.

The inspection of products and assessment of quality is connected with high costs and time effort in many industrial domains. This also applies to the forestry industry. Utilizing state-of-the-art deep learning models allows automizing the analysis of wooden piles in a vision-based manner. In this work a parallel two-step approach is presented for the segmentation and multi-facet classification of individual logs, according to the wood type and quality. The present approach is based on a preliminary log localization step and like this allows determining the quality, volume and also the value of individual logs, respectively the whole wooden pile. Using a YOLOv4 model for wood species classification for douglas firs, pines and larches results in an accuracy of 74.53%, while a quality classification model for spruce logs reaches 86.58%. In addition to that, the trained U-NET segmentation model reaches an accuracy of 93%. In the future, the underlying data set and models will be further improved and integrated to a mobile application for the on site analyzation of wooden piles by foresters.

This work discusses and contains content that may be offensive or unsettling. Hateful communication has always been part of human interaction, even before the advent of social media. Nowadays, offensive content is spreading faster and wider through digital communication channels. To help improve regulation of hate speech, we introduce ProperBERT, a fine-tuned BERT model for hate speech and offensive language detection specific to English. To ensure the portability of our model, five data sets from literature were combined to train ProperBERT. The pooled dataset contains racist, homophobic, misogynistic and generally offensive statements. Due to the variety of statements, which differ mainly in the target the hate is aimed at and the obviousness of the hate, a sufficiently robust model was trained. ProperBERT shows stability on data sets that have not been used for training, while remaining efficiently usable due to its compact size. By performing portability tests on data sets not used for fine-tuning, it is shown that fine-tuning on large scale and varied data leads to increased model portability.

Process mining techniques can be used to analyse business processes using the data logged during their execution. These techniques are leveraged in a wide range of domains, including healthcare, where it focuses mainly on the analysis of diagnostic, treatment, and organisational processes. Despite the huge amount of data generated in hospitals by staff and machinery involved in healthcare processes, there is no evidence of a systematic uptake of process mining beyond targeted case studies in a research context. When developing and using process mining in healthcare, distinguishing characteristics of healthcare processes such as their variability and patient-centred focus require targeted attention. Against this background, the Process-Oriented Data Science in Healthcare Alliance has been established to propagate the research and application of techniques targeting the data-driven improvement of healthcare processes. This paper, an initiative of the alliance, presents the distinguishing characteristics of the healthcare domain that need to be considered to successfully use process mining, as well as open challenges that need to be addressed by the community in the future.

Background: HL7 Austria is a non-profit association dedicated to improving electronic data communication and interoperability in healthcare using the HL7 international standards. Objectives: We aim to provide an open infrastructure to develop, manage, and maintain HL7 FHIR implementation guides. Methods: We utilize state-of-the-art open-source tooling developed by the FHIR community to support continuous integration. Results: The implementation guides can be published as static HTML websites and maintained using GitHub. Conclusion: The solution supports all steps of a standard’s lifecycle, from drafting and reviewing to balloting, publishing, and maintenance.

This work introduces a concept for rule based model verification using a graph database on the example of Neo4j and its query language Cypher. An approach is provided that allows to define verification rules using a graph query language to detect transformation errors within a given domain model. The approach is presented based on a running example, showing its capability of detecting randomly generated errors in a transformation process. Additionally, the method’s performance is evaluated using multiple subsets of the IMDb movie data with a maximum of 17,000,000 nodes and 41,000,000 relationships. This performance evaluation is carried out in comparison to the Object Constraint Language, showing advantages in the context of highly connected datasets with a high number of nodes. Another benefit is the utilization of a well established graph database as verification tool without any need for re-implementing graph and pattern matching logic.

The Hohe Tauern National Park has founded an interdisciplinary monitoring and research program for long-term observation of alpine ecosystems. This initiative provides – among other findings – an ongoing digital image archive in the form of strictly standardized (geo-static, colourfast), high-resolution (1 px. ≈ 0,1mm) nadir photos (view vertical to the ground) with high information content and great relevance in terms of documentation. These data, comparable to earth orthophotos, represent the basis for the project at hand. It focuses on developing a software prototype to automatically recognize plants from image data using computer vision and machine learning. The goals are threefold. First, the reliable recognition of individual plant species and their individuals, despite overlap with other plants or vegetation structures, is aimed. Secondly, the variation in nature and thus divergent appearance of a specimen is addressed. Thirdly, it should be possible to detect identical plants within a time series. In addition, this should allow the models to be updated when new data are available.

Nowadays there is an increased pressure on mobile app developers to take non-functional properties into account. An app that is too slow or uses much bandwidth will decrease user satisfaction, and thus can lead to users simply abandoning the app. Although automated software improvement techniques exist for traditional software, these are not as prevalent in the mobile domain. Moreover, it is yet unknown if the same software changes would be as effective. With that in mind, we mined overall 100 Android repositories to find out how developers improve execution time, memory consumption, bandwidth usage and frame rate of mobile apps. We categorised non-functional property (NFP) improving commits related to performance to see how existing automated software improvement techniques can be improved. Our results show that although NFP improving commits related to performance are rare, such improvements appear throughout the development lifecycle. We found altogether 560 NFP commits out of a total of 74,408 commits analysed. Memory consumption is sacrificed most often when improving execution time or bandwidth usage, although similar types of changes can improve multiple non-functional properties at once. Code deletion is the most frequently utilised strategy except for frame rate, where increase in concurrency is the dominant strategy. We find that automated software improvement techniques for mobile domain can benefit from addition of SQL query improvement, caching and asset manipulation. Moreover, we provide a classifier which can drastically reduce manual effort to analyse NFP improving commits.

Emergency responders or task forces can benefit from outdoor Mixed Reality (MR) trainings, as they allow more realistic and affordable simulations of real-world emergencies. Utilizing MR devices for outdoor situations requires knowledge of real-world objects in the training area, enabling the realistic immersion of both, the real, as well as the virtual world, based on visual occlusions. Due to spatial limitations of state-of-the-art MR devices recognizing distant real-world items, we present an approach for sharing geo-referenced 3D geometries across multiple devices utilizing the CityJSON format for occlusion purposes in the context of geospatial MR visualization. Our results show that the presented methodology allows accurate conversion of occlusion models to geo-referenced representations based on a quantitative evaluation with an average error according to the vertices’ position from 1.30E-06 to 2.79E-04 (sub-millimeter error) using a normalized sum of squared errors metric. In the future, we plan to also incorporate 3D reconstructions from smartphones and drones to increase the number of supported devices for creating geo-referenced occlusion models.

Healthcare processes have many particularities captured and described within standards for medical information exchange such as HL7 FHIR. BPMN is a widely used standard to create readily understandable processes models. We show an approach to integrate both these standards via an automated transformation mechanism. This will allow us to use the various tools available for BPMN to visualize and automate processes in the healthcare domain. In the future we plan to extend this approach to enable mining and analyzing executed processes.

Fitting real world data into a model for classification, is a challenging task. Modern approaches to classification are often resource intensive and may become bottlenecks. A microservice architecture that allows maintaining a model of real world data, and adding new information as it becomes available is presented in this paper. Updates to the model are handled via different microservices. The architecture and connected workflows are demonstrated in a use case of classifying text data in a taxonomy represented by a directed acyclic graph (DAG). The presented architecture removes the classification bottleneck, as multiple data points can be added independent of each other, and reading access to the model is not restricted. Additional microservices also enable a manual intervention to update the model.

Nowadays, news spread quickly, and it is not always clear to the reader whether an article is real or fake. Moreover, readers use only a few sources to read the news without knowing the quality of the source. This is due to a lack of up-to-date news or media rankings. Machine learning models can be used to automatically detect fake news. In this work, a Passive-Aggressive-Classifier, a Random-Forest, and an LSTM network are trained to distinguish between fake and non-fake (real) news. Moreover, these models are used to classify news sources according to the amount of possible Fake News they may spread. The models are tested on English and translated German articles. The best results for Fake News detection on English articles is reached with the Passive-Aggressive-Classifier. For automatic news ranking of translated German articles, Random-Forest provides the best result. The correlation of Random-Forest with an actual news ranking reached 0.68. This shows that automated classification can be extended to languages other than English, using this approach. In the future, other machine learning models and translators will be used to extend the approach.

The COVID-19 pandemic has highlighted some of the opportunities, problems and barriers facing the application of Artificial Intelligence to the medical domain. It is becoming increasingly important to determine how Artificial Intelligence will help healthcare providers understand and improve the daily practice of medicine. As a part of the Artificial Intelligence research field, the Process-Oriented Data Science community has been active in the analysis of this situation and in identifying current challenges and available solutions. We have identified a need to integrate the best efforts made by the community to ensure that promised improvements to care processes can be achieved in real healthcare. In this paper, we argue that it is necessary to provide appropriate tools to support medical experts and that frequent, interactive communication between medical experts and data miners is needed to co-create solutions. Process-Oriented Data Science, and specifically concrete techniques such as Process Mining, can offer an easy to manage set of tools for developing understandable and explainable Artificial Intelligence solutions. Process Mining offers tools, methods and a data driven approach that can involve medical experts in the process of co-discovering real world evidence in an interactive way. It is time for Process-Oriented Data scientists to collaborate more closely with healthcare professionals to provide and build useful, understandable solutions that answer practical questions in daily practice. With a shared vision, we should be better prepared to meet the complex challenges that will shape the future of healthcare.

Enterprise risk management is a well established methodology used in industry. This area relies heavily on risk owners and their expert opinion. In this work, we present an approach to a semi-automated risk detection for companies using anomaly detection. We present various anomaly detection algorithms and present an approach on how to apply them on multidimensional data sources like news articles and stock data to automatically extract possible risks. To do so, NLP methods, including sentiment analysis, are used to extract numeric values from news articles, which are needed for anomaly analysis. The approach is evaluated by conducting interview questionnaires with domain experts. The results show that the presented approach is a useful tooling that helps risk owners and domain expert to find and detect potential risks for their companies.

Threat Information exchange is a highly relevant topic in today’s environment of increasing data breaches, hacks and scams. Standardized formats for exchanging such information exist, but if and how they are used by an active community is determinant for gaining information from such provided information. We provide an in depth analysis of the current state of the Structured Threat Information Expression (STIX) standard, consisting of 5 different active threat information providers. Based on an analysis of 480,867 threat information objects, we find that the STIX standard is not used to its full capabilities, and lacks usefulness due to the quality and upto-dateness of the information. We give suggestions for future improvements of standards based threat information exchange, such as more adherence to the core standard, and fostering an active community.

We present Amaru, a framework for Genetic Improvement utilizing Abstract Syntax Trees directly at the interpreter and compiler level. Amaru also enables the mining of frequent, discriminative patterns from Genetic Improvement populations. These patterns in turn can be used to improve the crossover and mutation operators to increase population diversity, reduce the number of individuals failing at run-time and increasing the amount of successful individuals in the population.

Modern Internet of Things solutions using edge devices produce large amounts of raw data. In order to utilize this data, it needs to be processed, aggregated, and categorized to enable decision making for management and end-users. This data management is a non-trivial task, as the computational load is directly proportional to the amount of data. In order to tackle this issue, we provide an extensible and scalable microservice architecture that can receive, normalize, and filter the raw data and persist it in different levels of aggregation, as well as for time series analysis.

Time series data is created in a variety of application areas such as sensors in cars, smartwatches or IoT devices. This kind of data is often characterized by high resource demand due to the frequency the information is measured, with data points once a day, hour and even down to milliseconds. While real-time processing of such data is often sufficient, there are also many use cases, where batch processing and consequently the storage and managed access of measurements is required. For this reason, this work evaluates different database management systems in the context of storing time related data using different data models such as classical relational models, non-relational models using NoSQL database systems and the recently upcoming group of NewSQL databases. The evaluation shows that a highly optimized time series databases such as InfluxDB is able to outperform the other tested systems regarding write-throughput and RAM as well as disk utilization in a single server setup.

Elevators, a vital means for urban transportation, are generally lacking proper emergency call systems besides an emergency button. In the case of unconscious or otherwise incapacitated passengers this can lead to lethal situations. A camera-based surveillance system with AI-based alerts utilizing an elevator state machine can help passengers unable to initiate an emergency call. In this research work, the applicability of RGB-D images as input for instance segmentation in the highly reflective environment of an elevator cabin is evaluated. For object segmentation, a Region-based Convolution Neural Network (R-CNN) deep learning model is adapted to use depth input data besides RGB by applying transfer learning, hyperparameter optimization and re-training on a newly prepared elevator image dataset. Evaluations prove that with the chosen strategy, the accuracy of R-CNN instance segmentation is applicable on RGB-D data, thereby resolving lack of image quality in the noise affected and reflective elevator cabins. The mean average precision (mAP) of 0.753 is increased to 0.768 after the incorporation of additional depth data and with additional FuseNet-FPN backbone on RGB-D the mAP is further increased to 0.794. With the proposed instance segmentation model, reliable elevator surveillance becomes feasible as first prototypes and on-road tests proof.

The 3D silhouette reconstruction of a human body rotating in front of a monocular camera system is a very challenging task due to elastic deformation and positional mismatch from body motion. Nevertheless, knowledge of the 3D body shape is a key information for precise determination of one’s clothing sizes, e.g. for precise shopping to reduce the number of return shipments in online retail. In this paper a novel three step alignment process is presented, utilizing As-Rigid-As-Possible (ARAP) transformations to normalize the body joint skeleton derived from OpenPose with a CGI rendered reference model in A- or T-pose. With further distance-map accelerated registration steps, positional mismatches and inaccuracies from the OpenPose joint estimation are compensated thus allowing for 3D silhouette reconstruction of a moving and elastic object without the need for sophisticated statistical shape models. Tests on both, artificial and real-world data, generally proof the practicability of th is approach with all three alignment/registration steps essential and adequate for 3D silhouette reconstruction data normalization.

There is a lack of secure official communication channels for peer review and peer feedback on medical findings. Objectives: We aimed to utilize the existing Austrian eHealth infrastructure to enable review and feedback processes. Methods: We extended the IHE XDW workflow document to enable the exchange of text messages (i.e., comments on documents or images) over an XDS infrastructure. Results: The workflow enabled the exchange of comments on specific sections of CDA documents or radiological images and was verified in an XDS test environment. Conclusion: The presented solution is a proof of concept that could lead to the specification of a new IHE workflow definition.

Process mining has become its own research discipline over the last years, providing ways to analyze business processes based on event logs. In healthcare, the characteristics of organizational and treatment processes, especially regarding heterogeneous data sources, make it hard to apply process mining techniques. This work presents an approach to utilize established standards for accessing the audit trails of healthcare information systems and provides automated mapping to an event log format suitable for process mining. It also presents a way to simulate healthcare processes and uses it to validate the approach.

Further information can be found here

Since medical processes are hard to be designed by consensus of experts, the use of data available for creating medical processes is a recurrent idea in literature. Data-driven paradigms are named to be a feasible solution in this field that can support medical experts in their daily decisions. Behind this paradigm, there are frameworks specifically designed for dealing with process-oriented problems. This is the case of process mining.

With recent improvements in deep-learning architectures and availability of GPU hardware, state of the art deep learning (DL) has already manifested as powerful image processing technology in the clinical routine to provide segmentation results of high accuracy. As a drawback, it’s black-box nature does not naturally feature inspection and post-processing by medical experts. We present a Graph segmentation (GS) approach that derives it’s fitness function from arbitrary DL results in a generic way. To allow for efficient and effective post-processing by the medical experts, various interaction paradigms are presented and evaluated in this paper. The trade-off of GS compared to the initial DL results is marginal (delta JI= 0.196%), yet potential DL segmentation errors can be corrected in a reliable way. The intuitive approach shows a high level of both, inter and intra user reproducibility. Change propagation of corrected slices keeps the demand for user-interaction to a minimum when successfully correction potential weaknesses in the DL segmentations. Thereby, the formerly error-prone slice mini-batches get corrected in an automated way with the JI being significantly increased.

In industrial domains with time and cost intensive manual or semi-automated inspection the demand for automation is high. Utilizing state of the art deep learning models for localization in vision-based domains such as wood log analysis, the precision can be increased while reducing the demand for manual inspection. In this paper a YOLO network is trained on wood log images to allow for detection of single wood piles in images with hundreds and thousands instances. Due to the high variability in scale and large amount of wood logs within the images, common YOLO architectures are not applicable. Thus, tiling is necessitated to implicitly form a multi-resolution image pyramid. Due to lack in training data, besides common data augmentation modelling of different seasonal and weather conditions is applied. The wood log detection process can be run on a client/server architecture to allow for both, preview and refined results. Evaluation on real-world data sets shows an log detection accuracy of 82,9% utilizing a tiny YOLO model and 94,1% with a fully connected YOLO model, respectively.

With the widespread use and adoption of mobile platforms like Android a new software quality concern has emerged – energy consumption. However, developing energy-efficient software and applications requires knowledge and like-wise proper tooling to support mobile developers. To this aim, we present an approach to examine the energy evolution of software revisions based on their API interactions. The approach stems from the assumption that the utilization of an API has direct implications on the energy being consumed during runtime. Based on an empirical evaluation, we show initial results that API interactions serve as a flexible, lightweight, and effective way to compare software revisions regarding their energy evolution. Given our initial results we envision that in future using our approach mobile developers will be able to gain insights on the energy implications of changes in source code in the course of the software development life-cycle.

For the purpose of analyzing a building according to its accessibility or structural resilience, printed 2D floor plans are not sufficient because of the missing link to semantic information. This paper tackles this issue and introduces a concept for clustering classified lines of a floor plan and for creating semantically enriched contour elements based on different image processing, computer vision and machine learning algorithms. Based on a general line clustering approach, we introduce type specific methods for walls, windows, doors and stairs. The resulting clusters are in turn used for a contour creation, which uses minimal rotated rectangles. Those rectangles are transformed to polygons that are refined using post processing steps.The approach is evaluated via positive testing using a pixel-based comparison of the process’s result. For this, automatically generated as well as real world building plans are used. The final evaluation shows, that the concept reaches a confidence of >90% for door, stair and windows and only around 10% for stairs with the run-time linearly scaling with the size of the input.

The number of computer vision and image processing tasks has increased during the last years. Although Python is most of the time the first choice in this area, there are situations, where the utilization of another programming language such as Java should be preferred. For this reason, multiple Java based frameworks as e.g. OpenIMAJ, ND4J or multiple OpenCV wrappers are available. Unfortunately, these frameworks are not interoperable at all. In this work, the open-source Imaging Framework is introduced to solve exactly this problem. The project features a concept for combining multiple frameworks and provides an interoperable and extendable foundation to 9 image-related projects with 10 different image representations in Java.

A discussion of the relationship between the concept of healthcare information systems interoperability and process-oriented data analysis. The goal is to show that some of the current challenges of process mining in healthcare are also interoperability problems. By participating in solving these problems we can also improve our data sources.

Genetic Improvement (GI) can be used to give better quality software and to create new functionality. We show that GI can evolve the PowerPC open source GNU C runtime library square root function into cube root, binary logarithm log2 and reciprocal square root. The GI cbrt is competitive in run-time performance and our inverted square root x**-0.5 is far more accurate than the approximation used in the Quake video game. We use CMA-ES to adapt constants in a Newton-Raphson table, originally from glibc’s sqrt, for other double precision mathematics functions. Such automatically customised math libraries might be used for mobile or low resource, IoT, mote, smart dust, bespoke cyber-physical systems. Evolutionary Computing (EC) can be used to not only adapt source code but also data, such as numerical constants, and could enable a new way to conduct software data maintenance. This is an exciting opportunity for the GECCO and optimisation communities.

Further information can be found here

At a prevalence of almost 1%, potential epileptic seizures manifest a significant health risk for many juvenile patients. Thus, monitoring is essential to set early counteractive measurements to prevent from damage. The sensor-based monitoring systems mainly address epileptic seizures indicated by a change in the muscle tonus but cannot be utilized for patients that show Prévost’s-sign only. To monitor initiating Prévost’s-sign with opened-eyes as critical visual feature, the applicability of deep-learning eye detection systems on night vision images is evaluated in this paper as basis for modelling and classifying the eye state (closed, opened, not visible). A holistic research prototype is presented as proof of concept, showing the applicability of state-of-the-art face detection on night vision images as well as multi-variate feature analysis on Graph segmentation pre-fragmentation, applicable to detect the state of the eye in a robust way. Results show a single frame accuracy in face/eye detection of 73.91% and 94.44% for classification of the opened eyes as indication of a potentially initiating epileptic seizure. The monitoring system is based on a Raspberry computation unit with two ELP night vision cameras attached and a smart phone app for user-interaction and configuration besides on-demand visual monitoring. Future work will show that the single frame detection rate is sufficient for building up a rule-based monitoring state machine at user predefined sensitivity and specificity by analysing the visual content as time-series rather than single images.

Process mining can provide greater insight into medical treatment processes and organizational processes in healthcare. To enhance comparability between processes, the quality of the labelled-data is essential. A literature review of the clinical case studies by Rojas et al. in 2016 identified several common aspects for comparison, which include methodologies, algorithms or techniques, medical fields, and healthcare specialty. However, clinical aspects are not reported in a uniform way and do not follow a standard clinical coding scheme. Further, technical aspects such as details of the event log data are not always described. In this paper, we identified 38 clinically-relevant case studies of process mining in healthcare published from 2016 to 2018 that described the tools, algorithms and techniques utilized, and details on the event log data. We then correlated the clinical aspects of patient encounter environment, clinical specialty and medical diagnoses using the standard clinical coding schemes SNOMED CT and ICD-10. The potential outcomes of adopting a standard approach for describing event log data and classifying medical terminology using standard clinical coding schemes are further discussed. A checklist template for the reporting of case studies is provided in the Appendix A to the article.

We propose Knowledge-guided Genetic Improvement as a combination of Grammar-guided Genetic Programming with Tree-based Genetic Programming. Instead of utilizing a grammar directly, an operator graph based on that grammar is created, that is responsible for producing abstract syntax trees. Each operator contains knowledge about the grammar symbol it represents and returns only trees valid according to user-defined restrictions such as depth, complexity and approximated run-time performance. The expected benefits are a search space that excludes invalid individuals in an evolutionary run, ensuing a reduced overhead to evaluate invalid solutions and improving overall quality of the explored search space. The operator graph supports improvements based on previously run experiments and extensions towards further non-functional features.

With the continuous progress of Deep Learning (DL) powerful tools are now available for sophisticated segmentation tasks. Nevertheless, the generally very high demand for training data and precise reference segmentations in the medical domain often cannot be met when dealing with small and individual studies or acquisition protocols. As common strategies, reinforcement learning or transfer learning are applicable, but coherent with immense effort due to domain-specific adaptation. In this work, we evaluate the applicability of a U-grid cascade for training on a very small set of abdominal MRI datasets of the parenchyma and discuss strategies to compensate for the lack of training data. Although model accuracy is rather low when training on 13 MRI bands with achievable JI=89.41, the results are still good enough for annual post-processing using a graph-cut (GC) approach with moderate user interaction requirements. In this way, DL models are retrained as additional test data sets become available to subsequently improve classification accuracy. With only 2 additional GC post-processing datasets, the accuracy after model retraining is JI= 89.87. Furthermore, the applicability of Generative Adversarial Networks (GAN) in the medical field is evaluated, discussing to synthesize axial CT slices together with perfect ground truth reference segmentations. It is shown for abdominal CT slices of the parenchyma that in the absence of training data, synthesized slices that can be derived in arbitrary numbers can significantly improve the DL training process when only an insufficient amount of data is available. While training on 2,200 real-world images only leads to an accuracy of JI=88.75, enrichment with 2,200 additional images synthesized from a GAN trained on 5,000 datasets leads to an increase up to JI=92.02. Even when the DL model is trained exclusively on 4,400 computer-generated images, the classification accuracy on real-world data is remarkable with JI=90.81.

An automated and generally applicable method for segmentation is still in the focus of medical image processing research. For several years, artificial intelligence methods have shown promising results, especially with widely available scalable deep learning libraries. In this work, a five-layer hybrid U-network is developed for slice-wise segmentation of liver datasets. The training data is obtained from the Medical Segmentation Decathlon database, which contains 131 fully segmented volumes. A slice-based segmentation model is implemented using Deep Learning algorithms with adjustments for variable parenchyma shape along the stacking direction and similarities between adjacent slices. Both are transformed for coronal and sagittal views. The implementation is done on a GPU rack using TensorFlow and Keras. Standardized volume and surface metrics are used for a quantitative measure of segmentation accuracy. The results DSC=97.59, JI=95.29 and NSD=99.37 show correct segmentation comparable to 3D U-meshes and other state of the art U-meshes. The development of a 2D slice oriented segmentation we justified by the advantages of short training times and lower complexity and also massively reduces memory consumption. This work manifests the high potential of AI methods for general application in medicine. Segmentation as a fully or semi-automatic tool under the supervision of the expert user.

Modern compilers and interpreters provide code optimizations during compile and run time, simplifying the development process for the developer and resulting in optimized software. These optimizations are often based on formal proof, or alternatively stochastic optimizations have recovery paths as backup. The Genetic Compiler Optimization Environment (GCE) uses a novel approach, which utilizes genetic improvement to optimize the run-time performance of code with stochastic machine learning techniques. In this paper, we propose an architecture to integrate GCE, which directly integrates with low-level interpreters and compilers, with HeuristicLab, a high-level optimization framework that features a wide range of heuristic and evolutionary algorithms, and a graphical user interface to control and monitor the machine learning process. The defined architecture supports parallel and distributed execution to compensate long run times of the machine learning process caused by abstract syntax tree (AST) transformations. The architecture does not depend on specific operating systems, programming languages, compilers or interpreters.

Accurate human orientation estimation with respect to the POSE of a monocular camera system is a challenging task due to general aspects of camera calibration and the deformability of a moving human body. Therefore, novel deep learning approaches for precise object position determination in robotics are difficult to adapt for human body analysis. In this work, we present a hybrid approach for accurately estimating a human body relative to a camera system, significantly improving the results derived from poseNet by applying optical flow analysis in a frame-to-frame comparison. The human body, which rotates in the T-position in situ, is thereby center-aligned, with object tracking methods applied to compensate for translations of the body motion. After 2D skeletal extraction, optical flow is calculated for an ROI region aligned relative to the vertical skeletal junction representing the spine and compared frame by frame. To evaluate the suitability of clothing as a basis for good features, local pixel homogeneity is considered to constrain optical flow to heterogeneous regions with distinguishing features such as imprint patterns, buttons, or buckles in addition to local illumination change. Based on the mean optical flow with rough approximation of the axial body shape as an ellipse, accuracy between 0.1° and 2.0° is achieved for orientation estimation on a frame-to-frame comparison evaluated and validated on both CGI renderings and real videos of people wearing clothes with different features.

While most software automation research concentrates on programs’ code, we have started investigating if Genetic Improvement (GI) of data can assist developers by automating aspects of the maintenance of parameters embedded in source code. We extend recent GI work on optimising compile time constants to give new functionality and describe the transformation of a GNU C library square root function into the double precision reciprocal function, drcp. Multiplying by 1/x (drcp) allows division free division without requiring the hardware to support division. The evolution (6 seconds) and indeed the GI dp division (7.14 ± 0.012 nS) are both surprisingly fast.

This research covers generic approaches to determine the outdoor position and orientation of an augmented reality device due to the lack of outdoor suitability of depth or ambient sensing based devices currently available in the market. Orientation is primarily determined using an Attitude Heading Reference System (AHRS) for rough estimation. Based on a connected/integrated video camera, accuracy is improved for minor changes in orientation by using registration to evaluate orientation differences between two video frames, compensating for gyroscope drift errors. Position determination is performed using GPS with a real-time kinematic beacon system with rover and base station to achieve improved accuracy. The results show that based on the sensor application, AR hardware considered for indoor use can be retrofitted to work properly outdoors, at long distances, and even in moving vehicles. This will facilitate the future implementation of applications in various fields.

Background: Third-party software libraries often serve as fundamental building blocks for developing applications. However, depending on such libraries for development raises a new concern, energy consumption, which has become of increased interest for the software engineering community in recent years. Understanding the energy implications of software design choices is an ongoing research challenge, especially when working with mobile devices that are constrained by limited battery life. Aims: Our goal is to research approaches, which will support software developers to better comprehend the energy implications of software design choices on Android. For this study, we particularly focus on APIs from third-party libraries for which we research methods that will enable estimating the energy consumption without the need for specific measurement hardware and laborious measurements. Method: To achieve the stipulated goal we introduce System API Utilization Profiles (uAPI) which are based on the general assumption that the actual energy consumption of a library is directly tied to its utilization of the underlying System API. We provide a formal definition and implementation of the proposed uAPI profiles that are calculated based on dynamic call graphs obtained from a library under test. To further show the connection between our proposed uAPI profiles and energy consumption, we empirically examined their correlation using two experiments. The first one is dedicated to Android I/O operations, the second one examines uAPI profiles based on a popular open-source library for JSON document processing. Results: In our empirical evaluation, we collected 1052 individual call graphs which we used as an input to evaluate the our model and compare it with the actual energy consumption measured. For each call graph, we measured the energy consumption with special hardware and attributed the measurements to individual methods. Based on that data, we examined a strong linear correlation between the proposed uAPI profiles and the actual measured energy consumption. Conclusions: uApi profiles serve as a lightweight and feasible approach to characterize the energy characteristics of third-party libraries. Their computation does not involve any special hardware, and there are no alterations on the target mobile device required. For future work, we investigate possibilities to use the proposed uAPI profiles as a foundation for a regression model, which would allow us to predict the energy consumption development of a library over time.

In this paper we present a new approach applying computer vision methods to image data acquired with depth perception cameras to map the interior of the elevator, detect the position and the state of the door and to detect objects in the door area. The depth data is used to determine the elevator cabin as safety cube, i.e. the position of the door, layout of the elevator and so on, while color data further enhances the detection of new objects. The approach can detect the state of the elevator door as either opened or closed, while no object is blocking the view to the door, as well as successfully identify objects blocking an open door. This elevator monitoring proves to be relevant for determination of the elevator state, safety as well aspects of predictive maintenance.

Process mining can provide greater insight into medical treatment processes and organizational processes in healthcare. A review of case studies in the literature identified several different common aspects for comparison, which include methods, algorithms or techniques, medical domains, and healthcare specialties. However, from a medical perspective, clinical terms are not used in a consistent manner and do not follow a standardized clinical coding scheme. In addition, the characteristics of event log data are not always described. In this paper, we identified 38 clinically relevant case studies on process mining in healthcare published between 2016 and 2018 that described the tools, algorithms, and techniques used, as well as details about event log data. We then mapped the clinical aspects of the patient encounter environment, clinical specialty, and medical diagnoses using the standard SNOMED CT and ICD-10 clinical coding schemes. The possible results of adopting a standard approach for describing event log data and classifying medical terminology using standard clinical coding schemes are discussed.

We report the discussion session at the sixth international Genetic Improvement workshop, GI-2019 @ ICSE, which was held as part of the 41st ACM/IEEE International Conference on Software Engineering on Tuesday 28th May 2019. Topics included GI representations, the maintainability of evolved code, automated software testing, future areas of GI research, such as co-evolution, and existing GI tools and benchmarks.

With the advent of mobile application development a new software quality concern - energy consumption - was introduced. For mobile software developers knowledge about software and algorithm design choices and their implications on energy consumption is crucial. However, software developers either lack this knowledge or tools to support them in estimating the energy consumption of their applications and therefore are unable to reflect on their design choices. In this empirical study we examine the energy consumption of 12 sorting algorithms and the resulting energy impact when used with different data types. We propose a methodology to obtain energy readings and relate them to application execution traces. Our results show that the choice of data type together with algorithm design can have significant impact on the energy profile of an application.

BACKGROUND: Process mining is a relatively new discipline that helps to discover and analyze actual process executions based on log data. In this paper we apply conformance checking techniques to the process of surveillance of melanoma patients. This process consists of recurring events with time constraints between the events. OBJECTIVES: The goal of this work is to show how existing clinical data collected during melanoma surveillance can be prepared and pre-processed to be reused for process mining. METHODS: We describe an approach based on time boxing to create process models from medical guidelines and the corresponding event logs from clinical data of patient visits. RESULTS: Event logs were extracted for 1023 patients starting melanoma surveillance at the Department of Dermatology at the Medical University of Vienna between January 2010 and June 2017. Conformance checking techniques available in the ProM framework and explorative applied process mining techniques were applied. CONCLUSIONS: The presented time boxing enables the direct use of existing process mining frameworks like ProM to perform process-oriented analysis also with respect to time constraints between events.

When attempting to improve the non-functional requirements of software, specifically run-time performance of code, an important requirement is to preserve the correctness of the optimized code. Additionally when attempting to integrate Genetic Improvement into a compiler or interpreter, the large search spaces resulting from the amount of operators and operands a language provides needs to be dealt with. This publication explores dynamic fitness functions as a foundation for a use in Genetic Improvement to optimize programs. An approach of using a test suite to verify code correctness in the Truffle Framework [19, 20] and Graal Compiler [11] is presented. Two types of fitness functions are explored, which split the test suite according to their complexity and attempt to generate correct solutions with a growing set of increasingly complex tests. One of them increases the amount of tests sequentially over several iterations. The parallel fitness function attempts to split a test suite and to re-combine the results with increasingly large suites. The results show that these functions only marginally improve the fitness landscape on their own, but show that more partially correct solutions can be found with dynamic fitness functions. In the future, our approach may be improved by implementing specific crossover and mutator operations to accompany the dynamic fitness functions.

Background: For ethical, legal and administrative reasons, patients have to give explicit consent to a medical treatment. Objectives: This paper identifies the design requirements for electronic treatment consent (eConsent) architecture, and subsequently proposes a model for the eConsent architecture based on the HL7 FHIR® standard. Methods: Six requirements for the eConsent architecture were identified. A conceptual model for the system was then developed to address the identified requirements using HL7 FHIR. Results: The proposed concept makes use of the existing consent model of HL7 FHIR, and includes additional resources for presenting the information to the patient. Moreover, it uses the SNOMED CT terminology to enable semantic interoperability with other health information systems. Conclusions: The proposed eConsent architecture meets the identified requirements. That said, the system is limited by the low maturity of the implemented FHIR resources and the fact that the terminology is currently inexhaustive for the use case. Custom extensions of the used FHIR resources must be considered.

Modern compilers provide code optimizations before and during run-time, thus moving required domain knowledge about the compilation process away from the developer and speeding up resulting software. These optimizations are often based on formal proof, or alternatively have recovery paths as backup. Genetic improvement (GI), a field of science utilizing genetic programming, a stochastic optimization technique, has been previously utilized to fix bugs in software and improve non-functional software requirements.

BACKGROUND: Multidisciplinary team meetings (MDTMs) are already in use for certain areas in healthcare (e.g. treatment of cancer). Due to the lack of common standards and accessibility for the applied IT systems, their potential is not yet completely exploited. OBJECTIVES: Common requirements for MDTMs shall be identified and aggregated into a process definition to be automated by an application architecture utilizing modern standards in electronic healthcare, e.g. HL7 FHIR. METHODS: To identify requirements, an extensive literature review as well as semi-structured expert interviews were conducted. RESULTS: Results showed, that interoperability and flexibility in terms of the process are key requirements to be addressed. An architecture blueprint as well as an aggregated process definition were derived from the insights gained. To evaluate the feasibility of identified requirements, methods of explorative prototyping in software engineering were used. CONCLUSION: MDTMs will become an important part of modern and future healthcare but the need for standardization in terms of interoperability is imminent.

The demand for high quality mobile applications is constantly rising, especially in mission critical settings. Thus, new software engineering methodologies are needed in order to ensure the desired quality of an application. The research presented proposes a quality assurance methodology for mobile applications through test automation by optimizing test suites. The desired goal is to find a minimal test suite while maintaining efficiency and reducing execution cost. Furthermore to avoid invalidating an optimized test suite as the system under test evolves, the approach further proposes to extract patterns from the applied changes to an application. The evaluation plan comprises a combination of an empirical and an industrial case study based on open source projects and an industrial project in the healthcare domain. It is expected that the presented approach supports the testing process on mobile application platforms.

An alteration of the job shop scheduling problem, concerning advertisement scheduling on digital advertisement spaces, is presented. Dispatching Rules (DR), Iterated Local Search (ILS) and Genetic Algorithms (GA) are discussed and applied to the problem space. The results show that ILS is the best performing heuristic, and surpasses the other heuristics especially in large problem spaces (≥ 100 machines, ≥ 100 jobs). The results match previously made findings, which indicates that effects on large-scale problems should be further researched in conjunction with amalgam algorithms between DR, GA and ILS.

An agent-based elevator simulation was implemented to test the validity of an advertisement scheduling system. The elevator simulation imitates the elevators and their advertisement system, thus testing the different schedules and sending the results back to the scheduler. To validate the results, certain data sets of the simulation application are compared with expected values. First results have shown that the created data is in a valid range around the estimations. Finally, methods to further improve the simulation are presented. It was found that agent-based simulations are a good method to test systems, which are too complex or expensive to test in the real environment.

Prior studies as well as medical imaging data are crucial for a radiologist to diagnose a patient. In this paper the radiological workflow is analyzed from a patient’s perspective in order to gain knowledge on how possible existing prefetching strategies still can be applied in connection with a standardized distributed health information system conforming to architectures defined by IHE and ELGA. As a result an adaption to such architectures is proposed and further evaluated in a testing environment. Although the approach presented works in terms of prefetching relevant prior studies together with medical imaging data, additional research has to be carried out on how to apply intelligent search strategies in order to narrow retrieved results concerning their possible utilization for a specific diagnosis.

Recently Business Intelligence approaches like process mining are applied to the healthcare domain. The goal of process mining is to gain process knowledge, compliance and room for improvement by investigating recorded event data. Previous approaches focused on process discovery by event data from various specific systems. IHE, as a globally recognized basis for healthcare information systems, defines in its ATNA profile how real-world events must be recorded in centralized event logs. The following approach presents how audit trails collected by the means of ATNA can be transformed to enable process mining. Using the standardized audit trails provides the ability to apply these methods to all IHE based information systems.

Business Intelligence approaches such as process mining can be applied to the healthcare domain in order to gain insight into the complex processes taking place. Disclosing as-is processes helps identify room for improvement and answers questions from medical professionals. Existing approaches are based on proprietary log data as input for mining algorithms. Integrating the Healthcare Enterprise (IHE) defines in its Audit Trail and Node Authentication (ATNA) profile how real-world events must be recorded. Since IHE is used by many healthcare providers throughout the world, an extensive amount of log data is produced. In our research we investigate if audit trails, generated from an IHE test system, carry enough content to successfully apply process mining techniques. Furthermore we assess the quality of the recorded events in accordance with the maturity level scoring system. A simplified simulation of the organizational workflow in a radiological practice is presented. Based on this simulation a process miing task is conducted.

The continuous monitoring of vital signs has become an important supplement to traditional medical treatment to ensure the success of a therapy. Integrated health monitoring solutions based on existing health standards provide interoperability and enable healthcare providers and patients to exchange and access their data across institutional borders. This paper shows an integrated monitoring solution based on Continua and Integrating the Healthcare Enterprise, which has been tested by more than 130 patients and 14 healthcare institutions. According to user feedback, one recurring problem is the low battery life of smartphones due to high data traffic. Since the recently developed HL7 standard FHIR offers a resource efficient handling of web service connections, a possible approach to extend the monitoring solution to support FHIR.

The publication of the HL7-FHIR standard offers new possibilities for integrated applications in healthcare. Although trial implementations have only recently started, the application of FHIR in context of a Personal Health Monitoring solution is worth investigating. Most of the existing telemonitoring solutions in healthcare rely on guidelines defined by the Continua Health Alliance (CHA). This paper compares the requirements of CHA and HL7-FHIR with respect to data traffic between client devices and server side applications. Therefore an existing CHA-compliant telemonitoring solution is extended towards supporting HL7-FHIR. Both approaches were simultaneously evaluated in a live system with 68 participants. The results of the evaluation show that the FHIR approach offers the possibility of reducing data traffic in comparison to the CHA solution.

The popularity of mobile application platforms like IOS, Android or Windows Phone 7 increases from day to day. Affiliated app stores offer thousands of different applications (so called apps) for nearly any imaginable purpose. From a software developer’s point of view these app stores provide high potential: developers have the chance to reach the general public, thus having a much broader audience than with any other channel of distribution. Nevertheless, the increased amount of potential customers also increases expectations related to user experience. The user interface as the main interaction hub in mobile environments attracts a lot more attention. The success or failure of an app is often related to its user interface. Furthermore, users often demand the same user experience on different platforms. In order to allow for high quality in user interface development and to enable development for different platforms, a framework is presented that provides a platform independent definition of a user interface backed by constraints in form of rules. Using model driven development, the framework allows the creation of user interfaces that are not related to a specific target technology. Furthermore, it provides the ability to define a user interface and transform it into a destination user interface, whether this will be Android, iPhone or Windows Phone 7, with respect to given rules.