Fraunhofer-Institut für Angewandte Informationstechnik FIT

Filters

3 2
Reset filters

Settings
Now showing 1 - 5 of 5
  • Publication
    DeepKneeExplainer: Explainable knee osteoarthritis diagnosis from radiographs and magnetic resonance imaging
    ( 2021)
    Karim, Rezaul
    ;
    Jiao, Jiao
    ;
    Döhmen, Till
    ;
    Cochez, Michael
    ;
    Beyan, Oya
    ;
    Rebholz-Schuhmann, Dietrich
    ;
    Decker, Stefan
    Osteoarthritis (OA) is a degenerative joint disease, which significantly affects middle-aged and elderly people. Although primarily identified via hyaline cartilage change based on medical images, technical bottlenecks like noise, artifacts, and modality impose an enormous challenge on high-precision, objective, and efficient early quantification of OA. Owing to recent advancements, approaches based on neural networks (DNNs) have shown outstanding success in this application domain. However, due to nested non-linear and complex structures, DNNs are mostly opaque and perceived as black-box methods, which raises numerous legal and ethical concerns. Moreover, these approaches do not have the ability to provide the reasoning behind diagnosis decisions in the way humans would do, which poses an additional risk in the clinical setting. In this paper, we propose a novel explainable method for knee OA diagnosis based on radiographs and magnetic resonance imaging (MRI), which we called DeepKneeExplainer. First, we comprehensively preprocess MRIs and radiographs through the deep-stacked transformation technique against possible noises and artifacts that could contain unseen images for domain generalization. Then, we extract the region of interests (ROIs) by employing U-Net architecture with ResNet backbone. To classify the cohorts, we train DenseNet and VGG architectures on the extracted ROIs. Finally, we highlight class-discriminating regions using gradient-guided class activation maps (Grad-CAM++) and layer-wise relevance propagation (LRP), followed by providing human-interpretable explanations of the predictions. Comprehensive experiments based on the multicenter osteoarthritis study (MOST) cohorts, our approach yields up to 91% classification accuracy, outperforming comparable state-of-the-art approaches. We hope that our results will encourage medical researchers and developers to adopt explainable methods and DNN-based analytic pipelines towards an increasing acceptance and adoption of AI-assisted applications in the clinical practice for improved knee OA diagnoses.
  • Publication
    Towards Reusability in the Semantic Web. Decoupling Naming, Validation, and Reasoning
    ( 2020)
    Lipp, Johannes
    ;
    Gleim, Lars
    ;
    Decker, Stefan
    RDFS and OWL ontologies simultaneously define naming, hierarchy, syntactical data structure, and axioms. This strong coupling complicates the reusability of both ontological concepts and annotated data, due to logical pitfalls in RDFS and OWL semantics. The differences between OWL axioms and integrity constraints used for validation are often not clear to users and lead to confusing and unintended semantics in practice. To avoid these pitfalls, we revisit Tom Gruber's basic ontology definition and reimagine a more decoupled ontology design pattern, consisting of independent layers for naming, validation, and reasoning. We argue that such decoupling improves reusability because it clarifies the usage of the three layers during ontology creation and reuse. A naming layer built on synonym sets enables reusing named concepts in different contexts, detached from constraints or OWL axioms defined elsewhere. On top of that, we suggest a two-step approach of constraint checking and reasoning: Validate a term's integrity via constraints first, and only include it for reasoning if that validation succeeds. Our proposal is one step towards a clearer in-practice usage of naming, validation, and reasoning-and additionally supports this with a revised semantic layer model.
  • Publication
    Representing medication guidelines for use in production rule systems in the context of POLYCARE project
    ( 2018)
    Könning, Jonas W.
    ;
    Velasco, Carlos A.
    ;
    Mohamad, Yehya
    ;
    Decker, Stefan
    ;
    Beyan, Oya
    This paper presents an approach to represent medication guidelines in a machine readable form for its use within a production Home Care environment. Part of this work was developed under the scope of the POLYCARE project. The POLYCARE project aims at developing a patient-centred integrated care environment, supported by ICT systems to improve the quality of home hospitalization. Part of the project design is a decision support system for improving the medication of the elderly patients. For this, a machine actionable version of the medication guidelines to be used was needed. However, as none were freely available to the project, we used our own approach. Scope of this work is the design and implementation of such a rule base. Guidelines were selected based on their prominence in the domain and analyzed for their structure to allow for the creation of templates, which can be used in the generation of rules. The templates were designed to work with the Drools business rule management system. While still simple, the current rule base shows good performance and potential for future extensions. Freely available, open business rule management systems proved themselves sufficient for the task.
  • Publication
    Towards a FAIR sharing of scientific experiments: Improving discoverability and reusability of dielectric measurements of biological tissues
    ( 2017)
    Rezaul Karim, M.
    ;
    Heinrichs, Matthias
    ;
    Gleim, Lars C.
    ;
    Cochez, Michael
    ;
    Porter, Emily
    ;
    Gioia, Alessandra la
    ;
    Salahuddin, Saqib
    ;
    O'Halloran, Martin
    ;
    Decker, Stefan
    ;
    Beyan, Oya
    Experiments on the dielectric properties of biological tissues generate data that characterizes the interaction of human tissues with electromagnetic fields. This data is vital for designing electromagnetic-based therapeutic and diagnostic technologies, and for assessing the safety of wireless devices. Despite the importance of the data, poor reporting and lack of metadata impede its reuse and forgo interoperability. Recently, the minimum information model for reporting Dielectric Measurements of Biological Tissues (MINDER) has been developed as a common framework. In this work, we have developed a metadata model and implemented a data sharing framework to improve findability and reproducibility of experimental data inspired by FAIR principles. We define a process for sharing the reported data and present tools to support rich metadata generation based on existing community standards. The developed system is evaluated against competency questions collected from data consumers, and thereby proven to help to interpret and compare data across studies.
  • Publication
    Towards the use of graph summaries for privacy enhancing release and querying of linked data
    ( 2016)
    Heitmann, Benjamin
    ;
    Hermsen, Felix
    ;
    Decker, Stefan
    Linked Data has become an important standard to describe meta-data about open government data. At the same time, most government data is not released as Linked Data. One reason for this could be the difficulty of applying privacy enhancing technologies such as differential privacy and private information retrieval to Linked Data. We introduce the idea of graph summaries to function as a schema for Linked Data which is schema-less. This in turn can provide a conceptual bridge for applying differential privacy and private information retrieval to Linked Data.