1 - 10 of 22
-
PublicationDeepKneeExplainer: Explainable knee osteoarthritis diagnosis from radiographs and magnetic resonance imaging( 2021)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.
-
-
PublicationINGE - integrate4care. Eine erweiterte Unterstützung für Pflegekräfte, Pflegebedürftige und ihre Angehörigen( 2021)INGE steht für digitale Integrierte Gesundheits- und Pflegeversorgung mit IT-gestütztem Pflegeberatungsbesuch nach § 37 Abs. 3 SGB XI. Dahinter steckt die Idee ein Onlineportal und eine App zu entwickeln, Beobachtungen der häuslichen Pflegesituation und Erfassung von Problembereichen beinhalten und Pflegeberater*innen bei der Dokumentation der häuslichen Pflegesituation unterstützt. Gleichzeitig werden Unterstützungsangebote für den*die pflegende Angehörige aufgezeigt. Derzeit werden die Informationen jedoch meist nur partiell und in Papierform dokumentiert. Dies kann zu Informationsverlusten führen und die Weitergabe an andere Mitversorger*innen ist erschwert. An diesem Punkt setzt auch das Projekt INGE integrate4care an. Hier steht die Verbesserung der sektorübergreifenden Begleitung, also die Versorgung mit ambulanten und stationären Leistungen, von Pflegebedürftigen, die zu Hause von Angehörigen gepflegt werden, im Zentrum. INGE hat das Ziel, mittels einer App (INGE-App) die Dokumentation des gesetzlich verankerten Pflegeberatungsbesuches zu digitalisieren. Konkret bedeutet dies, dass die Pflegeberater*innen die INGE-App im Beratungsbesuch nutzen und alle Informationen digital erfassen.
-
PublicationFinding and analysing energy research funding data: The EnArgus system( 2021)
-
PublicationExtending the Automation Pyramid for Industrial Demand Response( 2019)Industrial demand response uses a multitude of energy flexibility measures. Their planning and control requires various production IT systems. A widely accepted approach to classify these inhouse IT systems are the levels of the automation pyramid in companies. This paper broadens the scope of this concept to overcome the limitation to companies' (virtual) borders by including required IT systems that refine and monetarize a company's energy flexibility, e.g. energy markets, aggregators, etc. Therefore, a holistic approach for the classification of functionalities for industrial demand response across companies and energy markets is developed and applied exemplarily.
-
PublicationHarnessing the Full Potential of Industrial Demand-Side Flexibility: An End-to-End Approach Connecting Machines with Markets through Service-Oriented IT Platforms( 2019)The growing share of renewable energy generation based on fluctuating wind and solar energy sources is increasingly challenging in terms of power grid stability. Industrial demand-side response presents a promising way to balance energy supply and consumption. For this, energy demand is flexibly adapted based on external incentives. Thus, companies can economically benefit and at the same time contribute to reducing greenhouse gas emissions. However, there are currently some major obstacles that impede industrial companies from taking part in the energy markets. A broad specification analysis systematically dismantles the existing barriers. On this foundation, a new end-to-end ecosystem of an energy synchronization platform is introduced. It consists of a business-individual company-side platform, where suitable services for energy-oriented manufacturing are offered. In addition, one market-side platform is established as a mediating service broker, which connects the companies to, e.g., third party service providers, energy suppliers, aggregators, and energy markets. The ecosystems aim at preventing vendor lock-in and providing a flexible solution, relying on open standards and offering an integrated solution through an end-to-end energy flexibility data model. In this article, the resulting functionalities are discussed and the remaining deficits outlined.
-
PublicationFlexible IT Platform for Synchronizing Energy Demands with Volatile Markets( 2018)Abandoning fossil and nuclear energy sources in the long run and increasing amount of renewable energies in electricity production causes a more volatile power supply. Depending on external realities, renewable energy production emphasizes the need for measures to guarantee the necessary balance of demand and supply in the electricity system at all times. Energy intensive industry processes theoretically include high Demand Response potentials suitable to tackle this increasing supply volatility. Nevertheless, most companies do not operate their production in a flexible manner due to multiple reasons: among others, the companies lack know-how, technologies and a clear business case to introduce an additional level of flexibility into their production processes, they are concerned about po ssible impacts on their processes by varying the electricity demand and need assistance in exploiting their flexibility. Aside from fostering knowledge in industry companies, an IT-solution that supports companies to use their processes Demand Response potential has become necessary. Its concept must support companies in managing companies energy-flexible production processes and monetarize those potentials at flexibility markets. This paper presents a concept, which integrates both companies and energy markets. It enables automated trading of companies Demand Response potential on different flexibility markets.
-
PublicationFlexible IT-platform to Synchronize Energy Demands with Volatile Markets( 2017)Based on the goal of exiting nuclear and fossil energies within the electricity generation, the percentage of renewable energies in the energy mix rises. Due to renewable energies' dependence on natural resources like sun or wind this development leads to a volatile energy supply on the markets. To satisfy their customers' needs even with a volatile energy supply, especially companies of the manufacturing sector need to consider this development. Production processes need to be developed further to be more energy efficient and to be adaptable in their energy demand to volatile supply. This includes being operable on various power levels or with different kinds of energy such as electricity or gas. Energy-flexible production processes need to be supported by flexible IT solutions. While there are already solutions for demand-side-management on the company side as well as on the market side, there are no holistic solutions yet, allowing for integration regardless of company or market boundaries. Therefore, this paper presents the concept of a service-oriented architecture for a flexible IT-platform to synchronize energy demands with volatile markets. A holistic approach allows for integration of companies as well as energy markets and enables an automated and efficient exploitation of demand response potentials.
-
PublicationEarly results of experiments with responsive open learning environments( 2011)Responsive open learning environments (ROLEs) are the next generation of personal learning environments (PLEs). While PLEs rely on the simple aggregation of existing content and services mainly using Web 2.0 technologies, ROLEs are transforming lifelong learning by introducing a new infrastructure on a global scale while dealing with existing learning management systems, institutions, and technologies. The requirements engineering process in highly populated test-beds is as important as the technology development. In this paper, we will describe first experiences deploying ROLEs at two higher learning institutions in very different cultural settings. The Shanghai Jiao Tong University in China and at the "Center for Learning and Knowledge Management and Department of Information Management in Mechanical Engineering" (ZLW/IMA) at RWTH Aachen University have exposed ROLEs to theirs students in already established courses. The results demonstrated to readiness of the technology for large-scale trials and the benefits for the students leading to new insights in the design of ROLEs also for more informal learning situations.
-
PublicationUsage-based object similarity( 2010)Recommender systems are widely used online to support users in finding relevant information. They can be based on different techniques such as content-based and collaborative filtering. In this paper, we introduce a new way of similarity calculation for item-based collaborative filtering. Thereby we focus on the usage of an object and not on the object's users as we claim the hypothesis that similarity of usage indicates content similarity. To prove this hypothesis we use learning objects accessible through the MACE portal where students can query several architectural repositories. For these objects, we generate object profiles based on their usage monitored within MACE. We further propose several recommendation techniques to apply this usagebased similarity calculation in real systems.
