TY - JOUR A1 - Rothe, Felix A1 - Berger, Jörn A1 - Welker, Pia A1 - Fiebelkorn, Richard A1 - Kupper, Stefan A1 - Kiesel, Denise A1 - Gedat, Egbert A1 - Ohrndorf, Sarah T1 - Fluorescence optical imaging feature selection with machine learning for differential diagnosis of selected rheumatic diseases JF - Frontiers in Medicine N2 - Background and objective: Accurate and fast diagnosis of rheumatic diseases affecting the hands is essential for further treatment decisions. Fluorescence optical imaging (FOI) visualizes inflammation-induced impaired microcirculation by increasing signal intensity, resulting in different image features. This analysis aimed to find specific image features in FOI that might be important for accurately diagnosing different rheumatic diseases. Patients and methods: FOI images of the hands of patients with different types of rheumatic diseases, such as rheumatoid arthritis (RA), osteoarthritis (OA), and connective tissue diseases (CTD), were assessed in a reading of 20 different image features in three phases of the contrast agent dynamics, yielding 60 different features for each patient. The readings were analyzed for mutual differential diagnosis of the three diseases (One-vs-One) and each disease in all data (One-vs-Rest). In the first step, statistical tools and machine-learning-based methods were applied to reveal the importance rankings of the features, that is, to find features that contribute most to the model-based classification. In the second step machine learning with a stepwise increasing number of features was applied, sequentially adding at each step the most crucial remaining feature to extract a minimized subset that yields the highest diagnostic accuracy. Results: In total, n = 605 FOI of both hands were analyzed (n = 235 with RA, n = 229 with OA, and n = 141 with CTD). All classification problems showed maximum accuracy with a reduced set of image features. For RA-vs.-OA, five features were needed for high accuracy. For RA-vs.-CTD ten, OA-vs.-CTD sixteen, RA-vs.-Rest five, OA-vs.-Rest eleven, and CTD-vs-Rest fifteen, features were needed, respectively. For all problems, the final importance ranking of the features with respect to the contrast agent dynamics was determined. Conclusions: With the presented investigations, the set of features in FOI examinations relevant to the differential diagnosis of the selected rheumatic diseases could be remarkably reduced, providing helpful information for the physician. KW - rheumatoid arthritis KW - osteoarthritis KW - connective tissue disease KW - gradient boosting machine KW - forward selection algorithm Y1 - 2023 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:kobv:526-opus4-17922 SN - 2296-858X VL - 10 PB - Frontiers ER - TY - CHAP A1 - Pulwer, Silvio A1 - Fiebelkorn, Richard A1 - Zesch, Christoph A1 - Steglich, Patrick A1 - Villringer, Claus A1 - Villasmunta, Francesco A1 - Gedat, Egbert A1 - Handrich, Jan A1 - Schrader, Sigurd A1 - Vandenhouten, Ralf T1 - Endoscopic orientation by multimodal data fusion T2 - Proc. SPIE 10931, MOEMS and Miniaturized Systems XVIII N2 - To improve the feasibility of endoscopic inspection processes we developed a system that provides online information about position, orientation and viewing direction of endoscopes, to support the analysis of endoscopic images and to ease the operational handling of the equipment. The setup is based on an industrial endoscope consisting of a camera, various MEMS and multimodal data fusion. The software contains algorithms for feature and geometric structure recognition as well as Kalman filters. To track the distal end of the endoscope and to generate 3D point cloud data in real time the optical and photometrical characteristics of the system are registered and the movement of the endoscope is reconstructed by using image processing techniques. KW - endoscope KW - orientation KW - Kalman filter KW - visual odometry KW - mems KW - gyroscope KW - accelerometer Y1 - 2019 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:kobv:526-opus4-11420 SN - 1996-756X ER - TY - GEN A1 - Gedat, Egbert A1 - Fechner, Pascal A1 - Fiebelkorn, Richard A1 - Vandenhouten, Jan A1 - Vandenhouten, Ralf T1 - Image recognition of multi-perspective data for intelligent analysis of gestures and actions T2 - Wissenschaftliche Beiträge 2018 N2 - The BERMUDA project started in January 2015 and was successfully completed after less than three years in August 2017. A technical set-up and an image processing and analysis software were developed to record and evaluate multi-perspective videos. Based on two cameras, positioned relatively far from one another with tilted axes, synchronized videos were recorded in the laboratory and in real life. The evaluation comprised the background elimination, the body part classification, the clustering, the assignment to persons and eventually the reconstruction of the skeletons. Based on the skeletons, machine learning techniques were developed to recognize the poses of the persons and next for the actions performed. It was, for example, possible to detect the action of a punch, which is relevant in security issues, with a precision of 51.3 % and a recall of 60.6 %. N2 - Das Projekt BERMUDA konnte im Januar 2015 begonnen und nach knapp drei Jahren im August 2017 erfolgreich abgeschlossen werden. Es wurden ein technischer Aufbau und eine Bildverarbeitungs- und Analysesoftware entwickelt, mit denen sich multiperspektivische Videos aufzeichnen und auswerten lassen. Basierend auf zwei in größerem Abstand gewinkelt positionierten Kameras wurden synchrone Videos sowohl im Labor als auch in realen Umgebungen aufgenommen. Die Auswertung umfasst die Hintergrundeliminierung, die Körperteilklassifikation, ein Clustering, die Zuordnung zu Personen und schließlich die Rekonstruktion der Skelette. Ausgehend von den Skeletten wurden Methoden des maschinellen Lernens zur Erkennung der Haltungen und darauf aufbauend zur Gestenerkennung entwickelt. Beispielhaft konnte die im Sicherheitskontext relevante Handlung des Schlagens mit einer Genauigkeit von 51,3 % und einer Trefferquote von 60,6 % erkannt werden. Y1 - 2018 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:kobv:526-opus4-10230 SN - 0949-8214 VL - 22 SP - 25 EP - 30 ER - TY - GEN A1 - Gedat, Egbert A1 - Fechner, Pascal A1 - Fiebelkorn, Richard A1 - Vandenhouten, Ralf T1 - Szenenanalyse und Unterscheidung der Skelette mehrerer Menschen in digitalen Bildern mit Graphentheorie durch eine k-kürzeste-disjunkte-Wege-Suche T2 - Wissenschaftliche Beiträge 2017 N2 - Das Erkennen von Personen auf Fotos und in Videos geschieht in aktuellen Klassifizierungsverfahren durch die Zuordnung eines Köperteils zu jedem Pixel. Anschließend werden die gefundenen Körperteile zu Skeletten zusammengefasst. Im Falle mehrerer abgebildeter Personen ergibt sich das Problem der Zuordnung der Körperteile zu den verschiedenen Skeletten. Es wurde in dieser Arbeit ein auf dem Suurballe-Algorithmus basierendes graphentheoretisches Verfahren entwickelt, das diese Aufgabe löst. Aufbauend auf im Wesentlichen abstandsabhängigen Kantengewichten wird eine k-kürzeste-disjunkte-Wege-Suche durchgeführt. Unter Einbezug von möglicherweise fehlenden Körperteilen durch Unsichtbar-Knoten und automatisches Aussortieren zu viel gefundener Körperteile werden so die k-optimalen Skelette gefunden. Die Methode wurde an idealisierten computergenerierten Bildern mit einer Trefferquote von 100 % gefundener Personen getestet. Ein Test mit realen Bilddaten lieferte eine Trefferquote von 71,7 %. N2 - Recognition of humans on photos and in videos is done per current classification procedures by assigning a body part to each pixel. The body parts thus allocated are subsequently ordered into skeletons. In the case the images indicate that multiple humans are present, the problem arises of assigning the body parts to different skeletons. A Suurballe algorithm-based method developed by us solves this task. By mainly relying on distance-related edge weights, a k shortest disjoint paths search is performed. Accounting for possibly missing body parts by invisible-nodes and automatically sorting out redundant body parts, the optimal k skeletons are found. The method was tested with computer-generated data, yielding a rate of 100 % persons detected. A test with real-world images yielded a recall ratio of 71.7 %. Y1 - 2017 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:kobv:526-opus4-9394 SN - 0949-8214 VL - 21 SP - 31 EP - 35 ER -