Publications

Evaluations of new dry, high-density EEG caps have only been performed so far with serial measurements and not with simultaneous (parallel) measurements. For a first comparison of gel-based and dry electrode performance in simultaneous high-density EEG measurements, we developed a new EEG cap comprising 64 gel-based and 64 dry electrodes and performed simultaneous measurements on ten volunteers. We analyzed electrode-skin impedances, resting state EEG, triggered eye blinks, and visual evoked potentials (VEPs). To overcome the issue of different electrode positions in the comparison of simultaneous measurements, we performed spatial frequency analysis of the simultaneously measured EEGs using spatial harmonic analysis (SPHARA). The impedances were 516 ± 429 kOhm (mean ± std) for the dry electrodes and 14 ± 8 kOhm for the gel-based electrodes. For the dry EEG electrodes, we obtained a channel reliability of 77%. We observed no differences between dry and gel-based recordings for the alpha peak frequency and the alpha power amplitude, as well as for the VEP peak amplitudes and latencies. For the VEP, the RMSD and the correlation coefficient between the gel-based and dry recordings were 1.7 ± 0.7 μV and 0.97 ± 0.03, respectively. We observed no differences in the cumulative power distributions of the spatial frequency components for the N75 and P100 VEP peaks. The differences for the N145 VEP peak were attributed to the different noise characteristics of gel-based and dry recordings. In conclusion, we provide evidence for the equivalence of simultaneous dry and gel-based high-density EEG measurements.

Autonome Beckennerven sind bei viszeralchirurgischen Eingriffen im Bereich des kleinen Beckens wie beispielsweise Rektumresektionen bei Rektumkarzinom einem sehr hohen Risiko der intraoperativen iatrogenen Schädigung ausgesetzt. Die Folgen von Schädigungen der Beckennerven sind postoperativ auftretende Funktionsstörungen der Beckenorgane wie Harninkontinenz, Stuhlinkontinenz und sexuelle Funktionsstörungen, die zu einer Beeinträchtigung der Lebensqualität der Patienten führen können. Das autonome Nervengeflecht des Beckens ist hochkomplex, sehr filigran, innerviert glatte Muskulatur und kann in seiner topografischen Lage interindividuell verschieden sein. Da die Nerven rein visuell nur sehr schwer identifiziert werden können, ist deren Schonung im Verlauf der chirurgischen Präparation ohne technische Hilfsmittel und ohne fundierte Kenntnisse der pelvinen Neuroanatomie nicht möglich. Aufgrund der Unterschiede im Erregungs- und Reizreaktionsverhalten zwischen dem somatischen und dem autonomen Nervensystem können bekannte elektrophysiologische Neuromonitoring-Methoden nicht direkt auf das Monitoring autonomer Nerven übertragen werden, wodurch die Entwicklung neuer Methoden notwendig ist. Im Rahmen dieser Arbeit wurde eine neue Neuromonitoring-Methode zur Lokalisation autonomer Nerven mittels elektrischer Stimulation im OP-Gebiet und Impedanzmessung an glatter Muskulatur konzipiert und erforscht. Die bei der Durchführung einer präklinischen, wie auch einer klinischen Studie erzielten Ergebnisse weisen sowohl die technische als auch die klinische Machbarkeit der Impedanzmessung als Verfahren des intraoperativen Neuromonitorings nach und lassen darauf schließen, dass eine nervenschonende Präparation und eine damit verbundene Verminderung des Risikos für das Auftreten von postoperativen Funktionsstörungen der Beckenorgane möglich ist. Die klinische Machbarkeit wurde bei 28 der 30 Patienten (93,3 %) nachgewiesen. Darüber hinaus gehen erste klinische Daten zur klinischen Sicherheit und zum klinischen Nutzen der Methode aus der Auswertung des funktionellen Outcomes der Patienten der klinischen Studie hervor. Der auf der Basis der gewonnenen Messdaten der klinischen Studie realisierte softwaregestützte Automatic Muscle Impedance and Nerve Analyzer (AMINA) zur Beurteilung der abgeleiteten Impedanzsignale soll eine assistierte Signalinterpretation für eine einfache und schnelle Identifikation funktionaler Nerven während der OP möglich machen. Die Validierung des Verfahrens anhand erster Messdaten ergab eine Sensitivität der Signalbeurteilung von 96,3 % und eine Spezifität von 91,2 %.

Scene analysis is essential for enabling autonomous systems, such as mobile robots, to operate in real-world environments. However, obtaining a comprehensive understanding of the scene requires solving multiple tasks, such as panoptic segmentation, instance orientation estimation, and scene classification. Solving these tasks given limited computing and battery capabilities on mobile platforms is challenging. To address this challenge, we introduce an efficient multi-task scene analysis approach, called EMSAFormer, that uses an RGB-D Transformer-based encoder to simultaneously perform the aforementioned tasks. Our approach builds upon the previously published EMSANet. However, we show that the dual CNN-based encoder of EMSANet can be replaced with a single Transformer-based encoder. To achieve this, we investigate how information from both RGB and depth data can be effectively incorporated in a single encoder. To accelerate inference on robotic hardware, we provide a custom NVIDIA TensorRT extension enabling highly optimization for our EMSAFormer approach. Through extensive experiments on the commonly used indoor datasets NYUv2, SUNRGB-D, and ScanNet, we show that our approach achieves state-of-the-art performance while still enabling inference with up to 39.1 FPS on an NVIDIA Jetson AGX Orin 32 GB.

Convolutional Neural Networks (CNNs) are used mainly for image classification because of their high accuracy and fast performance. Due to their complexity, their functionality is like a black box to the human user. Hence, this black box functionality may classify an image based on the wrong features, which can lead to severe consequences in critical applications, such as disease detection. Therefore, explainer methods provide the users with the reasoning behind each classification. However, selecting well-matching pairs of CNN models and explainers is challenging.This paper proposes a framework for automated explainability evaluation in CNN models, which follows a quantitative approach for assessing the model and explainer pairs. The proposed framework supersedes the previous attempts towards an explainability evaluation framework by replacing the time-consuming qualitative measure with a unified quantitative metric. This quantitative approach allows the users to assess several models and explainer pairs and compare the results based on two aspects on the one hand, selecting the most prominent model/explainer pair or, on the other hand, making compromises for better real-time performance. The framework is applied to a defect detection problem for printed circuit board assembly(PCBA) automatic optical inspection (AOI). The results are analyzed for thirteen built-in CNN models and ten built-in explainer methods. The results demonstrate the superiority of the "CoAtNet0" and "Grad-Cam selfmatch" pair with the 55.38% Explanation Score (ES) metric. The authors also provide a discussion on other criteria in the selection of a prominent pair using the proposed novel FACEE framework.

The availability of safety-critical vehicle systems is essential to ensure passengers' safety in the context of automated driving and X-by-wire systems. The resulting safety-related availability requirements aim to maintain a minimum level of functionality also in the presence of component failures. In addition to electrification and the evolution of cross-domain functionalities, they also increase overall design complexity. Therefore, automated design space exploration, embedding automated safety analysis, becomes crucial for optimal system design. This article proposes a framework for model-based optimization of safety-critical mechatronic systems facing two major challenges as follows. 1) Modeling and exploration of a large design space resulting from topology variants and associated safety measures. 2) Dynamic safety analysis of each variant considering all component failures and their effects on relevant functionalities. Due to the high computational complexity, a two-level modeling approach combining behavioral and logical modules is introduced to reduce the number of evaluation runs when automatically exploring the design space. Further, an algebraic approach is proposed to calculate relevant safety metrics with high accuracy and comparatively low calculation time. The framework is exemplified by optimizing an electric powertrain over a nested design space considering acquisition and operation cost as well as different safety-related availability requirements.