Nicolai Spicher: Explainable AI in biomedical signal processing
Guest lecturer at ELTE Faculty of Informatics, TKP21 NVA29
Explainable AI in biomedical signal processing: Do the features of Deep Neural Networks agree with clinical guidelines?
Dr. Nicolai Spicher (University Medical Center Göttingen, Germany)
March 8, 2024: 10:00 – 11:00 (TKP meeting, online)
Abstract of the talk
Deep learning (DL) models processing biomedical time series show remarkable results. Two examples, out of many areas where they are applied successfully, are electrocardiography classificiation and polysomnography sleep staging.
Traditionally both tasks are performed by medical experts based on visual inspection of the signals following rules defined in the clinical guidelines. Recently, DL models have started to challenge human experts in accuracy and reliability. While early models were based on handcrafted features, nowadays end-to-end DL pipelines demonstrated superiority but are so-called ''black boxes'': While their architecture is known, the shear number of parameters makes it impossible to derive direct relationships between signal input and model output.
In this talk, I will give insights in our research projects where we apply attribution methods to explain the decisions of trained DL models for both clinical tasks. I will focus on the relationship between model prediction and learned features and analyze in how far the model's decisions overlap with clinical guidelines based on medical knowledge.
Dr. Nicolai Spicher is a research group leader at the University Medical Center Göttingen, Germany, where he directs the Biosignal Processing Lab since 2022. He received a B.Sc. & M.Sc. degree in Computer Science and Ph.D. degree from University Duisburg/Essen where he was a member of the Erwin L. Hahn Institute for Magnetic Resonance Imaging. He completed postdoctoral training at the Peter L. Reichertz Institute for Medical Informatics at TU Brauschweig and Hannover Medical School. Nicolai is the recipient of various awards including the Johann Peter Süßmilch Medal of the German Association for Medical Informatics, Biometry and Epidemiology. His research interests include signal processing and machine learning for cardiovascular and multimodal applications