Author:

Abstract:

Biomedical signals are non-invasive data modalities that provide insight into processes taking place in the body and in specific organs. Electrocardiography (ECG) and electroencephalography (EEG) are two such modalities, meant to monitor the heart and brain respectively. Both provide a key tool for clinical experts in the diagnosis and follow-up of various pathologies, and for the general monitoring of the body. The signals capture changes in electric potential on the surface of the skin caused by specific physiological processes (such as the engagement of muscles or the firing of neurons). Part of the analysis of these signals is the detection of important patterns, often coming in the form of events of interest such as epileptic seizures which can be observed in the EEG. Unfortunately, sources outside of the organs under investigation can contribute to the potentials measured on the skin. Other muscles or electric inference outside of the body can interfere with the desired signal, causing disturbances. Many algorithms exist for the detection of such disturbances and signal events, tailor-made for specific types of targets and specific signal modalities. Designing and applying the right algorithm for a novel setting takes substantial domain expertise and engineering effort. This PhD develops novel, generic algorithms for the detection of disturbances and events in biomedical signals. Designing generic algorithms that can easily be adapted for a different setting reduces the need for domain expertise when designing tools for the analysis of biomedical signals. We rely on deep learning methods to automatically learn relevant characteristics and patterns for the task at hand.