ML Anomaly Detection for Ex Vivo Perfusion
Master's thesis (6.00/6.00) — deep learning for multivariate time-series anomaly detection during organ perfusion.
My Master’s thesis, carried out at the Institute for Dynamic Systems and Control (IDSC), ETH Zürich, in collaboration with the University Hospital of Zurich (USZ), investigated machine-learning methods for detecting anomalies during ex vivo perfusion of organs such as livers and limbs.
The work focused on multivariate time-series signals from the perfusion machine, using LSTM networks and autoencoders to flag deviations from healthy perfusion in real time — a step toward safer, more autonomous organ-preservation systems.
- Supervisor: Prof. Dr. Christopher Onder — IDSC, ETH Zürich
- Grade: 6.00 / 6.00
- Keywords: medical devices, time-series anomaly detection, LSTM, autoencoders, deep learning