In this project we received a wide range of medical measurements for various patients, including: ECG, blood pressure, chest volume, saturation, carbon dioxide emission and air flow. In the first step, the signals underwent preprocess of cleaning noise and artifacts. In the second stage, we extracted features with statistical and medical significance in consultation with Dr. Eytan. In the third stage, we applied an algorithm of dimensionality reduction called diffusion map. We applied it on each measurement separately using a Euclidean metric, and on the measurements together using Riemannian metric. In the final stage, we performed a visual analysis of the results by presenting the three most informative feature vectors. The analysis was based on prior knowledge of the time when the patients were disconnected from an external respiration. The hypothesis of the project was that after lowering the dimensionality, it would divide the diffusion map results into two manifolds according to the time of disconnection. In our case, we expected to receive the following separated manifolds: before disconnecting from the respirator and after disconnecting from the respirator. The results we received were inconclusive, some of the results we received were indeed characterized by good separation to two manifolds, but there were also results characterized by only one manifold.
Dr. Dani Eitan