Prediction and Tracking Changes in Bio-medical Sensor Data

Abstract

Both our preliminary works [1], [2], on biomedical sensor signal processing, have focused on abdominal tumor motion traces. In stereotactic radiotherapy for thoracic and abdominal tumors, respiratory motion management is crucial for improving efcacy of treatment, while minimizing risk to heathy tissue and organs. Since tumor motion exhibits dynamic variation in characteristics, between and within patients, our first work concentrated on predicting imminent anomalous or irregular tumor motion ahead of its occurrence, and our second work consists of analysis of behavioral distribution of tumor motion, used for patient grouping with the aim to improve treatment planning.As an immediate step forward for improving treatment planning for patients with abdominal tumor, we are currently focusing on developing patient profiles. The profiles will be in the form of a histogram, based on behaviors illustrated by a patient during a particular treatment session. We expect the patient profiles to advance interpretability of patient groping or clusters, which in return will improve the decision of choosing a treatment plan. We need to decide the best distance metric to use for histogram comparison and wish to reduce the number of working parameters needed to describe the profile.