Classification of cardiac dysfunctions using a 3-axis accelerometer and deep learning architectures

Abstract

This thesis explores the possibility of using a 3-axis accelerometer and deep learning architectures to classify cardiac dysfunctions. The thesis is built upon previous research performed at The Intervention Centre at Oslo University Hospital, where they have developed a novel technique for detecting cardiac heart dysfunctions using a 3-axis accelerometer. The purpose of this thesis is to examine if deep learning architectures can be used to automatically detect cardiac dysfunctions using the 3-axis accelerometer. Three experiments have been conducted. The first experiment explores what deep learning architectures that works best for extracting features from the accelerometer data, as well as building a bridge between deep learning and previous research performed at The Intervention Centre. The second experiment is performed to see if deep learning can be applied to classify heart functions using 3-axis accelerometer data. The last experiments is conducted to investigate wheter further enhancement of the classifier is possible by converting the accelerometer data into the image domain. The conclusion is that deep learning is an excellent tool for classifying cardiac heart dysfunctions, and can also be used as a tool for extracting clinical knowledge.