Energy Efficient Machine Learning-Based Classification of ECG Heartbeat Types

Abstract

To meet the accuracy, latency and energy efficiency requirements during real-time collection and analysis of health data, a distributed edge computing environment is the answer, combined with 5G speeds and modern computing techniques. Using the state-of-the-art machine learning based classification techniques plays a crucial role in creating the optimal healthcare system on the edge. This thesis first provides a background on the current and emerging edge computing classification techniques for healthcare applications, specifically for electrocardiogram (ECG) beat classification. We then present key findings from an extensive survey of over hundred studies on the topic while taxonomizing the literature with respect to key architectural differences, application areas and requirements. Leveraging the insights drawn from the extensive analysis of the pertinent literature we select a set of most promising machine learning based classification techniques for ECG beats, based on their suitability for implementation on a small edge device called a Raspberry Pi. After implementing these classification techniques on a Raspberry Pi based platform we perform a comparison of the performance of these classification techniques with respect to three key performance indicators (KPI) of interest for health care applications namely accuracy, energy efficiency, and latency. ECG measures the electrical activity of the heart and help healthcare professionals to evaluate heart conditions of a patient, sometimes diagnosing life-threatening conditions. The features of ECG signals are pre-processed and fed into the classification algorithms to detect and classify abnormal beat types. ECG classification requires low complexity but still high level of performance in terms of aforementioned three KPIs. The classification algorithms chosen, namely Naïve Bayes, Multilayer Perceptron (MLP), and distilled deep neural network (DNN) are all energy efficient methods hence suitable for implementation for small edge devices. The comparative multi-faceted evaluation presented in this thesis is a new contribution to research that exists on edge based classification as it offers comparison of selected classification algorithms in terms three KPIs instead of one while using real edge device based implementation. The performance of analyzed machine learning classification techniques is ranked according to each KPI. Benefiting from the results of the comparative analysis presented in this thesis a particular classification algorithm can be selected for optimal deployment in given scenario in healthcare system depending on the specific requirements of the given scenario. Edge computing paves the way for a new generation of health devices that can offer a higher quality of life for users if low-latency, low-energy, and high- performance requirements are addressed.