On Edge Wearable ECG Signal Quality Assessment using Residual Hermite Projection and Liquid State Machine, a Hierarchical Domain Adaptation Approach

Samiee, Kaveh and Kovács, Péter (2024) On Edge Wearable ECG Signal Quality Assessment using Residual Hermite Projection and Liquid State Machine, a Hierarchical Domain Adaptation Approach. In: 51st Computing in Cardiology. IEEE, Karlsruhe. (In Press)

Text CinC247.pdf - Published Version Restricted to Registered users only Download (1MB) | Request a copy

Abstract

This paper presents a novel technique for real-time quality assessment of Electrocardiogram (ECG) signals on the edge. First, a Variable Projection Neural Network (VPNN) with Hermite bases, in regressor mode, is trained on single-channel ECG epochs. Then, the residual between each input epoch and its reconstruction is passed through a Poisson rate encoder of length T and then fed to a Liquid State Machine (LSM). To classify ECG epochs into three classes: good, intermediate, and bad; a linear Support Vector Machine (SVM) is trained on LSM spiking outputs at time T to perform a crisp distinction between good-intermediate and bad classes. The distinction between good and intermediate is achieved by training another linear SVM on the subset of LSM outputs corresponding to epochs predicted as good-intermediate by the former SVM. By leveraging a pre-trained Hermite VPNN regressor as the feature extractor preceding the LSM, the average F1 scores of 99.1% and 78.6% are obtained in the crisp classification of one-second ECG epochs into goodintermediate vs bad and good vs intermediate, respectively. Due to the one-second temporal resolution and robust discriminatory power of the proposed technique, prediction results can be easily translated into some actionable technical alarms.

Actions (login required)

Edit Item