Compressed Sensing ECG using Restricted Boltzmann Machines

TL;DR

This paper introduces a novel compressed sensing method for ECG signals that leverages restricted Boltzmann machines to model sparsity patterns, resulting in fewer measurements and improved reconstruction accuracy.

Contribution

It presents a new approach using RBMs to exploit higher-order statistical dependencies in ECG sparsity patterns, enhancing compressed sensing performance.

Findings

Fewer measurements needed for accurate ECG reconstruction

RBM-based modeling captures complex sparsity dependencies

Improved reconstruction accuracy demonstrated experimentally

Abstract

Recently, it has been shown that compressed sensing (CS) has the potential to lower energy consumption in wireless electrocardiogram (ECG) systems. By reducing the number of acquired measurements, the communication burden is decreased and energy is saved. In this paper, we aim at further reducing the number of necessary measurements to achieve faithful reconstruction by exploiting the representational power of restricted Boltzmann machines (RBMs) to model the probability distribution of the sparsity pattern of ECG signals. The motivation for using this approach is to capture the higher-order statistical dependencies between the coefficients of the ECG sparse representation, which in turn, leads to superior reconstruction accuracy and reduction in the number of measurements, as it is shown via experiments.