Machine Learning Techniques for Source Localisation in Elastic Media

TL;DR

This paper explores machine learning methods to localize sources of acoustic signals in elastic media, aiming to improve non-invasive detection of coronary artery disease by analyzing surface signals caused by arterial occlusions.

Contribution

It evaluates seven ML algorithms for source localization in elastic media and identifies an ensemble model combining k-NN and Random Forest as the most effective.

Findings

Ensemble model outperforms individual algorithms.

Best performance measured by mean squared error and Euclidean distance.

ML techniques can accurately predict occlusion sources from surface signals.

Abstract

Coronary Artery Disease (CAD) results from plaque deposit in a coronary artery. Early diagnosis is imperative, so a non-invasive detection method is being developed to identify acoustic signals caused by partial occlusions in the artery. The blood flow in the artery is disturbed and imposes oscillatory stresses on the artery wall. The deformations caused by the stresses can be detected at the chest surface. Therefore, by using data simulating these surface signals, which arise from randomly assigned source positions, machine learning (ML) can be utilised to predict the source of the occlusion. Seven ML algorithms were investigated, and the results from this study found that an ensemble model combining k-Nearest Neighbours and Random Forest had the best performance. The metrics used to evaluate this was the mean squared error and Euclidean distance.