An Ensemble of Transfer, Semi-supervised and Supervised Learning Methods for Pathological Heart Sound Classification

Abstract

In this work, we propose an ensemble of classifiers to distinguish between various degrees of abnormalities of the heart using Phonocardiogram (PCG) signals acquired using digital stethoscopes in a clinical setting, for the INTERSPEECH 2018 Computational Paralinguistics (ComParE) Heart Beats Sub-Challenge. Our primary classification framework constitutes a convolutional neural network with 1D-CNN time-convolution (tConv) layers, which uses features transferred from a model trained on the 2016 Physionet Heart Sound Database. We also employ a Representation Learning (RL) approach to generate features in an unsupervised manner using Deep Recurrent Autoencoders and use Support Vector Machine (SVM) and Linear Discriminant Analysis (LDA) classifiers. Finally, we utilize an SVM classifier on a high-dimensional segment-level feature extracted using various functionals on short-term acoustic features, i.e., Low-Level Descriptors (LLD). An ensemble of the three different approaches provides a relative improvement of 11.13% compared to our best single sub-system in terms of the Unweighted Average Recall (UAR) performance metric on the evaluation dataset.

Machine Listening for Heart Status Monitoring: Introducing and Benchmarking HSS—The Heart Sounds Shenzhen Corpus

“…There were no winners in the INTERSPEECH 2018 COMPARE challenge heart sound sub-challenge. In addition to the official baseline paper [8], there are three published works which used our HSS database [46], [69], [70]. In the study of [69], the authors proposed a 1D-CNN time-convolution (tConv) layers based model pre-trained by the PhysioNet CinC Challenge database to learn higher representations from the heart sounds.…”

“…In addition to the official baseline paper [8], there are three published works which used our HSS database [46], [69], [70]. In the study of [69], the authors proposed a 1D-CNN time-convolution (tConv) layers based model pre-trained by the PhysioNet CinC Challenge database to learn higher representations from the heart sounds. In addition, they also investigated representation learning (RL) by sequence to sequence autoencoders.…”

“…The state-of-the-art deep learning based techniques can be promising in extracting some higher representations without any domain knowledge. However, the performance between the development set and the test set has a substantial gap [8], [46], [69]. One reasonable explanation for this phenomenon is that, due to the extremely limited data size, overfitting occurred in the development set.…”

Deep Learning Methods for Heart Sounds Classification: A Systematic Review

“…In addition, Humayun et al [71] proposed a 1D CNN in which transfer learning was used to learn the parameters of a 1D CNN model pre-trained on the PhysioNet HS Classification dataset. The flattened layer was transferred [72] to a new CNN architecture with a fully connected layer and three output neurons representing normal, mildly abnormal, and severely abnormal categories, respectively.…”

Towards Domain Invariant Heart Sound Abnormality Detection Using Learnable Filterbanks

“…data acquired from different stethoscopes (especially of lower cost) and diverse environments. The frequency characteristics of the stethoscope or sensor used for recording can cause machine learning models to be biased towards majority sources of training data [3], [4]. The visible clusters in Fig 1-(a) corresponding to different stethoscope models prove that feature distributions can be significantly different depending on which domain the data is from.…”