A Lightweight CNN Model for Detecting Respiratory Diseases from Lung Auscultation Sounds using EMD-CWT-based Hybrid Scalogram

Abstract

Listening to lung sounds through auscultation is vital in examining the respiratory system for abnormalities. Automated analysis of lung auscultation sounds can be beneficial to the health systems in low-resource settings where there is a lack of skilled physicians. In this work, we propose a lightweight convolutional neural network (CNN) architecture to classify respiratory diseases using hybrid scalogram-based features of lung sounds. The hybrid scalogram features utilize the empirical mode decomposition (EMD) and continuous wavelet transform (CWT). The proposed scheme's performance is studied using a patient independent train-validation set from the publicly available ICBHI 2017 lung sound dataset. Employing the proposed framework, weighted accuracy scores of 99.20% for ternary chronic classification and 99.05% for six-class pathological classification are achieved, which outperform well-known and much larger VGG16 in terms of accuracy by 0.52% and 1.77% respectively. The proposed CNN model also outperforms other contemporary lightweight models while being computationally comparable.

A Robust CNN Framework with Dual Feedback Feature Accumulation for Detecting Pneumonia Opacity from Chest X-ray Images

“…Nevertheless, against the magnificent performance of the deep works, their resource intensive training acts as a trade-off and ultimately make them unfeasible for applying in point-of-care remote setup and mobile platforms. To resolve this issue, a large fraction of contemporary medical diagnosis algorithms are being geared towards efficient approaches namely weight quantization [10], [11], low precision [12] and lightweight networks [13], [14]. Although numerous investigations on CXRs have been implemented to detect pneumonia opacity, the existing works have scarcely explored [6] the efficient approaches considering the trade-off between accuracy and number of parameters.…”

CardioXNet: A Novel Lightweight Deep Learning Framework for Cardiovascular Disease Classification Using Heart Sound Recordings

“…Advent of deep learning (DL) architectures namely Convolutional Neural Networks (CNNs), Unsupervised Pre-trained networks (UPNs), Recurrent and Recursive neural networks (RNNs) and their application in the the domain of PCG classification have intuitively resolved both the generalization and the accuracy concern by utilizing the inherent self-learning competency of these networks [14], [15]. However, these deep networks attain the automated feature extraction capability only after going through a computationally complex extensive training phase with a significantly large dataset [28]. The resource-intense requirement of DL-based frameworks makes them unbefitting to be deployed in the low-resource point-of-care locations of the developing and under-developed countries.…”

“…A handful of crossdomain studies have investigated the issue and introduced a few advanced strategies such as lightweight networks [29], [30], weight quantization [31] and low precision computation techniques [32]. Recently, some studies have successfully implemented these concepts in various biomedical applications including ECG classification [33], respiratory disease classification [28] and most importantly, PCG classification [15].…”