Automated Lung Sound Classification using Multi-Scale Contextual Attention with Adaptive Feature Scaling

Respiratory sound is one of the major things that give us information about healthy and unhealthy lungs through the breath that is inhaled and exhaled. If any abnormal sound is detected in the lungs, then it is a thing, that should be seriously dealt with. Lung disease is known to be the third most common cause, which makes the people die. So it is essential to detect and classify the abnormalities that are happening in the lungs through the signal waves. Nowadays, various artificial intelligence (AI) techniques are playing a crucial role in image and signal processing. This technique offers promising result in image (or) video recognition, and audio signal processing. In those AI techniques, deep learning has played a vital role over decades for accurately detecting the disease than the other learning models. Although, it is hard to classify the abnormalities present within it. Hence, this study proposes a Multi-scale contextual attention with an adaptive feature scaling algorithm to classify the lung sound and detect abnormalities. For filtering the signal, the total variation denoising technique is used. Following this Short-Term Fourier Transform along with Zero Crossing Rate is used for feature extraction and PCA with relief is used for reducing the dimensionalities. Further, the proposed classification algorithm is also compared with the other existing algorithms, such as R-CNN, LSTM, and E-LSTM for better results. Moreover, accuracy, precision, and recall comparisons are also made to evaluate the performance of the proposed model in the detection of lung sounds through signal waves.