Deep Learning for Lung Disease Detection: A CNN-Based Approach

Auscultation involves the use of a stethoscope to diagnose respiratory diseases and is a straightforward and non-intrusive diagnostic approach. However, the accuracy of the diagnosis relies heavily on the doctor’s expertise, making it challenging to achieve precise quantitative assessments. In this paper, we have used the difference signals of the actual lung sound signals to detect various lung diseases using a convolutional neural network. The difference signals can classify lung signals better than the raw signals. The raw signals attained an accuracy of 90%, whereas the difference signals resulted in an accuracy of 93%. In binary classification tasks, we have attained 100% accuracy, distinguishing between signals from the disease class and those from the normal class. The proposed architecture, comprising solely two convolution layers, is lightweight, facilitating practical implementation. Utilizing difference signals has yielded superior results compared to the existing techniques.