Lung Disease Diagnosis Using Radial STFT and a Lightweight Convolutional Neural Network

Lung diseases are among the leading causes of death worldwide. Nowadays, to detect lung diseases, a specialist uses auscultation to make a diagnosis. Newer auscultation devices based on stethoscopes allow these sounds to be recorded for later analysis. However, the diagnosis process is time-consuming and relies on medical expertise to generate an accurate diagnosis. For these reasons, automated and objective diagnostic systems are crucial for the early detection of lung diseases and preventing them from worsening. In this study, a computer-aided diagnostic system that integrates the Radial Short-Time Fourier Transform (RSTFT) with a Convolutional Neural Network (CNN) enhanced by attention mechanisms is presented. The RSTFT is employed to convert lung sound recordings from a public dataset into angular frequency representations, which are used as input to the CNN. The network automatically extracts discriminative features and classifies the recordings into five categories: Chronic Obstructive Pulmonary Disease (COPD), bronchiectasis, pneumonia, asthma, and healthy lungs. Experimental results demonstrate that the proposed method outperforms several state-of-the-art approaches in terms of accuracy, precision, recall, and F1-score. These findings indicate that the proposed RSTFT–CNN framework provides an effective and reliable solution for the automated diagnosis of lung diseases, offering valuable support for clinical decision-making and early intervention.