A crucial part of examining patients suspected of having respiratory diseases is auscultation. Doctors use a stethoscope to listen to breath sounds to diagnose lung conditions. These noises, such as wheezes and crackles, indicate different respiratory conditions. Nevertheless, distinguishing these sounds is frequently reliant on the doctor's expertise and is subject to interpretation, underscoring the necessity for a more impartial method. This research presents a method that utilizes deep learning to examine lung sounds. The setup includes an electronic stethoscope for collecting chest noises, a smartphone application for recording and tracking patient information, and a sophisticated artificial intelligence model for categorizing chest sounds. Data preprocessing for audio involves normalization, temporal segmentation, applying Short-Term Fourier Transform (STFT), and converting data into spectrograms suitable for CNN input. The CNN is trained using multilabel classification methods, utilizing categorical cross-entropy as the loss function and assessing metrics like accuracy, precision, recall, F1-score, and ROC curve examination. The mobile app, which is easy to use, utilizes Flutter for the frontend and MongoDB and Django for the backend and database, respectively, guaranteeing compatibility across platforms, as well as speed and scalability.

This study suggests a practical categorization of lung noises as wheezes, crackles, and normal states. The performance metrics of the model indicate its potential value in clinical environments, with a validation precision of 96.42%, a recall of 94.75%, and a validation loss of 0.15. The combination of an e-stethoscope, mobile app, and DL model in the automated analysis system for lung sounds shows great potential in enhancing the accuracy of diagnosing respiratory diseases. This approach has the potential to improve patient outcomes by reducing the subjectivity of traditional auscultation, leading to more accurate and timely intervention.