A deep CNN-based acoustic model for the identification of lung diseases utilizing extracted MFCC features from respiratory sounds

Machine learning algorithms have recently been increasingly used in medical data, particularly in healthcare areas where image processing techniques have played a crucial role. This study aims to utilize artificial intelligence (AI) techniques to forecast respiratory diseases by implementing a deep convolutional neural network (CNN) structure. The study employs an extensive dataset, specifically the Public Breathing Sound Database, which includes breathing sounds from 126 individuals with six different respiratory disorders. Furthermore, the main aim of this study is to tackle the difficulties related to the precise detection of lung disorders by creating a strong and effective model. The study examines the intricacies of pre-processing audio data, augmenting it, and extracting information from it. The primary focus is the utilization of Mel-frequency cepstral coefficients (MFCC) to identify significant characteristics of respiratory sounds. The suggested methodology utilizes a deep CNN structure to analyze retrieved characteristics and accurately identify diseases by detecting patterns and correlations. Moreover, the outcomes demonstrate a significant improvement in the precision of the model following the implementation of data balancing and augmentation strategies. The created model obtains a remarkable accuracy of 97.4% on the validation dataset, showcasing its effectiveness in training. Furthermore, it maintains a high accuracy of 95.1% on the independent test dataset. This research adds to the expanding collection of studies at the crossroads of AI and healthcare and shows great potential for promptly and precisely detecting respiratory disorders using acoustic signals. The results highlight the capacity of deep learning methods to transform diagnostic procedures in respiratory healthcare fundamentally.