RESPIRATORY SOUND CLASSIFICATION USING MACHINE LEARNING

Respiratory diseases are a major global health challenge, ranking among the leading cause for mortality and
disability worldwide. Conditions such as chronic obstructive pulmonary disease (COPD), asthma,
tuberculosis, and lung cancer contribute significantly to global mortality, with COPD alone responsible for
an estimated 3.91 million deaths annually. The burden is particularly severe in low-income regions with
limited healthcare infrastructure, where socioeconomic factors and a shortage of medical professionals
further hinder timely and accurate diagnoses. To tackle these issues, innovations in Artificial Intelligence
provide effective strategies to minimize misdiagnoses and facilitate precise treatments. A cost-efficient and
user-friendly algorithm was created to analyze respiratory sounds, ensuring compatibility with various
devices. The methodology integrates machine learning approaches, utilizing Gammatone Cepstral
Coefficients (GTCC) within a Convolutional Neural Network (CNN) and a CNN-LSTM hybrid model
incorporating GTCC and STFC features. Four datasets were prepared for classifying respiratory audio,
covering healthy versus pathological classification, rale/rhonchus/normal sound classification, singular
respiratory sound type classification, and audio type classification with all sound types.