Respiratory Sound Using Deep Learning
Respiratory sound classification is an important application in the field of medical diagnostics. Identifying and classifying different types of respiratory sounds, such as wheezes, crackles, or other abnormal lung sounds, can assist in diagnosing respiratory diseases like asthma, pneumonia, and chronic obstructive pulmonary disease (COPD). In this report, we explore both deep learning and machine learning approaches to automatically classify respiratory sounds using an annotated dataset. The main objective of this project is to build models that can accurately classify sounds into categories like "normal," "wheeze," "crackles," or "both" (both wheezes and crackles).
Dataset
Respiratory sounds are important indicators of respiratory health and respiratory disorders. The sound emitted when a person breathes is directly related to air movement, changes within lung tissue and the position of secretions within the lung. A wheezing sound, for example, is a common sign that a patient has an obstructive airway disease like asthma or chronic obstructive pulmonary disease (COPD). These sounds can be recorded using digital stethoscopes and other recording techniques. This digital data opens up the possibility of using machine learning to automatically diagnose respiratory disorders like asthma, pneumonia and bronchiolitis, to name a few. The Respiratory Sound Database was created by two research teams in Portugal and Greece. It includes 920 annotated recordings of varying length - 10s to 90s. These recordings were taken from 126 patients. There are a total of 5.5 hours of recordings containing 6898 respiratory cycles - 1864 contain crackles, 886 contain wheezes and 506 contain both crackles and wheezes. The data includes both clean respiratory sounds as well as noisy recordings that simulate real life conditions. The patients span all age groups - children, adults and the elderly. This Kaggle dataset includes: 920 .wav sound files 920 annotation .txt files A text file listing the diagnosis for each patient A text file explaining the file naming format A text file listing 91 names (filename_differences.txt ) A text file containing demographic information for each patient
Results
The results demonstrate the effectiveness of deep learning models, particularly CNNs, for respiratory sound classification. By optimizing the hyperparameters using Spider Wasp Optimization (SWO) and Improved Spider Wasp Optimization (ISWO), the model's performance improved significantly, achieving near-perfect classification results. Initial CNN Accuracy: 94% After SWO Optimization: 97% After ISWO Optimization: 100%
