Respiratory Sound Based Classification of Chronic Obstructive Pulmonary Disease: a Risk Stratification Approach in Machine Learning Paradigm | JoVE Visualize

Area of Science:

Biomedical Engineering
Respiratory Medicine
Artificial Intelligence

Background:

Chronic Obstructive Pulmonary Disease (COPD) diagnosis relies on clinical assessment and spirometry.
Respiratory sound analysis offers a non-invasive method for evaluating lung function.
Machine learning presents opportunities to improve the accuracy and efficiency of COPD diagnosis.

Purpose of the Study:

To investigate the efficacy of machine learning techniques in classifying normal and COPD subjects using respiratory sound analysis.
To evaluate the diagnostic performance of various machine learning classifiers based on lung sound and spirometry features.
To determine if combining respiratory sound features with spirometry data enhances COPD classification accuracy.

Main Methods:

Data from 30 COPD and 25 healthy subjects were collected, with 39 lung sound and 3 spirometry features extracted.
Machine learning classifiers including Support Vector Machine (SVM), k-nearest neighbor (KNN), logistic regression (LR), decision tree, and discriminant analysis (DA) were employed.
Feature relevance was assessed using parametric and nonparametric tests, and classification accuracy was evaluated.

Main Results:

The SVM classifier achieved a maximum accuracy of 83.6% using key lung sound parameters (median frequency and linear predictive coefficients).
Combining relevant lung sound features with spirometry parameters (Forced Vital Capacity - FVC and Forced Expiratory Volume in 1s - FEV1) resulted in 100% classification accuracy with SVM and LR classifiers.
The study demonstrated the superior performance of combined respiratory sound and spirometry data for COPD classification.

Conclusions:

Integrating lung sound analysis with spirometry data significantly improves the accuracy of COPD diagnosis.
The proposed machine learning approach can assist clinicians in routine practice, potentially leading to automated COPD diagnosis.
Future development of handheld medical systems incorporating lung sounds for COPD diagnosis is feasible and clinically significant.