Machine Learning-Driven Lung Sound Analysis: Novel Methodology for Asthma Diagnosis.

Area of Science:

• Respiratory Medicine
• Artificial Intelligence in Healthcare
• Diagnostic Technology

Background:

• Asthma is a chronic airway disease with variable symptoms, often presenting normally in well-controlled cases, complicating diagnosis.
• Traditional diagnostic methods like spirometry and bronchial provocation tests have limitations, including invasiveness and potential for inducing bronchoconstriction.

Purpose of the Study:

• To develop a non-invasive, objective, and reproducible diagnostic method for early asthma detection.
• To utilize machine learning-based lung sound analysis for identifying asthma, even during stable periods with normal spirometry.

Main Methods:

• A machine learning algorithm was developed to classify controlled asthma patients and healthy individuals using digital stethoscope recordings of respiratory sounds.
• 120 participants (60 asthmatic, 60 healthy) were enrolled. Respiratory sound segments were analyzed using Mel-Frequency Cepstral Coefficients (MFCCs) and Tunable Q-Factor Wavelet Transform (TQWT).
• Features selected with ReliefF were used to train Quadratic Support Vector Machine (SVM) and Narrow Neural Network (NNN) models.

Main Results:

• Pulmonary function tests showed lower FEV1 and FEV1/FVC ratios in the asthma group, though within normal ranges.
• The Quadratic SVM model achieved 99.86% accuracy, correctly classifying 99.44% of controls and 99.89% of asthma cases.
• The Narrow Neural Network model achieved 99.63% accuracy, with sensitivity, specificity, and F1-scores exceeding 99%.

Conclusions:

• The developed machine learning algorithm accurately diagnoses asthma, even in patients with normal spirometry and clinical findings.
• This approach offers a non-invasive, objective, and efficient diagnostic tool for early asthma detection.
• Lung sound analysis combined with machine learning shows significant potential for improving asthma diagnosis.