AI Model Achieves 90.3% Accuracy in Asthma and COPD Screening

Researchers have developed a multimodal artificial intelligence (AI) framework capable of screening asthma and chronic obstructive pulmonary disease (COPD) using respiratory sounds and clinical metadata. The system combines handcrafted acoustic features, cough and vowel sound embeddings, and patient information through a CatBoost-based machine learning model trained on the AIRS Kaggle benchmark dataset. The model achieved an overall accuracy of 90.3%, with F1-scores of 94.5% for healthy individuals, 91.5% for asthma, and 84.2% for COPD.

It also demonstrated strong reliability with a Matthews Correlation Coefficient of 0.856, Cohen’s Kappa of 0.849, and AUC values of 0.94, 0.92, and 0.99 for healthy, asthma, and COPD classifications, respectively. Ablation analyses revealed that multimodal feature fusion, attention mechanisms, data augmentation, and class-weighted learning were critical…