TLMACEA: design of a transfer learning model for correlative analysis of auscultation and clinical parameters via explainable AI-based recommender

Auscultations are commonly used to analyze lung conditions through signal processing and classification techniques. However, the efficiency of these models is often limited by factors like signal quality, sensor performance, and dataset size. Current models rely on approximations, making it difficult to pinpoint exact causes of lung conditions. To improve accuracy and interpretability, this study proposes a composite transfer learning model with explainable AI called TLMACEA. The model first converts auscultation data into 2D spectral and spatial feature vectors, which are then processed using an ensemble convolutional neural network (CNN) to identify initial lung conditions. These results are cross verified with clinical data such as lung function tests, patient demographics, smoking history, and symptoms (e.g., cough, wheezing). The data is processed through an ensemble classification layer combining random forest, support vector machine, linear regression, and Naïve Bayes models for effective lung condition prediction. The model's performance was evaluated on over 100 patients and compared to existing models. Results showed that TLMACEA outperformed state-of-the-art models, with 8.5% higher accuracy, 6.2% better precision, 7.9% improved recall, and 10.4% lower delay. The model's ensemble classification achieved 99.5% accuracy, making it suitable for real-time clinical use. The explainable AI layer also demonstrated over 98% precision, ensuring the clinical utility of the recommendations generated.