PPEA: Post-Position Encoding Attention for Imbalanced Lung Sound Classification

Early detection and continuous monitoring of respiratory diseases remain critical challenges in clinical practice. Although auscultation is a primary diagnostic tool, the interpretation of lung sounds requires significant expertise and is subject to inter-observer variability. This paper presents PPEA, a novel deep-learning framework designed specifically for automated respiratory sound classification. Our approach addresses the inherent challenge of imbalanced clinical data through a novel post-position encoding attention mechanism and layer-wise feature fusion strategy. Evaluations on the ICBHI dataset demonstrate superior performance, achieving a 0.9933 specificity and 0.7863 sensitivity across six respiratory conditions, significantly outperforming existing methods and handling few-shot settings. These results suggest PPEA's potential for supporting decision-making in respiratory care.