Flash-Attention Multiscale Stack Residual LSTM Framework for Chronic Obstructive Pulmonary Disease Detection and Classification

Recently, Chronic Obstructive Pulmonary Disease is considered a major health concern that requires early detection and classification. Various prior methods suffer from limited temporal capability and inaccurate prediction. To overcome such shortcomings, this paper proposes a novel Flash-Attention Multiscale Stack Residual Long Short Term Memory model for earlier detection. Analysis of the proposed model's effectiveness relies on two datasets. The proposed architecture integrates fine, mid, and coarse-scale residual Long Short-Term Memory blocks to learn hierarchical respiratory patterns. A Flash Attention mechanism is incorporated to enhance global temporal dependency among the datasets. Static patient attributes and sequential lung-function measurements are fused to provide a comprehensive representation of patient condition. Experimental evaluation on two benchmark datasets demonstrates that the proposed model consistently surpasses existing architectures. The proposed model attained a mean accuracy of 99.04 %, precision of 98.91 %, recall of 98.78 %, and$\mathbf{F 1}$-score of 98.84 %. Therefore, the proposed model efficient solution with strong potential for remote monitoring.