Deep Learning-Based Early COPD Detection and Severity Classification Using Respiratory Sound Analysis
Year: 2026
Type: article
Abstract: Chronic Obstructive Pulmonary Disease (COPD) is normally confirmed by spirometry, which needs calibrated equipment and trained staff that are scarce in rural and resource-limited settings.This paper presents a two-model deep learning framework for respiratory sound analysis on the public ICBHI 2017 database.Model 1 is a binary COPD screening network that passes a 128 × 157 log-Mel spectrogram through a frozen EfficientNetB0 backbone, a bidirectional LSTM and four-head self-attention.Model 2 is a four-stage residual CNN that assigns each annotated respiratory cycle to one of four coded acoustic categories, copd0 to copd3.The models are joined in a cascade in which Model 2 runs only when Model 1 predicts COPD.On a patient-level held-out test set of 1255 cycles, Model 1 reached 87.73 percent accuracy, 88.38 percent sensitivity, 83.80 percent specificity and an AUC of 0.9313.Model 2 reached 91.58 percent four-class accuracy and an ICBHI score of 92.56 percent.Because the binary test set is dominated by COPD cycles, Model 1 accuracy is only modestly above the 85.74 percent majority-class rate, and the four classes are acoustic labels rather than clinical severity grades.A React and Express web interface exposes the pipeline with an explicit research disclaimer.
Source: IJARCCE
Language: English
Subfield: Pulmonary and Respiratory Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W7236572075
