Optimizing Convolutional Neural Networks for Chronic Obstructive Pulmonary Disease Detection in Clinical Computed Tomography Imaging

2023-03-13

Computers in Biology and Medicine

AI Summary

To address insufficient accuracy in automatic detection and severity grading of chronic obstructive pulmonary disease (COPD) from computed tomography (CT) images, this paper proposes an end-to-end trainable convolutional neural network (CNN) framework. Methodologically: (i) a multi-scale feature fusion architecture with lung-region adaptive weighting is designed, explicitly embedding COPD pathological priors—such as emphysema spatial distribution patterns—into the network; (ii) an enhanced ResNet backbone is adopted, integrated with attention-gated modules, lung parenchyma segmentation preprocessing, and gradient clipping for optimization. Evaluated on a multi-center clinical CT dataset, the model achieves 92.4% accuracy and 89.7% AUC—outperforming baseline methods by 6.2%—and demonstrates markedly improved sensitivity for small lesions. Results are validated by dual-blinded assessment from radiologists, confirming strong potential for clinical deployment.

Problem

• Convolutional Neural Networks
• Chronic Obstructive Pulmonary Disease
• Accuracy Improvement

Innovation

• CT image enhancement
• DenseNet
• COPD detection