Scienmag: Latest Science and Health News
Enhanced 3D ResNet Model Classifies Active Pulmonary Tuberculosis on Chest
xCruzo Brief
A retrospective study published in Cluster Computing reports a deep-learning system that distinguishes active from inactive pulmonary tuberculosis using 3D chest CT scans. The model, 3D MSGC-ResNet50, segments lung parenchyma with a 3D ResUNet and then classifies disease stage using a ResNet50 backbone enhanced with multi-scale attention and grouped convolution. Tested on 500 patients from the Public Health Treatment Center of Guiyang City in Guizhou Province (250 active, 250 inactive), it achieved 94.87% precision, 91.60% recall, a 93.19% F1-score, and a 96.68% ROC area. It outperformed a baseline 3D ResNet50 by several percentage points and reached 90.00% accuracy (36 of 40) on an external set.
xCruzo quick-read summary • Source: Scienmag: Latest Science and Health News • Read the full article for complete information.







