<![CDATA[Tuberculosis (TB) is a chronic infectious disease and one of the leading causes of mortality worldwide. Conventional diagnostic processes are often hampered by high costs and technical complexity; consequently, Chest X-Ray (CXR) examinations combined with Artificial Intelligence (AI)-based computer detection systems have emerged as a more efficient alternative. This study aims to comparatively analyze the effectiveness of three fundamental CNN architectures AlexNet, ZFNet, and ResNet18 in detecting TB from chest X-ray images. The research methodology employs the Knowledge Discovery in Databases (KDD) framework on a public CXR dataset. Hyperparameter optimization was implemented using a Grid Search strategy integrated with 5-Fold Cross-Validation to systematically identify the optimal configuration. Experimental results indicate a significant positive correlation between architectural depth and diagnostic performance. Based on the optimal parameters identified through Grid Search specifically a learning rate of 0.0001 and a batch size of 32 the ResNet18 model demonstrated superior performance, achieving 99.28% scores for Accuracy, and 100% for precision, recall, F1-score and AUC-ROC. The superiority of ResNet18 lies in its residual learning mechanism, which effectively addresses the vanishing gradient problem and facilitates the extraction of complex pathological features. The combination of ResNet18 with Grid Search optimization demonstrates that the synergy of modern architecture and systematic tuning yields a highly reliable Computer-Aided Detection (CAD) system, surpassing the results of previous studies]]>
Copyrights © 2026
