Binary Classification of Tuberculosis CXR Images Across Diverse Range of CNN Architectures: A Comparative Study

This paper investigates the performance of widely used pre-trained CNN architectures (VGG16, MobileNetV3, DenseNet121, and RegNet040) across diverse datasets, particularly focusing on tuberculosis (TB) detection using Chest X-Rays (CXRs). Deep learning (DL) techniques applied to CXRs aid radiologists in promptly and accurately identifying TB, which is especially critical in low-income regions with constrained diagnostic resources. The research reveals that MobileNetV3 consistently demonstrates superior performance compared to other architectures.