Neural Perspectives on Chest X-rays: Comparative Evaluation of Deep Learning Models for Thoracic Disorder Detection

Thoracic diseases are viral infections that spread at a rapid rate, and they have affected millions of people around the globe. The increasing prevalence and rapid dissemination of the disease have placed healthcare professionals in a precarious position, impeding their ability to promptly identify and contain the illness. The swiftly developing domain of medical image analysis exhibits potential for effectively tackling this obstacle with enhanced accuracy. This study compares and contrasts a number of modern deep learning architectures, including CapsNet, AlexNet, VGGNet, GoogLeNet, ResNet, and DenseNet, in order to find and label a wide range of thoracic disorders. The study utilized a dataset consisting of 3043 chest X-ray images. Among the patients included were those who were deemed healthy (134 images), those who had tuberculosis (138 images), and those who had COVID-19 (217 images). CapsNet's accuracy of 98.48% serves as proof that it performs poorly when compared to other models based on perplexity and performance metrics.