Differential Diagnosis of Infectious Respiratory Diseases via Hybrid Deep Learning: Clinical Evaluation of COVID-19, Pneumonia, and Tuberculosis on Chest Radiography

Abstract

The timely and precise differentiation of infectious respiratory diseases is vital in ensuring opti-mal patient outcomes and reinforcing public health infection control measures. The present study presents a precise hybrid deep learning architecture for the classification of infectious respiratory diseases based on chest X-ray images. The model was able to classify images into COVID-19, Normal, Pneumonia, and Tuberculosis. The model was able to perform a comprehensive analysis of localized pathological features in the images through the employment of a seg-mental random patch-based sampling strategy. The model also employed a pretrained ResNet-18 model in com-bination with a Context Pooling layer and Long Short-Term Memory networks. The model was able to achieve a remarkable 96% overall accuracy on a comprehensive dataset of 7,132 images. Notably, the model achieved a remarkable level of diagnostic sensitivity for high-priority com-municable diseases, including a 99% F1-score for Tuberculosis and 95% for COVID-19 after only three training epochs. The results indicate that the inclusion of spatial-temporal sequence model-ing results in a potent and efficient Clinical Decision Support System (CDSS), which can aid in automated screening and triage in resource-constrained settings where access to radiological ex-pertise may be limited.