Attention Tub: Harnessing Deep Attention Network for Tuberculosis Detection in Chest X-Rays

Abstract

Pulmonary tuberculosis (TB) is a significance issue for public health worldwide, thus requiring accurate and prompt diagnosis for effective prevention and treatment Chest x-ray (CXR) is an essential tool a for diagnosis due to its high cost and non-invasive nature. However, models for deep learning in particular convolutional neural networks (CNNs) often face challenges such as overfitting and difficulty in capturing subtle differences in lesion characteristics and in order to overcome that on these issues a new approach is proposed, the Deep Attention Network (DANet). The model adds layers for feature extraction, followed by maximal pooling procedures to reduce spatial dimensions. Skip connections facilitate gradient propagation during training, while external attention introduced through the connection layer enhances feature representation by focusing on appropriate image locations Global maximal pooling for feature collection its classification is simple, and results in binary classification tasks 99% accuracy. Remarkable accuracy in findings suggest that DANet can be a valuable tool to aid in clinical decision making, providing important support radiologists and medical professionals.