Adaptive segmentation for accurate detection of tuberculosis disorder using multiscale residual densenet with attention mechanism by analyzing the chest X-ray images.

Tuberculosis (TB) remains a formidable global health issue, particularly in areas with restricted access to specialized healthcare services. Reliance on expert radiologists for interpreting chest X-rays (CXRs) poses a significant problem, especially in resource-constrained settings. This paper proposes an advanced DL framework with an attention mechanism for automatic TB detection from CXR images to alleviate this burden and enhance diagnostic accuracy. The methodology encompasses image collection, segmentation, and TB classification. In this study, we developed an adaptive segmentation approach, termed Adaptive SegUnet, by using the "VinDr-CXR" dataset and optimizing parameters with the TOA. Subsequently, TB classification leverages multiscale residual Densenet with attention mechanism (MResDen-AM) to enhance accuracy. The novelty of this methodology lies in its Adaptive Segmentation with Multiscale Residual Densenet and Attention Mechanism for TB detection. Experimental evaluation demonstrates significant advancements over existing techniques, achieving a remarkable accuracy rate of 98%. The methodology's superiority in sensitivity, specificity, and false positive rate compared to conventional approaches is particularly noteworthy. This study presents a robust automated detection system poised to revolutionize TB screening, offering a promising global solution to improve patient outcomes.