A Systematic Analysis of Tuberculosis prediction using Deep Learning Technique

Abstract

Tuberculosis (TB) ranks among the top ten causes of death attributed to infectious agents globally. Despite being a curable and preventable disease, untimely diagnosis and treatment delays can result in fatal outcomes for patients. Notably, strides in computer-aided diagnosis (CAD), particularly in the classification of medical images, play a pivotal role in the early detection of TB. Deep learning algorithms underpin the state-of-the-art CAD systems used for medical picture classification. But a significant shortcoming of these deep learning methods is that they frequently model using just one modality. This stands in stark contrast to clinical practice, where demographics, patient assessments, and laboratory test results are among the critical clinical data utilized for tuberculosis diagnosis; photos are just one component of this data. To tackle this discrepancy, we conduct a thorough literature analysis that explores different deep learning strategies and contrasts single-modal and multimodal approaches. These multimodal approaches provide a more comprehensive picture of tuberculosis diagnosis by combining additional clinical data with imaging data. To compile this comprehensive review, we systematically searched databases including Springer, PubMed, ResearchGate, and Google Scholar for original research leveraging deep learning in the context of pulmonary TB detection. By elucidating the landscape of single modal and multimodal deep learning methods, our review aims to contribute valuable insights for researchers, clinicians, and stakeholders invested in advancing the state of TB diagnostic technology and improving patient outcomes.