Weak Localization of Radiographic Manifestations in Pulmonary Tuberculosis from Chest X-ray: A Systematic Review.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Pulmonary tuberculosis (PTB) is a leading cause of infectious disease mortality globally.
• Chest radiography (CXR) is crucial for diagnosing PTB, but feature interpretation relies heavily on radiologist expertise.
• Deep learning (DL) shows promise for detecting PTB in CXRs, yet fully supervised methods require extensive, costly, and subjective annotations.

Approach:

• This review focuses on weakly supervised learning techniques for localizing PTB radiographic manifestations.
• It examines common public CXR datasets used for tuberculosis identification.
• The review discusses various approaches for weakly localizing PTB features in chest X-rays.

Key Points:

• Weakly supervised localization highlights image regions most influential for DL classification, aiding disease pinpointing.
• This approach mitigates the need for pixel-level annotations, reducing time and cost.
• Identifying radiographic features like cavitation, consolidation, and nodules is key to diagnosis.

Conclusions:

• Weakly supervised methods offer a viable alternative to fully supervised segmentation for PTB detection in CXRs.
• Further research is needed to address the limitations and challenges of these techniques.
• Advancing these methods can improve the accuracy and efficiency of PTB diagnosis.