YOLOv8's advancements in tuberculosis identification from chest images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Tuberculosis (TB) poses significant health risks due to diagnostic challenges.
• Traditional TB detection methods like X-rays and CT scans are time-consuming and often lack accuracy.
• Existing automated TB prediction models face issues with accuracy, overfitting, and speed.

Purpose of the Study:

• To improve the accuracy and efficiency of tuberculosis detection.
• To address limitations in current automated TB prediction models, particularly regarding small object detection.
• To enhance the YOLOv8 object detection model for medical image analysis.

Main Methods:

• Implementation of the Selection Focal Fusion (SFF) block within the YOLOv8 object detection model.
• Utilizing an attention mechanism to refine detection capabilities.
• Training and evaluation on the Kaggle TBX-11k dataset.

Main Results:

• The SFF block integration improved YOLOv8's detection performance.
• Reduced missed detection rates for small objects indicative of TB.
• Achieved high efficacy demonstrated through metrics like recall, precision, F1-score, and mAP.

Conclusions:

• The proposed SFF-enhanced YOLOv8 model offers a more accurate and efficient approach to TB detection.
• This research contributes a valuable tool for radiologists, potentially leading to earlier and more reliable TB diagnoses.
• The findings suggest a promising direction for AI-driven medical image analysis in combating tuberculosis.