Research on improved YOLOv8s model for detecting mycobacterium tuberculosis

Area of Science:

• Medical Imaging
• Computer Vision
• Microbiology

Background:

• Accurate identification of Mycobacterium tuberculosis (M. tuberculosis) is crucial for tuberculosis diagnosis.
• Current object detection methods face challenges with the diverse morphology and size of M. tuberculosis in sputum smear images, hindering precise localization.

Purpose of the Study:

• To develop an improved object detection model for accurate M. tuberculosis identification in sputum smear images.
• To address the limitations of existing methods in detecting M. tuberculosis with varied appearances and sizes.

Main Methods:

• An enhanced YOLOv8s model was proposed, incorporating an additional detection head for small targets.
• A multi-scale feature fusion module was introduced to handle variations in M. tuberculosis size.
• Modifications to the Coordinate Attention (CA) module, including a convolutional layer and a self-attention mechanism, were implemented to improve feature extraction and localization.

Main Results:

• The improved YOLOv8s model demonstrated strong performance on a public dataset.
• The model achieved an average precision of 85.7% for M. tuberculosis detection.
• The enhancements effectively improved the model's ability to identify M. tuberculosis with varied morphology and size.

Conclusions:

• The proposed enhanced YOLOv8s model significantly improves the accuracy of M. tuberculosis detection in sputum smear images.
• The model's effectiveness in handling morphological and size variations makes it a valuable tool for tuberculosis diagnosis.
• This work highlights the potential of advanced deep learning techniques in medical image analysis for infectious disease identification.