BCE YOLOv8: a novel YOLO model for brain tumor instance segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Precision brain tumor segmentation is vital for intelligent healthcare.
• Existing methods face challenges with low segmentation precision and missed detections.

Purpose of the Study:

• To propose an intelligent brain tumor segmentation method to address precision and detection challenges.
• To enhance the accuracy and efficiency of brain tumor instance segmentation.

Main Methods:

• Developed BCE YOLOv8, integrating Bidirectional Feature Pyramid Network (Bi-FPN), Coordinate Attention (CA), and Efficient-IoU (EIoU) with YOLOv8.
• Incorporated CA into the C2f module for prioritized feature extraction.
• Utilized weighted Bi-FPN for sophisticated feature fusion and EIoU loss for accelerated convergence.

Main Results:

• BCE YOLOv8 achieved a precision of 67.8%, mAP@0.5 of 64.6%, and mAP@0.5:0.95 of 50.6%.
• Demonstrated significant improvements over YOLOv8 (21.07% precision increase) and YOLOv11 (16.49% precision increase).
• Effectively mitigated missed detections in brain tumor segmentation.

Conclusions:

• BCE YOLOv8 enhances brain tumor instance segmentation precision and reduces missed detections.
• The proposed method offers optimal performance and technical support for intelligent brain tumor detection.
• This advancement contributes to alleviating physician workload and improving healthcare outcomes.