Bi-directional YOLOv10 with average convolution for brain tumor detection in MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Medical professionals use MRI scans for brain tumor evaluation, facing challenges with fatigue and subjective interpretation.
• Automated detection systems are crucial for improving diagnostic accuracy and efficiency.

Purpose of the Study:

• To enhance the YOLOv10 object detection model for improved brain tumor detection accuracy.
• To introduce and evaluate the effectiveness of the AvConv module and Bidirectional Pathway.

Main Methods:

• Integration of AvConv module and Bidirectional Pathway into the YOLOv10 architecture.
• Experiments conducted on two Kaggle datasets (1116 and 153 images) annotated for brain tumors.
• Comparison of the enhanced model against eight mainstream object detection methods.

Main Results:

• The AvConv module improved YOLOv10's mAP50 by 0.3% (88.5%) and mAP50-95 by 3.1% (77.7%).
• The Bi-directional detection model achieved 90.3% mAP50 and 83.5% mAP50-95, outperforming the baseline YOLOv10.
• The proposed modules demonstrated significant gains in detection accuracy and maintained competitive inference efficiency.

Conclusions:

• The integrated AvConv module and Bidirectional Pathway effectively enhance YOLOv10 for brain tumor detection.
• Customized deep learning architectures show promise for reliable, automated brain tumor diagnosis.
• The study supports the development of advanced AI tools to aid medical professionals in interpreting MRI scans.