Application research on YOLOv5 model based on Lightweight Atrous Attention Module in brain tumor MRI image...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate segmentation of brain tumors in MRI is crucial for diagnosis and treatment planning.
• Existing methods often face challenges in balancing segmentation accuracy with computational efficiency.
• Deep learning models, particularly convolutional neural networks, have shown promise but require optimization for clinical applications.

Purpose of the Study:

• To develop a novel Lightweight Atrous Attention Module (LAAM) for enhancing brain tumor segmentation in MRI.
• To integrate LAAM into the YOLOv5s model to improve accuracy and computational efficiency.
• To evaluate the performance of the enhanced YOLOv5s-LAAM model against existing architectures.

Main Methods:

• The study utilized two public MRI datasets (meningioma and glioma).
• A Lightweight Atrous Attention Module (LAAM) was designed, incorporating depthwise separable convolutions, dual attention, and residual connections.
• The LAAM was integrated into the YOLOv5s architecture, and the model was trained and validated using five-fold cross-validation.

Main Results:

• The YOLOv5s-LAAM model achieved high performance metrics: 92.3% precision, 90.4% recall, and an mAP@50 score of 0.925.
• Computational efficiency was improved, with a 15% reduction in GFLOPs compared to the YOLOv5s-ASPP baseline.
• The enhanced model demonstrated superior segmentation accuracy and maintained computational efficiency.

Conclusions:

• The LAAM significantly improves the YOLOv5s model's performance for brain tumor MRI segmentation.
• The lightweight design makes the model suitable for resource-constrained environments.
• This enhanced model offers a valuable tool for clinical diagnosis and treatment planning in neuro-oncology.