ESA-YOLOv5m: a lightweight spatial and improved attention-driven detection for brain tumor MRI analysis.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Early and accurate brain tumor detection is crucial for patient outcomes.
• Conventional deep learning models face challenges with small/low-contrast tumors and computational demands.
• There is a need for efficient and accurate automated brain tumor detection systems.

Purpose of the Study:

• To develop a lightweight and efficient framework for brain tumor detection in MRI scans.
• To enhance the performance of You Only Look Once v5 medium (YOLOv5m) using an Enhanced Spatial Attention (ESA) module.
• To improve tumor localization accuracy and model generalization for clinical applications.

Main Methods:

• An Enhanced Spatial Attention (ESA) module was integrated into the YOLOv5m architecture.
• The ESA module was strategically placed after the Spatial Pyramid Pooling-Fast (SPPF) layer.
• Experiments were conducted on the Figshare brain tumor MRI dataset (glioma, meningioma, pituitary).

Main Results:

• ESA-YOLOv5m achieved 90% Precision, 90% Recall, and 91% mAP@0.5, outperforming baseline YOLOv5m by 11-12%.
• Ablation studies confirmed optimal performance with ESA after the SPPF layer.
• The model demonstrated stable performance across cross-validation and classwise analyses with minimal computational overhead (<4.3% parameters, <10 ms latency).

Conclusions:

• Integrating a lightweight spatial attention mechanism significantly enhances brain tumor localization and model generalization.
• The ESA-YOLOv5m framework offers a reliable, scalable solution for automated brain tumor detection.
• The model is suitable for clinical decision-support systems and edge healthcare applications.