A lightweight attention-driven YOLOv5m model for improved brain tumor detection

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Oncology
• Neurosurgery and Neurology

Background:

• Brain tumors are aggressive diseases significantly impacting patient survival.
• Current MRI-based diagnosis relies on expert interpretation, facing challenges like variability and human error.
• Automated diagnostic tools are crucial for advancing brain tumor detection accuracy.

Purpose of the Study:

• To enhance the YOLOv5m model for improved brain tumor detection using magnetic resonance imaging (MRI).
• To integrate an Enhanced Spatial Attention (ESA) layer into the YOLOv5m architecture for precise feature analysis.
• To improve the model's ability to differentiate between common brain tumor types: meningioma, pituitary, and glioma.

Main Methods:

• Modification of the YOLOv5m object detection model by incorporating an Enhanced Spatial Attention (ESA) layer.
• Training and validation of the enhanced model on a dataset of 3064 T1-weighted contrast-enhanced MRI images from 233 patients.
• Comparative analysis of the modified YOLOv5m's performance against the standard YOLOv5m model.

Main Results:

• The enhanced YOLOv5m model demonstrated superior performance metrics compared to the standard YOLOv5m.
• The ESA layer improved the model's focus on salient features, enhancing differentiation of tumor types.
• The modified model showed increased detection reliability and minimized false positives in brain tumor classification.

Conclusions:

• The ESA-enhanced YOLOv5m model offers a robust and precise tool for automated brain tumor diagnosis from MRI scans.
• This advancement has significant potential for clinical applications, improving diagnostic efficiency and accuracy.
• The study highlights the efficacy of attention mechanisms in deep learning for medical image analysis.