Application of MRI image segmentation algorithm for brain tumors based on improved YOLO.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Accurate segmentation of brain tumors in magnetic resonance imaging (MRI) is crucial for diagnosis and treatment planning.
• Deep learning models offer potential for automating and improving the speed and accuracy of this process.

Purpose of the Study:

• To investigate the feasibility of the YOLOv5s deep learning algorithm for brain tumor segmentation in MRI.
• To optimize and enhance the YOLOv5s model for improved segmentation detection and clinical identification of brain tumor types.

Main Methods:

• Utilized two public Kaggle datasets (meningioma and glioma) comprising 3,439 MRI images.
• Annotated 3,000 images from Dataset 1 for training/validation (7:3 ratio); used remaining images and Dataset 2 as test sets.
• Optimized the YOLOv5s model by incorporating Atrous Spatial Pyramid Pooling (ASPP), Convolutional Block Attention Module (CBAM), and Coordinate Attention (CA), creating versions like YOLOv5s-ASPP.
• Trained the original YOLOv5s, five optimized variants, and YOLOv8s models for 100 epochs, evaluating performance on test sets.

Main Results:

• All seven models successfully segmented and recognized brain tumor MRI images.
• The YOLOv5s-ASPP model achieved a precision of 93.5% and a recall of 85.3% on the validation set.
• The optimized YOLOv5s-ASPP model significantly outperformed the original YOLOv5s model in image segmentation ability on the test set.

Conclusions:

• The YOLOv5s-ASPP model demonstrates significantly enhanced brain tumor MRI segmentation capabilities compared to the original YOLOv5s.
• This improved model aids in assisting clinical diagnosis and treatment planning for brain tumors.