A successive framework for brain tumor interpretation using Yolo variants.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Manual analysis of brain tumors in MRI is time-consuming and prone to variability.
• Existing automated methods like R-CNN and earlier YOLO versions have high computational costs and limited segmentation accuracy.

Purpose of the Study:

• To evaluate and compare the performance of YOLOv9, YOLOv10, and YOLOv11 for brain tumor detection and segmentation.
• To identify the most effective YOLO variant for accurate and efficient analysis of brain tumor MRI datasets.

Main Methods:

• Utilized the Figshare Brain Tumor and BraTS2020 datasets for training and testing.
• Applied preprocessing techniques including log transformation, histogram equalization, and edge-based ROI extraction.
• Trained YOLOv9, YOLOv10, and YOLOv11 models on 80% of the combined dataset and evaluated on the remaining 20%.

Main Results:

• YOLOv11 achieved superior performance, with classification accuracies of 96.22% (BraTS2020) and 96.41% (Figshare).
• YOLOv11 demonstrated excellent segmentation metrics: F1-score (0.990), recall (0.984), mAP@0.5 (0.993), and mAP@[0.5:0.95] (0.801).
• Achieved a rapid inference time of 5.3 ms with a balanced precision-recall curve.

Conclusions:

• YOLOv11 is a highly effective and efficient model for brain tumor detection and segmentation in MRI.
• The proposed framework offers a robust, real-time solution suitable for clinical deployment.
• YOLOv11 addresses the limitations of previous methods, improving diagnostic accuracy and speed.