YOLOv8 and SAM-Based CNN Feature Fusion Framework for Brain Tumor Detection in MR Image

Brain tumor identification using Magnetic Resonance Imaging (MRI) is essential for diagnosis. We present an optimized hybrid deep learning framework to detect and classify brain tumors using MRI with a Convolutional Neural Network (CNN), You Only Look Once version 8 (YOLOv8), and a Segment Anything Model (SAM). We used ResNet50 as a backbone with multi-scale feature extractions and a Feature Pyramid Network (FPN) to improve contextual learning. YOLOv8 was used for tumor region localization and classification, while SAM improved object boundaries for segmentation. The framework was trained on the Kaggle dataset of brain MRI with 1,956 images and evaluated using standard metrics (Precision, Recall, Accuracy, and Mean Average Precision (mAP)). The results showed that CNN-YOLOv8-SAM achieved an mAP of 96.5%, surpassing previous YOLOv5 and YOLOv7 studies, demonstrating potential for automated tumor identification.