Research on MRI Object Detection Using an Enhanced YOLOv8 Framework

Abstract

Early diagnosis and accurate localization of brain tumors are crucial for improving patient survival rates. Among these, automated detection methods based on deep learning have become a research hotspot. However, tumor detection, especially for complex shapes and small targets, remains challenging. To address this, this study proposes an improved YOLO model—YOLO-CDF—aimed at enhancing the detection accuracy for complex and small targets in brain tumor MRI images. The model builds upon YOLO by incorporating the BRASPPF module (A combination of bi-level routing attention mechanism and spatial pyramid pooling), dilated convolution, and small object detection layers. Experimental results show that the YOLOv8-CDF model achieves a good balance between precision and recall, with an overall [email protected] of 0.929 and an F1 score reaching 0.90, demonstrating excellent detection performance. When detecting tumors, the model's precision values are 0.974, 0.964, and 0.851, respectively. Validation results show that the model can provide accurate predictions at both high and low confidence levels, with strong detection capabilities and good generalization ability, making a significant contribution to the identification of brain tumors.