Optimizing YOLOv8 for Enhanced Medical Image Analysis: A Deep Learning Approach to Brain Tumor Detection

This study presents a comprehensive optimization of the YOLOv8 architecture for enhanced brain tumor detection in magnetic resonance imaging (MRI) data, aiming to improve both accuracy and efficiency in medical image analysis. To ensure compatibility with input data characteristics, the model's input layer was resized to 256×256 pixels prior to training and testing. The dataset, comprising approximately 7,000 MRI scans, was partitioned into training (80%) and testing (20%) subsets to ensure robust model evaluation. The YOLOv8 model from the Ultralytics framework was fine-tuned with a training schedule of 20 epochs. Due to high memory demands, the batch size was constrained to 2 to prevent out-of-memory errors during training. The Adam optimizer was employed with a default learning rate of 0.001, while the dropout rate was maintained at 0.0, as batch normalization effectively mitigates overfitting in YOLO's architecture, rendering dropout unnecessary. Data augmentation and preprocessing techniques-including grayscale conversion, Gaussian blurring, thresholding, erosion, dilation, and contour-based cropping-were applied to enhance feature visibility and model generalization. Experimental results demonstrate exceptional performance, with an overall accuracy of 99.31%, precision of 98.4%, recall of 98.8%, and an F1-score of 98.6%. Notably, the model achieved near-perfect precision (99.33%) and recall (99.33%) for pituitary tumors, alongside a minimal average inference time of 0.0061 seconds per image, highlighting its suitability for real-time clinical diagnostics. These findings affirm that optimized YOLOv8 is a highly effective, fast, and reliable solution for intelligent brain tumor detection, offering significant potential for integration into computer-aided diagnostic systems.