Optimizing Hyperparameters of YOLO to Improve Performance of Brain Tumor Detection in MRI Images

Brain tumor is one of the disorders of the central nervous system (CNS) caused by the growth of abnormal tissue in the brain. Magnetic Resonance Imaging (MRI) is the most popular electronic modality used by doctors to diagnose brain tumors. Early detection of brain tumors based on MRI images can help doctors provide the right treatment, thus increasing the patient's chances of recovery. Recently, deep learning algorithms, especially CNN, have been widely used in the medical field and show good performance for medical images analysis. In this study, we propose the detection of brain tumor lesions on MRI images using YOLOv7. This study uses a 2D MRI images dataset extracted from BraTS2020 in the axial plane of the T1CE sequence. To improve model performance, we try to find the best set of hyperparameters using three methods, including random search (RS), genetic algorithm (GA), and Bayesian optimization (BO). The results show that Bayesian optimization is the most efficient method for finding the optimal hyperparameter combination, where BO is 1.5 times faster than RS and 4 times faster than GA. By using the best hyperparameter obtained, the performance of YOLOv7 improved by 8% compared to the original model, achieving the best performance at 0.916 mAP. These results also outperform previous research using the same dataset. The research results indicate that hyperparameter optimization can enhance the model's performance.