Automatic Detection of Brain Tumor on MRI Images Using a YOLO-Based Algorithm

This paper presents a study on the application of transfer learning and fine-tuning techniques to a deep learning model for the purpose of detecting three specific types of brain tumors from MRI images. The proposed approach utilizes the YOLO algorithm for automatic diagnosis. Specifically, the YOLOv4-tiny model, which is a smaller version of the YOLOv4 algorithm, was trained and evaluated due to its improved performance. The dataset utilized in this research is obtained from the figshare data repository, which comprises of labeled MRI images. The division of the dataset resulted in 80% for training, 10% for validation, and 10% for testing purposes. Additionally, a pre-processing technique was devised to enhance the features in the MRI images. The outcomes of the implementation demonstrate that the YOLOv4-tiny model obtained a mean average precision (mAP) of 0.8074 for the raw data and 0.8324 for the processed data.