DEEP TRANSFER LEARNING–BASED FRAMEWORK FOR AUTOMATED BRAIN TUMOR DETECTION, CLASSIFICATION AND SEGMENTATION USING THE YOLOV8 ARCHITECTURE

The presence of complex and heterogeneous tumour characteristics is still a decisive problem in the diagnosis of brain tumours that makes it a significant issue in neuropathology. It is necessary to detect early and properly along with effective planning of treatment and better patient survival. This study gives a comprehensive transfer learning framework of automated detection, classification, and segmentation of brain tumours using the YOLOv8 architecture. The proposed system will be able to address the dynamic neuropathic features to incorporate the real-time object recognition and enhanced features extraction and multi-scale learning functions. The model was trained and evaluated using a hybrid dataset of primary MRI data that was gathered in the University of Nigeria Teaching Hospital (UNTH), Enugu, and secondary data that was sourced in the Kaggle repository. The data was a collection of 7,243 MRI images divided into four classes in the form of glioma, meningioma, pituitary tumour, and no tumour. The system has been built with the help of the Extreme Programming (XP) in order to deal with the refinements and flexibility in an iterative way. Instead, transfer learning using the pre-trained YOLOv8 weights was used and the model was hyperparameter tuned and improved feature fusion methods. It was shown that the suggested YOLOv8-based model reached a precision of 0.85, recall 0.75, and a mean Average Precision (mAP50) of 0.80, which is a good result in the context of brain tumour detection and classification. The software system was integrated and it was able to provide real-time tumour localization, segmentation and labelling that generates clinical decision-making. The results show the usefulness of deep transfer learning in automated brain tumour detection and emphasise the possibilities of the proposed system as a useful diagnostic aiding tool in a healthcare setting.