BRAIN TUMOR IMAGE (BTI) DETECTION AND CLASSIFICATION USING AN ENHANCED YOU ONLY LOOK ONCE VERSION 8 (YOLOV8) ARCHITECTURE

Brain tumours have been known to be detected early and accurately as a constituent of effective clinical intervention and better patient survival. Nevertheless, the current diagnostic systems based on medical imaging have difficulties with the detection of heterogeneity of tumours, tumour size, dynamic growth, and the lack of labelled clinical data. This paper suggests a better deep transfer learning model to detect and classify brain tumour images (BTI) with a modified You Only Look Once Version 8 (YOLOv8) model. The proposed model combines Bi-Directional Feature Pyramid Network (Bi-FPN) to build a better multi-scale fusion of features, a Proposed Feature Optimizer (PFO) to build a better discriminative feature representation and an extra P2 detection layer to build a better sensitivity to small and early-stage tumours. A mixed dataset of clinical MRI data of the University of Nigeria Teaching Hospital (UNTH), Enugu, and a publicly available Kaggle brain MRI dataset of 7,243 images of four classes ( glioma, meningioma, pituitary tumour, and no tumour ) were used to train and evaluate the system. Pre-trained weights were used in transfer learning to overcome the problem of data scarcity and enhance generalisation. Performance on the model was evaluated based on standard evaluation measures such as precision, recall, mean Average Precision (mAP) as well as loss measures. Using experimental outcomes, it is proven that the enhanced YOLOv8 model is much better in comparison with the initial YOLOv8 framework, having a precision of 0.98, recall of 0.95, mAP50 of 0.98, and mAP5095 of 0.54, as well as lower localization and classification losses. These results show that there is improved robustness, accuracy, and reliability of brain tumour detection and classification. The suggested system is practical and scalable in early brain tumour screening and has high chances of being incorporated into the real clinical diagnostic systems.