Abstract
Identifying brain tumors accurately through medical imaging is vital in supporting computer-aided diagnostic systems, playing an essential role in early disease identification and effective treatment planning. Manual analysis of medical scans, like MRI scans, often slow and susceptible to human error, emphasizing the growing demand for automated, efficient, and precise detection systems. In the proposed study, we present an enhanced approach to fine-tuning an object detection model for accurately identifying brain tumors, demonstrating the capabilities of deep learning techniques in medical imaging analysis. The proposed method leverages the YOLOv10 architecture, a state-of-the-art model recognized for its high detection speed and precision. Due to the limited availability of extensive labeled medical imaging datasets, a transfer learning approach is adopted by initializing the model with parameters trained on the COCO dataset. These parameters are then fine-tuned using a brain tumor-specific dataset to significantly enhance the model's detection performance. The fine-tuned model gains a mean Average Precision (mAP) of 96.1% and a precision of 96.8%, surpassing the baseline performance of the original YOLOv10 model. These results highlight the efficacy of applying transfer learning techniques to medical imaging problems, particularly when dealing with scarce data resources. Furthermore, our approach demonstrates how modern object detection architectures can be efficiently adapted for specialized clinical tasks, offering promising pathways for future advancements in computer-aided diagnosis and healthcare applications.