Deep Learning for Brain Tumor Detection Leveraging YOLOv10 for Precise Localization

Brain tumors are some of the most dangerous and life-threatening diseases that require early detection to increase survival and prognosis. Detecting patients of brain tumors with great spatial resolution and time efficiency remains a challenge due to the variation in the size, shape, and location of tumors. The issues mentioned above are addressed in the present work using YOLOv10 and four of its variants, includes YOLOv10-S, YOLOv10-M, YOLOv10-N, and YOLOv10-B due to their improved real-time detection performances. The dataset comprises 5249 MRI images obtained from Kaggle, including four distinct tumor categories such as glioma, pituitary, meningioma, and no tumor. The models are evaluated with precision, recall, and mean average precision mAP@50. The variations YOLOv10-S, YOLOv10-M, YOLOv10-N, and YOLOv10-B achieved median average precision (mAP50) values of 0.964, 0.949, 0.951, and 0.955, respectively. The findings presented here suggest that YOLOv10-S outperforms the other versions studied in this work in identifying brain tumors, thus demonstrating its potential to be used in more accurate and faster diagnostic tools in the healthcare domain.