MRI-based Brain Tumor detection using Deep Learning Techniques

Abstract

Automated brain tumor detection is vital for early identification of tumors, enabling timely medical intervention and improving overall tumor detection outcomes for patients. This study proposes a novel approach that utilizes the advanced DL technique YOLOv7 object detection framework, to achieve precise and real-time identification of brain tumors using MRI images. The manual review method is laborious and requires specialized knowledge to prevent human errors. Hence, the necessity for an automated brain tumor detection system arises to facilitate timely diagnosis of the disease. The YOLOv7 model underwent training using a dataset of 7023 MRI images that were pre-processed and labeled. An effective collection of characteristics for brain tumor identification was created by employing transfer learning and utilizing pre-trained weights from the MSCOCO dataset. The model achieved a mean average precision of 81.7% for glioma, 98.6% for meningioma, 98.1% for pituitary, and 98.6% for brain without a tumor. The results demonstrated a superior performance of the YOLO detection models compared to prior versions and other studies that employed bounding box detections. The mean average precision achieved was 93.14%, with a precision of 90.34%, recall of 88.58%, and F1-Score of 89.45%. Based on the results, it has been determined that the YOLOv7 model is capable of effectively and automatically detecting brain tumors at a fast pace by utilizing appropriate fine-tuning and transfer learning techniques. The primary purpose of the research is to assist healthcare practitioners in identifying brain tumors by utilizing imaging techniques.