A deep learning approach for brain tumour classification and detection in MRI images using YOLOv7

Abstract

The medical imaging field has grown tremendously due to the latest digital imaging and artificial intelligence (AI) advancements. These advancements have improved tumour classification accuracy, time, cost efficiency, etc. Radiologists utilize an MRI scan due to its exceptional capacity to identify even the most minor alterations in brain activity. This research uses YOLOv7, a Deep Learning (DL) model, to classify and detect brain tumours and to conduct a detailed analysis of the frequently used structures for tumour identification. The study uses a brain MRI dataset from Roboflow with 2870 labelled pictures divided into four types of tumours. Our brain tumour dataset has four distinct classes: pituitary, gliomas, meningiomas, and no tumours. This preprocessed sample was used to assess the performance of deep learning models on identifying and classifying brain tumours. Throughout the preprocessing stage, aspect ratio normalization and resizing algorithms are applied to improve tumour localization for bounding box-based detection. YOLOv7 performs admirably, with a recall score of 0.813 and a box detection accuracy of 0.837. Remarkably, the mAP value for the 0.5 IoU threshold is 0.879. During box identification within the extended IoU spectrum of 0.5 for a to 0.95, the mAP value was 0.442.