Multiclass Light Weight Brain Tumor Classification and Detection using Machine Learning Model Yolo 5

Abstract

Early brain tumor identification is a critical challenge for neurologists and radiologists. Manually identifying brain tumors through magnetic resonance imaging (MRI) is difficult and prone to mistakes. The diagnosis of tumor is a complex job when performed in a traditional manner. Brain abnormalities can be fatal, lowering a patient's quality of life and adversely harming their overall health. Brain tumors vary in nature based on where they are situated and how rapidly they develop inside the skull. Tumors are a proliferation of abnormal nerve cells that form a mass. Some brain tumors begin in the cells that support the brain's nerve cells. This paper proposes a machine learning algorithm known as YOLO v5 SSD (single shot detection) to detect and classify such tumors namely meningioma, glioma, and pituitary gland with 88% accuracy. For this purpose, data augmentation was applied to the publically available dataset from Kaggle. MRI of different classes including 396 glioma images, 397 meningioma, 380 no tumor, and 399 images of pituitary tumors were employed. The current study presents false negative, true positive false positive, and true negative, which were used to test the YOLO v5 (You Only Look Once) classifier performance. It was determined that the YOLO v5 model is giving 88% accuracy.