Binary and Multi-class Classification of Brain Tumors using MRI Images

Abstract

A dangerous and potentially fatal condition is a brain tumor. Early detection of this disease is critical for determining the best course of treatment. Tumor detection and classification by human inspection is a time consuming, error-prone task involving huge amounts of data. Computer-assisted machine learning and image analysis techniques have achieved significant results in image processing. In this study, we use supervised and deep learning classifiers to detect and classify tumors using the MRI images from the BRATS 2020 dataset. At the outset, the proposed system classifies images as healthy or normal brains and brain having tumorous growth. We employ four supervised machine learning classifiers SVM, Decision tree, Naïve Bayes and Linear Regression, for the binary classification. Highest accuracy (96%) was achieved with SVM and DT, with SVM giving a better Recall rate of 98%. Thereafter, categorization of the tumor as Pituitary adenoma, Meningioma, or Glioma, is performed using supervised (SVM, DT) classifiers and a 6-layer Convolution Neural Network. CNN performs better than the other classifiers, with a 93% accuracy and 92% recall rate. The suggested system is employable as a powerful decision-support tool to assist radiologists and oncologists in clinical diagnosis without requiring invasive procedures like a biopsy.