An Ensembling Approach using DL & Non-DL Techniques for Detecting Brain Tumors using MRI Scans

Abstract

A brain tumor is an abnormal growth of brain cells and tissues, which may further lead to life-threatening diseases, one of them being cancer. Doctors use Magnetic Resonance Imaging (MRI) scans to identify such tumors. However, the primary concern of detecting and extracting these infectious tumors using MRI is highly tedious and complex, hindering the accuracy of trained clinicians and radiologists. Thus, to address these challenges and categorize whether a tumor is present inside the brain, researchers have begun employing various Machine Learning (ML) and Deep Learning (DL) algorithms. Apart from these techniques, other non-DL algorithms have also been used to solve the problem. Thus, in this paper, the proposed scheme is based on utilizing various non-DL features - statistical, image-based, and the novel technique of Topological Data Analysis (TDA) and ensembling them with DL-based techniques. These non-DL features are initially embedded in the original MRI scan and passed on through a CNN-based classifier. The proposed model is able to achieve higher accuracies, precision, and recall when compared to the existing VGGNet16 and ResNet50 models.