Brain Tumor Diagnosis and Classification based on AutoML and Traditional Analysis

Abstract

Analysis of brain tumors is a challenging task. Due to the complexities of the brain structure, the analysis of a brain MRI (Magnetic Resonance Imaging) requires sophisticated tools and vast knowledge in the field of neuroscience. Detecting the presence of Malignant brain tumors can be done by using ML (Machine Learning) techniques. Moreover, recent changes in automation with the advent of tools like AutoML (Automated Machine Learning) have created significant room for research. When training ML models, there are several hyperparameters such as the number of clusters in the case of KNN, or the number of hidden layers in the case of Artificial Neural Networks. Different methods and models are implemented on the brain MRI scan dataset. The images were preprocessed and ran on all the models discussed in the paper individually. The validation accuracies are then compared with the performance results obtained by using AutoML. The comparison of the obtained validation accuracies assists in discerning the most optimal segmentation method along with the corresponding preprocessing technique used. The best model can then be used for detecting the presence of brain tumors in any given MRI scan.