Multi-classification of Brain Tumor MRI Images Using Deep CNN Features and ML Classifiers

Abstract

Abnormal accumulation of unusual cells in brain parts may lead to brain tumor that may affect normal brain functionalities. The early diagnosis of brain tumors will help health professionals greatly in treating tumors in the beginning stage itself. Also, the patient’s risk of death will be widely reduced. Brain tumor (BT)-phase categorization plays an important job in tumor determination and its operative treatment. Manual or semi-manual evaluation of Magnetic Resonance Imaging (MRI) images for tumor diagnosis involve extensive computations. In this study, the performance with a range of ML & DL technologies for BT-phase categorization is evaluated by Deep Convolutional Neural Network (DCNN) features and ML classifiers. In our work, the ten different pre-trained DCNN feature models are compared with five varied ML classifiers on tumor datasets for evaluating the accuracy of the system. In that, the pre-trained CNN system with SVM (RBF) kernel performance is the best on the tumor dataset and also the deep features from the DenseNet architectures perform better than the other DCNN features.