A comparative analysis and classification of cancerous brain tumors detection based on classical machine learning and deep transfer learning models

Brain tumor can be fatal for human life. Therefore, proper and timely diagnosis and treatment is important to save human lives. The similarity and variety between the normal and tumor tissues make it difficult for diagnosis through human assisted techniques. In the recent past, machine learning and deep learning techniques have been applied for the classification and segmentation of brain tumor. These techniques have shown promising results by improving the accuracy of classification and segmentation. In this paper, we proposed to implement various classical machine-learning techniques support vector machine, Naive Bayes classifier, K- Nearest Neighbor, random forest, and deep-learning CNN-based models Xception, Inceptionv3, VGG19, and DenseNet201 techniques to classify gliomas, meningiomas, and pituitary tumors and compare their performance. Further, we proposed to modify Xception model to improve the performance of classification and segmentation. In this research, we used the standard Figshare dataset consisting of 3064 images of size 112x112\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$112\times 112$$\end{document} each. The performance of these models is measured and compared in terms of precision, recall, F1-score, and accuracy. The classical Machine learning model gives scores varying from 88% to 93%. For the above four metrics. However, all the deep learning models give their scores of more than 96% for the above metrics. Our proposed modified Xception model gives scores more than 98% in all the above four metrics, which is comparable to the best scores reported in the literature.