Performance Analysis of Brain Tumor Detection Based on Gradient Boosting Machine -CNN model

Abstract

Tumors of the brain are one of the most prevalent and severe diseases, due to high mortality rate. Thus, The categorization of brain tumors is crucial for accurate clinical diagnosis and treatment. The aim of this research is to develop a prototype system for detecting brain tumors using the gradient boost algorithm and deep learning model in order to analysis the efficiency. In our proposed method was tested on a data set of 3000 MRI images, and we divided them into (80%)for training and (20%)testing, and then based on the idea of transfer learning and the use of deep convolutional neural networks that have been pre-trained to extract deep features from brain MRI.For the classification stage, At first we applied the concept of deep learning, that is, we used CNN as a classifier and as a feature extractor, and then use same the features with gradient boosting (Gradient Boost, XGBoost, LightGBM, CatBoost) algorithms as a classifier. And for the purpose of analyzing the performance of these algorithms, we have implemented other machine learning algorithms (LogisticRegression, Support Vector Machines, Neural network, Adaptive Boosting, Random Forest, Decision Tree, K-Nearest Neighbors) and finally we have compared all the classifiers.the results showed that most of the proposed detection models, whether independent (CNN) or composite models , have high performance in detecting brain tumor and this is one of our goals in this study. Our main goal is to analyze the performance of Gradient Boosting algorithms. Whereas, the Gradient Boosting algorithm was the least effective in classifying with an accuracy of 0.952 and a time of (17m and 3s), followed by LightGBM with an accuracy of 0.979 and a time of (8m and 51s), and CatBoost with an accuracy of 0.974 and a time of(20m and40s), while the XGBoost algorithm had the best performance in this category, with an accuracy of 0.981 but with a larger time of (23m and 3s).