Computer Aided Detection of Brain Tumors using Convolutional Neural Network based Analysis of MRI Data

Abstract

Abstract A significant topic of investigation in the area of medical imaging is brain tumor classification. The manual examination of medical imaging may result in imprecise findings and is also a time-consuming and laborious task. With the emergence of artificial intelligence, the research community has started providing automated solutions for smart detection of different types of brain tumors. So, the present paper also presents a computer-aided diagnostic technique, which makes use of the proposed architecture of a 16-layer convolutional neural network (CNN) model for accurate detection of different brain tumor types through the analysis of MR images. In this approach, the MR images are resized and normalized first. Then, a hybrid method of synthetic minority oversampling technique and edited nearest neighbour has been employed to provide a balanced dataset. Afterwards, these images are fed to the proposed CNN model for classification. In this work, a CNN-based feature extractor has also been used in association with machine learning-based classification, which includes random forest, kNN, support vector machine, naïve Bayes and decision tree algorithms. The thorough analysis of the proposed CNN model and the hybrid models of CNN and machine learning algorithms demonstrate that the proposed CNN model provides an accurate classification of different brain tumor types with maximum accuracies of 98.88% and 97.83% for binary classification of tumor detection and three class classification of meningioma, glioma, pituitary tumor types using two different datasets. From this analysis, it is evident that the proposed 16-layer CNN model appears to be an efficient method for accurate detection of brain tumors as well as identification of different types of tumors.