Multimodal Brain Tumor Classification using Capsule Convolution Neural Network with Differential Evolution Optimization Process

Abstract

Manual identification of brain tumors is error-prone and time-consuming for radiologists. Therefore, automation of the process is crucial. Although binary classification, such as distinguishing between malignant and benign tumors, is often straightforward, radiologists face significant challenges when classifying multimodal brain tumors. In this study, we present an automated approach that uses deep learning to classify brain tumor types using many types of data. The proposed method consists of three sequential phases. First, the median filter is used to eliminate any noise. For feature extraction in the second stage, linear contrast enhancement is used on VGG-16. The meningioma, glioma, and pituitary images are identified in the third stage of the brain tumor classification (BTC) process, which uses a modified capsule convolution neural network (CNN) design. The experimental results show that the brain tumor detection technique presented in this study successfully identifies the locations of tumor lesions. The results obtained were notably superior, with an accuracy of 98.34 %, a precision of 97.84 %, a recall of 05.34 %, and an F1-score of 94.56 %.