Enhanced Brain Tumor Grade Classification with ConvNext: Performance Analysis

Abstract

The grade of a brain tumor is a crucial component of its diagnosis and aids in treatment planning. Biopsies and manual review of medical images are examples of traditional diagnostic techniques that are either invasive or may produce incorrect results. Using a contemporary convolutional neural network (CNN) architecture named ConvNext that receives magnetic resonance imaging (MRI) data, this study suggests a metho d for classifying brain tumor grades. In order to diagnose brain cancers consistently, accurately, and non-invasively, deep learning based metho ds are taking the place of invasive treatments. Data scarcity is a well-known issue with applying deep learning architectures to medical imaging. In order to attain the required accuracy and prevent overfitting, modern architectures need enormous datasets and contain enormous trainable parameters. As a result, researchers that use medical imaging data frequently use transfer learning. CNNs have recently lost ground to transformer-based designs for picture data. However, by implementing specific modifications influenced by vision transformers, freshly proposed CNNs have attained greater accuracy. A method for extracting features from the ConvNext architecture and feeding them into a fully connected neural network for final classification was presented in this article. Using pre-trained ConvNext, the proposed study obtained state-of-the-art performance on the BraTS 2019 dataset. When three MRI sequences were fed into the pre-trained CNN as three channels, the highest accuracy of 99.5% was attained.