Automated Glioma Grading and IDH Mutation Status Prediction Using CNN-Based Deep Learning Models

The objective of this work is to determine the isocitrate dehydrogenase (IDH) status and grade of gliomas using various deep learning models based on convolutional neural networks (CNN) using T1-weighted (T1), T2-weighted (T2), and fluid-attenuated inversion recovery (FLAIR) stacking images. The Cancer Genome Atlas (TCGA), consisting of 34 grade IV and 37 grade III scans, was used for the analysis. On the stacked image of the T1, T2, and FLAIR modalities, various CNN models, including VGG16, ResNet101, DenseNet121, EfficientNetV2S, and InceptionV3 models, were applied. The InceptionV3 CNN multi-task architecture achieved impressive results, achieving test accuracy of 99.14% for grade classification tasks and 99.71% for IDH classification. The study’s findings demonstrate that the multi-task InceptionV3 model may be employed in clinical workflows and was very accurate in predicting both the IDH status and the grade of gliomas. Gradient-weighted Class Activation Mapping (GradCAM) is used to generate gradient class activation maps, which aid in locating the image’s most discriminative area.