A Survey of Methods and Architectures for Deep ConVGNet: Efficient Framework for Brain Tumour Classification with Masked-attention Mask Transformer based Segmentation

Abstract

Brain tumour diagnosis is a critical and complex challenge in medical imaging, requiring precise classification and segmentation of MRI data to support effective clinical decision-making. Traditional methods based on manual analysis and handcrafted features are often limited in accuracy and scalability. Recent advances in deep learning, particularly convolutional neural networks and transformer-based models, have significantly improved automated brain tumour analysis. This survey presents a comprehensive review of the Deep ConVGNet framework, a hybrid architecture that integrates VGG-inspired convolutional networks with a Masked-attention Mask Transformer for unified tumour classification and segmentation. The convolutional backbone captures hierarchical spatial features, while the transformer-based segmentation module enhances localization by focusing attention on relevant regions, improving boundary delineation for complex tumour structures. The framework is evaluated across benchmark datasets such as BraTS, Figshare, and TCGA-GBM, demonstrating superior performance in terms of classification accuracy, Dice Similarity Coefficient, and Intersection over Union compared to conventional CNN and transformer models. Multi-task learning further enhances generalization by combining classification and segmentation objectives within a single pipeline. Despite strong performance, challenges remain in model interpretability, dataset variability, and clinical deployment.This review highlights the effectiveness of hybrid deep learning architectures and outlines future directions for developing scalable, accurate, and clinically applicable brain tumour diagnostic systems.