Deep Learning and Optimization Approaches in Deep ConVGNet: Efficient Framework for Brain Tumour Classification with Masked-attention Mask Transformer based Segmentation: A Review

Abstract

Brain tumour classification and segmentation are critical challenges in medical imaging, requiring accurate and efficient automated systems to support clinical decision-making. Traditional diagnostic methods based on manual MRI interpretation are time-consuming and prone to variability, highlighting the need for advanced artificial intelligence solutions.  This paper presents a comprehensive review of the Deep ConVGNet framework integrated with a Masked-attention Mask Transformer for unified tumour classification and segmentation. The architecture builds upon convolutional neural networks with VGG-inspired depth, incorporating residual connections, depthwise separable convolutions, and multi-scale feature aggregation to effectively capture complex tumour characteristics across MRI modalities. The Masked-attention Mask Transformer enhances segmentation by focusing attention on relevant regions, improving boundary delineation while reducing computational complexity. This hybrid CNN-transformer design enables accurate pixel-wise segmentation alongside robust classification of tumour types such as glioma, meningioma, and pituitary tumours.The framework is evaluated on benchmark datasets including BraTS and Figshare, demonstrating strong performance across metrics such as Dice Similarity Coefficient and classification accuracy. Optimization strategies, including mixed-precision training, data augmentation, and adaptive learning rates, further enhance model robustness and efficiency.Overall, this review highlights the potential of hybrid deep learning architectures for developing scalable, accurate, and clinically applicable brain tumour analysis systems.