A Survey of Methods and Architectures for Multi-classification of Brain Tumour MRI Images Using Deep Dynamic Parallel Convolutional Neural Network with Fully Termite Alate Optimization Algorithm

Abstract

Brain tumour classification using Magnetic Resonance Imaging (MRI) is a critical task for early diagnosis and treatment planning. Traditional diagnostic methods rely heavily on manual interpretation, which is time-consuming and prone to errors. Recent advancements in deep learning, particularly Convolutional Neural Networks (CNNs), have significantly improved the accuracy of automated brain tumour classification systems. Multi-class classification approaches can effectively distinguish tumour types such as glioma, meningioma, and pituitary tumours, enhancing clinical decision-making.  Modern architectures such as deep dynamic parallel convolutional neural networks (DDPCNN) leverage multiple parallel pathways to capture both local and global features from MRI images, improving classification performance. Additionally, optimization algorithms play a crucial role in enhancing deep learning models by tuning hyperparameters, improving convergence, and avoiding local minima. Metaheuristic approaches, including swarm intelligence and bio-inspired algorithms, have been widely adopted for this purpose. This review presents a comprehensive analysis of recent deep learning and optimization approaches for brain tumour MRI classification. It focuses on hybrid frameworks combining CNN architectures with optimization techniques such as termite alate optimization. The study highlights key advancements, compares methodologies, and identifies challenges such as computational complexity and data imbalance, providing future research directions for intelligent diagnostic systems.