ADAPTIVE NEURO-EVOLUTIONARY TRANSFORMER FRAMEWORK FOR EFFICIENT MRI-BASED EARLY DETECTION AND MULTI-STAGE CLASSIFICATION OF ALZHEIMER’S DISEASE
: https://doi.org/10.65725/JCISE/2/3/003
JOURNAL OF COMPUTATIONAL INTELLIGENCE SCIENCE AND ENGINEERING (JCISE) 
ISSN: 3107-8168
Volume 2 Issue 3, Jul-Sep 2026

Abstract: 

Alzheimer’s disease is one of the most prevalent neurodegenerative disorders that is affecting cognitive memory, and daily functioning. Recent studies have integrated deep learning models with the meta-heuristic optimization algorithms for the improving classification accuracy. However, many of the conventional approaches have relied on complex hybrid optimization strategies that is increasing the computational cost. In addition, where various frameworks have focused on isolated brain regions rather than capturing comprehensive structural relationships across the MRI data. The current Alzheimer’s disease diagnostic models often is experiencing limitations that is that includes a higher computational complexity, dependency on large multimodal datasets, and insufficient combination between the feature learning and optimization strategies. Some segmentation-driven frameworks also struggle in maintaining the consistent classification performance when the MRI images is containing the noise, structural variability, or class imbalance. These challenges are showing the need for the efficient learning framework that is combining the adaptive feature extraction, optimized learning capability, and reduced computational overhead for the accurate multi-stage Alzheimer’s disease diagnosis. This study is proposing an Adaptive Neuro-Evolutionary Transformer Optimization (ANETO) framework for the MRI-based Alzheimer’s disease detection and classification. The proposed framework initially is performing adaptive contrast normalization and noise suppression for the improving MRI image quality. A dynamic region clustering module segments the significant brain tissues such as gray matter, white matter, and hippocampus. After segmentation, a transformer-based feature encoder is extracting the spatial relationships from the MRI slices. The neuro-evolutionary optimization mechanism then allows the adjustment of transformer attention weights and feature importance through the adaptive mutation and crossover operations that tend to maximize classification fitness. The optimized feature representation is feeding a multi-stage classifier that categorizes MRI scans into Cognitive Normal (CN), Mild Cognitive Impairment (MCI), and Alzheimer’s Disease (AD) classes. The experimental evaluation is showing that the proposed framework is achieving a superior classification performance compared with the several conventional hybrid deep learning models. The model is achieving a 98.4% accuracy, 97.9% precision, 97.4% recall, and 97.6% F1-score, while achieving an AUC value of 0.999 for the tumor classification.

Authors: Dr K S Thirunavukkarasu, J.Johnsirani
Keywords: Alzheimer’s disease diagnosis, MRI image analysis, neuro-evolutionary optimization, transformer deep learning, medical image classification.