A LIGHTWEIGHT AND RESOURCE-ADAPTIVE DEEP LEARNING FRAMEWORK FOR ALZHEIMER’S DISEASE CLASSIFICATION ON LIMITED MRI DATASETS

Abstract

Alzheimer’s disease is a progressive neurological disorder that affects memory, cognitive abilities, and daily functioning. Early and accurate classification of the disease is important for supporting clinical assessment and timely medical intervention. However, developing reliable deep learning models for Alzheimer’s disease classification remains challenging because of the limited availability of labeled Magnetic Resonance Imaging (MRI) datasets and the high computational requirements of complex neural networks. This paper proposes “A Lightweight and Resource-Adaptive Deep Learning Framework for Alzheimer’s Disease Classification on Limited MRI Datasets” to address these challenges through an efficient and computationally economical classification approach. The proposed framework involves MRI image preprocessing, data augmentation, feature extraction, and lightweight deep learningbased classification to identify relevant patterns associated with different stages of Alzheimer’s disease. Resource-adaptive mechanisms are incorporated to optimize model complexity and computational requirements according to the available dataset and hardware resources. Transfer learning and regularization techniques can be employed to improve feature learning and reduce overfitting when training data are limited. The framework aims to classify MRI images into clinically relevant categories, such as cognitively normal, mild cognitive impairment, and Alzheimer’s disease, depending on the selected dataset and its available labels. Its performance can be evaluated using accuracy, precision, recall, F1-score, model size, inference time, and computational resource utilization. By balancing classification performance with computational efficiency, the proposed approach seeks to provide a practical solution for Alzheimer’s disease classification in resource-constrained environments. The framework may support research in computer-aided neurological assessment, although its predictions require clinical validation before use in real-world medical decision-making.