Towards Real Time Alzheimer's Diagnosis: A PSO‐GA‐Driven Deep Learning Solution for Telemedicine

• Published in: International journal of imaging systems and technology Vol. 35; no. 5
• Main Authors: Kumar, Anupam, Ahmad, Faiyaz, Alam, Bashir
• Format: Journal Article
• Language: English
• Published: Hoboken, USA John Wiley & Sons, Inc 01.09.2025; Wiley Subscription Services, Inc
• Subjects: Alzheimer's disease; Classification; computer‐aided diagnosis (CAD); Deep learning; Diagnosis; Feature extraction; Genetic algorithms; Machine learning; medical image analysis; Medical imaging; nature inspired algorithms; neurodegenerative disease detection; Particle swarm optimization; scalable medical AI; Telemedicine
• ISSN: 0899-9457; 1098-1098
• DOI: 10.1002/ima.70180

ABSTRACT Alzheimer's disease (AD) is a progressive neurodegenerative disorder characterized by cognitive decline and structural brain deterioration, with its global prevalence projected to exceed 125 million by 2050. Early and accurate diagnosis—particularly the differentiation of mild cognitive impairment (MCI) from normal aging—is critical for effective intervention; yet it remains challenging due to subtle anatomical changes and high‐dimensional imaging data. This study presents a telehealth‐compatible computer‐aided diagnosis (CAD) framework for multi‐class AD classification using structural MRI (sMRI) images from the publicly available ADNI dataset. The framework integrates transfer learning with DenseNet121 (pre‐trained on RadImageNet) for deep feature extraction and employs a hybrid bio‐inspired particle swarm optimization–genetic algorithm (PSO‐GA) for feature selection and dimensionality reduction. This optimized pipeline reduces the original high‐dimensional feature space to 16 key features, improving classification accuracy from 88.48% to 99.78% using AdaBoost. The proposed PSO‐GA‐DenseNet framework delivers a lightweight, scalable solution suitable for remote diagnostic settings. Compared to existing state‐of‐the‐art models, it offers enhanced computational efficiency and robust cross‐site adaptability. Future research will focus on improving generalizability across imaging modalities and incorporating longitudinal data to enable real‐time, cross‐modal, and large‐scale deployment in clinical and telehealth environments.