O-26 Multimodal AI for early Alzheimer’s detection: accuracy and optimal combinations

Methods: A systematic review of 30 studies published (2015–2025) from PubMed, Scopus, and ResearchGate was conducted, using PRISMA guidelines. This indicated the use of artificial intelligence (AI) for the early detection of Alzheimer’s disease, especially the mild cognitive impairment (MCI) stage. The assessed modalities (MRI, PET, retinal, and clinical biomarkers) facilitated insights into Alzheimer’s progression, and AI models (CNNs, fusion networks, and deep learning systems) were assessed for their ability to process complex datasets and enhance predictive modeling. Outcomes (AUC, sensitivity) were reviewed to ensure consistency for diagnostic accuracy. Merits included improved sensitivity, multimodal integration, and non-invasive potential, while limitations involved dataset variability, generalizability, and computational demands. Findings highlight both the promise and challenges of AI-driven strategies in advancing early Alzheimer’s detection.

Results: Multistage MRI-based convolutional neural networks (CNNs) demonstrated the highest diagnostic performance, achieving 98.24% accuracy in dementia detection and 99.70% accuracy in Alzheimer’s disease stage subclassification (Ali et al., 2024). PET-based artificial intelligence (AI) systems demonstrated diagnostic accuracies ranging from 85% to 93%, with the notable advantage of identifying pathological changes up to six years before clinical diagnosis (Saad et al., 2024; Athanasopoulos et al., 2025). Multimodal AI systems integrating MRI and PET, with or without retinal imaging, consistently outperformed singlemodality approaches by 10-15%, achieving area under the curve (AUC) values greater than 0.95 and classification accuracies exceeding 98% (Zhang et al., 2023; Shao et al., 2024). In contrast, lower-cost AI models based on EEG signals and routine clinical data demonstrated moderate diagnostic accuracy (80–88%), supporting their potential role as scalable screening tools, particularly in healthcare settings with limited access to advanced neuroimaging technologies (Al-Saegh et al., 2024; Giunta et al., 2024). 

Conclusion: Multimodal AI (MRI+PET+retinal biomarkers) achieves over 98% accuracy in the early detection of Alzheimer’s disease (MCI stage), demonstrating costeffectiveness. Lower-cost EEGs, screening approaches, and biomarkers would be more cost-effective but offer moderate accuracy comparatively.