Leakage-Controlled Multimodal Learning for Diagnosis and Progression Prediction in Alzheimer's Disease Research

2026-10-07

AI Summary

This study addresses the challenges of irregular clinical visits, data heterogeneity, and missing modalities in Alzheimer’s disease prediction by proposing a multi-task longitudinal framework that integrates SFCN-MRI encoding, a causal Transformer, and ODE-GRU. The method innovatively incorporates Task-DRO to balance multi-task losses and Group-CVaR to optimize subgroup risks, while enforcing strict input controls to prevent data leakage. Generalization is further enhanced through LoRA fine-tuning and hybrid ensembling. Internal validation achieves a diagnostic AUROC of 0.935 with significantly reduced external calibration error, and MMSE prediction yields an MAE of only 1.61 points. These results demonstrate precise and robust disease progression forecasting.

Abstract

Alzheimer's disease prediction involves irregular visits, heterogeneous measurements and incomplete modalities. This study presents a multimodal multitask framework combining an adapted SFCN MRI encoder, four causal clinical Transformers, shared fusion and task-specific ODE-GRU dynamics. Fine-tuning and LoRA adapt the final two MRI blocks. Task-DRO balances task losses, while Group-CVaR targets cohort and comorbidity strata. Branch-specific input controls, subject-grouped partitions and empirical causality checks support longitudinal evaluation. Across 2,649 subjects and 17,317 visits from ADNI, OASIS-2 and MIRIAD, internal validation yields diagnosis, stage-1 progression and first-stage-1-visit progression AUROCs of 0.935 +/- 0.002, 0.884 +/- 0.003 and 0.870 +/- 0.005, respectively (mean +/- SD across three seeds). Corresponding hybrid AUROCs are 0.951, 0.909 and 0.896. Next-visit MMSE mean absolute error (MAE) is 1.61 points; worst-stratum diagnosis AUROC is 0.827 +/- 0.008. Sampled ADNI explanations identify task-specific input dependence. OASIS-3 external validation yields network and hybrid diagnosis AUROCs of 0.763 and 0.767, hybrid next-visit progression AUROC of 0.764, diagnosis calibration error decreasing from 0.197 to 0.052, and next-visit MMSE MAE of 0.86. Seed-42 paired ablations of six components yield pooled diagnosis and progression AUROC differences between -0.004 and +0.004; removing clinical encoder inputs lowers diagnosis AUROC by 0.272. The framework integrates longitudinal prediction, missing-modality handling, auxiliary comorbidity modelling and subgroup evaluation within a common pipeline.

Problem

• Alzheimer's disease
• multimodal learning
• progression prediction
• irregular visits
• incomplete modalities

Innovation

• Multimodal Multitask Learning
• ODE-GRU
• LoRA Fine-tuning
• Task-DRO
• Leakage Control