AT-Attn: Temporal-Aware Cross-Attention for Longitudinal Multimodal Alzheimer's Disease Diagnosis

2026-07-08 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition · Depth: Expert, quick

Summary

The AT-Attn framework is proposed for longitudinal multimodal Alzheimer's disease (AD) diagnosis, addressing challenges posed by irregularly collected clinical and imaging data. This temporal-aware system integrates structural MRI with longitudinal clinical information, including six cognitive-scale trajectories and seven static clinical variables. AT-Attn employs Change-and-Time encoding, time-biased asymmetric cross-attention, and gated fusion to prevent performance degradation from noisy or intermittently unavailable MRI. Evaluated on an MRI-retained ADNI cohort of 1,520 patients using patient-level five-fold cross-validation, the model achieved an accuracy of 0.719+/-0.024, macro F1 of 0.721+/-0.023, ROC-AUC of 0.873+/-0.013, and PR-AUC of 0.783+/-0.018. These results demonstrate AT-Attn's superior performance over unimodal and naive multimodal fusion baselines, confirming that a constrained, temporal-aware fusion strategy enhances MRI's contribution to patient-level AD diagnosis support.

Key takeaway

For AI Scientists developing diagnostic models for diseases like Alzheimer's with irregular longitudinal data, you should prioritize temporal-aware fusion strategies. Implementing techniques such as time-biased asymmetric cross-attention and gated fusion can significantly improve diagnostic accuracy by effectively integrating noisy or intermittently available imaging data with clinical trajectories. This approach helps ensure that complementary information from all modalities contributes positively, rather than degrading, your model's performance.

Key insights

Temporal-aware fusion of irregular multimodal data improves Alzheimer's diagnosis by mitigating noise and data gaps.

Principles

• Temporal awareness is crucial for longitudinal data fusion.
• Constrained fusion prevents performance degradation from noisy modalities.
• Asymmetric cross-attention can selectively integrate modalities.

Method

AT-Attn combines Change-and-Time encoding, time-biased asymmetric cross-attention, and gated fusion to integrate MRI with longitudinal clinical data for AD diagnosis.

In practice

• Apply gated fusion to manage noisy imaging data.
• Use time-biased attention for irregular longitudinal inputs.
• Integrate cognitive trajectories with static clinical variables.

Topics

• Alzheimer's Disease Diagnosis
• Multimodal Data Fusion
• Longitudinal Analysis
• Cross-Attention Networks
• Structural MRI
• ADNI Cohort

Best for: Computer Vision Engineer, AI Scientist, Research Scientist

Related on AIssential

• Alzheimer's Disease Diagnosis using a Multimodal Approach with 3D MRI and PET — Input-adaptive multimodal modeling with gated fusion and Mixture-of-Experts significantly enhances Alzheimer's disease diagnosis using MRI and PET.
• Forecasting Medium-Horizon Alzheimer's Disease Progression: Residual Gap-Aware Transformers for 24-Month CDR-SB Change from ADNI Clinical and Biomarker Histories — Combining statistical anchoring with gap-aware residual learning improves medium-horizon Alzheimer's disease progression prediction.
• Fair Cognitive Impairment Detection Through Unlearning — Multimodal fusion and gradient reversal unlearning can significantly reduce demographic bias in MCI detection models while improving overall performance.
• Multilingual Detection of Alzheimer's Disease from Speech: A Cross-Linguistic Transfer Learning Approach — Cross-language transfer learning effectively detects Alzheimer's Disease from speech across multiple languages, achieving strong generalization and rapid inference.
• A Quantitative Analysis of Multimodal Biomarkers in Alzheimer's Disease — The study systematically characterizes multimodal AD biomarker relationships to improve interpretability and selection.
• Interpretability of multimodal neural networks for prediction of visual acuity in patients with branch retinal vein occlusion — Multimodal neural networks can predict 12-month BCVA in BRVO, but require rigorous validation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.