Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Collaborative Attention and Consistent-Guided Fusion of MRI and PET for Alzheimer's Disease Diagnosis

Delin Ma, Menghui Zhou, Jun Qi, Yun Yang, Po Yang

TL;DR

This work tackles Alzheimer's disease diagnosis by fusing MRI and PET data through a unified multimodal framework. It introduces a Triple-Collaborative Attention FE module, a Cross-modal Consistent Feature Enhancement with Learnable Parameter Representations, and a Shared–Specific Feature Fusion scheme to jointly learn modality-specific and shared representations. The approach employs a FCC-based consistency loss and an MSE constraint to align cross-modal distributions, achieving state-of-the-art accuracy on the ADNI dataset for CN vs AD and AD vs MCI tasks. The results demonstrate that preserving modality-specific cues while leveraging cross-modal consistency improves diagnostic performance, with potential for more reliable clinical screening and early intervention.

Abstract

Alzheimer's disease (AD) is the most prevalent form of dementia, and its early diagnosis is essential for slowing disease progression. Recent studies on multimodal neuroimaging fusion using MRI and PET have achieved promising results by integrating multi-scale complementary features. However, most existing approaches primarily emphasize cross-modal complementarity while overlooking the diagnostic importance of modality-specific features. In addition, the inherent distributional differences between modalities often lead to biased and noisy representations, degrading classification performance. To address these challenges, we propose a Collaborative Attention and Consistent-Guided Fusion framework for MRI and PET based AD diagnosis. The proposed model introduces a learnable parameter representation (LPR) block to compensate for missing modality information, followed by a shared encoder and modality-independent encoders to preserve both shared and specific representations. Furthermore, a consistency-guided mechanism is employed to explicitly align the latent distributions across modalities. Experimental results on the ADNI dataset demonstrate that our method achieves superior diagnostic performance compared with existing fusion strategies.

Collaborative Attention and Consistent-Guided Fusion of MRI and PET for Alzheimer's Disease Diagnosis

TL;DR

This work tackles Alzheimer's disease diagnosis by fusing MRI and PET data through a unified multimodal framework. It introduces a Triple-Collaborative Attention FE module, a Cross-modal Consistent Feature Enhancement with Learnable Parameter Representations, and a Shared–Specific Feature Fusion scheme to jointly learn modality-specific and shared representations. The approach employs a FCC-based consistency loss and an MSE constraint to align cross-modal distributions, achieving state-of-the-art accuracy on the ADNI dataset for CN vs AD and AD vs MCI tasks. The results demonstrate that preserving modality-specific cues while leveraging cross-modal consistency improves diagnostic performance, with potential for more reliable clinical screening and early intervention.

Abstract

Alzheimer's disease (AD) is the most prevalent form of dementia, and its early diagnosis is essential for slowing disease progression. Recent studies on multimodal neuroimaging fusion using MRI and PET have achieved promising results by integrating multi-scale complementary features. However, most existing approaches primarily emphasize cross-modal complementarity while overlooking the diagnostic importance of modality-specific features. In addition, the inherent distributional differences between modalities often lead to biased and noisy representations, degrading classification performance. To address these challenges, we propose a Collaborative Attention and Consistent-Guided Fusion framework for MRI and PET based AD diagnosis. The proposed model introduces a learnable parameter representation (LPR) block to compensate for missing modality information, followed by a shared encoder and modality-independent encoders to preserve both shared and specific representations. Furthermore, a consistency-guided mechanism is employed to explicitly align the latent distributions across modalities. Experimental results on the ADNI dataset demonstrate that our method achieves superior diagnostic performance compared with existing fusion strategies.

Paper Structure

This paper contains 14 sections, 10 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodes
  4. Overview
  5. Feature Extraction
  6. Cross-modal Consistent Feature Enhancement
  7. Specific versus Shared Fusion
  8. Experiments And Results
  9. Materials and Preprocessing
  10. Experimental Settings and Evaluation
  11. Comparison with State-of-the-art Methods
  12. Loss Values Curves of Different Tasks
  13. Ablation Study
  14. Conclusion

Figures (2)

  • Figure 1: The proposed framework for Alzheimer’s disease diagnosis. The network first extracts modality-specific and shared representations through dedicated/shared feature extraction (FE) blocks. Subsequently, the Cross-modality Consistent Feature Enhancement (CCFE) module employs a learnable parameter representation (LPR) block to achieve MRI and PET information complementation at the feature level, while imposing a feature-level cross-correlation (FCC) consistency constraint to reduce modality discrepancies. Finally, the Shared–Specific Feature Fusion (SSFF) combines the enhanced shared features with the original modality-specific features and feeds them into a classifier for disease prediction.
  • Figure 2: The loss value trends in the three classification tasks—(a) CN vs. AD, (b) CN vs. MCI, and (c) AD vs. MCI—illustrate the model’s training dynamics over the epochs. For each task, the loss curves of the 10 cross-validation folds show a consistent decline, indicating effective convergence. The mean loss curve, highlighted in black, represents the overall training trend, while the individual colored lines reflect fold-specific variations, demonstrating the model’s robustness and stability across different data splits.