HA-HI: Synergising fMRI and DTI through Hierarchical Alignments and Hierarchical Interactions for Mild Cognitive Impairment Diagnosis

TL;DR

HA-HI tackles early cognitive impairment detection from dual-modal MRI by proposing a hierarchical framework that aligns and fuses fMRI and DTI features across temporal, regional, and connectivity levels. The Dual-Modal Hierarchical Alignments (DMHA) synchronize multi-scale DFC, bridge static and dynamic connectivity, and align functional with structural regional information, while Dual-Domain Hierarchical Interactions (DDHI) fuse regions and connectivities through fine-grained and global attention. Synergistic Activation Mapping (SAM) provides interpretability, revealing networks such as DMN, FPN, and CON and key regions implicated in cognitive decline. Evaluations on GUTCM and ADNI, plus extensive ablations, show HA-HI achieves superior or competitive accuracy and F1 scores, with ablations confirming the necessity of hierarchical alignments and interactions, offering an interpretable, dual-modal diagnostic tool for MCI/SCD.

Abstract

Early diagnosis of mild cognitive impairment (MCI) and subjective cognitive decline (SCD) utilizing multi-modal magnetic resonance imaging (MRI) is a pivotal area of research. While various regional and connectivity features from functional MRI (fMRI) and diffusion tensor imaging (DTI) have been employed to develop diagnosis models, most studies integrate these features without adequately addressing their alignment and interactions. This limits the potential to fully exploit the synergistic contributions of combined features and modalities. To solve this gap, our study introduces a novel Hierarchical Alignments and Hierarchical Interactions (HA-HI) method for MCI and SCD classification, leveraging the combined strengths of fMRI and DTI. HA-HI efficiently learns significant MCI- or SCD- related regional and connectivity features by aligning various feature types and hierarchically maximizing their interactions. Furthermore, to enhance the interpretability of our approach, we have developed the Synergistic Activation Map (SAM) technique, revealing the critical brain regions and connections that are indicative of MCI/SCD. Comprehensive evaluations on the ADNI dataset and our self-collected data demonstrate that HA-HI outperforms other existing methods in diagnosing MCI and SCD, making it a potentially vital and interpretable tool for early detection. The implementation of this method is publicly accessible at https://github.com/ICI-BCI/Dual-MRI-HA-HI.git.

Figures (6)

• Figure 1: Workflow of HA-HI: This framework enhances cognitive impairment identification via dual-modal hierarchical alignments (DMHA) and dual-domain hierarchical interactions (DDHI). DMHA aligns diverse features from fMRI and DTI modalities horizontally, and DDHI optimizes feature fusion across regional and connectivity domains vertically.
• Figure 2: The technical details of HA-HI, developed for cognitive impairment detection using fMRI and DTI inputs, capitalize on DMHA's strengths in performing hierarchical alignments between dynamic temporal scales, integrating dynamic and static networks, and correlating functional with structural features, and on DDHI's role in conducting hierarchical interactions from the fine-grained to the global level, to fuse features across regional and connectivity domains.
• Figure 3: Strategy for Multiscale DFC alignment.
• Figure 4: MRI synergy effects reflected by multi-scale DFC features. A) Activation maps. B) Weighted feature maps, calibrated with the activation maps. C) Top five significant connectivities. D) Five key regions with notable ROI-wise connectivity strength. Networks represented include CER (cerebellum network), CON (cingulo-opercular network), DMN (default mode network), EMO (emotional network), OCC (occipital network), FPN (frontoparietal network), and SEN (sensorimotor network)
• Figure 5: MRI synergy effects from an SFC Perspective. A) Activation maps from the SAM technique. B) Top five connectivities impacted by cognitive impairment. C) Five regions with notable abnormal ROI-wise connectivity strength.