Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cross-Modality Translation with Generative Adversarial Networks to Unveil Alzheimer's Disease Biomarkers

Reihaneh Hassanzadeh, Anees Abrol, Hamid Reza Hassanzadeh, Vince D. Calhoun

TL;DR

This work tackles the challenge of missing modalities in multimodal neuroimaging by introducing a bidirectional cross-modality translation between resting-state FNC maps and T1-weighted MRI in Alzheimer's disease using a Cycle-GAN. The model extends Cycle-GAN with weak supervision from paired data via an identity loss, combining adversarial, cycle-consistency, and identity objectives to preserve disease-specific patterns. Quantitatively, generated data achieve $SSIM$ around $0.894$ for T1 and Pearson correlation around $0.707$ for FNC, and qualitatively replicate AD-associated connectivity and hippocampal-temporal atrophy patterns, outperforming a standard Cycle-GAN baseline. These results suggest the approach preserves diagnostically relevant patterns and could support multimodal analyses and downstream diagnosis with synthesized modalities, particularly when one modality is missing.

Abstract

Generative approaches for cross-modality transformation have recently gained significant attention in neuroimaging. While most previous work has focused on case-control data, the application of generative models to disorder-specific datasets and their ability to preserve diagnostic patterns remain relatively unexplored. Hence, in this study, we investigated the use of a generative adversarial network (GAN) in the context of Alzheimer's disease (AD) to generate functional network connectivity (FNC) and T1-weighted structural magnetic resonance imaging data from each other. We employed a cycle-GAN to synthesize data in an unpaired data transition and enhanced the transition by integrating weak supervision in cases where paired data were available. Our findings revealed that our model could offer remarkable capability, achieving a structural similarity index measure (SSIM) of $0.89 \pm 0.003$ for T1s and a correlation of $0.71 \pm 0.004$ for FNCs. Moreover, our qualitative analysis revealed similar patterns between generated and actual data when comparing AD to cognitively normal (CN) individuals. In particular, we observed significantly increased functional connectivity in cerebellar-sensory motor and cerebellar-visual networks and reduced connectivity in cerebellar-subcortical, auditory-sensory motor, sensory motor-visual, and cerebellar-cognitive control networks. Additionally, the T1 images generated by our model showed a similar pattern of atrophy in the hippocampal and other temporal regions of Alzheimer's patients.

Cross-Modality Translation with Generative Adversarial Networks to Unveil Alzheimer's Disease Biomarkers

TL;DR

This work tackles the challenge of missing modalities in multimodal neuroimaging by introducing a bidirectional cross-modality translation between resting-state FNC maps and T1-weighted MRI in Alzheimer's disease using a Cycle-GAN. The model extends Cycle-GAN with weak supervision from paired data via an identity loss, combining adversarial, cycle-consistency, and identity objectives to preserve disease-specific patterns. Quantitatively, generated data achieve around for T1 and Pearson correlation around for FNC, and qualitatively replicate AD-associated connectivity and hippocampal-temporal atrophy patterns, outperforming a standard Cycle-GAN baseline. These results suggest the approach preserves diagnostically relevant patterns and could support multimodal analyses and downstream diagnosis with synthesized modalities, particularly when one modality is missing.

Abstract

Generative approaches for cross-modality transformation have recently gained significant attention in neuroimaging. While most previous work has focused on case-control data, the application of generative models to disorder-specific datasets and their ability to preserve diagnostic patterns remain relatively unexplored. Hence, in this study, we investigated the use of a generative adversarial network (GAN) in the context of Alzheimer's disease (AD) to generate functional network connectivity (FNC) and T1-weighted structural magnetic resonance imaging data from each other. We employed a cycle-GAN to synthesize data in an unpaired data transition and enhanced the transition by integrating weak supervision in cases where paired data were available. Our findings revealed that our model could offer remarkable capability, achieving a structural similarity index measure (SSIM) of for T1s and a correlation of for FNCs. Moreover, our qualitative analysis revealed similar patterns between generated and actual data when comparing AD to cognitively normal (CN) individuals. In particular, we observed significantly increased functional connectivity in cerebellar-sensory motor and cerebellar-visual networks and reduced connectivity in cerebellar-subcortical, auditory-sensory motor, sensory motor-visual, and cerebellar-cognitive control networks. Additionally, the T1 images generated by our model showed a similar pattern of atrophy in the hippocampal and other temporal regions of Alzheimer's patients.
Paper Structure (7 sections, 5 equations, 2 figures, 3 tables)

This paper contains 7 sections, 5 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Cross-Modality Transformation Framework
  4. Model Architecture
  5. Datasets and Experiments
  6. Results
  7. Conclusion

Figures (2)

  • Figure 1: (A) Translation framework. (B) Architecture of generators and discriminators. Numbers indicate the output channel and linear size.
  • Figure 2: (A) Real (first column) and generated (second column) FNC maps and the group mean differences. (B) Real and generated T1 images and $t$-values. Numbers in parentheses represent the sample size, and U and L indicate the upper and lower triangular matrix, respectively. Note that, to compare real and generated samples, we visualized only the samples for which both FNC and T1 data were available.