[Paper Review] Brain Imaging-to-Graph Generation using Adversarial Hierarchical Diffusion Models for MCI Causality Analysis

Qiankun Zuo, Tian, Hao+4 more|arXiv (Cornell University)|May 18, 2023

Functional Brain Connectivity Studies4 citations

TL;DR

This paper proposes a Multi-resolution Spatiotemporal Enhanced Transformer Denoising (MSETD) network with adversarial functional diffusion to generate brain effective connectivity (BEC) from 4D fMRI for mild cognitive impairment (MCI) analysis. By leveraging a conditional diffusion process and a multi-scale transformer generator and discriminator, the model achieves superior BEC prediction and identifies clinically consistent MCI-related causal connections.

ABSTRACT

Effective connectivity can describe the causal patterns among brain regions. These patterns have the potential to reveal the pathological mechanism and promote early diagnosis and effective drug development for cognitive disease. However, the current methods utilize software toolkits to extract empirical features from brain imaging to estimate effective connectivity. These methods heavily rely on manual parameter settings and may result in large errors during effective connectivity estimation. In this paper, a novel brain imaging-to-graph generation (BIGG) framework is proposed to map functional magnetic resonance imaging (fMRI) into effective connectivity for mild cognitive impairment (MCI) analysis. To be specific, the proposed BIGG framework is based on the diffusion denoising probabilistic models (DDPM), where each denoising step is modeled as a generative adversarial network (GAN) to progressively translate the noise and conditional fMRI to effective connectivity. The hierarchical transformers in the generator are designed to estimate the noise at multiple scales. Each scale concentrates on both spatial and temporal information between brain regions, enabling good quality in noise removal and better inference of causal relations. Meanwhile, the transformer-based discriminator constrains the generator to further capture global and local patterns for improving high-quality and diversity generation. By introducing the diffusive factor, the denoising inference with a large sampling step size is more efficient and can maintain high-quality results for effective connectivity generation. Evaluations of the ADNI dataset demonstrate the feasibility and efficacy of the proposed model. The proposed model not only achieves superior prediction performance compared with other competing methods but also predicts MCI-related causal connections that are consistent with clinical studies.

Motivation & Objective

• To address the limitations of existing methods that rely on manual fMRI preprocessing and shallow models in capturing complex causal brain connectivity.
• To develop an end-to-end framework that maps raw fMRI to effective connectivity without dependency on empirical time-series extraction.
• To improve the quality, diversity, and efficiency of generated effective connectivity graphs using adversarial and diffusion-based denoising.
• To identify MCI-related causal connections consistent with clinical literature for potential biomarker discovery.

Proposed method

• Employs a conditional diffusion probabilistic model (DDPM) to progressively denoise blurred fMRI time series into clean, high-quality ROI-based time series.
• Introduces a multi-resolution enhanced transformer generator with spatial/temporal multi-head attention to extract local and global spatiotemporal features at multiple scales.
• Designs a multi-scale diffusive transformer discriminator to capture temporal patterns across different scales and stabilize the adversarial training process.
• Uses the AAL90 atlas to parcellate 3D fMRI into 90 regions-of-interest (ROIs), enabling graph-based BEC representation with directional edges.
• Applies an adversarial strategy to accelerate the slow denoising process of DDPM while preserving sample diversity and quality.
• Trains the model end-to-end to generate effective connectivity graphs directly from 4D fMRI, avoiding intermediate preprocessing bottlenecks.

Experimental results

Research questions

• RQ1 Can a conditional diffusion-based generative model effectively reconstruct high-quality effective connectivity from noisy fMRI time series without manual preprocessing?
• RQ2 How do multi-resolution spatiotemporal attention mechanisms improve the fidelity and diversity of generated brain connectivity graphs?
• RQ3 To what extent does the adversarial training strategy enhance the efficiency and quality of the diffusion-based denoising process?
• RQ4 Are the generated MCI-related causal connections biologically plausible and consistent with clinical findings in Alzheimer’s disease and MCI?

Key findings

• The proposed MSETD model achieves superior prediction performance on ADNI datasets compared to competing shallow and deep learning methods.
• Removal of the adversarial strategy (MSETD w/o MDT) results in a significant drop in classification performance, confirming its critical role in improving generation quality and speed.
• The model successfully identifies 12 key MCI-related effective connections, including diminished HIP.L to PCG.R and enhanced HIP.L to MOG.R, consistent with clinical literature on memory and cognitive decline.
• The effective connection from SOG.R to PCG.R weakens from NC to LMCI, aligning with clinical reports of memory-related network disruption.
• The AMYG, HIP, and ANG regions show altered connectivity patterns that correlate with known MCI and Alzheimer’s disease pathology.
• The model demonstrates feasibility and efficacy in generating biologically interpretable, clinically relevant effective connectivity patterns from raw fMRI.