Brain Network Diffusion-Driven fMRI Connectivity Augmentation for Enhanced Autism Spectrum Disorder Diagnosis

TL;DR

This paper tackles data scarcity in rs-fMRI-based ASD diagnosis by introducing Brain-Net-Diffusion, a transformer-driven latent diffusion model for generating functional connectivity. The approach combines a VQ-VAE latent space, a diffusion transformer conditioned on diagnosis, and a real-sample guidance mechanism to produce high-quality, in-distribution FC matrices for augmentation. Key contributions include distribution normalization to align latent and noise distributions and a conditional contrastive loss to improve conditioning effectiveness, along with extensive ablations and biomarker interpretations. Empirical results on ABIDE-I show meaningful classification gains and improved realism of generated FC, underscoring the method's potential to enhance neuroimaging-based ASD diagnostics and its applicability to other disorders.

Abstract

Functional magnetic resonance imaging (fMRI) is an emerging neuroimaging modality that is commonly modeled as networks of Regions of Interest (ROIs) and their connections, named functional connectivity, for understanding the brain functions and mental disorders. However, due to the high cost of fMRI data acquisition and labeling, the amount of fMRI data is usually small, which largely limits the performance of recognition models. With the rise of generative models, especially diffusion models, the ability to generate realistic samples close to the real data distribution has been widely used for data augmentations. In this work, we present a transformer-based latent diffusion model for functional connectivity generation and demonstrate the effectiveness of the diffusion model as an augmentation tool for fMRI functional connectivity. Furthermore, extended experiments are conducted to provide detailed analysis of the generation quality and interpretations for the learned feature pattern. Our code will be made public upon acceptance.

Figures (9)

• Figure 1: The architecture of Brain-Net-Diffusion framework for functional functional connectivity generation
• Figure 2: Transformer-based node auto-encoding with vector quantization
• Figure 3: Numerical distribution of latent space
• Figure 4: Diffusion Transformer Block
• Figure 5: Accuracy for various number of synthetic samples for each augmentation methods.