DINO-BOLDNet: A DINOv3-Guided Multi-Slice Attention Network for T1-to-BOLD Generation

2025-12-09

AI Summary

To address the challenge of downstream functional brain analysis when BOLD fMRI data are missing or corrupted, this paper proposes, for the first time, a direct reconstruction method that synthesizes mean BOLD images from T1-weighted structural MRI. Methodologically, we employ a frozen DINOv3 self-supervised encoder to extract robust anatomical representations, incorporate a multi-slice attention mechanism to model cross-layer functional dependencies, and integrate a multi-scale decoder with a DINO-perceptual loss to ensure high-fidelity generation. Evaluated on a clinical dataset comprising 248 subjects, our approach significantly outperforms conditional GAN baselines, achieving state-of-the-art performance in both PSNR and MS-SSIM metrics. This work establishes a novel, interpretable, and high-accuracy paradigm for structure-to-function mapping, enabling reliable functional inference from structural scans alone.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Bridging structured and unstructured data
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Generating BOLD images from T1w images offers a promising solution for recovering missing BOLD information and enabling downstream tasks when BOLD images are corrupted or unavailable. Motivated by this, we propose DINO-BOLDNet, a DINOv3-guided multi-slice attention framework that integrates a frozen self-supervised DINOv3 encoder with a lightweight trainable decoder. The model uses DINOv3 to extract within-slice structural representations, and a separate slice-attention module to fuse contextual information across neighboring slices. A multi-scale generation decoder then restores fine-grained functional contrast, while a DINO-based perceptual loss encourages structural and textural consistency between predictions and ground-truth BOLD in the transformer feature space. Experiments on a clinical dataset of 248 subjects show that DINO-BOLDNet surpasses a conditional GAN baseline in both PSNR and MS-SSIM. To our knowledge, this is the first framework capable of generating mean BOLD images directly from T1w images, highlighting the potential of self-supervised transformer guidance for structural-to-functional mapping.

Problem

• Generates BOLD images from T1w images
• Recovers missing BOLD information for downstream tasks
• Integrates DINOv3 for structural-to-functional mapping

Innovation

• Uses frozen DINOv3 encoder for structural representation
• Employs slice-attention module for cross-slice context fusion
• Applies DINO-based perceptual loss for feature consistency