This work introduces a novel surface-based diffusion bridge framework DB-SUiT for MRI-to-PET translation that operates natively on the cortical manifold and demonstrates high-fidelity synthesis that substantially outperforms other baselines.
Abstract
Cortical hypometabolism measured by Fluorodeoxyglucose Positron Emission Tomography (FDG-PET) is a highly sensitive biomarker for dementia diagnosis. However, high costs, radiation exposure, and limited accessibility constrain its clinical utility. While cross-modal synthesis from Magnetic Resonance Imaging (MRI) offers a promising alternative, existing volumetric generation methods do not explicitly account for the highly folded cortical geometry, where disease-related patterns predominantly reside. To address this, we introduce a novel surface-based diffusion bridge framework DB-SUiT for MRI-to-PET translation that operates natively on the cortical manifold. A conditional Spherical U-shaped vision Transformer (SUiT) is specifically designed to model the intricate cross-modal relationships while preserving surface topology. It combines spherical convolutional encoders for multi-scale surface feature extraction with bottleneck Transformers to capture long-range spatial dependencies, while incorporating demographic and subcortical conditions to refine the synthesis. Evaluated on two datasets, including subjects with different dementia types, DB-SUiT demonstrates high-fidelity synthesis that substantially outperforms other baselines. In automated dementia classification, synthesized PET surfaces improve performance over MRI by 14.2% and PET volumes by 11.3%, approaching the performance of real PET surfaces. In a blinded reader study, synthetic PET achieved 85.5% diagnostic accuracy, compared with 75.8% for MRI and 95.2% for real PET. This further demonstrates cross-cohort and cross-pathology generalization, as the model was evaluated without retraining on an external cohort that included a dementia subtype not represented during training. Our code is available at https://github.com/ai-med/DB-SUiT.
A structural-guided latent diffusion framework for synthesizing PET from structural MRI while jointly learning disease-discriminative representations is proposed, providing an effective solution for multimodal data completion and highlighting the clinical potential of structural-guided diffusion models.
Nina Cheng, Zhenghua Guan, Jian-Qing Zheng et al.· Journal of Physics, Conferen...· 0 citations
Deep learning models for Alzheimer disease (AD) classification routinely report near-perfect discrimination, yet few are shown to rest on AD-relevant neurobiology rather than on dataset artifacts, subject-level leakage, or non-brain image content. We present a fusion network combining T1 MRI and FDG PET across axial, c...
Cross-modal PET-to-CT translation is an important problem in medical image computing. PET encodes the functional distribution of a radiotracer, whereas CT encodes anatomical electron density, and the two are separated by a substantial modality gap. Existing approaches have well-known limitations: generative adversarial...
Tian-Yu Shi, Xiao-Yang Gu· AI Medicine· 0 citations
UniPET is introduced, a unified approach for whole-body CT-to-PET translation based on curriculum learning that provides a promising approach for augmenting CT-based workflows with synthetic metabolic information, with potential applications in screening triage, in low-resource environments where PET is unavailable, an...
Francesco Di Feola, V. Guarrasi, Mikael Johansson et al.· Frontiers in Artificial Inte...· 0 citations
Experimental results demonstrate that the proposed lesion-robust framework for longitudinal brain CT registration based on MRguided image translation consistently outperforms conventional intensity-based registration methods as well as existing image-translation–assisted strategies in both quantitative metrics and visu...
Zijun Cheng, Hui-Xiang Zhuang, Yue Guan et al.· International Conference on...· 0 citations
This survey organizes transformer-based fusion models for healthcare along four axes:modality, pairing, fusion level, and attention and architecture design, to give researchers and clinicians a structured entry point into this area.
N. Jagtap, D. J. Saini, Prasun Chakrabarti· Journal of Data Science and...· 0 citations
Known for his clear and elegant writing style, Bertsekas shaped fields from control and optimization to large-scale computation and artificial intelligence.