EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Pilot Study on PET-to-CT Cross-Modal Translation: Systematic Validation of Brownian Bridge Diffusion Models

Xiaoyang Gu, Tianyu Shi

AI Medicine · 2026

DOI: 10.53941/aim.2026.100009 ↗

Kurz gefasst

Fachartikel in der Zeitschrift AI Medicine, 2026. Verfasst von Xiaoyang Gu und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: the National Natural Science Foundation of China, the Education Department of Liaoning Provincial Youth Scientist.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

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 networks (GANs) suffer from unstable training and anatomical artifacts, while standard denoising diffusion probabilistic models (DDPMs) converge slowly because the reverse process starts from pure noise. In this work, we systematically evaluate the Brownian Bridge diffusion model (BBDM) for prostate-specific membrane antigen (PSMA) PET-to-CT translation under a patient-disjoint experimental setting. BBDM constructs a stochastic bridge between the source and target domains, so that the reverse diffusion trajectory is initialized from the PET source endpoint rather than from pure Gaussian noise. The processed dataset was partitioned at the patient level into 287 training patients, 32 validation patients, and 35 independent test patients. Quantitative evaluation was performed on 3741 test slices using complementary image-quality metrics, and statistical significance was assessed using patient-level MAE values. Compared with the matched DDPM baseline using the same UNet backbone and training budget, BBDM achieved substantially lower reconstruction error, supporting the utility of source-anchored bridge diffusion relative to conventional noise-initialized diffusion for PET-to-CT synthesis. After adding SelfRDB-L1 as a recent diffusion-bridge baseline, BBDM remained competitive across the broader metric set, although SelfRDB-L1 achieved a lower patient-level MAE. These results suggest that BBDM is a competitive bridge-based formulation for PET-conditioned CT synthesis, while further clinical validation is required before diagnostic use.

Förderung

  • the National Natural Science Foundation of China
  • the Education Department of Liaoning Provincial Youth Scientist

Zitieren

APA

Gu, X., & Shi, T. (2026). A Pilot Study on PET-to-CT Cross-Modal Translation: Systematic Validation of Brownian Bridge Diffusion Models. AI Medicine. https://doi.org/10.53941/aim.2026.100009

Vancouver

Gu X, Shi T. A Pilot Study on PET-to-CT Cross-Modal Translation: Systematic Validation of Brownian Bridge Diffusion Models. AI Medicine. 2026. doi:10.53941/aim.2026.100009

BibTeX
@article{Gu2026,
  author = {Gu, Xiaoyang and Shi, Tianyu},
  title = {A Pilot Study on PET-to-CT Cross-Modal Translation: Systematic Validation of Brownian Bridge Diffusion Models},
  journal = {AI Medicine},
  year = {2026},
  doi = {10.53941/aim.2026.100009}
}
Kostenlos starten →