Related Experiment Video
Updated: May 22, 2026
07:13
Multimodal Cross-Device and Marker-Free Co-Registration of Preclinical Imaging Modalities
Published on: October 27, 2023
A text-guided Brownian bridge diffusion model for unified multiphase contrast-enhanced CT synthesis
Li Yu1, Jiaqing Liu1, Rahul Kumar Jain2
1Graduate School of Information Science and Engineering, Ritsumeikan University, 2-150 Iwakuracho, Ibaraki, Osaka 567-8570, Japan.
Biomedical Physics & Engineering Express
|May 20, 2026
Summary
Synthesizing contrast-enhanced CT (CECT) from non-contrast CT (NCCT) aids liver lesion assessment. A novel Text-Guided Brownian Bridge Diffusion Model (TGBBDM) generates arterial and portal-venous phases in one go, improving image quality and offering phase control.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Healthcare
- Image Processing
Background:
- Multiphase contrast-enhanced CT (CECT) is crucial for evaluating liver lesions by reflecting vascularity and perfusion.
- Limitations of CECT include risks from iodinated contrast agents and radiation exposure from repeated scans.
- Current methods for synthesizing CECT from non-contrast CT (NCCT) often use separate models for each phase or require multiphase inputs, hindering unified and controllable synthesis.
Purpose of the Study:
- To develop a unified framework for synthesizing multiphase CECT from NCCT.
- To enable phase-controllable generation of arterial (ART) and portal-venous (PV) CECT phases using a single model.
- To address the limitations of existing NCCT-to-CECT conversion methods.
Main Methods:
- Proposed the Text-Guided Brownian Bridge Diffusion Model (TGBBDM), a text-conditioned image-to-image diffusion framework.
- Utilized a Brownian-bridge formulation for image synthesis.
- Incorporated phase-specific text prompts encoded and injected into the denoiser for guiding phase-aware generation within a single model.
Main Results:
- TGBBDM demonstrated superior performance in synthesizing both ART and PV phases compared to a baseline model on a retrospective liver CT dataset.
- Achieved improved whole-image Peak Signal-to-Noise Ratio (PSNR) from 24.14/23.38 to 24.49/23.99 for ART/PV phases.
- Showcased improved whole-image Pearson Correlation Coefficient (PCC) from 0.9670/0.9680 to 0.9695/0.9731 and competitive Structural Similarity Index Measure (SSIM) of 0.8457/0.8397.
Conclusions:
- Text-guided, phase-controllable NCCT-to-CECT synthesis using TGBBDM shows promise for liver lesion assessment.
- The proposed method offers a unified approach for generating multiple CECT phases, overcoming limitations of existing techniques.
- Further validation through larger, multi-center studies is necessary before widespread clinical adoption.
