Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Structure-constrained Language-informed Diffusion Model for Unpaired Low-dose Computed Tomography Angiography Reconstruction

Genyuan Zhang, Zihao Wang, Zhifan Gao, Lei Xu, Zhen Zhou, Haijun Yu, Jianjia Zhang, Xiujian Liu, Weiwei Zhang, Shaoyu Wang, Huazhu Fu, Fenglin Liu, Weiwen Wu

TL;DR

The paper tackles the safety risk of iodinated contrast in CT angiography by reconstructing normal-dose images from low-dose scans using a Structure-constrained Language-informed Diffusion Model (SLDM). SLDM fuses a topology-constrained diffusion prior with semantic guidance from a CTA-focused language model (CTA-CLIP) and a subtraction angiography refinement module (SAEM) to preserve anatomy while enhancing vascular contrast in unpaired data. It introduces a topology-aware extended diffusion process and a two-stage training regime, achieving superior structural fidelity and contrast over state-of-the-art baselines as judged by radiologists and quantitative metrics. The approach holds clinical promise for reducing contrast dose without sacrificing diagnostic power and offers a pathway toward practical deployment with controllable vascular enhancement.

Abstract

The application of iodinated contrast media (ICM) improves the sensitivity and specificity of computed tomography (CT) for a wide range of clinical indications. However, overdose of ICM can cause problems such as kidney damage and life-threatening allergic reactions. Deep learning methods can generate CT images of normal-dose ICM from low-dose ICM, reducing the required dose while maintaining diagnostic power. However, existing methods are difficult to realize accurate enhancement with incompletely paired images, mainly because of the limited ability of the model to recognize specific structures. To overcome this limitation, we propose a Structure-constrained Language-informed Diffusion Model (SLDM), a unified medical generation model that integrates structural synergy and spatial intelligence. First, the structural prior information of the image is effectively extracted to constrain the model inference process, thus ensuring structural consistency in the enhancement process. Subsequently, semantic supervision strategy with spatial intelligence is introduced, which integrates the functions of visual perception and spatial reasoning, thus prompting the model to achieve accurate enhancement. Finally, the subtraction angiography enhancement module is applied, which serves to improve the contrast of the ICM agent region to suitable interval for observation. Qualitative analysis of visual comparison and quantitative results of several metrics demonstrate the effectiveness of our method in angiographic reconstruction for low-dose contrast medium CT angiography.

Structure-constrained Language-informed Diffusion Model for Unpaired Low-dose Computed Tomography Angiography Reconstruction

TL;DR

The paper tackles the safety risk of iodinated contrast in CT angiography by reconstructing normal-dose images from low-dose scans using a Structure-constrained Language-informed Diffusion Model (SLDM). SLDM fuses a topology-constrained diffusion prior with semantic guidance from a CTA-focused language model (CTA-CLIP) and a subtraction angiography refinement module (SAEM) to preserve anatomy while enhancing vascular contrast in unpaired data. It introduces a topology-aware extended diffusion process and a two-stage training regime, achieving superior structural fidelity and contrast over state-of-the-art baselines as judged by radiologists and quantitative metrics. The approach holds clinical promise for reducing contrast dose without sacrificing diagnostic power and offers a pathway toward practical deployment with controllable vascular enhancement.

Abstract

The application of iodinated contrast media (ICM) improves the sensitivity and specificity of computed tomography (CT) for a wide range of clinical indications. However, overdose of ICM can cause problems such as kidney damage and life-threatening allergic reactions. Deep learning methods can generate CT images of normal-dose ICM from low-dose ICM, reducing the required dose while maintaining diagnostic power. However, existing methods are difficult to realize accurate enhancement with incompletely paired images, mainly because of the limited ability of the model to recognize specific structures. To overcome this limitation, we propose a Structure-constrained Language-informed Diffusion Model (SLDM), a unified medical generation model that integrates structural synergy and spatial intelligence. First, the structural prior information of the image is effectively extracted to constrain the model inference process, thus ensuring structural consistency in the enhancement process. Subsequently, semantic supervision strategy with spatial intelligence is introduced, which integrates the functions of visual perception and spatial reasoning, thus prompting the model to achieve accurate enhancement. Finally, the subtraction angiography enhancement module is applied, which serves to improve the contrast of the ICM agent region to suitable interval for observation. Qualitative analysis of visual comparison and quantitative results of several metrics demonstrate the effectiveness of our method in angiographic reconstruction for low-dose contrast medium CT angiography.
Paper Structure (37 sections, 21 equations, 11 figures, 2 tables, 1 algorithm)

This paper contains 37 sections, 21 equations, 11 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Virtual Contrast Enhanced CT Algorithm
  4. Visual Language Models Assisted Image Generation
  5. Preliminaries in Image Restoration SDE
  6. METHODS
  7. The Overall of Problem and Method
  8. Overview of the ICM Enhancement Process
  9. Architecture of SLDM
  10. Overall Clarification
  11. Topologically Constrained IR-SDE
  12. CTA-CLIP Assistance Phase
  13. Subtraction Angiography Enhancement Module
  14. EXPERIMENTS
  15. Datasets
  16. ...and 22 more sections

Figures (11)

  • Figure 1: The proposed method employs three strategies to cope with the corresponding problems. (a) Structural differences in low-match data affect reconstructed structures; (b) False enhancements caused by entanglement between enhanced and non-enhanced regions; (c) Intensity differences of subjects affect the reconstructed grayscale.
  • Figure 2: Overview of the proposed SLDM. (a) The IR-SDE constitutes the foundational framework, topological constraint score (green) and text supervision score (blue) are integrated throughout both training and testing phases to ensure structural fidelity; (b) The pre-trained CTA-CLIP encoder, designed to facilitate fine-grained feature extraction and semantic alignment within the model; (c) SAEM refines the reconstruction process through targeted optimization of subtraction angiography features.
  • Figure 3: Validation semantic datasets of the proposed SLDM. (a) An example of generating the LDCT image-text tuple via LLAVA-med; (b) The image description text of NDCT included information on the type of organ, location information and the HU value of contrast agent area.
  • Figure 4: (a) The visualization comparison between our method and LDCT and NDCT from axial views; (b) The coronary volume rendering (VR) image of LDCT, Ours and NDCT.
  • Figure 5: The visualization comparison between our method and state-of-the-art methods. Synthesized images from competing methods are displayed along with LDCT and NDCT (reference) images for representative tasks. The display window for the reconstructed image is set to [-300, 700] HU. The portion of the image in the green and red boxes is the region of interest (ROI). Compared to baselines, proposed method maintains higher anatomical fidelity.
  • ...and 6 more figures