ProDM: Synthetic Reality-driven Property-aware Progressive Diffusion Model for Coronary Calcium Motion Correction in Non-gated Chest CT

Xinran Gong, Gorkem Durak, Halil Ertugrul Aktas, Vedat Cicek, Jinkui Hao, Ulas Bagci, Nilay S. Shah, Bo ZhouarXiv 2025 · DOI · arXiv

Introduction

Coronary artery calcium (CAC) scoring is a critical clinical metric for predicting cardiovascular disease (CVD) risk. While ECG-gated CT scans are the established gold standard for CAC scoring due to their ability to synchronize imaging with the cardiac cycle, they are costly and not routinely used for general screening. In contrast, non-gated chest CT scans are performed significantly more frequently for various indications but suffer from severe motion artifacts. These artifacts "smear" calcium deposits, leading to inaccurate Agatston scores and unreliable clinical assessments.

The Challenge of Motion in Non-gated CT

The primary obstacle in correcting motion artifacts in non-gated CT is twofold. First, there is a lack of large-scale, paired "motion-corrupted" and "motion-free" datasets required for supervised learning. Second, standard image restoration models, such as U-Nets, often prioritize global image smoothness, which can inadvertently "wash out" small, high-intensity calcium lesions—the very features essential for accurate scoring.

Methodology: The ProDM Framework

To address these challenges, the authors propose the Property-aware Progressive Diffusion Model (ProDM). This framework integrates a physics-informed data generation engine with a property-aware diffusion process.

1. CAC Motion Simulation Data Engine

To overcome the data scarcity issue, ProDM utilizes a physics-based simulation engine. Starting with clean, gated CT scans, the engine applies 21 diverse motion profiles—including jitters, drifts, and oscillations—to simulate realistic cardiac movement. These motion-perturbed states are then processed through Radon transforms to generate synthetic sinograms, which are reconstructed into motion-corrupted CT images. This approach ensures that the synthetic data captures the complex artifacts inherent in the CT reconstruction process.

Overview of the CAC motion simulation engine and comparison between simulated and real motion profiles.

2. Property-Aware Learning and Calcium Consistency

A key innovation of ProDM is the inclusion of a Property-aware loss function. Since the clinical Agatston score is non-differentiable, the researchers developed a differentiable Volume Score surrogate. This loss function forces the model to preserve the specific physical properties of calcium lesions, such as intensity and spatial distribution, during the restoration process. This prevents the over-smoothing common in traditional deep learning models.

3. Progressive Diffusion Backbone

ProDM employs a diffusion-based architecture that progressively refines a motion-corrupted image into a motion-free state. The model uses a 2.5D strategy, processing 3-slice patches to maintain essential 3D anatomical context while remaining computationally efficient. During training, a U-Net denoiser learns to predict the noise added to the motion-free patches. During inference, the model iteratively removes artifacts to reconstruct a high-fidelity, motion-corrected image.

The ProDM framework illustrating the training process with calcium consistency loss and the iterative inference process.

Experimental Results and Clinical Validation

The model was trained and evaluated using the Stanford COCA dataset and a large-scale synthetic benchmark of over 9,000 scans.

  • Quantitative Performance: ProDM significantly outperformed baseline models. It achieved a Mean Absolute Error (MAE) in Agatston scores of 9.39, a stark improvement over the 37.88 MAE of uncorrected scans. Notably, standard U-Nets performed worse than uncorrected scans (MAE 45.16) due to their tendency to over-smooth small calcifications.
  • Risk Stratification: The model demonstrated a 95.3% accuracy in classifying patients into correct clinical risk categories (e.g., Minimal, Mild, Moderate, Severe), which is vital for clinical decision-making.
  • Radiologist Evaluation: In a blinded study involving 30 real-world non-gated scans, clinical readers rated ProDM-corrected images significantly higher than original scans in terms of motion suppression (2.97 vs. 1.42) and overall clinical usability (2.78 vs. 1.44) on a 5-point Likert scale.

Conclusion and Future Directions

ProDM represents a significant step toward making routine, non-gated chest CTs a viable tool for accurate CAC scoring. By bridging the gap between low-cost routine imaging and gold-standard gated scans, this technology could enable widespread, opportunistic screening for cardiovascular disease. Future work will focus on optimizing inference speeds and expanding the motion simulation engine to encompass even more complex patient-specific movement patterns.

Summary updated

Abstract

Coronary artery calcium (CAC) scoring from chest CT is a well-established tool to stratify and refine clinical cardiovascular disease risk estimation. CAC quantification relies on the accurate delineation of calcified lesions, but is oftentimes affected by artifacts introduced by cardiac and respiratory motion. ECG-gated cardiac CTs substantially reduce motion artifacts, but their use in population screening and routine imaging remains limited due to gating requirements and lack of insurance coverage. Although identification of incidental CAC from non-gated chest CT is increasingly considered for it offers an accessible and widely available alternative, this modality is limited by more severe motion artifacts. We present ProDM (Property-aware Progressive Correction Diffusion Model), a generative diffusion framework that restores motion-free calcified lesions from non-gated CTs. ProDM introduces three key components: (1) a CAC motion simulation data engine that synthesizes realistic non-gated acquisitions with diverse motion trajectories directly from cardiac-gated CTs, enabling supervised training without paired data; (2) a property-aware learning strategy incorporating calcium-specific priors through a differentiable calcium consistency loss to preserve lesion integrity; and (3) a progressive correction scheme that reduces artifacts gradually across diffusion steps to enhance stability and calcium fidelity. Experiments on real patient datasets show that ProDM significantly improves CAC scoring accuracy, spatial lesion fidelity, and risk stratification performance compared with several baselines. A reader study on real non-gated scans further confirms that ProDM suppresses motion artifacts and improves clinical usability. These findings highlight the potential of progressive, property-aware frameworks for reliable CAC quantification from routine chest CT imaging.
BibTeX
@misc{gong2025prodm,
  author = {Xinran Gong and Gorkem Durak and Halil Ertugrul Aktas and Vedat Cicek and Jinkui Hao and Ulas Bagci and Nilay S. Shah and Bo Zhou},
  title = {{ProDM: Synthetic Reality-driven Property-aware Progressive Diffusion Model for Coronary Calcium Motion Correction in Non-gated Chest CT}},
  year = {2025},
  eprint = {2512.24948},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2512.24948},
}
Figure