Enhancing Plaque Segmentation in CCTA with Prompt- based Diffusion Data Augmentation

Ruan Yizhe, Xuangeng Chu, Ziteng Cui, Yusuke Kurose, JUNICHI IHO, Yoji Tokunaga, Makoto Horie, YUSAKU HAYASHI, Keisuke Nishizawa, Yasushi Koyama, Tatsuya Harada

Abstract: Coronary computed tomography angiography (CCTA) is essential for non-invasive assessment of coronary artery disease (CAD). However, accurate segmentation of atherosclerotic plaques remains challenging due to data scarcity, severe class imbalance, and significant variability between calcified and non-calcified plaques. Inspired by DiffTumor’s tumor synthesis and PromptIR’s adaptive restoration framework, we introduce PromptLesion, a prompt-conditioned diffusion model for multi-class lesion synthesis. Unlike single-class methods, our approach integrates lesion-specific prompts within the diffusion generation process, enhancing diversity and anatomical realism in synthetic data. We validate PromptLesion on a private CCTA dataset and multi-organ tumor segmentation tasks (kidney, liver, pancreas) using public datasets, achieving superior performance compared to baseline methods. Models trained with our prompt-guided synthetic augmentation significantly improve Dice Similarity Coefficient (DSC) scores for both plaque and tumor segmentation. Extensive evaluations and ablation studies confirm the effectiveness of prompt conditioning.

Submission Length: Regular submission (no more than 12 pages of main content)

Changes Since Last Submission:

We would like to extend our sincere gratitude to the Area Chair and the reviewers for their insightful feedback and constructive suggestions, which have significantly improved the quality and clarity of our manuscript.

We are pleased to confirm that we have thoroughly addressed all the required minor revisions. Crucially, we have implemented all three primary recommendations outlined by the Area Chair to ensure the paper's reproducibility, transparency, and consideration of broader impacts.

Below is a summary of the key changes:

1. Reproducibility Package:

In line with the AE's request, we have now included a public link to our full implementation, including preprocessing scripts, model weights, training & evaluation codes. The link is provided in a footnote on the first page of the main paper.

2. Clear Statement on CCTA Dataset and Benchmark Protocol:

To address the AE's request for transparency, we have now fully documented our CCTA data and established a clear protocol for comparison. Specifically, in Section 4.1 ("Datasets"), we have detailed our private dataset, including patient numbers, splits, annotation procedure, and IRB approval. Additionally, we introduced a new dedicated Section 4.3 ("Benchmarking Protocol for CCTA") that outlines the requirements for fair future comparisons.

3. Broader Impact / Clinical Considerations:

To address the omission noted by the reviewers and the AE, we have added a new section titled "Broader Impact and Clinical Considerations" following the Conclusion. This section discusses the intended use of our method as a training-time tool, potential clinical risks, and data privacy considerations.

We believe these revisions fully satisfy the conditions for acceptance. We thank you once again for your valuable guidance in strengthening this work and are confident that the manuscript is now ready for publication in TMLR.

Code: https://github.com/RuanYizhe-77/PromptLesionTMLR

Assigned Action Editor: ~Gustavo_Carneiro1

Submission Number: 4737