Awakening ImageCAS:
Automated Coronary Ostia Localization Framework and Centerlines Open Dataset

Matteo Leccardi1 ORCID Google Scholar, Marco Morandini1, Michele Brembilla1,
Alessandro Nicosia1, Andrea Penserini1, Christian Brandini1,
Marco Marcon1 ORCID Google Scholar, Andrea Moglia1 ORCID Google Scholar,
Luca Mainardi1 ORCID Google Scholar, Pietro Cerveri1,2 ORCID Google Scholar

1: Politecnico di Milano, 2: Università di Pavia
For correspondence: matteo.leccardi@polimi.it

Paper arXiv GitHub Dataset (Zenodo) Model weights
Overview of the work: datasets, annotation protocol, benchmark and public release

Abstract

Accurate localization of coronary ostia and the coronary artery centerline tree supports diagnostic assessment, pre-operative planning and intra-operative navigation, and automated coronary CT angiography (CCTA) analysis pipelines more broadly. However, no publicly available dataset has been specifically curated for coronary ostia localization, and no comparably large-scale public dataset of coronary centerline graphs exists either. This paper extends ImageCAS, the largest publicly available CCTA dataset to date, with two new sets of annotations covering its complete 1000-patient cohort. Coronary ostia landmarks are annotated manually in the first 500 volumes by six annotators, organized in rotating groups of three and consolidated by consensus (mean distance of the individual annotators from the consensus: 1.45 mm for the right and 1.95 mm for the left ostium), and semi-automatically in the remaining 500, by correcting the candidates generated by a model trained on the first batch, which required about 5 person-hours instead of the approximately 144 of fully manual annotation. These annotations are reconciled with the operational definitions of the ostium of CAT08 (32 annotated scans) and ASOCA (40 annotated scans). Coronary centerline graphs, rooted at the annotated ostia, are extracted from the ImageCAS segmentation masks by a deterministic minimal-path tree-growing procedure, completed in 999 of the 1000 patients, and validated against the expert centerlines of ASOCA, where they cover 99.5% of the reference length (median) at a mean distance of 0.22 mm. With 1000 patients, this is, to the authors' knowledge, the largest curated dataset on coronary centerline to date.

Building on these annotations, an open-weight benchmark for automatic coronary ostia localization is established, comprising SwinUNETRv2, ResNet50, ResNet101, MedNeXt and nnResUNet models trained with a differentiable coordinate-regression loss and evaluated on a test set of 58 volumes from the three datasets. Localization errors are graded against thresholds derived from the ostial anatomy and from the variability between annotators. The best models, SwinUNETRv2 and nnResUNet, localized all the ostia without any failure, with median errors of 0.74 mm (right) and 0.85 mm (left) for SwinUNETRv2 and of 0.70 mm and 0.87 mm for nnResUNet.

The extended dataset annotations, the annotation protocol, the code and the model weights are made publicly available, providing the community with a common, reproducible reference point for coronary ostia localization and centerline extraction, in place of the private datasets and closed implementations that have characterized much of the prior work in this area.

At a glance

1000ImageCAS patients annotated
6annotators, rotating groups of three
~5 hinstead of ~144 h for the semi-automatic half
999centerline graphs out of 1000 patients
5open-weight benchmark models

Ostia annotations

The coronary ostia are the points where the left and right coronary arteries originate from the aorta. ImageCAS provides 1000 CCTA volumes with coronary segmentation masks but no ostium annotation; we add one left and one right ostium landmark to every volume.

Ostia annotation example (coming soon)
Example of left and right ostium landmarks on a CCTA volume.

From annotation to benchmark

How the ostia dataset was assembled and how the benchmark was trained and evaluated, step by step: the two annotation stages, the assistant model, the test set, the benchmark models and the ablation studies. Use the chapter buttons to jump to a step.

0:00
Schematic illustration: every square stands for one CCTA volume, and the order of the annotator groups is illustrative.

Centerline graphs

Centerline graphs rooted at the annotated ostia are extracted from the ImageCAS segmentation masks with a deterministic minimal-path tree-growing procedure. They are validated against the expert centerlines of ASOCA.

Centerline extraction pipeline
Centerline extraction pipeline: from the segmentation mask and the annotated ostia to one rooted tree per side, with per-node position, radius, topological type and side.

Centerline tracking, in brief

A simplified 2D illustration of how a centerline tree is traced: from the ostia placed by the model, through the distance to the vessel wall, to the rooted graph.

0:00
Illustration on a synthetic vessel tree. The actual procedure works in 3D on the ImageCAS masks and includes further refinement steps that are left out here.

Centerline tracking, in detail

The same procedure with every step of the pipeline: mask repair, ostium and gap bridging, distance and cost maps, the two minimal-cost fronts, the two growing passes, tip trimming, smoothing and the final typed graph. Use the chapter buttons to jump to a step.

0:00
Illustration on a synthetic vessel tree with a 0.5 mm voxel, using the parameters of the procedure (cost exponent, 5 mm minimum branch length, 2 mm trimming, 2 mm smoothing sigma). The real procedure works in 3D on the ImageCAS masks.

Benchmark

We release open-weight models for automatic ostia localization: SwinUNETRv2, ResNet50, ResNet101, MedNeXt and nnResUNet, trained with a differentiable coordinate-regression loss and evaluated on a test set of 58 volumes from ImageCAS, CAT08 and ASOCA. Errors are graded against thresholds derived from the ostial anatomy and from the variability between annotators.

The best models localized all the ostia without any failure. Median errors are 0.74 mm (right) and 0.85 mm (left) for SwinUNETRv2, and 0.70 mm and 0.87 mm for nnResUNet.

Resources

Citation

Paper

@article{leccardi_awakening_imagecas,
  title   = {Awakening ImageCAS: Automated Coronary Ostia Localization Framework and Centerlines Open Dataset},
  author  = {Leccardi, Matteo and Morandini, Marco and Brembilla, Michele and Nicosia, Alessandro and
             Penserini, Andrea and Brandini, Christian and Marcon, Marco and Moglia, Andrea and
             Mainardi, Luca and Cerveri, Pietro},
  note    = {Under review}
}

Dataset (Zenodo)

@dataset{leccardi_awakening_imagecas_dataset,
  title     = {Awakening ImageCAS: coronary ostia and centerline annotations for ImageCAS},
  author    = {Leccardi, Matteo and Morandini, Marco and Brembilla, Michele and Nicosia, Alessandro and
               Penserini, Andrea and Brandini, Christian and Marcon, Marco and Moglia, Andrea and
               Mainardi, Luca and Cerveri, Pietro},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.XXXXXXX},
  note      = {DOI placeholder}
}

Original ImageCAS dataset

Our annotations extend ImageCAS: please also cite the original dataset paper.

@article{zeng2023imagecas,
  title   = {ImageCAS: A large-scale dataset and benchmark for coronary artery segmentation based on computed tomography angiography images},
  author  = {Zeng, An and Wu, Chunbiao and Lin, Guisen and Xie, Wen and Hong, Jin and Huang, Meiping and
             Zhuang, Jian and Bi, Shanshan and Pan, Dan and Ullah, Najeeb and Khan, Kaleem Nawaz and
             Wang, Tianchen and Shi, Yiyu and Li, Xiaoming and Xu, Xiaowei},
  journal = {Computerized Medical Imaging and Graphics},
  volume  = {109},
  pages   = {102287},
  year    = {2023},
  doi     = {10.1016/j.compmedimag.2023.102287}
}