A set of MATLAB tools for interacting with a normative biventricular pediatric cardiac shape atlas built from 93 healthy patients aged 10-21. This atlas is constructed at both end-diastole (ED) and end-systole (ES) using principal component analysis (PCA) and can be used to:
- Explore and visualize the modes of shape variation in a healthy pediatric population
- Quantify how a new patient's cardiac morphology deviates from healthy pediatric norms
- Project patient models onto the atlas to obtain shape mode Z-scores
- Compare pediatric cardiac shape across disease populations (e.g., Tetralogy of Fallot)
Standard clinical metrics (volumes, ejection fractions, wall thickness) provide a limited view of cardiac geometry. This atlas captures global and regional variation in ventricular size, shape, and function in a single, quantative framework - offering a richer characterization of pediatric cardiac morphology than scalar indices alone.
The atlas was derived from 93 healthy pediatric patients (ages 10–21; 62% male; mean age 15.4 ± 2.3 years) imaged at Rady Children's Hospital San Diego. CMR images were acquired at 1.5T using standard ECG-gated, breath-held SSFP cine sequences. Biventricular models were generated using an automated deep learning pipeline and refined by expert review.
The atlas was built using PCA on Procrustes-aligned, height-corrected biventricular surface meshes. 11 shape modes together explain 83.1% of total shape variance in the cohort. A non-height-corrected version is also provided for comparisons against atlases from other populations (e.g., adults).
Cohort summary:
| Characteristic | Value (n=93) |
|---|---|
| Age (years) | 15.4 ± 2.3 |
| Sex (male / female) | 62% / 38% |
| Height (cm) | 166.2 ± 11.9 |
| Weight (kg) | 62.5 ± 17.5 |
The atlas has been validated against a cohort of 65 pediatric Tetralogy of Fallot (ToF) patients. ToF patients showed significant deviations from the healthy pediatric mean, particularly in modes associated with right ventricular dilation and sphericity. The pediatric atlas better characterized the ToF cohort than an adult reference (mean Mahalnobis distance 4.18 vs. 4.74, p < 0.001).
The shape variations captured by each of the first 11 modes of the height-corrected atlas are visualized below (± 3 standard deviations from the mean shape):
| Modes | Anterior View | Basal View | Posterior View |
|---|---|---|---|
| 1 |  |
 |
 |
| 2 |  |
 |
 |
| 3 |  |
 |
 |
| 4 |  |
 |
 |
| 5 |  |
 |
 |
| 6 |  |
 |
 |
| 7 |  |
 |
 |
| 8 |  |
 |
 |
| 9 |  |
 |
 |
| 10 |  |
 |
 |
| 11 |  |
 |
 |
Prerequisites
- MATLAB R2021a or later (tested on R2025a)
- The Statistics and Machine Learning Toolbox is needed for some analyses
Clone the repository:
git clone https://github.com/cmrg-lab/Pediatric-Atlas.git
cd Pediatric-AtlasThe atlas .mat files are not included in the repository due to file size. Download them from one of two sources:
- Google Drive: Height-corrected atlas and non-height-corrected atlas.
- Zenodo: both atlas files and additional MATLAB files needed for some of the scripts are archived at the Zenodo repository linked to this project.
Place the downloaded .mat files into the Data/ directory before running scripts.
For adult atlas comparisons: A biventricular adult reference atlas (derived from the UK Biobank) can be downloaded from the Cardiac Atlas Project website.
Full documentation for all scripts is in Scripts/README.md. The core workflows are:
1. Atlas Creation
To create a new atlas, 3D model files must first be added in a Models/ folder. The path in extractPoints.m should be updated. These scripts extract and align models, and creates the atlas using PCA.
% From the Scripts/directory:
run('extractPoints.m')
run('alignPoints.m')
run('genAtlas.m')2. Explore and visualize the atlas
Loads the atlas and visualizes the mean shape and shape mode variations at ED and ES. Plots the variance explained by the shape modes. Derives z-scores for models used to generate the atlas. Analyzes correlations between atlas shape modes and clinical metrics. Cohorts within the atlas separated by sex or bmi can also be compared mode-wise.
run('atlasViewer.m')
run('plotVarianceExplained.m')
run('genZScores.m')
run('analyzeCorrelations.m')
run('compareCohorts.m')3. Project a new model onto the atlas
Provide your own biventricular model as an input (an example one is included in Models/). This function will return per-mode Z-scores indicating morphological deviation from the healthy pediatric mean.
run('projectOntoAtlas.m')4. Compare populations
Computes Mahalanobis distances and mode-wise Z-scores distributions across separate cohorts.
run('distanceAnalysis.m')├── Data/
│ └── MeshFiles
├── Images/
│ ├── disease_comparison
│ └── pedatlas_characterization
├── Models/
├── Scripts/
│ ├── helpers
│ └── README.md
├── .gitignore
├── LICENSE
└── README.mdIndividual patient data (CMR images, clinical data, point clouds) can be made available through the Cardiac Atlas Project subject to a Data Use Agreement. The CAP IDs for both the pediatric healthy and diseased participants used to in this work are listed on the Pediatric Atlas page for registered users.
Anna Qi - anqi@ucsd.edu
If you use the pediatric atlas or these tools in your research, please cite:
Qi, A. et al. (2025). A Pediatric Cardiac Shape Atlas: Insights into the Structure of Young Healthy Hearts. In: Chabiniok, R., Zou, Q., Hussain, T., Nguyen, H.H., Zaha, V.G., Gusseva, M. (eds) Functional Imaging and Modeling of the Heart. FIMH 2025. Lecture Notes in Computer Science, vol 15673. Springer, Cham. https://doi.org/10.1007/978-3-031-94562-5_14
A follow-up manuscript with an expanded cohort (n=93) and Tetralogy of Fallot validation is under review. Please check back for an updated citation.
Distributed under the Apache 2.0 License. See LICENSE for details.