Cardiac: Getting Started

Two practical examples

Example	Inputs	What it demonstrates
CardioVAE	Chest X-ray + ECG	Multimodal representation learning and a cardiac endpoint
LS-EMVAE	Multi-lead ECG	Lead-specific representations, transfer to a subset of leads and endpoint prediction

Both can generate synthetic cohorts when no data path is supplied. They are pipeline checks, not clinically validated experiments.

Prepare the repository

Use the KaleCardiac repository with Python 3.10, 3.11 or 3.12. Install a compatible PyTorch build first; the core installation includes PyKale. Follow the Cardiac Quick Start to obtain a checkout, then run from its root:

pip install -e .
python -c "import kalecardiac, torch, kale; print(kalecardiac.__version__)"

Optional extras cover ECG file formats (.[ecg]), imaging (.[imaging]), interpretation (.[interpret]) and development (.[dev]). Starter examples do not require participant datasets.

Avoid mixing the checkout with an older installed package. Confirm the supported session environment with the organisers before setup.

Choose a synthetic example

CardioVAE: chest X-ray + ECG

python -m examples.cardiovae_multimodal.main --cfg examples/cardiovae_multimodal/configs/quick.yaml

Continue to CardioVAE Hands-on.

LS-EMVAE: multi-lead ECG

python -m examples.lsemvae_ecg.main --cfg examples/lsemvae_ecg/configs/quick.yaml

Continue to LS-EMVAE Hands-on.

The quickstart describes CPU-capable examples that write predictions, metrics and configuration records under outputs/. Exact runtime depends on hardware and dependencies.

Results and technical references

See Results and Interpretation. For experiment-specific requirements, consult the repository’s quickstart, examples overview, CardioVAE README and LS-EMVAE README.

Note

The software continues to evolve. Final session exercise selection remains to be confirmed. Synthetic results are not clinical findings.