CardioVAE hands-on

CardioVAE hands-on#

Combining ECG and chest X-ray information#

The cardiovae_multimodal example pairs ECG and chest X-ray modalities. Its repository README describes product-of-experts multimodal variational representation learning, followed by fine-tuning on a haemodynamic endpoint.

Run the quick configuration#

From the Cardiac repository root:

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

Warning

With no dataset path, the example uses generated cohorts. Any resulting metric is a synthetic pipeline check, not a clinical result.

Proposed participant prompts#

  1. Identify the input modalities. Identify the two input modalities.

  2. Find the configuration file. Find the configuration file examples/cardiovae_multimodal/configs/quick.yaml.

  3. Identify the endpoint and saved results. Identify the endpoint and locate the saved results.

  4. Discuss missing modalities. Discuss what could happen when one modality is missing.

For experiment-specific data requirements and output file names, see examples/cardiovae_multimodal/README.md in the Cardiac repository.

Note

These activities are provisional; the final session exercises are still to be confirmed.

Getting started · LS-EMVAE hands-on · Results and interpretation.