LS-EMVAE hands-on#
Different ECG leads as complementary views#
The lsemvae_ecg example treats individual ECG leads as separate modalities. Its example README describes representation pretraining on multi-lead recordings and fine-tuning using a subset of leads for a pulmonary hypertension task.
Run the quick configuration#
From the Cardiac repository root:
python -m examples.lsemvae_ecg.main --cfg examples/lsemvae_ecg/configs/quick.yaml
Without a dataset path, this generates synthetic cohorts. It does not reproduce the published clinical findings.
Proposed participant prompts#
Identify configuration. Identify where lead names and fine-tuning lead subsets are configured.
Compare training stages. Compare the meaning of pretraining without labels and fine-tuning with an endpoint.
Discuss limited lead availability. Discuss why a model might need to work with fewer available leads.
Examine output descriptions. Examine the output descriptions in
examples/lsemvae_ecg/README.md, including lead importance and ablation.
Optional configuration exploration#
The example README documents an optional comparison of lead sets:
python -m examples.lsemvae_ecg.main --leads I,I+II,I+II+III
Warning
Configuration experiments are optional. Confirm the supported lead-set comparisons with your facilitator before running them.
Getting started · CardioVAE hands-on · Results and interpretation.