riccardolunelli/xECG_base_model_v1

This model has been pushed to the Hub using the PytorchModelHubMixin integration:

• Code: https://github.com/dlaskalab/bench-xecg/
• Paper: BenchECG and xECG: a benchmark and baseline for ECG foundation models

To use this model, first copy to your code xECG.py and ensure the requirements in requirements.txt are inatalled and then:

from xECG import xECG

model = xECG.from_pretrained("riccardolunelli/xECG_base_model_v1")

Important: when you pass the signal to the model ensure the leads respect the following order: ['i', 'ii', 'iii', 'avr', 'avl', 'avf', 'v1', 'v2', 'v3', 'v4', 'v5', 'v6']. For applications where less leads are available, just set them to zero.

This model returns both the set of final patches and a pooled representation.

We provide a class for ECG task with a head on the pooled representations, downstream_models.xECGClassification: use this for signal level class, e.g. classification, regression...

Moreover we provide another class where the head is appended to the each patch, downstream_models.xECGFeatureClassification: use this for task like segmentation (head output of the same size of the original patch, 25) or for heartbeat level classification.