Included Models

The models in our catalog all fit into the criteria below:

• Models developed or trained using machine learning, deep learning, foundation-model, or generative-AI methods for a cardiovascular-specific task
• Have a cardiovascular-specific output, target phenotype, intended application, or primary validated use. Input does not neccesarily have to be cardiology-specific
• Cardiovascular-relevant tasks including, but not limited to, detection, classification, diagnosis, prognosis, functional measurement, segmentation, localization, quantification, risk or outcome prediction, treatment-response prediction, procedural planning, phenotyping, information extraction, reconstruction, representation learning, or generation
• The following are not accepted:
• Raw datasets
• Clinical tools or APIs without an identifiable qualifying underlying model
• Non-learned classical algorithms
• Models without a cardiovascular-specific target, intended use, or validated application
• Agent frameworks alone—they are not a specific pre-trained model
• Generic LLMs fed a cardiovascular prompt (although ones fine-tuned on cardiovascular-specific text are okay)

We have tried to include all applicable models with either public training code or model weights, and have selected some paper-only models.