Medial EarlySign's Machine Learning infrastructure and development toolkit for building, validating, and deploying predictive clinical models using Electronic Medical Records (EMR).
This award-winning framework used in production across multiple sites and recognized in the CMS AI Health Outcomes Challenge offers an end-to-end solution for high-stakes medical AI.
The repository provides the:
- Wiki: a complete knowledge base for our proprietary and open sourced ML infrastructure and workflows.
- medpython: The infrastructure library to train, test, deploy our models
- MR_Tools: Utilities that uses medpython library and were used directly to develop our products
Our system is designed to handle the unique challenges of clinical data, providing robust tools for every phase of model development:
| Phase | Component & Focus | Key Capabilities |
|---|---|---|
| Ingestion | RepProcessors (Data Preparation) |
Cleans, aggregates, and standardizes raw EMR signals into meaningful, temporal records. |
| Modeling | FeatureGenerators & MedAlgo |
Transforms data into complex features (e.g., embeddings) and handles training, calibration, and feature importance analysis (e.g., XGBoost, MedPredictor). |
| Deployment Prep | PostProcessors |
Essential steps for production readiness: score calibration, explainability analysis, and fairness adjustment. |
| Validation | Medial Tools (Evaluation) |
Utilities like AutoTest and bootstrap_analysis ensure model quality, robustness, and performance over time. |
Start your exploration by reviewing the high-level [Wikimedial Overview] and the [Installation Guide] for setting up the environment and the Python API.
| Section | Description | Link |
|---|---|---|
| Wikimedial Overview | The essential starting point for all users. | View Overview |
| Tutorials | Deep dive into the core technical components. | Tutorials |
| Installation | Environment setup (Python API & C++ Library). | View Installation |