PyHealth Healthcare AI

About

PyHealth is a specialized toolkit designed to streamline the complex workflow of healthcare AI development, from initial data processing to clinical deployment. It offers a modular five-stage pipeline that handles electronic health records (EHR), medical coding translations, and physiological signal analysis with high efficiency. By providing access to standardized healthcare datasets like MIMIC and eICU alongside over 33 specialized models—including healthcare-specific architectures like RETAIN and SafeDrug—it empowers developers to build reliable prediction models for mortality, hospital readmission, and medication recommendation while ensuring clinical validity through fairness and interpretability tools.

Key Features

• Support for 10+ healthcare datasets including MIMIC-III/IV, eICU, and OMOP
• 2,066 GitHub stars
• Seamless translation between medical coding systems like ICD-9/10, NDC, and RxNorm
• Extensive library of 33+ healthcare-specific and general deep learning models
• Modular five-stage pipeline for standardized data loading, task definition, and model training
• Advanced clinical evaluation tools for fairness, calibration, and interpretability

Use Cases

• Predicting clinical outcomes such as patient mortality, length of stay, or hospital readmission
• Developing medication recommendation systems with drug-drug interaction constraints
• Classifying physiological signals such as sleep stages from EEG or seizure detection