MILA_public/medax_pipeline
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 2 Wiki Activity

2 Releases 2 Tags

RSS Feed
  • release faf1ab1c99

    MeDaX Pipeline v1.0.0 Stable

    gebhardtt released this 2025-04-24 15:37:50 +02:00 | 4 commits to main since this release

    Release Notes - MeDaX Pipeline v1.0.0 (Initial Release)

    Overview

    We are pleased to announce the first release of our MeDaX pipeline, enabling seamless transformation of hospital FHIR data into a Neo4j graph database, thereby enabling innovative medical data exploration at german Data Integration Centres. This pipeline facilitates:

    • Local setup for controlled data handling
    • Efficient data exploration and analysis across connected healthcare data
    • Improved query capabilities for complex healthcare relationships

    Key Features

    Easy Deployment

    • Containerised setup using Docker Compose for straightforward deployment and configuration
    • Flexible configuration options for connecting to any FHIR server through customisable URL and proxy settings
    • Built-in support for Open Access HAPI FHIR server, enabling immediate testing and validation
    • Simple environment variable configuration through .env file

    Data Processing

    • Validated with real hospital data, ensuring production readiness
    • Implemented property convolution and reference path reduction for efficient graph size reduction
    • Manually curated graph schema to enable semantic enrichment with ontological information, currently using BioLink (BioCypher default)
    • Automated schema extension for unspecified input data to maintain compatibility with evolving FHIR resources
    • Batching of input data to process large-scale data sets
    • Support for patient-centric data retrieval using FHIR's $everything operation

    Extensibility

    • Developed using BioCypher framework, enabling modular architecture
    • Support for additional data source adapters, allowing future expansion to different resources
    • Flexible architecture consists of:
      1. FHIR Import module for data retrieval
      2. Reference Processor for relationship management
      3. Property convolution for complexity reduction
      4. BioCypher Adapter for Neo4j integration

    Installation & Usage

    For detailed installation instructions and information how to cite this work, please refer to the README document included in the repository. Basic setup involves:

    1. Clone the repository
    2. Configure the environment variables
    3. Run docker compose up --build
    4. Access Neo4j at http://localhost:8080/

    Known Issues and Limitations

    Performance Considerations

    • Processing large hospital datasets requires significant computational resources and time due to data complexity
    • Complete pipeline restart required when modifying graph reduction parameters
    • Initial load time may be extensive for large datasets

    User Interface

    • Currently limited to standard Neo4j browser interface
    • Default UI may not be optimal for specialised healthcare use cases

    Technical Requirements

    • Docker and Docker Compose
    • Sufficient storage and computational resources for processing FHIR datasets
      • Currently the memory is the bottleneck, 12GB RAM recommended for a batch size of 200
    • Network access to FHIR server

    Next Steps

    We are actively working on:

    • Testing large-scale data sets
    • Integrating fitting visualisation interfaces
    • Implementing incremental update capability
    • Integration with BRO (Biomedical Resource Ontology) through curated schema mapping for standardised terminology
    • Performance optimisations for handling larger datasets

    Feedback and Contributions

    We welcome feedback, bug reports, and contributions! Please submit issues and pull requests through our repository.

    © 2025 MeDaX Project Team | Released under MIT License

    Downloads