Skip to content


The model registration and evaluation platform for ABCDS Oversight @ Duke


Algorithms used for clinical decision support, whether for diagnostic, or therapeutic use cases, have significant implications for quality and safety. Despite this, regulation of these algorithms and models (what the FDA calls “Software as a Medical Device) is still in the early stages. In order to both foster a more responsible environment for clinical algorithm use and prepare for potential formal regulation, a new governance body called Algorithm-Based Clinical Decision Support (ABCDS) was formed at Duke. As the committee began putting in place a formal model registration & review process, ABCDS soon identified a need for a tool that would simplify the process for both researchers and review committee members. 

Model Registration Diagram
ABCDS Oversight Model Checkpoints


Registration of all the facts related to an algorithm, including its clinical and operational stakeholders, its indication, its methodology, its version, among other metadata


Transparency of the process to those proposing or submitting algorithms

Retention + Provenance

Retention and provenance of records and artifacts related to the algorithms themselves as well as the ABCDS governance process around the algorithms


Continuous surveillance of registered algorithms


Working closely with ABCDS stakeholders, Crucible has developed the Machine Intelligence Tracking Platform (MITRA) for algorithmic governance at Duke. MITRA leverages existing Duke systems to ensure the model registration process is intuitive and familiar for Duke researchers. 

  • Public and Duke-internal catalog views of registered models
  • User profile functionality to track and update your registered models
  • Review and comment tool for checkpoint review committee
Architecture of a previous MITRA iteration
Developed with