The MONAI Deploy Application SDK offers a framework and associated tools to design, verify and analyze the performance of AI-driven applications in the healthcare domain.

It contains the following elements:

  • Pythonic framework for app development

  • A mechanism to package an app in a “MONAI Application Package” (MAP), a container image that is self-describing

  • A mechanism to locally run a MAP via App Runner

  • A set of sample applications

  • API documentation

To further accelerate the development of medical imaging AI inference application with DICOM imaging network integration, a set of domain specific functionalities are provided by this Application SDK,

  • generic application classes to automate the inference with MONAI Bundle

  • DICOM study parsing and selection classes, as well as DICOM instance to volume image conversion

  • DICOM instance writers to encapsulate AI inference results in these DICOM OID,

    • DICOM Segmentation

    • DICOM Basic Text Structured Report

    • DICOM Encapsulated PDF

    • and more to come


  • Starting with release v0.6, the core modules are dependent on the Python SDK of Nvidia Holoscan, for its enterprise grade optimized libraries for data processing and AI, and core microservices to run surgical video, ultrasound, medical imaging, and other applications anywhere, from embedded to edge to cloud. A list of its modules are directly exposed through this SDK’s namespace, monai.deploy. Domain specific operators from previous releases, e.g. DICOM parsing and writing, remain naive to this App SDK, and have been updated to be compatible with the holoscan based core modules.

  • The software and the imaging AI results generated are for research use only, and per applicable laws and regulations, are not for clinical use.

  • The App SDK, and specifically the DICOM instance writers, rely upon the underlying operating system to provide accurate and consistent time, and have no direct control over how the OS performs time synchronization, e.g. NTP on Ubuntu. It would be incumbent on the developer to appropriately use the available system time service and reflect the correct time in those objects properly. Additionally, the DICOM instance generated by the App SDK built-in classes set the Timezone Offset From UTC with the underlying operating system’s timezone setting.