MONAI Label – An emergent standard for AI medical imaging

Monai Label has arrived to leverage the integration of AI models in medical imaging, with the objective of continuously reducing the need for manual annotations, a time-consuming task, while abstracting complex models through simple interfaces, suitable for non-experts. It has received a growth of interest, and is being adopted by companies such as Google, that has integrated it and all the MONAI ecosystem in its Medical Imaging Suite.





MONAI is an open-source project that aids the build and application of AI models in all medical imaging stages, including production, constantly evolving to improve medical care. The MONAI ecosystem includes two more categories:

  • MONAI Core, that ranges from functionalities, such as pre, post-processing methods for images, architectures and metrics, the definition of state-of-the-art pipelines or even the use of AutoML. 
  • MONAI Deploy, oriented for testing, packaging and deployment of the AI models, supporting also clinical workflows, aimed for clinical production. 

The advantages of using the MONAI Label Platform are very clear:

  • AI pipelines are abstracted from practitioners.
  • Common ground to generate evidence for medical research and ease of reproducibility. 
  • Reduced workload for annotators.
  • Faster annotation time, which leads to a more rapid diagnosis.
  • Customizable services, creation of new AI bundles.
  • DICOM capabilities. 

The workflow in MONAI Label follows an Active Learning Loop, that is repeated while we don’t have a perfectly adapted model for our use case. It is characterized with the following steps:

  • Fetch of a dataset image to be annotated: given several strategies, that can be customized according to the use case. It might pretend to fetch the images that scored the lowest given an accuracy metric.
  • Annotation process: in the beginning, this requires a more manual intervention, while later in the training cycle, it is possible that we already have a somewhat accurate model that can be used to do semi-automatic annotations. 
  • Training of the model(s).

Currently, at BMDSoftware, we are interested in how we can integrate MONAI Label in our medical imaging software, such as PACScenter Viewer and our archive Dicoogle.

Since MONAI Label also supports DICOM images in its training loop, we aim at creating a full architecture that can communicate by taking advantage of the DICOM standard. We look forward to giving more information about this process! In the meantime, if you are interested in taking advantage of MONAI Label and a fully integrated medical imaging workflow, contact us at

Comments are closed.