MONAI Label: A framework for AI-assisted interactive labeling of 3D medical images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Annotated datasets are crucial for supervised machine learning models.
• Manual data annotation is time-consuming and expensive, hindering AI development.
• A need exists for efficient tools to accelerate the creation of annotated medical datasets.

Purpose of the Study:

• Introduce MONAI Label, a framework to reduce annotation time for radiology datasets.
• Enable researchers to develop and deploy AI annotation applications.
• Facilitate collaboration between researchers and clinicians.

Main Methods:

• Developed MONAI Label as a free, open-source framework.
• Integrated support for interactive and non-interactive AI labeling applications.
• Included active learning strategies to enhance segmentation algorithm training.
• Enabled deployment of AI apps as services compatible with 3D Slicer and OHIF frontends.

Main Results:

• Demonstrated significant reductions in annotation time using MONAI Label's interactive model.
• Validated performance on two public radiology datasets.
• Provided plug-and-play labeling applications for immediate use.

Conclusions:

• MONAI Label effectively addresses the bottleneck of manual annotation in radiology.
• The framework empowers researchers to build and share AI-driven annotation tools.
• Facilitates faster development and deployment of AI models in medical imaging.