nnInteractive: A Breakthrough in AI-Driven 3D Interactive Segmentation

The Helmholtz Imaging Engineering and Support Unit at DKFZ, together with partners from DKFZ, has launched nnInteractive, a state-of-the-art AI-driven 3D interactive segmentation tool designed to revolutionize volumetric annotation in imaging. It is the first method integrated into widely used image viewers (e.g., Napari, MITK), ensuring broad accessibility for real-world clinical and research applications.

Why nnInteractive Stands Out

Current interactive segmentation models, like Meta’s SAM, excel in 2D applications but struggle with volumetric consistency, usability, and interactivity in the medical domain. nnInteractive addresses these limitations by introducing:

• Diverse Prompting Options: Users can interact with the model using points, scribbles, bounding boxes, and a novel lasso prompt, ensuring precise segmentation control.
• Seamless 3D Segmentation: Unlike conventional models that require per-slice annotations or unintuitive 3D bounding boxes, nnInteractive transforms intuitive 2D interactions into full 3D segmentations.
• Unmatched Generalization: Trained on 120+ volumetric datasets, including CT, MRI, PET, and 3D Microscopy, nnInteractive adapts to a wide range of structures and modalities.
• Broad Accessibility for Real-World Applications: nnInteractive is embedded in widely used image viewers, making it accessible for both clinical and research workflows.

A New Era in 3D Interactive Segmentation

Extensive benchmarking demonstrates that nnInteractive far surpasses existing methods, setting a new state-of-the-art standard for AI-driven volumetric segmentation. The tool’s efficiency and precision significantly reduce annotation time while maintaining expert-grade accuracy. Whether for tumor delineation in MRI scans or microscopic cellular analysis, nnInteractive adapts to user intent with minimal effort.

nnInteractive is now available for Napari and MITK users, bringing next-generation AI-powered segmentation directly into established workflows.

Links

Paper (submitted on arXiv)
Napari Plugin
MITK Integration
Python backend