Uploaded July 2026 | Updated November 2026, 3 hours ago
Speaker: Fabian Isensee (German Cancer Research Center)
Abstract: Manual annotation of three-dimensional biomedical image data remains a major bottleneck in research and clinical practice. nnInteractive addresses this challenge with a universal interactive 3D segmentation framework that generates coherent volumetric masks from intuitive 2D user prompts. Unlike existing tools that often require slice-wise annotation, cumbersome 3D interactions, or task-specific models, nnInteractive supports points, scribbles, bounding boxes, and freeform lasso prompts placed on ordinary 2D image views while producing full 3D segmentations. Trained at scale on diverse volumetric datasets spanning radiology, microscopy, and other biomedical imaging modalities, the model generalizes to unseen structures, anatomies, species, and acquisition settings. In extensive evaluations, nnInteractive outperforms established 2D and 3D interactive segmentation baselines and reaches high-quality segmentations with substantially less combined user and compute time. With integrations into platforms such as Napari, MITK, and OHIF, nnInteractive is designed to make accurate 3D segmentation practical for everyday biomedical image analysis
Speaker: Fabian Isensee (German Cancer Research Center)
Abstract: Manual annotation of three-dimensional biomedical image data remains a major bottleneck in research and clinical practice. nnInteractive addresses this challenge with a universal interactive 3D segmentation framework that generates coherent volumetric masks from intuitive 2D user prompts. Unlike existing tools that often require slice-wise annotation, cumbersome 3D interactions, or task-specific models, nnInteractive supports points, scribbles, bounding boxes, and freeform lasso prompts placed on ordinary 2D image views while producing full 3D segmentations. Trained at scale on diverse volumetric datasets spanning radiology, microscopy, and other biomedical imaging modalities, the model generalizes to unseen structures, anatomies, species, and acquisition settings. In extensive evaluations, nnInteractive outperforms established 2D and 3D interactive segmentation baselines and reaches high-quality segmentations with substantially less combined user and compute time. With integrations into platforms such as Napari, MITK, and OHIF, nnInteractive is designed to make accurate 3D segmentation practical for everyday biomedical image analysis










