🚧 Under construction

The MITK v2026.06 release notes are currently under construction.

News for MITK Workbench users

We are proud to announce the release of MITK v2026.06. Thank you to everyone who contributed!

Over the past six months, we resolved around 120 tasks across approximately 1,000 commits. 🚀

The complete changelog can be found at the end of these release notes. As always, we'll start by highlighting some of the most notable improvements and new features in this release.

🚀 nnInteractive v2.5

We integrated the latest version of the groundbreaking nnInteractive segmentation tool and are already prepared for upcoming model checkpoints as they become available. Compared to nnInteractive v1, the new version delivers improved speed, higher segmentation quality, and introduces long-awaited support for remote sessions.

Multiple MITK instances can now connect to a shared GPU server, enabling lightning-fast interactive 3D segmentation even on lightweight laptops and Macs.

We've also significantly optimized MITK's integration of nnInteractive to minimize memory consumption and maximize performance. Installing or updating nnInteractive is now just a single click, and a range of new configuration options let you tailor nnInteractive to your segmentation workflow.

The scribble interaction of nnInteraction is used to quickly segment several organs and vertebra.

🌈 TotalSegmentator v2.14

We completely redesigned our TotalSegmentator integration to make it easier than ever to set up and use. Install it with a single click, choose a segmentation task, and start it with another click. Once the results are ready, simply review and confirm them.

If you have a TotalSegmentator license, you can also enter your license key to unlock additional restricted segmentation tasks.

Installing and using TotalSegmentator in MITK is just a few clicks.

🕶️ Improved live 3D visualization of segmentations

In MITK v2025.12, we introduced live 3D visualization of segmentation labels. For MITK v2026.06, we completely redesigned the rendering pipeline by replacing volume rendering with blazing-fast polygonal mesh generation.

The result is dramatically improved visual quality, significantly higher rendering performance, and a much more responsive interactive experience.

📷 Screenshot Maker Overhaul

We simplified and modernized the Screenshot Maker to make creating high-resolution screenshots easier than ever. Choose a custom background color (or use transparency) and capture publication-ready images without any enforced logo watermark.

Drag the right screenshot over another image to see that the transparency of the liver is preserved.

📈 Standardized PET SUV quantification

The PET SUV computation view is now part of the official release build, together with a companion command-line app for batch and headless processing. It supports multiple SUV normalization strategies and automatic acquisition-time deduction with manual override down to the slice level.

To support standardized and reproducible quantification, MITK's SUV computation was evaluated against the IBSI SUV reference, using its suv computation reference as the benchmark (@IBSI thank you for this great set of DROs🙏).

🍏💄 Improved graphical user interface on macOS

The MITK Workbench has now a much better overall appearance on macOS, matching our Windows and Linux ports. It also finally supports the native fullscreen mode.

🍏⚠️ Running the MITK Workbench app on macOS

In particular since macOS 15 Sequoia, macOS imposes strict restrictions on executing the MITK Workbench application, as we are not a registered Apple developer. You may need to manually remove the MITK Workbench from quarantine to run it, e.g., allow the application to start in the security settings or run the following command on the MITKWorkbench.app file (adapt its file path as needed):

:::bash
xattr -d com.apple.quarantine /Applications/MITKWorkbench.app

Note that the very first launch of the MITK Workbench on macOS can take up to 2 minutes to complete all security checks.

News for developers

Recent advances in coding LLMs have significantly accelerated development, enabling us to deliver new features and fixes faster than ever. While AI has become an important productivity tool, we continue to apply the same engineering standards as before, with thorough code reviews, testing, and long-term maintainability taking precedence over pure vibe coding.

This accelerated pace of development also means that this release includes substantial changes under the hood, including several API-breaking changes. We believe that, in the age of AI-assisted development, adapting existing code to such changes is easier than ever. The resulting improvements in code quality, consistency, and modernization outweigh the temporary cost of resolving these incompatibilities.

To make migration as straightforward as possible, we have ensured that our commit messages document the rationale and details behind these changes. They provide valuable context not only for developers but also for coding LLMs, making them an excellent resource when updating existing code to the new APIs.

In case you experience any trouble while migrating to the latest version of MITK, please do not hesitate to create an issue on our GitHub page.

🐍 Python bindings: mitk-python package on PyPI available now!

The mitk-python package exposes the data structures and core utilities of MITK to Python. It ships as a self-contained wheel that bundles the compiled MITK runtime together with its native dependencies like ITK, VTK, CppMicroServices, and DCMTK, so you can install it with pip install mitk-python and start working with medical images without building MITK from source.

Note that the wheels on PyPI are currently available only for Python 3.12. The documentation is hosted on Read the Docs.

:::python
import mitk
import numpy as np

img = mitk.Image(np.zeros((64, 64, 64), dtype=np.float32), spacing=(1.0, 1.0, 2.5))
img.save("output.nrrd")

loaded = mitk.Image("output.nrrd")
print(loaded.shape, loaded.spacing, loaded.origin)

Remote control the MITK Workbench from Python

MITK v2026.06 introduces a REST API that lets external programs drive a running Workbench, together with a companion Python client, mitk-workbench-remote, that wraps it. Install it with pip install mitk-workbench-remote, connect to a live session, and script it end to end: push data into the DataStorage, navigate scenes, control the Standard Display and MxN multi-widget editors, drive the crosshair and per-window cameras, and capture screenshots — all from an external process, without writing a line of C++.

​```python import mitk_workbench_remote as mw

wb = mw.connect("http://localhost:8080") wb.show(image, name="case_001") # push data into the running Workbench png = wb.mxn.screenshot() # capture the MxN multi-widget editor ​```

A full example suite goes from a first "connect and show" up to a cohort screenshot factory that turns the Workbench into a scriptable visualization service: it applies an MxN layout once, then loops over a cohort and its segmentation labels, points the crosshair and cameras at each lesion, captures editor screenshots, and assembles a Markdown QA report — the kind of batch job that used to mean clicking through every case by hand. Combined with the layout-as-code DSL from the mitk package (mitk.mxn.layout), entire review layouts can be defined in Python and reused across cases.

Explore the examples, starting with Connect and Show and building up to the cohort screenshot factory.

🛠 Third-party dependency changes

Dependency	Before (v2025.12)	Now (v2026.06)
ACVD	83a0bd0b (2025-03-27)	3ca0b532 (2025-11-14)
Boost	v1.85.0 (patched)	v1.91.0
CMake (minimum)	v3.22	v3.28 (v3.31 on Windows)
cpp-httplib	v0.15.3	v0.27.0
CTK	adab52fb (2024-11-26)	db30f292 (2026-03-15)
dcmqi	v1.3.2	145b7b49 (2026-06-01)
DCMTK	v3.6.7	2dd54ca1 (2026-05-08)
GDCM	v3.0.14	v3.2.2
HDF5	v1.14.3	Use ITK's HDF5 instead
ITK	v5.4.3 (patched)	v5.4.6 (patched)
lz4	v1.9.4	v1.10.0
MatchPoint	c6188c95 (2025-04-10)	d02f5fbe (2026-06-11)
nlohmann_json	v3.11.3	v3.12.0
nnInteractive	[v1.1.2, v2.0.0)	[v2.5.0, v3.0.0)
oneTBB	v2022.1.0	v2022.3.0
OpenSSL	v3.0	v3.5
Poco	v1.14.1	v1.14.2
pybind11		v3.0.1
Python	v3.12.11	v3.12.12
Qt 6 (minimum)	v6.6	v6.10
Qwt (Qt6Qwt6)	149ab80d (2023-07-21)	611fda68 (2026-07-02)
SWIG	v4.3.0
tinyxml2	v8.0.0	v11.0.0
TotalSegmentator	2.9.0	2.14.0
VTK	40ef44cb (2025-04-09)	v9.5.2
zlib	v1.2.13	v1.3.1

Changelog since MITK v2025.12

Segmentation

• Added "Remove Result" button to several segmentation utilities
• Added context menu action to Data Manager to shuffle colors of selected binary masks
• Added hints to segmentation selection widgets when segmentations are filtered because of mismatching geometry
• Added plausability check for "Convert to Segmentation" context menu action
• Added support for 3D+t images to nnInteractive
• Added support for nnInteractive remote sessions
• Added support for nnInteractive v2.5
• Added support for nnInteractive's undo feature
• Added support for nnInteractive's model download feature
• Added the selection segmentation tool also to the 3D tools tab
• Added widgets to control opacity and visibility of preview segmentations
• Algorithm-type provenance on segmentation labels is now recorded
• Fixed hidden label suggesions when auto-filtering is active
• Fixed several issues in and with nnInteractive
• Improved conversions of labels to surfaces with vtkSurfaceNets3D
• Improved install, update, and uninstall user experience for nnInteractive
• Improved label color generation
• Minimized RAM usage of nnInteractive in MITK
• Reduced default opacity of newly created segmentations
• Removed legacy AI tools: Segment Anything, MedSAM, nnU-Net
• Rewrote 3D rendering of segmentations based on vtkSurfaceNets3D
• Rewrote TotalSegmentator segmentation tool and added support for v2.14
• Temporarily removed MONAL Label tools

REST API and remote control

• Added a new REST server and plugin for interaction with the workbench
• Added a REST endpoint for remote control of the data storage
• Added a REST endpoint for scene navigation and render-windows
• Added a REST endpoint for StdMulit editor and MxN editor
• Added OpenAPI/Swagger documentation

Python bindings and packaging

• Added PythonWheel build configuration to build wheels for PyPI
• Added a reusable Python package installer
• Completely rewrote Python bindings based on pybind11 instead of SWIG
• Fixed various findings in mitk::PythonContext
• Fixed venv path hash on Windows

Workbench UI, rendering, and interaction

• Added "Show crosshair in 3D" render window menu option
• Allow mouse interactions for manipulating manual registrations
• Fixed intermittent crash when reordering child nodes via drag-and-drop in Data Manager
• Fixed measurement selection via render-window click
• Fixed potential crash when a data node is removed from the Data Manager
• Made bounding shape handles work on oblique images
• Overhauled Screenshot Maker
• selected state of data nodes no longer persisted in MITK scene files

Core, data model, and preferences

• Added AllocateZeroedVolume() helper to mitk::Image
• Added a core service for getting the active Datastorage
• Added preference page for undo limit
• Allow preferences to be patched at workbench start
• Allow preferences to be temporarily overriden
• Fixed several bugs in CppMicroServices
• Fixed several potential bugs in MitkCore
• Fixed several potential bugs in MitkPlanarFigure
• Fixed wrong uses of the itkCloneMacro/mitkCloneMacro
• Improved preference page for Standard Display
• Replaced remaining boost::posix_time / boost::gregorian usages with std::chrono

I/O and scene format

• Fixed and improved DICOM SEG IO handling based on properties
• Fixed DICOM-SEG crash
• Fixed inconsistent path setting in mitk::IOUtil
• Introduced new developer-friendly scene file format
• Segmentation task lists accept standalone .mitkscene.json scenes now

PET

• Added the PET SUV computation view to the release configuration
• Added a PET SUV computation CLI app
• Reworked PET SUV computation view to offer functionalities (e.g. different SUV computation strategies, automatic time deduction and custom timing down to slice level...)
• Successully evaluate SUV code against IBSI SUV

MxN Display

• Added a layout-building DSL for MxN setups to mitk-python and to the REST rendering API
• Added persistence of MxN render-window layouts in the new scene format (ApplyLayout/SerializeLayout)
• Added strict MxN layout parsing, a scene migration script, and post-load state restoration
• Fixed sync-group assignment so top-row MxN windows no longer show duplicate images
• Refactored the synchronization-group lifecycle to a canonical API with named groups

Build, platforms, dependencies, and modernization

• Added auto-discovery of H_FILES, QRC_FILES, and UI_FILES
• Assimilated CppMicroServices as regular MITK module
• Fixed floating point precision issues on macOS ARM64
• Fixed font size and GUI in general on macOS
• Fixed full-screen mode on macOS
• Made MITK compatible to CMake v4
• Made MITK compatible to Doxygen v1.10 and later
• Made MITK compatible to Ubuntu 26.04
• Made MITK compatible to Visual Studio 2026
• Modernized CMake build system
• Removed deprecated code
• Unified include directives and guards, include/, src/, and resource/ folders of modules
• Unified UI file usage in C++ classes

Documentation

• Added common CLAUDE.md to repo
• Added Sphinx-based API documentation for the mitk Python package
• Fixed outdated URLs
• Revisited API documentation in general with AI assistence

Tests

• Fixed failed mitkIOUtilTest in ~ 1:2700 runs
• Fixed viewport rendering tests for CI clients