Delineate
A 3D Slicer extension for liver internal radiotherapy: AI segmentation, image registration and voxel dosimetry brought together in a single clinical tool.
Context
Y-90 radioembolization is a liver internal radiotherapy treatment that relies on a long imaging chain: liver and tumour segmentation, registration between diagnostic MRI and workup imaging (SPECT/CT, PET/CT), then dose computation across tissue. Each of these three steps had been the subject of a separate research thesis within the IFTIM team, but remained standalone prototypes never designed for routine clinical use.
The goal of this fixed-term position was to consolidate that research into a single, intuitive tool that clinicians could actually use, built directly into 3D Slicer, the open-source medical imaging platform widely used in clinical research.
What the tool does
Delineate ships as three Slicer modules, usable independently or as a pipeline:
Segmentation
Registration
Dosimetry
- Segmentation — automatic deep-learning detection of liver and tumours, with standard RTstruct export.
- Registration — automatic alignment between diagnostic MRI and workup imaging, through a sequence of increasingly precise steps.
- Dosimetry — dose computation across tissue using two complementary physical models, with dose-volume histogram generation.
The architecture consistently separates the Slicer interface from the business logic, itself isolated in Python modules independent from Slicer — making them testable outside the application and enabling multiple deployment modes (local workstation, remote server, or hybrid).
Reliability & validation
A tool meant for clinical use has to be checkable, not just demonstrable. This is where I invested the most effort: every module has its own test suite, runnable both inside Slicer and standalone (GitLab CI), against an explicit quality threshold rather than a one-off result.
Systematic DICOM export (RTstruct, RTdose) is not a minor detail: it is what lets a result produced by Delineate be read directly by hospital treatment planning systems and PACS, with no manual conversion step to introduce errors.
Current status
Delineate is now listed in the official 3D Slicer extension catalog (stable release 5.12). The code is public on IN2P3's GitLab, with detailed user documentation describing the full clinical protocol, from diagnostic MRI through post-treatment verification.