Problem
Research models die in notebooks. Clinicians and collaborating researchers already live in 3D Slicer. If inference is not a module in that environment, it does not get used.
What shipped
Qt-based Slicer extensions that orchestrate deep-learning inference: model loading, preprocessing, and a run button, without leaving the imaging workspace.
Lab researchers and clinical collaborators use it to run segmentation and synthesis models on patient volumes. The interesting engineering is the boring part — wrapping a research-grade checkpoint behind the GUI people already trust.
This is the deployment half of the ISTAR imaging work. MAISI-US is the model; this is how someone who is not me actually runs it.