Medical Imaging

• Viewer: multi-planar reconstruction, window/level presets (CT brain, lung, bone, etc.)
• Annotations: ROI, bounding box, polygon, mask, freehand measurements
• AI models: pre-trained classifiers and segmenters; bring your own via ONNX
• Interoperability: export annotations as DICOM-SEG, Structured Report (SR), or RT Structure Set (RTSTRUCT) for a TPS / 3D Slicer round-trip
• Radiomics: extract IBSI-aligned features (first-order, shape, GLCM, GLRLM, GLSZM) from any ROI
• Collaboration: real-time multi-user annotation with presence indicators

Train a model from your annotations

The DICOM-SEG export panel includes one-click training actions that turn your annotations into a model on the Deep Learning worker fleet — no round-trip through an external trainer.

1. 1
   Train detector from boxes

   Your rectangle/box annotations become an object-detection model (Faster R-CNN).
2. 2
   Train segmenter from regions

   Your polygon/freehand/brush regions are rasterised into masks and become a segmentation model (U-Net).
3. 3
   Train classifier from labels

   Each image is labelled by its dominant annotation class and becomes an image classifier.

Clicking any of these takes you straight to Deep Learning → Training, where the run resumes live — no manual navigation. (Segmentation needs at least one region; classification needs at least two classes.) The same in-browser annotator is available there for general, non-DICOM images.