Open work, shipping.
Two releases from our team. A public medical-imaging viewer hosted at Johns Hopkins — a working example of what our platform delivers. And RadThinking, the first dataset in our commercial pipeline, introduced in a 2026 preprint that shows what the platform produces. Both are open.
Two releases. One platform.
The viewer below is a working example of what our platform delivers today. The paper introduces the first dataset in our commercial pipeline.
BodyMaps imaging viewer
A public, interactive medical-imaging viewer hosted at Johns Hopkins. Browse 9,901 CT cases from the PanTS pancreatic-imaging release with 28 organ-level segmentations, three orthogonal MPR views, and a 3D label-coloured render — a working example of the imaging foundations our platform delivers.
- 9,901 cases · train + test splits
- 28 organ-level segmentations per case
- 6 segmentation models, selectable for uploaded scans
RadThinking — Longitudinal clinical reasoning in radiology
A 2026 preprint introducing RadThinking, a Visual Question Answering dataset that teaches AI to reason about cancer screening — observing findings, comparing across prior scans, integrating clinical context, and reaching a diagnostic conclusion. The first dataset in our commercial pipeline, and a peer-reviewed demonstration of what the platform produces.
- 20,362 CT scans · 9,131 patients
- 43 cancer groups · 19 organ targets
- Compositional VQA with chain-of-thought + verifiable rewards