Models that read the body — and show their working.
Benchmarks are easy. Surviving a regulator, a hospital's IT security review and a sceptical radiologist on the same afternoon is the actual problem. That's what this division is built around.
Where we operate
Imaging画像診断
CT, MR and plain film. Detection, segmentation and prioritisation for the reporting queue.
Pathology病理
Whole-slide inference at tile level, with region evidence surfaced to the reporting pathologist.
Waveform波形
ECG, EEG and continuous bedside monitoring. Sub-10ms inference on local silicon.
Multimodal統合
Imaging fused with labs, notes and history — the context a single scan can't carry.
scan wall lit blue, clinician silhouetted
Built to be overruled
A model that can't be disagreed with doesn't get used twice. Every Xvian prediction arrives with the evidence behind it and a one-click override that is logged, counted and fed back into calibration.
- ◈Evidence surfacesHeatmaps and region references, generated at inference.
- ◈Confidence bandsExplicit uncertainty, not a bare score.
- ◈Override loggingDisagreements are data, and they're anchored too.
- ◈Case-mix reportingMonthly performance against your population, not the paper's.
Clinical model registry
| Model | Domain | AUC | p95 | Anchor | Status |
|---|---|---|---|---|---|
| thorax-v4.2 | Chest CT | 0.992 | 41ms | 0x7a1f… | CLEARED |
| retina-v3.8 | Fundus | 0.981 | 28ms | 0x3c92… | CLEARED |
| histo-v2.1 | Pathology | 0.964 | 112ms | 0xb14e… | REVIEW |
| cardio-sig-v1.6 | ECG stream | 0.958 | 9ms | 0x9dd0… | CLEARED |
| onco-fusion-v0.9 | Multimodal | 0.941 | 210ms | 0x2f77… | TRIAL |
CLEARED indicates a clearance held for that model and indication in at least one jurisdiction. REVIEW and TRIAL models are research-use only and must not inform clinical decisions. Jurisdiction detail available under NDA.
Bring a question, not a dataset
The best deployments start from a bottleneck in your workflow. Tell us where the queue backs up.
Talk to the clinical team →