From ICU sensor streams to an interpretable clinical graph.
NabzGraph turns continuous multi-modal signals (ECG, PPG, arterial pressure, respiration) into a patient-specific knowledge graph of SNOMED concepts and their measured relationships. Every node and edge traces back to the exact signal windows behind it.
See NabzGraph build a patient graph.
Pick a patient, build the graph, then click any concept node to trace it to the raw signal. Prerendered for the demo; the platform runs against real waveform data.
Signal evidence
Build the graph, then select any concept node to trace it back to the raw signal windows that activated it.
Edge types measured, not imposed.
Every relationship is derived from signal evidence rather than imposed by an ontology author.
Three panels, one reasoning chain.
Move from clinical structure to the source signals that substantiate it.
Graph view
Patient-specific concepts, persistence tiers, and measured edge types in one view.
Signal evidence
Open the exact waveform windows that activated any concept node.
Patient timeline
Trace concept density across the ICU stay and jump to a moment in time.
Interpretability you can measure.
Published protocols test concept fidelity, explanation faithfulness, and causal replication.
Mean held-out AUROC for the concept bottleneck against physiologically grounded reference labels.
Comprehensiveness: removing the cited windows degrades the concept activation substantially, while those windows alone reproduce it (sufficiency near zero).
Granger edges derived on one half of a patient’s windows reappear on the held-out half.
Edges from source-window overlap reappear under the same split-half test.
Temporal (precedes/follows) edges replicate weakly on the same test, at 0.214. We report that as a limitation of the method and have not patched it: single-patient activation series appear too short and too bursty for stable cross-correlation lags, while Granger inference on the continuous features is better powered. Figures are from the single-seed held-out test split, with the multi-seed run in progress. Cohort: 200 MIMIC-IV-WDB patients, 167 with built graphs; V2 targets full MIMIC-IV with external validation.
Coverage across modalities, concepts, and edge types.
- ECGLive
- PPGLive
- Arterial BPLive
- RespirationLive
- SpO₂Roadmap
- Tachycardia · bradycardia (ECG)Live
- Tachycardia (PPG)Live
- Tachypnea · bradypnea (RESP)Live
- Hypotension · hypertension (ABP)Live
- ICD-10 concept extensionRoadmap
- Temporal edgesLive
- Co-occurrence edgesLive
- Granger edgesLive
- Retrospective analysisLive
- Real-time streamingRoadmap
Clinical AI you can deploy with confidence.
Enterprise safeguards designed for sensitive clinical workflows.
Questions we get a lot.
Bring your signals. See the graph behind the patient.
Talk to us about a retrospective cohort, research collaboration, or the next signal modality.