A 3-layer clinical intelligence stack for hospital deployment.
From computer vision monitor reading to physiological reasoning and role-aware escalation.
From data extraction to reasoning to action, ASTA operates as one clinical intelligence loop.
Computer Vision Layer
Reads existing bedside monitors and converts displayed numerics into structured vital streams without APIs, hardware taps, or vendor-specific integration.
- Extracts HR, BP, SpO2, and RR in real time
- Supports 15+ OEM monitor brands
- No API or hardware tap required
- Deploys without vendor lock-in
PPLM Reasoning Layer
A 10–20B parameter physiological language model reasons over continuous vital trajectories to surface deterioration context. Live deployment integration coming soon.
- 10–20B parameter physiological language model
- Trajectory-aware reasoning over continuous windows
- Ranked deterioration context
- Evidence-linked interpretation before output
Clinical Output Layer
Frames evidence-linked outputs and escalation guidance so the receiving nurse or clinician gets usable bedside context.
- Evidence-linked bedside outputs
- Role-aware escalation support
- Recommended next-step framing
- Explainable action at the endpoint
End-to-end signal flow from screen to alert.
How vital numerics move from bedside monitor displays to structured clinical action in real time.
Monitor display
ASTA starts with the bedside monitor already in use.
Structured extraction
Computer vision converts displayed numerics into structured, time-stamped vitals.
Physiological embedding
Vital windows are normalized into machine-readable physiological state embeddings.
Deterioration interpretation
Alerts trigger on vitals exceeding threshold and on camera obstruction detection. PPLM ranks deterioration context above the raw alert.
Clinical output
Vitals are captured continuously every 5 seconds, not static snapshots. Output reflects current patient trajectory, framed for the receiving clinical role.
Monitor-agnostic computer vision vital extraction.
Reads vital sign numerics directly from bedside displays across 15+ OEM monitor brands without digital interfaces.
Reads existing bedside monitors
ASTA works with the monitor already at the bed rather than requiring a new fleet or a vendor feed.
Real-time structured vital extraction
Displayed numerics become structured HR, BP, SpO2, and RR streams for downstream reasoning.
No monitor-side IT dependency
Monitor reading starts without hospital IT integration, proprietary APIs, or hardware taps.
Monitor-facing capture, not patient surveillance
The capture point is the monitor display. The workflow is built around numerical data extraction, not patient imagery.
Screen locate
Identify the bedside monitor region in the capture frame.
Layout classify
Recognize OEM-specific layouts and select the right reading template.
Value extract
Apply OCR and parsing to vitals and displayed numerics.
Signal normalize
Time-stamp and standardize values for downstream reasoning.
Vendor-agnostic data extraction makes ASTA additive to the infrastructure already in place.
Works with your current monitors and systems.
No hardware replacement. ASTA reads existing monitor displays via computer vision.
Physiological reasoning beyond threshold alerts.
Evaluates vital sign trajectories over time to distinguish clinically meaningful deterioration from temporary spikes.
Continuous trajectories
Looks at how vitals move across time windows, not one reading at a time.
Pattern interpretation
Detects deterioration signatures that threshold logic may miss or detect late.
Ranked clinical context
Surfaces likely deterioration context instead of sending a raw alert alone.
Evidence-linked output
Carries the reasoning pathway into the output so escalation remains explainable.
Escalate after a single value crosses a line.
Show status snapshots but do not reason over change over time.
Extract numbers accurately but stop before clinical interpretation.
Interprets trajectories, adds deterioration context, and produces evidence-linked output that is more useful than a simple alert.
This is where ASTA moves beyond raw data extraction into clinically useful reasoning.
Clinical validation and accuracy metrics.
Rigorously benchmarked in real hospital environments across edge scenarios, lighting variations, and monitor display types.
| Dimension | Proof | Implication |
|---|---|---|
| Data extraction | 98% CV accuracy across 15+ OEM monitor brands | Shows cross-device data extraction rather than one-vendor optimization. |
| Training scale | 100M+ labeled monitor frames | Supports layout robustness across real display conditions. |
| Real-world operation | 10+ hospital deployments and 25 live devices | Shows field usage in hospital environments, not just lab validation. |
For platform evaluation, ASTA leads with extraction accuracy, device breadth, real hospital operation, and low-latency output.
The live ward view for nursing, clinical leadership, and operations.
One shared product surface for the whole unit: monitored beds, structured vitals, escalation queue, and audit trail in a single ward view.
| Bed | HR (bpm) | SpO₂ (%) | BP (mmHg) | Status |
|---|---|---|---|---|
| B-01 | 76 bpm | 98% | 122/80 | ok |
| B-02 | 91 bpm | 94% | 138/88 | warn |
| B-03 | 68 bpm | 99% | 118/76 | ok |
| B-04 | 108 bpm | 89% | 142/92 | alert |
| B-05 | 72 bpm | 97% | 126/82 | ok |
| B-06 | 83 bpm | 96% | 130/84 | ok |
SpO2 89% - 6 min downward trend - PPLM flags respiratory deterioration - escalate to nurse on duty
HR 91 bpm - rising over 3 min - physiological watchlist only - no escalation yet
BP 142/92 - above ward range - trend stable - continue observation
Rapid hospital deployment in 4 steps.
Designed to deploy on existing hospital equipment without complex hardware installation.
Bedside capture
Monitor-facing capture reads displayed numerics from the bedside monitor already in use.
Local extraction
Vital streams are extracted, normalized, and buffered for continuous downstream reasoning.
Controlled deployment posture
Hospitals can adopt on-prem, hybrid, or managed cloud deployment under their own governance model.
Interoperable outputs
Structured outputs and escalation context can map into HL7/FHIR-aligned workflows when needed.
Numerical data only
No patient imagery is required in the monitoring workflow. The design centers on monitor numerics, not patient video.
Hospital-defined data residency
Retention and environment posture are defined by the hospital and enforced by the selected deployment model.
Flexible infrastructure posture
ASTA can operate in on-prem, hybrid, or managed cloud environments depending on site requirements.
Additive integration model
Monitor reading starts without hospital IT integration. Downstream workflow connectivity can be added when the site is ready.
Data extraction starts without hospital IT integration. Workflow and EMR connectivity can be added as later layers, not day-one prerequisites.