Clinical Validation Roadmap

AiVet is built on the premise that non-verbal patients produce measurable signals that can be objectively extracted and clinically validated. This is our research methodology and validation pathway.

Validation Targets

SignalMethodTarget MetricStatus
Feline Grimace Scale (FGS)48-node facial landmark geometryInter-rater κ > 0.82In Progress
Remote PhotoplethysmographyGreen-channel intensity extraction±5 BPM vs. contact sensorFunctional
Spectral Audio ClassificationFrequency-band isolation + spectral fluxSensitivity > 0.85 for stridorIn Progress
Bayesian Multimodal FusionWeighted confidence integrationAUC > 0.90 for pain detectionPlanned
Behavioral Activity ClassificationMotion intensity + position heuristicsAccuracy > 0.80 (6 classes)MVP

Methodology

Signal Extraction

Client-side pre-processing reduces bandwidth by extracting only clinically relevant features (mean green intensity, spectral peaks) before server-side analysis. This hybrid approach enables real-time operation on mobile devices while maintaining analytical depth.

Species-Specific Models

Vital ranges, grimace geometry, and vocalization patterns are parameterized per species. Current support: canine, feline, equine, avian. Each species has distinct normal ranges for heart rate, respiratory rate, and distress thresholds.

Validation Protocol

Ground truth is established via concurrent contact-sensor measurement (pulse oximetry, ECG) and expert veterinary scoring. Inter-rater reliability is measured using Cohen's kappa for categorical outputs and Bland-Altman analysis for continuous measurements.

One Health Translation

The signal extraction methodology is mathematically species-agnostic. Veterinary validation provides a lower-regulatory-barrier pathway to demonstrate safety and efficacy before potential translation to non-verbal human populations (pediatric, geriatric).

References & Prior Art

Evangelista et al. (2019)

Feline Grimace Scale: Development and validation of a novel facial expression-based pain assessment tool.

Verkruysse et al. (2008)

Remote plethysmographic imaging using ambient light. Foundational work on camera-based heart rate extraction.

Poh et al. (2010)

Non-contact, automated cardiac pulse measurements using video imaging and blind source separation.

WHO One Health Joint Plan of Action (2022)

Framework for integrated surveillance across human, animal, and environmental health domains.

Collaborate With Us

We are seeking academic partners for formal validation studies. If you work in veterinary science, applied ethology, or digital health — we want to hear from you.

Contact Research Team