How AiVet Works

AiVet is a multimodal clinical framework that extracts objective signals from non-verbal patients. It combines computer vision, acoustic analysis, and sensor fusion to produce structured clinical data where previously only subjective observation existed.

The Signal Pipeline

Vision Layer

Remote photoplethysmography (rPPG) extracts heart rate from video via green-channel intensity analysis. Grimace scoring uses facial landmark geometry for species-specific pain assessment.

Audio Layer

Spectral analysis of vocalizations classifies respiratory patterns (stridor, wheeze, stertor) and distress signatures. Species-specific frequency bands isolate clinically relevant signals.

Fusion Layer

Bayesian triage combines vision and audio confidence scores into a unified pain probability. The Doolittle Protocol translates multimodal signals into structured clinical documentation.

System Architecture

Client (AiVet Console)

  • rPPG green-channel extraction at 30fps
  • WebAudio capture and spectral pre-processing
  • AIVA offline buffer (localStorage EMR)
  • WebBluetooth HRM integration with HRV
  • Telemedicine sync via Firebase Firestore

Server (Doolittle Engine)

  • Hybrid vision analysis (local + cloud)
  • Audio classification via librosa
  • Multimodal Bayesian fusion
  • WebSocket real-time streaming
  • Clinical report generation (JSON/HTML/PDF)

Clinical Disclaimer

AiVet is a decision-support tool in active development. It is not validated clinical software and is not a replacement for veterinary examination. All outputs should be interpreted by a qualified veterinarian. The system is prone to errors, particularly with limited input data.