US20220207902A1 — System and method for automatically discovering, characterizing, classifying and semi-automatically labeling animal behavior and quantitative phenotyping of behaviors in animals

Assignee: Harvard University · Inventors: Sandeep Robert Datta, Alexander B. Wiltschko Filed: 2013-05-10 (EP13788526.5A; US17/581,326, filed 2022-03-23) · Published: 2022-06-30 · Status: Withdrawn

Abstract

A method for studying animal behavior in an experimental area: stimulate the animal, collect data (depth camera + touch-sensitive device), analyze collected data, and develop quantitative behavioral primitives. Automatically discovers, characterizes, and classifies animal behavior via 3D video + clustering.

Key technical features

• Brain recording: NO — depth camera (95 mentions) + touch-sensitive device (33 mentions) + video (64 mentions); no EEG/electrodes
• Species: Animal (rodent/mouse) — 230 “animal”; 109 “mouse”; 9 “rodent”; 3 “primate”; 37 “human” (generic “human” in claims); 23 “subject”; zero dog/cat
• Classification: G06V40/10 (animal body recognition); A01K29/005 (monitoring animal activity); A01K67/00 (breeding animals); A61B5/11 (body movement measurement); A61B5/103 (body shape/pattern/movement)
• Method: 3D depth camera → background removal → contour detection (Kalman filter tracking) → multi-dimensional parameter extraction (area + depth: perimeter, surface, rotation, height, width, velocity, spine curvature, limb position) → clustering (PCA/SVD/ICA/LLE dimensionality reduction) → behavioral primitives → semi-automatic labeling
• Family: EP4198926A1 (same invention, European publication)

Significance

The Datta Lab (Harvard) automated animal behavior phenotyping system — uses depth cameras and ML to discover, classify, and label mouse behaviors without human annotation. This is a quantitative phenotyping tool for laboratory animals, not cognitive testing per se, but directly relevant to the preclinical cognitive testing landscape because it provides the automated behavioral measurement infrastructure that operant chamber / touchscreen experiments rely on. Withdrawn (not granted), but the underlying work became the commercial platform DeepLabCut and the Wiltschko/Datta behavioral phenotyping ecosystem.