PixelPaws: a low-cost, open-source integrated platform for automated scoring of mouse behaviors

Abstract

Behavioral scoring is central to preclinical pain research but is typically performed manually, which is inherently low-throughput and subject to variability between observers. Automated alternatives exist, but cost and programming expertise limit adoption beyond the laboratories that build them. Here we describe PixelPaws, an open-source platform combining a near-infrared imaging enclosure built from consumer parts for under $300, multi-camera acquisition software, a bundled pose-estimation network, and eight classifiers in a graphical user interface. We evaluated the platform across standard assays of pain, itch, and opioid withdrawal. Licking, scratching, and jumping classifiers were validated against trained human observers by leave-one-session-out evaluation. In the formalin assay, classifier-scored licking followed the expected biphasic time course and was reduced dose-dependently by oxycodone, with a phase 2 potency estimate comparable to blinded manual scoring. Chloroquine roughly tripled scratching, and naloxone precipitated a burst of jumping in oxycodone-treated mice. From the same recordings we derived a set of paw contour metrics. Contour intensity fell after formalin and tracked the oxycodone dose series, and contour area shifted toward symmetry after oxycodone in nerve-injured mice. Behavioral sequencing revealed distinct changes in the transi