PixelPaws is an open-source, integrated platform designed to automate scoring of mouse behaviors for preclinical research. The system aims to address limitations of manual scoring—low throughput and observer variability—by providing hardware and software that can be assembled from consumer parts and used without extensive programming expertise. The authors evaluated PixelPaws across assays commonly used to study pain, itch, and opioid withdrawal.
The platform combines a near-infrared imaging enclosure built from consumer parts reported to cost under $300 with multi-camera acquisition software. The software bundle includes a pose-estimation network and a graphical user interface that hosts eight classifiers for automated detection of distinct behaviors. The design emphasis is on low cost and accessibility to broaden adoption beyond labs that develop bespoke automated systems.
PixelPaws includes classifiers for key actions relevant to pain and related paradigms, explicitly including licking, scratching, and jumping. The authors validated these classifiers against trained human observers using a leave-one-session-out evaluation approach, comparing automated scores to those produced by manual, blinded scoring. This validation method assesses classifier generalizability across sessions by iteratively leaving one session out of training and testing on it.
In the formalin pain assay, classifier-based scoring of licking reproduced the classic biphasic time course expected for this model. When oxycodone was administered, classifier-detected licking decreased in a dose-dependent manner. The authors report that the potency estimate for oxycodone in phase 2 of the formalin response, derived from classifier scoring, was comparable to the potency estimate produced by blinded manual scoring.
These findings indicate that PixelPaws can capture both temporal dynamics and pharmacological modulation of nocifensive behaviors in a commonly used pain assay.
PixelPaws detected an approximately threefold increase in scratching behavior after chloroquine administration, demonstrating sensitivity to an itch-inducing stimulus. In an opioid withdrawal paradigm, naloxone precipitated a burst of jumping behavior in mice that had received oxycodone; this event was identified in the same recordings used for other analyses. These results show that the platform can monitor behaviors spanning different domains—itch and opioid-related withdrawal—within the same experimental recordings.
Beyond discrete behavior classifiers, the authors extracted continuous paw contour metrics from the recordings. Reported observations include a fall in contour intensity after formalin administration and tracking of oxycodone dose series by contour intensity. In nerve-injured mice treated with oxycodone, contour area shifted toward symmetry, suggesting the potential for contour-derived metrics to capture subtler changes in limb use or posture.
The team performed behavioral sequencing analyses that reportedly revealed distinct changes in transition patterns between behaviors. The source text is truncated at that sentence and does not report the full details of those sequencing results; therefore, comprehensive specifics of transition alterations are not available from the source. Where methods and outcomes are fully described above (classifier validation, assay-specific results, contour metrics), PixelPaws produced results consistent with established expectations in these assays.
PixelPaws is presented as an open-source solution that integrates low-cost, consumer-part hardware with bundled software including pose estimation and multiple classifiers in a GUI. The stated hardware cost of under $300 and the software packaging are intended to lower the technical and economic barriers that limit the broader adoption of automated behavioral scoring systems.
The authors declare no competing interests. Funding sources listed in the source include grants from the National Institute on Drug Abuse (NIDA), the National Institute of Neurological Disorders and Stroke (NINDS), and the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK). The manuscript is posted on bioRxiv with DOI https://doi.org/10.64898/2026.09.19.752787 and was posted September 24, 2026.
All factual statements in this summary are drawn from the provided source text. The source text truncates one sentence describing behavioral sequencing results; detailed outcomes and metrics for that analysis were not reported in the provided source and therefore are not included here.