ARBEL: A Machine Learning Tool with Light-Based Image Analysis for Automatic Classification of 3D Pain Behaviors

Abstract A detailed analysis of pain-related behaviors in rodents is essential for exploring both the mechanisms of pain and evaluating analgesic efficacy. With the advancement of pose-estimation tools, automatic single-camera video animal behavior pipelines are growing and integrating rapidly into quantitative behavioral research. However, current existing algorithms do not consider an animal’s body-part contact intensity with- and distance from- the surface, a critical nuance for measuring certain pain-related responses like paw withdrawals (‘flinching’) with high accuracy and interpretability. Quantifying these bouts demands a high degree of attention to body part movement and currently relies on laborious and subjective human visual assessment. Here, we introduce a supervised machine learning algorithm, ARBEL: Automated Recognition of Behavior Enhanced with Light, that utilizes a combination of pose estimation together with a novel light-based analysis of body part pressure and distance from the surface, to automatically score pain-related behaviors in freely moving mice in three dimensions. We show the utility and accuracy of this algorithm for capturing a range of pain-related behavioral bouts using a bottom-up animal behavior platform, and its application for robust drug-screening. It allows for rapid objective pain behavior scoring over extended periods with high precision. This open-source algorithm is adaptable for detecting diverse behaviors across species and experimental platforms. Competing Interest Statement L. B. Barrett and C. J. Woolf have issued patents on the data acquisition technology. L. B. Barrett and C. J. Woolf are founders of Blackbox Bio, the company which has licensed the patent on the technology from Boston Children's Hospital. Z. Zhang now works for Blackbox Bio.