RMScratcher: An edge-AI platform for high-throughput analysis of rodent scratching behavior

Automated platform enables detailed and unbiased quantification of scratching behavior. • High-throughput processing increases efficiency and reduces manual effort. • Improved YOLOv8s balances efficiency and precision for fine-grained behavioral detection. • Immediate analytics support on-the-fly experimental monitoring and adjustments • Compact edge-computing design ensures scalable and easy deployment in laboratories Background Quantitative analysis of rodent scratching is essential for studying somatosensory and neuroimmune mechanisms, yet scalable and high-throughput analysis remains challenging. New method We developed RMScratcher, a compact platform combining multi-channel acquisition with Jetson TX2–based edge computing and an enhanced YOLOv8s framework. Incorporating GhostConv and Context Aggregation modules, the system achieves efficient and accurate detection of bouts, touch, duration and scratching region. Results RMScratcher achieved > 95% accuracy in distinguishing scratching from other behaviors, reliably quantified bout, touch and duration, and scratching region. Performance was robust across multiple rodent models and experimental conditions. Comparison with existing methods Manual scoring is slow and prone to bias, while existing automated tools often lack throughput and generalizability. Crucially, there is no known literature that identifies the scratching region. RMScratcher provides these capabilities with minimal hardware requirements. Conclusions RMScratcher extends the methodological toolkit for behavioral neuroscience by integrating video recording with intelligent computation. The platform facilitates reproducible studies of somatosensory function, neuroimmune interactions, behavioral phenotyping, and the neural mechanisms underlying abnormal sensory behaviors. Download: Download high-res image (169KB) Download: Download full-size image