Worm-PostureNet: An Integrated Deep Learning Framework for High-Throughput Pose Estimation and Behavioral Profiling in Caenorhabditis elegans

• End-to-end deep learning framework for automated C. elegans video analysis. • Unifies landmark detection, tracking, and behavior quantification in one workflow. • Enables high-throughput phenotyping for neurogenetic and pharmacological research. • Open-source platform for robust, interpretable biomedical behavior analysis. The analysis of Caenorhabditis elegans ( C. elegans ) behavior constitutes a fundamental pillar of neurogenetic and biomedical inquiry. Traditional approaches, however, typically depend on multi-stage image processing pipelines that are not only complex but also prone to inefficiency, cumulative errors, and inadequate representation of the organism's non-rigid locomotion. To address these limitations, this study introduces Worm-PostureNet, an integrated deep learning pipeline featuring an end-to-end pose estimation module that directly regresses anatomical landmarks from raw images, followed by tracking and behavioral quantification modules for automated, high-throughput analysis of C. elegans behavior. The framework incorporates three seamlessly integrated components: a landmark detection module utilizing YOLO-POSE for the direct regression of anatomical landmarks; a tracking module based on OC-SORT, specifically tailored to accommodate nonlinear motion and frequent occlusions, thereby ensuring reliable identity persistence; and a multi-tier behavioral quantification module that systematically derives metrics such as motion trajectories, velocity, directional dynamics, body curvature, and behavioral sequences. Validation performed on publicly available C. elegans datasets indicates that Worm-PostureNet maintains landmark detection accuracy on par with conventional multi-stage methods while, owing to its fully integrated architecture, delivering substantially accelerated processing, a streamlined analytical workflow, and enhanced robustness under challenging imaging conditions. This work offers an open-source, efficient, and scalable computational platform for high-throughput behavioral phenotyping in C. elegans , with promising implications for advancing neuroethological research and genetic screening applications.