Combining dual-view fusion pose estimation and multi-type motion feature extraction to assess arthritis pain in mice

Objective: In studies on arthritis pain, mice are the most important and frequently -used experimental animal models. Currently, in mouse pain assessment, the artificial observation is highly subjective, and common automatic analysis systems utilize single -view video from which the extracted features are not diverse. Therefore, this study aims to develop an efficient mouse arthritis pain assessment system by combining dual -view fusion pose estimation and multi -type motion feature extraction. Methods: First, a video acquisition device is designed and videos of mice with arthritis pain are collected. After mouse image annotation, a dual -view fusion framework for pose estimation is proposed to accurately locate the mouse keypoints. Then, outliers in motion trajectories are detected and corrected through a two -stage search method. On this basis, multi -type motion features are extracted to construct an efficient arthritis pain evaluation system. Finally, the effectiveness of constructed system is verified by evaluating the pain status of experimental mice. Results: The dual -view fusion pose estimation achieves an average precision of 0.83 and 0.87 in vertical and upward view, respectively. Statistical analysis shows that after arthritis modeling surgery, features with significant differences increases over time, which account for 100% at the 12th week after surgery. Conclusion: The proposed pain evaluation system obtains a highest classification accuracy of 0.8, indicating its effectiveness in measuring the mouse arthritis pain status. Significance: The development of automated pain assessment system is promising to help promote the research of pain pathology mechanisms and the development of therapeutic drugs.