Automated Analysis of Behavioral Phenotypes Correlating with the Severities of Parkinsonian Cynomolgus Monkey

Parkinson's disease (PD) is a neurodegenerative disorder characterized by tremor, bradykinesia, muscle tonus, postural instability, and dysfunctional motor synergies. PD primate models are an important vehicle for efficacy evaluation as well as translational medicine research. Previous methods for evaluating PD primate models have been clinical examination by neurologists using semiquantitative rating scales, but they are affected by subjectivity and the results of the evaluations are biased. Computer vision systems based on machine learning are capable of automating the analysis of complex behaviors. Cynomolgus monkeys have behavioral movement characteristics similar to humans, so the PD cynomolgus monkey model was used, and a computerized automated motion-posture acquisition and analysis system was employed to analyze the characteristic motion-posture behaviors of the PD cynomolgus monkey model in order to assess PD behavioral characteristics. We describe the use of principal component analysis (PCA) and hierarchical clustering methods to analyze large datasets and apply the PrimateScan systematic approach to screen for correlated behaviors and unbiased behavioral groups in a model of Parkinson's disease in cynomolgus monkeys. In addition, frame-by-frame image analysis of the target video is performed by the Ethovision system, and the tracked objects are distinguished from the background according to the gray level (brightness) or hue and saturation (color) value pixel points of the experimental objects. After the object is detected, the features of the experimental object are extracted, and these features are calculated to obtain the quantification of the overall movement of the animal, and ultimately, the motor activity status of the PD cynomolgus monkey can be analyzed, including the distance of the movement, the duration of the movement, and other indicators. Finally, the specific metrics of these two automated analyses were correlated with the PD clinical scores, respectively. The results of automated analyses showed a significant increase in non-activity time (P < 0.05), a significant decrease in postural stability and limb coordination (P < 0.01), and a significant decrease in distance moved and duration of movement (P < 0.05) in the Parkinson's disease cynomolgus monkeys compared with baseline. Correlation analysis between the results of the automated analysis and the PD clinical scores showed that there was no correlation between the non-activity time and the PD clinical scores, and that postural stability, limb coordination, distance moved, and movement duration all correlated with the PD clinical scores, suggesting that the four metrics of postural stability, limb coordination, distance moved, and movement duration used to differentiate between PD and non-PD are meaningful. The automated analysis method should be extendable to other behaviors, such as dyskinesia or seizures in stroke model studies, as well as social behaviors involving two or more animals with freedom of behavior. In conclusion, our study shows that machine learning-based and computer vision-based techniques are expected to provide new more objective and effective methods for judging and predicting the clinical severity and behavioral characteristics of PD.