Deep Learning on Parkinson's Rat Behavior Point Sets in Open Field.

Parkinson's disease (PD) is one of the common progressive neurodegenerative disorder with motor deficits. A substantial volume of research has reported various methods to characterize Parkinson's rat model, but the performances are unsatisfactory. Because of the complexity of behavior data, we can hardly recognize the non-trivial behavioral traits and distinguish the difference between ill rat and the healthy. In this paper, we construct a deep learning model to analyze PD rat behavior data, which is obtained from an optical motion capture system. This system is widely used in movement analysis, robotics, making animations, and can provide high resolution of spatiotemporal information. By using the three-dimensional motion capture technology with retro-reflective optical markers placed in locations that represents rat head and body axis, we can get real-time three-dimensional coordinate information of rat's motion in open field. After the preprocessing, we fed the data set to our model to get a global feature via three abstraction layers and then process them by fully connected layers to obtain the final classification result. The results have shown that this method is robust enough to identify biomechanical and kinematic changes in response to Parkinson's rat with unilateral Striatum lesion.