An application of Gaussian mixture models for medical characteristics analysis of nystagmus signals

Aiming at the problems of imperfectness and poor robustness to current analysis methods of videonystagmography (VNG), a medical characteristics analysis method for displacement vectors of nystagmus is proposed in this paper. Firstly, the video images of nystagmus are captured with infrared video cameras and the motion trajectories of the eyeballs are obtained through pupil localization. Then, the 2D displacement vectors between any two adjacent frames are acquired with computing the differences of their positions. The statistical results of the displacement vectors are demonstrated to match with a joint distribution of several Gaussian models, and each Gaussian model may be related to a unique element of the Nystagmus. Finally, Gaussian Mixture Models (GMMs) are applied to analyze these vectors. The parameters, such as mean values, covariance matrices and prior probabilities, are inferred with EM algorithm. More quantitative details and crytic features other than the conventional information are achieved for further clinical diagnosis and medical devices development. Moreover, the ant-jamming ability is improved.