Computerized Evaluation Of Facial Palsy Based On A Customized Camera System

Peripheral facial paralysis(PFP), also known as facial nerve palsy, the main characteristics of which is motor dysfunction of facial expression muscles. The main clinical symptom is unilateral of facial expression muscles completely paralyzed, which cause the abnormal facial appearance. The abnormal facial appearance makes the emotion of patient anxiety, dull and depressed. Therefore, most of patients with facial paralysis desire rehabilitation as soon as possible. Generally speaking, facial paralysis could be fully recovered with early detection, timely treatment and appropriate measures. The H-B facial palsy grading system is used to assessment the facial palsy in clinical, and the measures will be decided by the H-B grading score. Thus, an appropriate measure is the key to assessment of facial palsy. However, the H-B facial palsy grading system is subjective, vague description and overlapped grades. Different clinical doctors may gain different H-B score to the same patient that scoring by clinicians might include subjective biases. A study of facial structure symmetry is proposed in this paper. A consistency H-B score will be gained by the objective quantification of facial structure symmetry. The result will help the clinical doctor make an appropriate measure easily.For this study, two image capture devices are designed to capture the facial image. We cooperation with Shenzhen Sixth People's Hospital to capture 90 subjects of facial paralysis during March 2015 to January 2016. With a lot of literature reading and comparing, the supervised descent method is chosen to build a facial landmark detection model for building facial structure.61 features of facial structure symmetry based on facial landmark detection are proposed innovative in this thesis. These features reflect to H-B facial palsy grading system objectively. And with a lot of experiment result, these features are proved that reflect to H-B facial palsy grading system effectively. In this paper, dividing of facial carved lines is also studied. A best facial carved line is designed to build the facial structure. Then a facial paralysis grading classifier supervised by H-B grading score is built firstly in this study. This facial paralysis grading classifier is used to gain H-B grading score objectively. With the clinical data, the classification results are consistent with H-B grading system. With the premise of the data-augmentation, this classifier can achieve the classification accuracy in 94.4%. The results exploratory prove this study is promising in treatment of facial paralysis.