红外热成像辅助面神经功能自动评估方法研究

Automatic Assessment of Facial Nerve Function Based on Infrared Thermal Imaging

面瘫是一种多发的面神经疾病,表现为患侧面神经功能失调,严重影响患者的正常生活和人际交往。面神经功能自动评估方法对于面瘫的诊治是至关重要的。面部神经功能受损导致体表温度分布发生改变,可以通过红外热成像采集患者的面部温度分布不对称特征,基于红外热成像提出一种面神经功能自动评估新方法,融合温度特异性和边缘检测自动将面部红外热像划分为左右对称的八个区域,提取面部温度不对称特征,包括温差、有效热面积比和温度分布不对称度,采用径向基神经网络作为面神经功能自动分类器。实验收录了390幅单侧患病的面瘫患者正面红外热像图,结果显示:采用径向基神经网络的红外热成像面神经功能自动分类器的平均分类准确率为94.10%,比采用K近邻分类器和支持向量机分类器分别提高了9.31%和4.87%,优于传统的House-Brackmann面神经功能评估方法,对面神经功能的分类精度完全符合临床应用标准,可以有效评估面瘫患者的面神经功能,有助于面瘫的临床诊断与治疗。

Facial paralysis which is mainly caused by facial nerve dysfunction is a common clinical entity.It seriously devastates apatient＇s daily life and interpersonal relationships.A method of automatic assessment of facial nerve function is of critical importance for the diagnosis and treatment of facial paralysis.The contralateral asymmetry of facial temperature distribution is one of the newly symptoms of facial paralysis patients which can be captured by infrared thermography.This paper presents a novel framework for objective measurement of facial paralysis based on the automatic analysis of infrared thermal image.Facial infrared thermal image is automatically divided into eight regional areas based on facial temperature distribution specificity and edge detection,the facial temperature distribution features are extracted automatically,including the asymmetry degree of facial temperature distribution,effective thermal area ratio and temperature difference.The automatic classifier is used to assess facial nerve function based on radial basis function neural network（RBFNN）.This method comprehensively utilizes the correlation and specificity of the facial temperature distribution,extracts efficiently the facial temperature contralateral asymmetry of facial paralysis in the infrared thermal imaging.In our experiments,390 infrared thermal images were collected from subjects with unilateral facial paralysis.The results show：the average classification accuracy rate of our proposed method was 94.10%.It has achieved a better classification rate which is above 9.31% than K nearest neighbor（kNN）classifier and 4.87% above Support vector machine（SVM）.This experiment results is superior to traditional House-Brackmann facial neural function assessment method.The classification accuracy of facial nerve function with the method is full compliance with the clinical application standard.A complete set of automated techniques for the computerized assessment of thermal images has been developed to assess thermal dysfunction caused by facial paralysis,and the clinical diagnosis and treatment of facial paralysis also will benefit by this method.