基于深度神经网络的面部运动三维动态定量研究

The lack of an objective, accurate, and uniform facial muscles evaluation system that meets the clinical needs is an important factor that restricts facial paralysis diagnosis and treatment and research. In recent years, artificial intelligence technology has developed rapidly and gradually applied to the medical and health field, providing new ideas for objective, quantitative, and convenient evaluation of facial muscle movement. With the previous facial three-dimensional dynamic quantitative analysis system, the current project is planned to make the automated facial marker identification and reference frame measurement procedures based on the artificial neural network technology based on deep neural network. Screening and optimization of quantitative evaluation indicators of facial nerve paralysis three-dimensional measurement, including static, dynamic, complications (face muscle linkage and spasticity) indicators, aiming to find the characteristics and patterns of the interaction of the different regions of the face. It can provide theoretical and technical support for evaluating the degree, treatment effect, prognosis of facial paralysis and facial reconstruction. The current project is based on the clinical demand, with the data got by the new technology, the facial index curve will be finished, and then we can explore the prognosis patterns of the disease. It will provide an objective basis for the evaluation of effectiveness of facial paralysis treatment and establish a more scientific and reasonable facial paralysis grading system.

缺乏客观、准确、统一的符合临床需求的面肌运动评价系统是制约面瘫诊疗和研究的重要原因。近年来，人工智能技术飞速发展并逐渐应用于医疗健康领域，为客观、定量、便捷的评价面肌运动提供了新的思路。本课题拟在前期开发成功的基于动作捕捉的面部三维动态定量分析系统的基础上，利用基于深度神经网络的人工智能技术，完成自动化的面部标记点识别与参照系测量程序。筛选并优化面神经麻痹三维测量定量评价指标，包括静态、动态、并发症(面肌联动和痉挛)各项指标，发现相应的特点，揭示面部各区域间相互作用规律，为面瘫程度评价、治疗效果、预后判定以及面部重建提供理论与技术支撑。本项目研究立足于临床需求，充分考虑新技术应用，利用所获得数据绘制面部指标的变化曲线，探索疾病转归特点，为评价药物或手术等治疗手段的有效性和建立更加科学合理的面瘫分级系统供客观依据。

缺乏客观、准确、统一的符合临床需求的面肌运动评价系统是制约面瘫诊疗和研究的重要原因。近年来，人工智能技术飞速发展并逐渐应用于医疗健康领域，为客观、定量、便捷的评价面肌运动提供了新的思路。.本课题在前期开发成功的基于动作捕捉的面部三维动态定量分析系统的基础上进行在硬件、软件两方面升级，探索利用深度神经网络算法、三维重建及结构光深度摄像机，实现自动化的人脸关键点检测及动态测量，根据临床需求生成测量软件1项，配合相应运行环境后，成功搭建新一代面部智能面部运动定量评估样机。样机可自动识别面瘫观测点，自动检测其运动距离、速度及加速度，生成静态、动态三维测量定量评价指标，为面瘫程度评价、治疗效果、预后判定以及面部重建提供理论与技术支撑。探索智能分析仪准确度检测方法，完成重测信度检测。采集并分析健康人群面部运动数据情况，初步探索应用于采集面瘫患者的面部运动。.项目资助发表核心论文1篇，SCI 2篇，待发表1篇。获得专利1项，获得软件著作权1项，搭建深度神经网络的面肌运动三维动态定量分析试验样机1台，参与学术交流3次，采集健康人群及面瘫面部表情80例，拟发表英文论著1篇。培养在站临床博士后1名。项目资助金额25万元，支持23.2990万元，各项支出基本与预算相符。剩余经费1.7010万元，剩余经费计划用于本项目研究后续支出。