摘要
在神经影像研究中,患者的面部特征有时可以通过3维表面重建技术从影像中复原,这使得患者身份隐私信息泄漏存在潜在可能。为了解决这一问题,提出一种自动化面部特征剔除算法,从海量多模态大脑核磁共振影像中自动剔除患者身份相关的面部特征信息。该方法基于一种新提出的多分辨分层特征向量匹配方法来准确定位3维影像中的解剖学点标记,并通这种匹配方法从多模态磁共振影像中确定患者面部特征相关的解剖结构的空间位置,并以此为基础估计出一个最优3维剔除平面来剔除患者面部特征信息。最后,通过实验验证了该方法的有用性和可靠性。
In neuroimaging studies,subject's identity can sometimes be recovered from volumetric brain MR images via three-dimensional surface reconstruction or volume rendering techniques and directly leads to the violation of privacy protection regulations in medical applications. To address these concerns,a novel method for facial de-identification was developed to automatically remove facial feature from multi-modality brain MR images. A multi-resolution hierarchical feature vector based matching framework was proposed and applied to accurately locate several facial feature-related key points in the 3D brain MR images. An optimal 3D plane which cut through these detected key points was estimated and used to remove facial voxels from MR images. Experiments were conducted to validate the usefulness and applicability of the proposed method.
出处
《四川大学学报(工程科学版)》
EI
CAS
CSCD
北大核心
2013年第5期51-56,共6页
Journal of Sichuan University (Engineering Science Edition)
基金
美国国家卫生院阿兹海默神经影像倡议(ADNI
NIHGrant U01 AG024904)
国家自然科学基金资助项目(81173356)
关键词
核磁共振
大脑
点标记
数据驱动
面部特征
3维重建
magnetic resonance imaging
brain
point landmark
data-driven
facial feature
three-dimensional reconstruction