Ben Ayed Ismail, Lu Yingli, Li Shuo, Ross Ian
GE Healthcare, London, ON, Canada.
Med Image Comput Comput Assist Interv. 2008;11(Pt 1):1025-33. doi: 10.1007/978-3-540-85988-8_122.
This study investigates overlap priors for tracking the Left Ventricle (LV) endo- and epicardium boundaries in cardiac Magnetic Resonance (MR) sequences. It consists of evolving two curves following the Euler-Lagrange minimization of two functionals each containing an original overlap prior constraint. The latter measures the conformity of the overlap between the nonparametric (kernel-based) intensity distributions within the three target regions-LV cavity, myocardium and background-to a prior learned from a given segmentation of the first frame. The Bhattacharyya coefficient is used as an overlap measure. Different from existing intensity-driven constraints, the overlap priors do not assume implicitly that the overlap between the distributions within different regions has to be minimal. Although neither shape priors nor curve coupling were used, quantitative evaluation showed that the results correlate well with independent manual segmentations and the method compares favorably with other recent methods. The overlap priors lead to a LV tracking which is more versatile than existing methods because the solution is not bounded to the shape/intensity characteristics of a training set. We also demonstrate experimentally that the used overlap measures are approximately constant over a cardiac sequence.
本研究探讨了用于在心脏磁共振（MR）序列中跟踪左心室（LV）内膜和外膜边界的重叠先验。它包括通过对两个泛函进行欧拉 - 拉格朗日最小化来演化两条曲线，每个泛函都包含一个原始的重叠先验约束。后者测量三个目标区域（LV腔、心肌和背景）内非参数（基于核）强度分布之间的重叠与从第一帧给定分割中学习到的先验的一致性。使用巴氏系数作为重叠度量。与现有的强度驱动约束不同，重叠先验并不隐含地假设不同区域内分布之间的重叠必须最小。尽管既未使用形状先验也未使用曲线耦合，但定量评估表明结果与独立的手动分割结果相关性良好，并且该方法与其他近期方法相比具有优势。重叠先验导致的LV跟踪比现有方法更具通用性，因为解决方案不受训练集的形状/强度特征限制。我们还通过实验证明，所使用的重叠度量在心脏序列中近似恒定。
