基于Polar1DMLP模型的CCTA冠脉管腔分割方法研究

冠状动脉数字图像造影(CCTA)是一种有效的无创评估冠脉血管狭窄等病变情况的成像技术,对CCTA的自动筛查评估依赖于冠脉管腔的高精度分割。为探索能够分割出高质量的冠脉官腔的算法,本文进行基于深度学习的端到端分割实验以及基于中心线先验信息结合CCTA灰度特征的冠脉管腔分割实验,其中基于深度学习回归方法的Polar1DMLP模型能够结合中心线先验信息得到较好的分割效果。基于公开数据集Coronary Artery Stenoses Detection and Quantification Evaluation Framework中的78组冠脉截段数据进行训练与验证,在16段数据的验证集上得到MSD(mean surface distance)为0.169 mm,DICE为0.796。结果表明本文提出的以中心线为导向信息的Polar1DMLP模型能够较好地整合血管CCTA灰度特征,回归出较为准确的冠脉血管内壁管腔轮廓半径,得到较为平滑的冠脉管腔表面模型,本方法有着较大的潜力以及拓展空间。

Polar1DMLP: A Coronary Artery Lumen Segmentation Network in CCTA

As a reliable and non-invasive medical imaging method,Coronary Computed Tomography Angiography(CCTA) has been used to detect the stenoses and other lesions in coronary arteries.However,effective and automated CCTA imaging examination is based on precise coronary arteries lumen segmentation technology.The purpose of this study was to investigate a model which can get the high-quality 3D surface model of the coronary lumen.Here we proposed the deep learning based 1D Polar1MLP model,which can make good use of the complicated information of the coronary tree centerline information.We trained and evaluated our model with the publicly available Coronary Artery Stenoses Detection and Quantification Evaluation Framework(Rotterdam) including 78 coronary segments with experts'manual contour labels of them,and got the result with a Dice similarity coefficient of 0.796,mean surface distance(MSD) of 0.169 mm in the validation dataset with 16 segments.The result of the study indicated that the 1DPolarMLP model with consideration of the CT gray-level information and centerline guideline information,can predict more precise and smoother 3D surface model of the coronary.