基于超声心动图的二叶式主动脉瓣钙化病灶自动分割算法

Automatic segmentation of bicuspid aortic valve calcification algorithm based on echocardiography

目的 提出一种基于超声心动图(echocardiography, ECHO)的二叶式主动脉瓣钙化病灶(bicuspid aortic valve calcification,BAVC)自动分割算法,以提高BAVC识别效率。方法 选取BAVC患者的ECHO,应用迭代均值滤波器、一二阶全变分和指定直方图均衡化算法对其进行降噪、灰度平衡化的图像预处理;根据钙化阈值特征,应用阈值滤波器、8邻域原理获取钙化病灶的初始轮廓,并根据ECHO中主动脉瓣的大小和位置特征,自适应地选择初始轮廓作为初始种子区域;根据钙化阈值、迭代终止条件、自适应调整迭代次数等相关参数,自动分割BAVC区域。将自动分割的结果与手动分割结果进行比较,并分析ECHO中不同的椒盐噪声(salt and pepper noise,SPN)及斑点噪声(speckle noise,SN)对分割结果的影响。结果 所建立的自动分割算法的平均处理时间、平均迭代次数分别达到了4.747 s、31,像素精确度(pixel accuracy,PA)、交并比(intersection-over-union,IoU)、Dice系数(Dice coeffcient,DC)分别达到了0.9615、0.9559、0.9773。与MorphGAC半自动算法相比,自动分割算法的平均处理时间提高了45.61%,平均PA、IoU、DC分别提高了13.25%、11.02%、7.69%。结论 该算法能有效地分割出ECHO中的BAVC,并解决人工参与分割个性差异大、识别效率低且易受到噪声干扰的问题.

ObjectiveTo propose an automatic segmentation algorithm of bicuspid aortic valve calcification( BAVC) based on echocardiography( ECHO), and to improve the efficiency of BACC identification.MethodsThe ECHO of BAVC patients was selected, the iterative mean filter and first-and second-order total variation and the specified histogram equalization algorithm was applied to perform image preprocessing of noise reduction and gray balance;the threshold filter and the principle of 8 neighborhood were used to obtain the initial contour of the calcification lesion according to the calcification threshold characteristics, and the initial contour was adaptively selected as the initial seed area according to the size and position characteristics of the aortic valve in ECHO;according to the calcification threshold and iteration termination conditions, adaptive adjustment of related parameters such as the number of iterations, and automatic segmentation of BAVC regions. The results of automatic segmentation were compared with the results of manual segmentation, and the influence of different salt and pepper noise(SPN) and speckle noise( SN) in ECHO on the segmentation results was analyzed.ResultsThe average processing time and the average number of iterations of the established automatic segmentation algorithm reached4.747s and 31 respectively, and the average pixel accuracy(PA), intersection-over-union(IoU), and Dice coefficient( DC) reached 0.9615, 0.9559, and 0.9773 respectively. Compared with MorphGAC semiautomatic algorithms, the average processing time of the automatic segmentation algorithm had increased by 45.61%, and the average PA, IoU, and DC had increased by 13.25%, 11.02%, and 7.69% respectively.ConclusionsThe established algorithm can effectively segment the BAVC in ECHO, and solve the problems of large individual differences, low recognition efficiency, and susceptibility to noise interference in manual segmentation.