[1]金海燕,张 锦,王海鹏,等.基于控制点特征学习的前列腺组织轮廓线提取方法[J].计算机技术与发展,2023,33(07):27-33.[doi:10. 3969 / j. issn. 1673-629X. 2023. 07. 004]
　JIN Hai-yan,ZHANG Jin,WANG Hai-peng,et al.Extraction of Prostate Tissue Contour Based on Control Points Feature Learning[J].,2023,33(07):27-33.[doi:10. 3969 / j. issn. 1673-629X. 2023. 07. 004]
点击复制

基于控制点特征学习的前列腺组织轮廓线提取方法()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
33
期数:
2023年07期
页码:
27-33
栏目:
媒体计算
出版日期:
2023-07-10

文章信息/Info

Title:
Extraction of Prostate Tissue Contour Based on Control Points Feature Learning
文章编号:
1673-629X(2023)07-0027-07
作者:
金海燕1; 2 ; 张 锦1 ; 王海鹏1 ; 肖照林1; 2* ; 王 刚3 ; 陈 晶3 ; 张 雨4 ; 白志明3
1. 西安理工大学 计算机科学与工程学院,陕西 西安 710048;
2. 陕西省网络计算与安全技术重点实验室,陕西 西安 710048;
3. 海口市人民医院,海南 海口 570208;4. 海南大学,海南 海口 570228
Author(s):
JIN Hai-yan1; 2 ; ZHANG Jin1 ; WANG Hai-peng1 ; XIAO Zhao-lin1; 2* ; WANG Gang3 ; CHEN Jing3 ; ZHANG Yu4 ; BAI Zhi-ming3
1. School of Computer Science and Engineering,Xi’ an University of Technology,Xi’ an 710048,China;
2. Shaanxi Key Laboratory for Network Computing and Security Technology,Xi’ an 710048,China;
3. Haikou People’ s Hospital,Haikou 570208,China:4. Hainan University,Haikou 570228,China
关键词:
前列腺组织; 弥散加权成像; 控制点学习; U-Net 网络; 特征学习
Keywords:
prostate tissue; diffusion weighted imaging; control points learning; U-Net network; feature learning
分类号:
TP391. 41
DOI:
10. 3969 / j. issn. 1673-629X. 2023. 07. 004
摘要:
前列腺疾病检测和诊断的重要手段之一是分析核磁共振加权成像( T2 Weighted Imaging,T2WI) 与弥散加权成像(Diffusion Weighted Imaging,DWI) 的结果。 对前列腺组织图像识别和标注的工作依赖医生经验且效率较低,大量就诊数据的高效高精度处理成为该领域一大挑战。 目前,在 T2WI 图像上提取轮廓的深度学习图像分割算法已有报道,但在DWI 图像上提取前列腺组织轮廓,仍存在边缘模糊导致的轮廓线提取难题。 针对该问题,提出一种前列腺轮廓控制点的深度学习检测方法。 与直接检测轮廓线不同,该文提出一种 U 型卷积神经网络对轮廓线控制点进行特征学习,以降低由不同患者前列腺轮廓差异导致的特征歧义性问题。 在大量已标注数据集上,采用监督学习方式,提出一种结合控制点概率与空间分布的加权损失函数以优化神经网络收敛速度与检测性能。 在控制点高精度检测的基础上,采用曲线保凸拟合得到最终的前列腺组织轮廓线。 在实验部分,采用前列腺就诊临床数据测试了所提方法的性能,并与直接检测轮廓线方法、多种经典图像分割方法进行了对比。 在实验数据的测试结果表明,该方法在相似性系数指标及豪斯多夫距离指标等方面优于现有其他医学网络分割方法。 此外,该方法由于仅学习轮廓控制点,因此其在小样本数据集上的学习能力显著优于直接检测轮廓线的深度学习方法。
Abstract:
Prostate tissue contour labeling has always been an important step in diagnosis of prostate diseases using magnetic resonanceimaging ( MRI) . The traditional manual labeling relies on clinical experiences,and it is also a time consuming and costly process. Therapid growth of patients demands an automatic prostate tissue contour labeling. In recent years,deep learning image segmentation hasbeen used to extract the contour of prostate tissue in T2 ( Transverse relaxation time) weighted imaging ( T2) . However,there are fewreports on the effective processing of automatic prostate tissue contour extraction on diffusion-weighted imaging ( DWI) images,in whichboundaries and details are blur than T2 images,instead of extracting prostate tissue contours directly. We propose a new learning-basedsolution to detect the control points of prostate tissue contour,and then to generate the contour lines by applying the convex fitting. Based on the labeled contours,a bunch of control points are generated by analyzing the shape characteristic. To precisely detect control points,aU-net structural convolutional neural network is proposed to learn features of control points from a large number labeled samples. Byconsidering both position and distribution information,a weighted loss function is then introduced to optimize the proposed network performance and convergence. With detected control points, a convex - persevering fitting algorithm is finally applied to generate finalprostate tissue contours. In the experiment part,the contour extraction is tested on some real clinical datasets,the proposed method cangenerate precise results in quantitative evaluation, which indicates that the effectiveness of proposed solution when dealing with DWIimage blur.
更新日期/Last Update: 2023-07-10