一种基于活动轮廓模型的PET-CT肺肿瘤分割方法

宗静静; 邱天爽; 朱广文

doi:10.11999/JEIT200891

一种基于活动轮廓模型的PET-CT肺肿瘤分割方法

doi: 10.11999/JEIT200891

宗静静^{1, 2},
邱天爽^2, ,,
朱广文³

1.
大连交通大学计算机与通信工程学院大连 116028
2.
大连理工大学电子信息与电气工程学部大连 116024
3.
大连医科大学附属第一医院核医学科大连 116011

基金项目: 国家自然科学基金(61671105)，辽宁省教育厅科学研究项目(JDL2020029)

详细信息

作者简介:
宗静静：女，1981年生，博士，讲师，主要研究方向为医学图像处理

邱天爽：男，1954年生，博士，教授，博士生导师，主要研究方向为信号处理、医学图像处理

朱广文：男，1968年生，博士，主任医师，主要研究方向为PET/CT与SPECT/CT肿瘤显像、肿瘤分子核医学、放射性核素治疗

通讯作者:
邱天爽　qiutsh@dlut.edu.cn

中图分类号: TN957.52; TP391.41
计量
- 文章访问数: 766
- HTML全文浏览量: 567
- PDF下载量: 85
- 被引次数: 0
出版历程
- 收稿日期: 2020-10-16
- 修回日期: 2021-09-21
- 网络出版日期: 2021-10-25
- 刊出日期: 2021-12-21

A PET-CT Lung Tumor Segmentation Method Based on Active Contour Model

1.
School of Computer and Communication Engineering, Dalian Jiaotong University, Dalian 116028, China
2.
Faculty of Electronic Information and Electrical Engineering, Dalian University of Technology, Dalian 116024, China
3.
Department of Nuclear Medicine, First Affiliated Hospital, Dalian Medical University, Dalian 116011, China

Funds: The National Natural Science Foundation of China (61671105), The Scientific Research Project of Department of Education, Liaoning Province (JDL2020029)

摘要

摘要: 针对PET-CT肺肿瘤分割中存在的没有充分将医生临床经验融入到算法设计的问题，该文利用PET高斯分布先验，结合区域可伸缩拟合(RSF)模型和最大似然比分类(MLC)准则，提出一种基于变分水平集的混合活动轮廓模型RSF_ML。进一步，借鉴人工勾画肺肿瘤过程中融合图像的重要价值，提出了基于RSF_ML的PET-CT肺肿瘤融合图像分割方法。实验表明，所提出方法较好地实现了有代表性的非小细胞肺肿瘤(Non-Small Cell Lung Cancer, NSCLC)的精确分割，主客观结果优于对比方法，可为临床提供有效的计算机辅助分割结果。
- 活动轮廓模型 /
- 肺肿瘤分割 /
- 变分水平集 /
- 最大似然比分类
Abstract: To solve the problem that the doctors' clinical experience is not fully integrated into the algorithm design in PET-CT lung tumor segmentation, a hybrid active contour model named RSF_ML based on variational level set is proposed by combining with the PET Gaussian distribution prior, Region Scalable Fitting (RSF) model and Maximum Likelihood ratio Classification (MLC) criterion. Furthermore, referring to the important value of fusion image in the process of lung tumor manual delineation, a segmentation method for PET-CT lung tumor fusion image based on RSF_ML is proposed. Experiments show that the proposed method can achieve accurate segmentation of representative Non-Small Cell Lung Cancer (NSCLC), and the subjective and objective results are better than the comparison method, which can provide effective computer-aided segmentation results for clinic.
- Active contour model /
- Lung tumor segmentation /
- Variational level set /
- Maximum Likelihood ratio Classification (MLC)

HTML全文

图 1 PET, CT数据预处理：配准和融合

下载: 全尺寸图片幻灯片

图 2 基于RSF_ML模型的PET-CT肺肿瘤分割方法

下载: 全尺寸图片幻灯片

图 3 PET-CT肺肿瘤图像多模态分割结果

下载: 全尺寸图片幻灯片

表 1 基于RSF_ML模型的图像分割

输入：待分割图像像
输出：分割后的图像
步骤1：将水平集函数$ \scriptstyle\phi $初始化为二进制阶跃函数　　　　 $\scriptstyle \varphi (x,t=0)= $$\scriptstyle{\left\{\begin{aligned}& r,\quad\;\; x{\rm{在} }C{\text{内部} }\\ & -r,\quad {\text{其它} } \end{aligned}\right.}$，其中常量$ \scriptstyle r > 0 $；
步骤2：初始化水平集演化参数，如：迭代次数，$ \scriptstyle{\lambda _1} $, $\scriptstyle {\lambda _2} $, $\scriptstyle \mu $, 　　　　　 $ \scriptstyle\nu $, w, $ \scriptstyle\sigma $, r, $ \scriptstyle\Delta t $等；
步骤3：根据式(10)—式(13)更新均值和方差；
步骤4：基于式(18)更新水平集函数$ \phi $；
步骤5：如果达到收敛，则终止迭代，否则返回步骤3。

下载: 导出CSV

表 2 各方法的DSC

图像	RSF_ML			RSF			RSF&LoG
图像	PET-CT	PET	CT	PET-CT	PET	CT	PET-CT	PET	CT
I₁	0.9585	0.9109	0.9390	0.9616	0.9112	0.9427	0.9627	0.9113	0.9543
I₂	0.9945	0.8298	0.9941	0.9482	0.8301	0.9389	0.9660	0.8320	0.9394
I₃	0.9970	0.8120	0.9800	0.9636	0.7859	0.9542	0.9821	0.8147	0.9689
I₄	0.9898	0.8238	–	0.9635	0.7992	–	0.9661	0.8242	–
I₅	0.9823	0.8313	–	0.9112	0.6206	–	0.9144	0.6459	–
I₆	0.9498	0.9267	–	0.9191	0.8709	–	0.9214	0.8795	–
I₇	0.9899	0.9493	–	0.8825	0.9043	–	0.9273	0.9239	–
I₈	0.9814	0.7475	–	0.6731	0.7363	–	0.7795	0.7441	–
I₉	0.9902	0.8477	–	0.8874	0.7366	–	0.9637	0.8347	–
mean± std	0.9815± 0.0164	0.8532± 0.0640	0.9710± 0.0286	0.9011± 0.0912	0.7995± 0.0933	0.9453± 0.0080	0.9315± 0.0617	0.8234± 0.0859	0.9542± 0.0147

下载: 导出CSV

表 3 各方法的HD

图像	RSF_ML			RSF			RSF&LoG
图像	PET-CT	PET	CT	PET-CT	PET	CT	PET-CT	PET	CT
I₁	0.0693	0.0871	0.0681	0.0739	0.0881	0.0865	0.0817	0.0855	0.0879
I₂	0.0117	0.0884	0.0137	0.0781	0.0864	0.0820	0.0576	0.0864	0.0806
I₃	0.0020	0.0580	0.0239	0.0408	0.0531	0.0439	0.0203	0.0595	0.0298
I₄	0.0257	0.2119	–	0.0825	0.2016	–	0.0775	0.2096	–
I₅	0.1362	0.3594	–	0.2976	0.6549	–	0.2692	0.6444	–
I₆	0.0192	0.0275	–	0.0273	0.0511	–	0.0273	0.0452	–
I₇	0.0164	0.0325	–	0.0869	0.0699	–	0.0688	0.0523	–
I₈	0.0568	0.2503	–	0.3604	0.2607	–	0.2744	0.2486	–
I₉	0.0364	0.2044	–	0.1859	0.3187	–	0.1022	0.2061	–
mean± std	0.0415± 0.0416	0.1466± 0.1149	0.0352± 0.0289	0.1371± 0.1185	0.1983± 0.1971	0.0708± 0.0234	0.1088± 0.0959	0.1820± 0.1899	0.0661± 0.0317

下载: 导出CSV

参考文献(17)

[1]	COSMA L, SOLLAKU S, FRANTELLIZZI V, et al. Early ¹⁸F-FDG PET/CT in COVID-19[J]. Journal of Medical Imaging and Radiation Oncology, 2020, 64(5): 671–673. doi: 10.1111/1754-9485.13099
[2]	SONG Yang, CAI Weidong, and FENG D D. Global context inference for adaptive abnormality detection in PET-CT images[C]. Proceedings of the 9th IEEE International Symposium on Biomedical Imaging (ISBI), Barcelona, Spain, 2012: 482–485. doi: 10.1109/ISBI.2012.6235589.
[3]	DONG Yunyun, YANG Wenkai, WANG Jiawen, et al. An improved supervoxel 3D region growing method based on PET/CT multimodal data for segmentation and reconstruction of GGNs[J]. Multimedia Tools and Applications, 2020, 79(3): 2309–2338. doi: 10.1007/s11042-019-08250-4
[4]	JU Wei, XIANG Deihui, ZHANG Bin, et al. Random walk and graph cut for Co-segmentation of lung tumor on PET-CT images[J]. IEEE Transactions on Image Processing, 2015, 24(12): 5854–5867. doi: 10.1109/TIP.2015.2488902
[5]	LIAN Chunfeng, RUAN Su, DENŒUX T, et al. Joint tumor segmentation in PET-CT images using co-clustering and fusion based on belief functions[J]. IEEE Transactions on Image Processing, 2019, 28(2): 755–766. doi: 10.1109/TIP.2018.2872908
[6]	LI Laquan, LU Wei, TAN Yihua, et al. Variational PET/CT tumor co-segmentation integrated with PET restoration[J]. IEEE Transactions on Radiation and Plasma Medical Sciences, 2020, 4(1): 37–49. doi: 10.1109/TRPMS.2019.2911597
[7]	FOSTER B, BAĞCI U, MANSOOR A, et al. A review on segmentation of positron emission tomography images[J]. Computers in Biology and Medicine, 2014, 50: 76–96. doi: 10.1016/j.compbiomed.2014.04.014
[8]	LI Laquan, ZHAO Xiangming, LU Wei, et al. Deep learning for variational multimodality tumor segmentation in PET/CT[J]. Neurocomputing, 2020, 392: 277–295. doi: 10.1016/j.neucom.2018.10.099
[9]	ZHAO Xiangming, LI Laquan, LU Wei, et al. Tumor co-segmentation in PET/CT using multi-modality fully convolutional neural network[J]. Physics in Medicine & Biology, 2019, 64(1): 015011. doi: 10.1088/1361-6560/aaf44b
[10]	叶锋, 李婉茹, 陈家祯, 等. 基于显著性区域检测和水平集的图像快速分割算法[J]. 电子与信息学报, 2017, 39(11): 2661–2668. doi: 10.11999/JEIT170214 YE Feng, LI Wanru, CHEN Jiazhen, et al. Image fast segmentation algorithm based on saliency region detection and level set[J]. Journal of Electronics &Information Technology, 2017, 39(11): 2661–2668. doi: 10.11999/JEIT170214
[11]	SONG Yangyang, PENG Guohua, SUN Dongwei, et al. Active contours driven by Gaussian function and adaptive-scale local correntropy-based K-means clustering for fast image segmentation[J]. Signal Processing, 2020, 174: 107625. doi: 10.1016/j.sigpro.2020.107625
[12]	JIN Ri and WENG Guirong. Active contour model based on improved fuzzy c-means algorithm and adaptive functions[J]. Computers & Mathematics with Applications, 2019, 78(11): 3678–3691. doi: 10.1016/j.camwa.2019.06.010
[13]	GUO Lu, SHEN Shuming, HARRIS E, et al. A tri-modality image fusion method for target delineation of brain tumors in radiotherapy[J]. PLoS One, 2014, 9(11): e112187. doi: 10.1371/journal.pone.0112187
[14]	BALLANGAN C, WANG Xiuying, FULHAM M, et al. Lung tumor delineation in PET-CT images using a downhill region growing and a Gaussian mixture model[C]. Proceedings of the 18th IEEE International Conference on Image Processing, Brussels, Belgium, 2011: 2173–2176. doi: 10.1109/ICIP.2011.6116042.
[15]	LI Chunming, KAO C Y, GORE J C, et al. Minimization of region-scalable fitting energy for image segmentation[J]. IEEE Transactions on Image Processing, 2008, 17(10): 1940–1949. doi: 10.1109/TIP.2008.2002304
[16]	ZONG Jingjing, QIU Tianshuang, LI Weidong, et al. Automatic ultrasound image segmentation based on local entropy and active contour model[J]. Computers & Mathematics with Applications, 2019, 78(3): 929–943. doi: 10.1016/j.camwa.2019.03.022
[17]	DING Keyan, XIAO Linfang, and WENG Guirong. Active contours driven by region-scalable fitting and optimized Laplacian of Gaussian energy for image segmentation[J]. Signal Processing, 2017, 134: 224–233. doi: 10.1016/j.sigpro.2016.12.021