基于新型支持向量机的影像组学在肝脏结节分类中的应用

doi:10.11938/cjmr20212916

摘要/Abstract

摘要：

肝癌是最常见的恶性肿瘤之一，亚洲地区最为常见的肝癌演变过程为肝炎-肝硬化结节-异型增生结节-肝细胞性肝癌．判断肝脏结节在演变过程所处分期，并采取干预措施，对降低肝癌的发生率非常关键．本文针对影像组学提出了更精确的支持向量机（SVM）分类算法——LFOA-F-SVM，用于对120名患者的腹部动态增强磁共振图像的肝脏结节进行四分类．该算法利用了考虑半径与几何间距的F-SVM，并结合莱维飞行策略（LF）的果蝇优化算法（FOA）寻求超参．为了验证方法的有效性，本文另外添加了5个UCI分类数据集（心脏、帕金森疾病、虹膜、葡萄酒和动物园），并与SVM、PSO-SVM、FOA-SVM、F-SVM进行比较．结果表明，在6个分类数据集（包括肝脏结节数据集和5个UCI分类数据集）中，相对于其他分类算法，LFOA-F-SVM的分类准确率最高，在肝脏结节数据集中的四分类精确率和查全率也较高．

关键词: 肝脏结节, 分类, 影像组学, LFOA-F-SVM

Abstract:

Liver cancer is one of the most common malignant tumors. In Asia, liver cancer often develops on a background of cirrhosis caused by chronic hepatitis. The procedure of hepatitis, cirrhotic nodules, dysplastic nodules, and then hepatocellular carcinoma is the most common liver cancer evolutionary process. Judging the stage of hepatic nodules in the evolution process and taking intervention measures are critical for reducing the incidence of liver cancer. In this paper, a more accurate support vector machine (SVM) classification algorithm, LFOA-F-SVM, was proposed for radiomics to classify hepatic nodules from 120 patients into four categories based on dynamic enhanced magnetic resonance images. The algorithm uses radius-margin-based F-SVM, and combines the fruit fly optimization algorithm (FOA) of Levy flight (LF) strategy to optimize the parameters. To verify the effectiveness of the method, five UCI classification data sets (hearts, Parkinson’s disease, iris, wine and zoo) were added and compared with SVM, PSO-SVM, FOA-SVM, F-SVM. The results showed that LFOA-F-SVM has the highest classification accuracy in six data sets compared to the other methods. And in the hepatic nodules data set, the classification precision and recall are relatively high.

Key words: hepatic nodules, classification, radiomics, LFOA-F-SVM

中图分类号:

O482.53

李笛, 霍雷, 万梦云, 贾宁阳, 王丽嘉. 基于新型支持向量机的影像组学在肝脏结节分类中的应用[J]. 波谱学杂志, 2022, 39(3): 278-290.

Di LI, Lei HUO, Meng-yun WAN, Ning-yang JIA, Li-jia WANG. Application of Radiomics Based on New Support Vector Machine in the Classification of Hepatic Nodules[J]. Chinese Journal of Magnetic Resonance, 2022, 39(3): 278-290.

图/表 10

图1

表1

表2

肝脏结节数据集提取的影像组学特征

特征种类（个数）	特征
一阶统计特征（18）	能量、总能量、熵、最小值、像素10%、像素90%、最大值、均值、中位数、四分位间距、像素范围、平均绝对偏差、鲁棒平均绝对偏差、均方根、偏度、峰度、方差、均匀性
形状特征（14）	网格体积、像素体积、表面积、表面积体积比、球形度、最大3D直径、最大2D直径、最大2D直径列、最大2D直径行、长轴、短轴、最小轴、伸长率、平坦度
灰度共生矩阵（GLCM，24）	自相关、聚类突出度、聚类阴影、聚类趋势、对比度、相关性、差异均值、差异熵、差异方差、相关均值、相关能量、相关熵、相关性的非正式度量1、相关性的非正式度量2、逆差分矩、归一化逆差分矩、逆差分、归一化逆差分、逆方差、最大相关系数、最大概率、求和平均值、求和熵、求和方差
灰度游程长度矩阵（GLRLM，16）	灰度不均匀性、归一化灰度不均匀性、灰度方差、强调高灰度运行、强调长期游程、强调长期游程高灰度级别、强调长期游程低灰度级别、强调低灰度游程、强调短期游程、强调短期高灰度级、强调短期低灰度级、游程熵、游程长度非均匀性、标准化游程长度非均匀性、游程百分比、游程方差
灰度大小区域矩阵（GLSZM，16）	强调小区域、强调大区域、灰度非均匀性、归一化灰度非均匀性、尺寸区域非均匀性、归一化尺寸区域非均匀性、区域百分比、灰度方差、区域方差、区域熵、强调低灰度级区域、强调高灰度级、强调小区域低灰度、强调小区域高灰度、强调大区域低灰度、强调大区域高灰度
邻域灰度差分矩阵（NGTDM，5）	繁忙度、复杂度、对比度、粗糙度、强度
灰度依赖矩阵（GLDM，14）	依赖熵、依赖不均匀性、归一化依赖不均匀性、依赖方差、灰度不均匀性、灰度方差、强调大依赖性、强调高灰度级、强调大依赖性高灰度级、强调低灰度级、强调大依赖性低灰度级、强调小依赖性、强调小依赖性高灰度级、强调小依赖性低灰度级
高阶统计特征（744）	小波（LHL、LHH、HLL、LLH、HLH、HHH、HHL、LLL）变换提取特征，指在三个维度中的每一个维度应用高通（H）或低通（L）滤波器的组合变换后，进而提取的一阶统计和纹理特征

表2

图2

图3

图4

表3

图5

图6

表4

参考文献 32

1	郑荣寿, 孙可欣, 张思维, 等2015年中国恶性肿瘤流行情况分析[J].中华肿瘤杂志,2019,41(1):19-28.
	ZHENG R S, SUN K X, ZHANG S W, et alAnalysis on the prevalence of malignant tumors in China in 2015[J].Chinese Journal of Oncology,2019,41(1):19-28.
2	LAMBIN P, RIOS-VELAZQUEZ E, LEIJENAAR R, et alRadiomics: extracting more information from medical images using advanced feature analysis[J].Eur J Cancer,2012,48(4):441-446. doi: 10.1016/j.ejca.2011.11.036
3	XUE L Y, JIANG Z Y, FU T T, et alTransfer learning radiomics based on multimodal ultrasound imaging for staging liver fibrosis[J].Eur Radiol,2020,30(5):2973-2983. doi: 10.1007/s00330-019-06595-w
4	WU J J, LIU A L, CUI J J, et alRadiomics-based classification of hepatocellular carcinoma and hepatic haemangioma on precontrast magnetic resonance images[J].BMC Med Imaging,2019,19(1):23. doi: 10.1186/s12880-019-0321-9
5	BAEK J, SWANSON T A, TUTHILL T, et al. Support vector machine (SVM) based liver classification: fibrosis, steatosis, and inflammation[C]// International Ultrasonics Symposium. Las Vegas, NV. USA: IEEE, 2020: 1-4.
6	ANJU KRISHNA M, EDWIN D, HARIHARAN S. Classification of liver tumor using modified SFTA based multi class support vector machine[C]// International Conference on Current Trends in Computer, Electrical, Electronics and Communication, Mysore, India. USA: IEEE, 2017, 854-859.
7	LI H Z, GUO S, ZHAO H R, et alAnnual electric load forecasting by a least squares support vector machine with a fruit fly optimization algorithm[J].Energies,2012,5(11):4430-4445. doi: 10.3390/en5114430
8	苗续芝, 陈伟, 毕方明, 等基于改进FOA-SVM的矿井火灾图像识别[J].计算机工程,2019,45(4):267-274.
	MIAO X Z, CHEN W, BI F M, et allMine fire image recognition based on improved FOA-SVM[J].Computer Engineering,2019,45(4):267-274.
9	CHU D L, HE Q, MAO X HRolling bearing fault diagnosis by a novel fruit fly optimization algorithm optimized support vector machine[J].J Vibroeng,2016,18(1):151-164.
10	VAPNIK V, CHAPELLE OBounds on error expectation for support vector machines[J].Neural Comput,2000,12(9):2013-2036. doi: 10.1162/089976600300015042
11	WU X H, ZUO W M, LIN L, et alF-SVM: Combination of feature transformation and SVM learning via convex relaxation[J].IEEE T Neur Net Lear,2018,29(11):5185-5199. doi: 10.1109/TNNLS.2018.2791507
12	CORTES C, VAPNIK V J M LSupport-vector networks[J].Mach Learn,1995,20(3):273-297.
13	石钰阳, 何平, 刘奕, 等基于PSO-SVM模型的多区域多参数MRI脑胶质瘤MGMT分类[J].科学技术创新,2021,17,15-18.
	SHI Y Y, HE P, LIU Y, et alMulti-region and multi-parameter classification of MRI MGMT gliomas based on PSO-SVM model[J].Scientific and Technological Innovation,2021,17,15-18.
14	XU Y Y, YIN W TA block coordinate descent method for regularized multiconvex optimization with applications to nonnegative tensor factorization and completion[J].SIAM J Imaging Sci,2013,6(3):1758-1789. doi: 10.1137/120887795
15	CHEN H L, YANG B, WANG S J, et alTowards an optimal support vector machine classifier using a parallel particle swarm optimization strategy[J].Appl Math Comput,2014,239,180-197.
16	THARWAT A, HASSANIEN, ABOUL EChaotic antlion algorithm for parameter optimization of support vector machine[J].Appl Intell,2018,48(3):670-686. doi: 10.1007/s10489-017-0994-0
17	单黎黎, 张宏军, 王杰, 等一种改进粒子群算法的混合核ε-SVM参数优化及应用[J].计算机应用研究,2013,30(6):1636-1639.
	SHAN L L, ZHANG H J, WANG J, et alParameters optimization and implementation of mixed kernels r epsilon-SVM based on improved PSO algorithm[J].Application Research of Computers,2013,30(6):1636-1639.
18	PAN W TA new fruit fly optimization algorithm: Taking the financial distress model as an example[J].Knowledge-Based Systems,2012,26,69-74. doi: 10.1016/j.knosys.2011.07.001
19	王林, 吕盛祥, 曾宇容果蝇优化算法研究综述[J].控制与决策,2017,32(7):1153-1162.
	WANG L, LV S X, ZENG Y RLiterature survey of fruit fly optimization algorithm[J].Control and Decision,2017,32(7):1153-1162.
20	张水平, 王丽娜果蝇优化算法的进展研究分析[J].计算机工程与应,2021,57(6):22-29.
	ZHANG S P, WANG L NResearch and analysis on progress of fruit fly optimization algorithm[J].Computer Engineering and Applications,2021,57(6):22-29.
21	YANG X S, SUASH D. Cuckoo search via Lévy flights[C]//2009 World Congress on Nature & Biologically Inspired Computing (NaBIC), Coimbatore, India. USA: IEEE, 2009: 210-214.
22	SHEN L M, CHEN H L, YU Z, et alEvolving support vector machines using fruit fly optimization for medical data classification[J].Knowl-Based Syst,2016,96,61-75.
23	GU Q H, CHANG Y X, LI X H, et alA novel F-SVM based on FOA for improving SVM performance[J].Expert Syst Appl,2021,165,113713.
24	ZHANG Y H, WEN J H, WANG X B, et alSemi-supervised learning combining co-training with active learning[J].Expert Syst Appl,2014,41(5):2372-2378.
25	HARALICK R M, SHANMUGAM K, DINSTEIN ITextural features for image classification[J].IEEE T Syst Man Cy-S,1973,SMC-3(6):610-621.
26	GILLIES R, KINAHAN P, HRICAK H J RRadiomics: Images are more than pictures, they are data[J].Radiology,2016,278,563-577.
27	ZOU H, HASTIE TRegularization and variable selection via the elastic net[J].J Roy Stat Soc B,2005,67(5):768.
28	魏志宏, 闫士举, 韩宝三, 等基于多输出的3D卷积神经网络诊断阿尔兹海默病[J].波谱学杂志,2021,38(1):92-100.
	WEI Z H, YAN S J, HAN B S, et alDiagnosis of Alzheimer's disease based on multi-output three-dimensional convolutional neural network[J].Chinese J Magn Reson,2021,38(1):92-100.
29	刘颖, 陈静聪, 胡小洋, 等基于Mask RCNN的桥小脑角区脑膜瘤与听神经瘤分类定位研究[J].波谱学杂志,2021,38(1):58-68.
	LIU Y, CHEN J C, HU X Y, et alClassification and localization of meningioma and acoustic neuroma in cerebellopontine angle based on mask RCNN[J].Chinese J Magn Reson,2021,38(1):58-68.
30	谭章禄, 陈孝慈改进的分类器分类性能评价指标研究[J].统计与信息论坛,2020,35(9):3-8.
	TAN Z L, CHEN X CStudy on evaluation index of improved classifier classification performance[J].Statistics & Information Foroum,2020,35(9):3-8.
31	秦锋, 杨波, 程泽凯分类器性能评价标准研究[J].计算机技术与发展,2006,16(10):85-88.
	QIN F, YANG B, CHENG Z KResearch on measure criteria in evaluating classification performance[J].Computer Technology and Development,2006,16(10):85-88.
32	KIM T K, LEE K H, JANG H J, et alAnalysis of gadobenate dimeglumine-enhanced MR findings for characterizing small (1-2-cm) hepatic nodules in patients at high risk for hepatocellular carcinoma[J].Radiology,2011,259(3):730-738.

数据集	方法
数据集	SVM	PSO-SVM	FOA-SVM	F-SVM	LFOA-F-SVM
心脏	86.77%(0.004 s)	87.74%(0.0035 s)	87.41%(0.0047 s)	88.71%(4.0868 s)	89.35%(6.9537 s)
帕金森疾病	93.27%(0.0021 s)	92.74%(0.0022 s)	93.79%(0.0019 s)	93.68%(8.5848 s)	96.32%(5.215 s)
虹膜	96.00%(0.0006 s)	96.67%(0.0006 s)	96.67%(0.0005 s)	98.00%(0.035 s)	98.67%(0.0237 s)
葡萄酒	96.67%(0.0013 s)	97.78%(0.0011 s)	98.33%(0.0019 s)	98.89%(0.0045 s)	99.44%(0.0057 s)
动物园	93.64%(0.0015 s)	91.82%(0.0012 s)	96.36%(0.0015 s)	97.27%(0.0395 s)	99.09%(0.0092 s)
肝脏结节	74.72%(0.0097 s)	76.11%(0.0092 s)	76.67%(0.0196 s)	76.94%(21.5069 s)	81.00%(14.8110 s)

		SVM	PSO-SVM	FOA-SVM	F-SVM	LFOA-F-SVM
F1	Precision	0.72	0.79	0.82	0.69	0.81
	Recall	0.67	0.70	0.67	0.74	0.78
F2	Precision	0.63	0.72	0.71	0.68	0.77
	Recall	0.59	0.78	0.89	0.70	0.85
F3	Precision	0.80	0.87	0.87	0.78	0.88
	Recall	0.74	0.74	0.74	0.67	0.81
F4	Precision	0.86	0.86	0.83	0.79	0.89
	Recall	0.93	0.93	0.89	0.81	0.89

[1]	魏志宏, 闫士举, 韩宝三, 宋成利. 基于多输出的 3D 卷积神经网络诊断阿尔兹海默病[J]. 波谱学杂志, 2021, 38(1): 92-100.
[2]	刘颖, 陈静聪, 胡小洋, 章浩伟. 基于 Mask RCNN 的桥小脑角区脑膜瘤与听神经瘤分类定位研究[J]. 波谱学杂志, 2021, 38(1): 58-68.
[3]	刘可文, 刘紫龙, 汪香玉, 陈黎, 李钊, 吴光耀, 刘朝阳. 基于级联卷积神经网络的前列腺磁共振图像分类[J]. 波谱学杂志, 2020, 37(2): 152-161.
[4]	孙京文, 闫士举, 韩勇森, 宋成利. 基于脑部磁共振图像三维局部模式变换特征提取进行阿尔茨海默病病程预测分类[J]. 波谱学杂志, 2019, 36(3): 268-277.
[5]	孙伯勤, 邓克俊. 核磁自旋回波串的液体分量分解快速反演法[J]. 波谱学杂志, 2010, 27(3): 298-309.