基于CT征象量化分析的肺结节恶性度分级

陈皓; 段红柏; 郭紫园; 强永乾

doi:10.11999/JEIT200167

基于CT征象量化分析的肺结节恶性度分级

doi: 10.11999/JEIT200167

陈皓^{1, 2, ,},
段红柏¹,
郭紫园¹,
强永乾³

1.
西安邮电大学计算机学院西安 710121
2.
陕西省网络数据分析与智能处理重点实验室西安 710121
3.
西安交通大学第一附属医院西安 710061

基金项目: 国家自然科学基金(61876138, 61203311)，陕西省自然科学基金 (2019JM-365)，陕西省教育厅自然科学专项(17JK0701)，陕西省网络数据分析与智能处理重点实验室开放课题基金(XUPT-KLND(201804))，西安邮电大学创新基金(CXJJLI2018017)

详细信息

作者简介:
陈皓：男，1978年生，博士，副教授，硕士生导师，主要研究方向为医疗大数据

段红柏：男，1993年生，硕士生，研究方向为数据挖掘和模式识别

郭紫园：女，1996年生，硕士生，研究方向为数据挖掘和进化计算

强永乾：男，1965年生，博士，副教授，硕士生导师，研究方向为医学影像学

通讯作者:
陈皓　chenhao@xupt.edu.cn

中图分类号: TN911.73; TP391.41
计量
- 文章访问数: 1572
- HTML全文浏览量: 287
- PDF下载量: 60
- 被引次数: 0
出版历程
- 收稿日期: 2020-03-13
- 修回日期: 2020-09-25
- 网络出版日期: 2020-10-16
- 刊出日期: 2021-05-18

Malignancy Grading of Lung Nodules Based on CT Signs Quantization Analysis

1.
School of Computer, Xi`an University of Posts & Telecommunications, Xi’an 710121, China
2.
Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing, Xi’an University of Post and Telecommunications, Xi’an 710121, China
3.
First Affiliated Hospital of Xi`an Jiaotong University, Xi’an 710061, China

Funds: The National Natural Science Foundation of China (61876138, 61203311), The Natural Science Basic Research Program of Shaanxi Province (2019JM-365), The Scientific Research Program Funded by Shaanxi Provincial Education Department (17JK0701), The Science Foundation of the Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing (XUPT-KLND (201804)), The Innovation Funds of Xi'an University of Posts and Telecommunications (CXJJLI2018017)

摘要

摘要: 为了提高肺结节恶性度分级的计算精度及可解释性，该文提出一种基于CT征象量化分析的肺结节恶性度分级方法。首先，融合影像组学特征和通过卷积神经网络提取的高阶特征构造分析CT征象所需的特征集; 接着，在混合特征集的基础上利用进化搜索机制优化集成学习分类器，实现对7种肺结节征象的识别和量化打分; 最后，将7种CT征象的量化打分输入到一个利用差分进化算法优化产生的多分类器，实现肺结节恶性度的分级计算。在实验研究中使用LIDC-IDRI数据集中的2000个肺结节样本进行进化集成学习器和恶性度分级器的训练和测试。实验结果显示对7种CT征象的识别准确率可达0.9642以上，肺结节恶性度分级的准确率为0.8618，精确率为0.8678，召回率为0.8617，F1指标为0.8627。与多个典型算法的比较显示，该文方法不但具有较高的准确率，而且可对相关CT征象进行量化分析，使得对恶性度的分级结果更具可解释性。
- 肺恶性度分级 /
- CT征象 /
- 进化集成学习 /
- 量化分析 /
- 可解释性
Abstract: In order to improve the accuracy and interpretability of the grading of malignant nodules in the lung, a method is proposed to achieve grading automatically for lung nodules by using (Computed Tomography, CT) signs. Firstly, features sets are extracted of CT signs by combing the radiomics features with the higher-order features extracted by convolutional neural network. Then, the ensemble classifier is optimized by the evolutionary search mechanism based on the mixed feature sets, and it is used to realize quantitative scores for 7 CT signs. Finally, 7 quantitative scores are input to the optimized multi-classifier to achieve the grading of malignant nodules in the lung. In the experience, 2000 samples of lung nodules in LIDC-IDRI data set are used to train and test the proposed method. The results show that the recognition accuracy of the 7 CT signs can reach more than 0.9642, the grading accuracy reaches 0.8618, the precision reaches 0.8678, the recall reaches 0.8617, and the F1 index reaches 0.8627. With respect to typical algorithms, the proposed method not only has high accuracy, but also can quantitatively analyze the CT signs that make the grade result of malignancy more interpretive.
- Lung malignancy grading /
- Computed Tomography (CT) signs /
- Evolutionary ensemble learning /
- Quantization analysis /
- Interpretability

HTML全文

图 1 肺结节恶性度分级流程

下载: 全尺寸图片幻灯片

图 2 多特征融合流程

下载: 全尺寸图片幻灯片

图 3 基于进化搜索构造集成学习器的流程

下载: 全尺寸图片幻灯片

图 4 进化集成学习器的优化过程

下载: 全尺寸图片幻灯片

图 5 不同特征集合的聚类结果可视化对比

下载: 全尺寸图片幻灯片

表 1 不同征象的有效特征

特征	阈值	特征总数量	影像组学特征数	CNN特征数
精细度	0.0050	69	28	41
球形度	0.0030	136	23	113
边缘	0.0035	100	28	72
分叶征	0.0009	203	34	169
毛刺征	0.0060	55	23	22
纹理征	0.0070	39	12	27
钙化征	0.0300	68	20	48

下载: 导出CSV

表 2 恶性度分级和CT征象量化的精度

指标	精细度	球形度	边缘	分叶征	毛刺征	纹理征	钙化征	恶性度
ACC	0.9668	0.9764	0.9642	0.9693	0.9792	0.9728	0.9844	0.8618
Pre	0.9648	0.9772	0.9659	0.9522	0.9695	0.9733	0.9683	0.8678
Rec	0.9651	0.9785	0.9346	0.9497	0.9708	0.9736	0.9674	0.8619
F1	0.9649	0.9778	0.9499	0.9509	0.9701	0.9735	0.9678	0.8627

下载: 导出CSV

表 3 恶性度分级模型的权重系数

恶性度等级	精细度	球形度	边缘	分叶征	毛刺征	纹理征	钙化征
1	–0.0703	–0.8560	0.3835	0.2935	0.2751	0.0848	–0.7668
2	–0.0452	0.3054	0.5043	–0.0062	0.0069	–0.6435	–0.4618
3	–0.4187	–0.3507	0.4151	0.3337	0.17592	–0.0367	0.5758
4	0.1984	–0.1948	0.4221	0.0369	0.3217	0.0113	0.2751
5	0.8548	–0.4759	0.2756	–0.2990	0.4641	–0.1176	0.5337

下载: 导出CSV

表 4 不同集成学习器的量化计算结果对比

对比分类器	指标	精细度	球形度	边缘	分叶征	毛刺征	纹理征	钙化征
ET	ACC	0.9638	0.9526	0.9422	0.9603	0.9372	0.9572	0.9512
	Pre	0.9637	0.9558	0.9506	0.9616	0.9372	0.9574	0.9507
	Rec	0.9646	0.9532	0.8994	0.9590	0.9374	0.9572	0.9517
	F1	0.9638	0.9541	0.9197	0.9598	0.9365	0.9573	0.9511
	树个数	112	100	88	108	152	76	76
XGBoost	ACC	0.9621	0.9428	0.9452	0.9560	0.9275	0.9312	0.9498
	Pre	0.9619	0.9442	0.9504	0.9571	0.9276	0.9311	0.9489
	Rec	0.9625	0.9437	0.9266	0.9566	0.9289	0.9311	0.9506
	F1	0.9621	0.9438	0.9377	0.9566	0.9282	0.9311	0.9496
	树个数	188	180	192	188	176	110	80
RF	ACC	0.9585	0.9411	0.9422	0.9491	0.9490	0.9637	0.9471
	Pre	0.9583	0.9449	0.9399	0.9497	0.9492	0.9642	0.9466
	Rec	0.9593	0.9422	0.9166	0.9488	0.9502	0.9637	0.9477
	F1	0.9586	0.9432	0.9270	0.9491	0.9488	0.9639	0.9470
	树个数	128	172	188	136	128	128	72
本文方法	ACC	0.9668	0.9764	0.9642	0.9693	0.9792	0.9728	0.9844
	Pre	0.9648	0.9772	0.9659	0.9522	0.9695	0.9733	0.9683
	Rec	0.9651	0.9785	0.9346	0.9697	0.9708	0.9736	0.9674
	F1	0.9649	0.9778	0.9499	0.9509	0.9701	0.9735	0.9678
	树个数	77	67	78	54	70	60	23

下载: 导出CSV

表 5 相关文献的量化结果对比

文献	精细度	球形度	边缘	分叶征	毛刺征	纹理征	钙化征	恶性度
文献[11]	0.7190	0.5222	0.7250	/	/	0.8340	0.9080	0.8420
文献[22]	0.7163	0.7392	0.7332	0.7248	0.7422	0.7677	0.7390	0.8194
文献[23]	0.7431	0.7622	0.7013	0.8001	0.7828	0.9002	/	0.7556
文献[24]	0.8933	0.8933	0.8933	0.8933	0.8933	0.8933	0.8933	0.8933
本文方法	0.9668	0.9764	0.9642	0.9693	0.9792	0.9728	0.9844	0.8618

下载: 导出CSV

表 6 不同特征集合的聚类结果对比

特征集	均一性	v-measure	互信息
影像组学特征	0.32610	0.3233	0.3179
CNN特征	0.44342	0.4282	0.4118
融合特征	0.60850	0.5934	0.5771

下载: 导出CSV

参考文献(24)

[1]	MCWILLIAMS A, TAMMEMAGI M C, MAYO J R, et al. Probability of cancer in pulmonary nodules detected on first screening CT[J]. New England Journal of Medicine, 2013, 369(10): 910–919. doi: 10.1056/NEJMoa1214726
[2]	NAIDICH D P, BANKIER A A, MACMAHON H, et al. Recommendations for the management of subsolid pulmonary nodules detected at CT: A statement from the fleischner society[J]. Radiology, 2013, 266(1): 304–317. doi: 10.1148/radiol.12120628
[3]	GOULD M K, DONINGTON J, LYNCH W R, et al. Evaluation of individuals with pulmonary nodules: When is it lung cancer?: Diagnosis and management of lung cancer, 3rd ed: American college of chest physicians evidence-based clinical practice guidelines[J]. Chest, 2013, 143(5S): e93S–e120S. doi: 10.1378/chest.12-2351
[4]	TRAVIS W D, BRAMBILLA E, NOGUCHI M, et al. International association for the study of lung cancer/American thoracic society/European respiratory society: International multidisciplinary classification of lung adenocarcinoma[J]. Proceedings of the American Thoracic Society, 2011, 8(5): 381–385. doi: 10.1513/pats.201107-042ST
[5]	RODRIGUES M B, DA NÓBREGA R V M, ALVES S S A, et al. Health of things algorithms for malignancy level classification of lung nodules[J]. IEEE Access, 2018, 6: 18592–18601. doi: 10.1109/ACCESS.2018.2817614
[6]	DA NÓBREGA R V M, PEIXOTO S A, DA SILVA S S P, et al. Lung nodule classification via deep transfer learning in CT lung images[C]. The IEEE 31st International Symposium on Computer-Based Medical Systems, Karlstad, Sweden, 2018: 244–249.
[7]	ZUO Wangxia, ZHOU Fuqiang, LI Zuoxin, et al. Multi-resolution CNN and knowledge transfer for candidate classification in lung nodule detection[J]. IEEE Access, 2019, 7: 32510–32521. doi: 10.1109/ACCESS.2019.2903587
[8]	SHEN Shiwen, HAN S X, ABERLE D R, et al. An interpretable deep hierarchical semantic convolutional neural network for lung nodule malignancy classification[J]. Expert Systems with Applications, 2019, 128: 84–95. doi: 10.1016/j.eswa.2019.01.048
[9]	WANG Huafeng, ZHAO Tingting, LI Lihong, et al. A hybrid CNN feature model for pulmonary nodule malignancy risk differentiation[J]. Journal of X-Ray Science and Technology, 2018, 26(2): 171–187. doi: 10.3233/XST-17302
[10]	褚征, 于炯. 基于随机森林的流处理检查点性能预测[J]. 电子与信息学报, 2020, 42(6): 1452–1459. doi: 10.11999/JEIT190552 CHU Zheng and YU Jiong. Performance prediction based on random forest for the stream processing checkpoint[J]. Journal of Electronics &Information Technology, 2020, 42(6): 1452–1459. doi: 10.11999/JEIT190552
[11]	NISHIO M, NISHIZAWA M, SUGIYAMA O, et al. Computer-aided diagnosis of lung nodule using gradient tree boosting and bayesian optimization[J]. PLoS One, 2018, 13(4): e0195875. doi: 10.1371/journal.pone.0195875
[12]	吴艇帆, 张仁寿. 机器学习在肺内恶性磨玻璃密度结节的应用研究[J]. 广州大学学报: 自然科学版, 2018, 17(3): 33–39. WU Tingfan and ZHANG Renshou. Research on the application of machine learning in the malignant grinding glass density nodules of lung[J]. Journal of Guangzhou University:Natural Science Edition, 2018, 17(3): 33–39.
[13]	FERREIRA JR J R, OLIVEIRA M C, and DE AZEVEDO-MARQUES P M. Characterization of pulmonary nodules based on features of margin sharpness and texture[J]. Journal of Digital Imaging, 2018, 31(4): 451–463. doi: 10.1007/s10278-017-0029-8
[14]	WU Wenhao, HU Huihui, GONG Jing, et al. Malignant-benign classification of pulmonary nodules based on random forest aided by clustering analysis[J]. Physics in Medicine & Biology, 2019, 64(3): 035017. doi: 10.1088/1361-6560/aafab0
[15]	SHAUKAT F, RAJA G, ASHRAF R, et al. Artificial neural network based classification of lung nodules in CT images using intensity, shape and texture features[J]. Journal of Ambient Intelligence and Humanized Computing, 2019, 10(10): 4135–4159. doi: 10.1007/s12652-019-01173-w
[16]	李双双, 侯震, 刘娟, 等. 影像组学分析与建模工具综述[J]. 中国医学物理学杂志, 2018, 35(9): 1043–1049. doi: 10.3969/j.issn.1005-202X.2018.09.010 LI Shuangshuang, HOU Zhen, LIU Juan, et al. Review of radiomic analysis and modeling tools[J]. Chinese Journal of Medical Physics, 2018, 35(9): 1043–1049. doi: 10.3969/j.issn.1005-202X.2018.09.010
[17]	SHI Zhenghao, HAO Huan, ZHAO Minghua, et al. A deep CNN based transfer learning method for false positive reduction[J]. Multimedia Tools and Applications, 2019, 78(1): 1017–1033. doi: 10.1007/s11042-018-6082-6
[18]	刘家辰, 苗启广, 曹莹, 等. 基于混合多样性生成与修剪的集成单类分类算法[J]. 电子与信息学报, 2015, 37(2): 386–393. doi: 10.11999/JEIT140161 LIU Jiachen, MIAO Qiguang, CAO Ying, et al. Ensemble one-class classifiers based on hybrid diversity generation and pruning[J]. Journal of Electronics &Information Technology, 2015, 37(2): 386–393. doi: 10.11999/JEIT140161
[19]	ELYAN E and GABER M M. A fine-grained random forests using class decomposition: An application to medical diagnosis[J]. Neural Computing and Applications, 2016, 27(8): 2279–2288. doi: 10.1007/s00521-015-2064-z
[20]	MIAO Fen, CAI Yunpeng, ZHANG Yuxiao, et al. Predictive modeling of hospital mortality for patients with heart failure by using an improved random survival forest[J]. IEEE Access, 2018, 6: 7244–7253. doi: 10.1109/ACCESS.2018.2789898
[21]	PAUL A, MUKHERJEE D P, DAS P, et al. Improved random forest for classification[J]. IEEE Transactions on Image Processing, 2018, 27(8): 4012–4024. doi: 10.1109/tip.2018.2834830
[22]	WANG Qingfeng, CHENG Jiezhi, LIU Zhiqin, et al. Multi-order transfer learning for pathologic diagnosis of pulmonary nodule malignancy[C]. 2018 IEEE International Conference on Bioinformatics and Biomedicine, Madrid, Spain, 2018: 2813–2815.
[23]	YANG Jing, LI Na, FANG Shuai, et al. Semantic features prediction for pulmonary nodule diagnosis based on online streaming feature selection[J]. IEEE Access, 2019, 7: 61121–61135. doi: 10.1109/ACCESS.2019.2903682
[24]	WU Botong, ZHOU Zhen, WANG Jianwei, et al. Joint learning for pulmonary nodule segmentation, attributes and malignancy prediction[C]. The IEEE 15th International Symposium on Biomedical Imaging, Washington, USA, 2018: 1109–1113.