基于数据挖掘的触诊成像乳腺癌智能诊断模型和方法

doi:10.11959/j.issn.2096-0271.2019005

摘要/Abstract

摘要：

为了辅助医护人员利用触诊成像技术判定乳腺癌，提出了触诊成像乳腺癌智能诊断模型和方法。采用乳腺癌早期筛查及风险评估的临床数据，以触诊成像诊断结果为对比数据，通过决策树等机器学习算法以及投票法，对乳腺肿瘤的良恶性质进行判定。使用SMOTE算法对数据进行处理，建立了诊断模型和方法，自动完成对乳腺肿瘤性质的诊断。实验结果表明，乳腺癌正确筛查的准确性达到98%，提出的方法具有很好的应用价值。

关键词: 智能诊断, 临床数据, 机器学习, SMOTE算法

Abstract:

In order to assist the medical staff to diagnose breast cancer more effectively by palpation imaging technology, intelligent diagnosis model and method of palpation imaging breast cancer were established. Based on clinical data for early breast cancer screening and risk assessment, machine learning algorithms of decision tree, neural network, SVM, logistic regression, Bayesian network and five voting methods were adopted to distinguish breast tumor, or positive and negative outcome in algorithms. The positive sample data was incremented by the SMOTE algorithm, intelligent diagnosis model was established, and model can automatically diagnose breast tumors. Palpation imaging intelligent diagnosis model of breast cancer correctly screens all cases of breast cancer confirmed by pathology, and the accuracy of the model is as high as 98%. The intelligent diagnosis model is excellent as a screening modality for the detection of breast cancer.

Key words: intelligent diagnosis, clinical data, machine learning, SMOTE algorithm

中图分类号:

TP18

张旭东, 孙圣力, 王洪超. 基于数据挖掘的触诊成像乳腺癌智能诊断模型和方法[J]. 大数据, 2019, 5(1): 68-76.

Xudong ZHANG, Shengli SUN, Hongchao WANG. Intelligent diagnosis model and method of palpation imaging breast cancer based on data mining[J]. Big Data Research, 2019, 5(1): 68-76.

图/表 14

图1

图2

图3

表1

图4

表2

表3

表4

表5

图5

表6

表7

表8

图6

参考文献 16

[1]	郝淳敏 . 乳腺癌二级预防筛查模式的研究[J]. 中华健康管理学杂志， 2012,6(2): 131-132.
	HAO C M . Research on secondary preventive screening model for breast cancer[J]. Chinese Journal of Health Management, 2012,6(2): 131-132.
[2]	毕晓峰，徐志坚 . 乳腺癌筛查技术研究现状与展望[J]. 中华健康管理学杂志， 2016(4): 321-325.
	BI X F , XU Z J . Technology of breast cancer screening: current situation and prospect[J]. Chinese Journal of Health Management, 2016(4): 321-325.
[3]	王本忠，付君 . 乳腺触诊成像影像诊断学[M]. 北京：科学出版社， 2016.
	WANG B Z , FU J . Diagnostics of breast palpation imaging[M]. Beijing: China Science Publishing House, 2016.
[4]	宋颖，李静，张仁知，等. 乳腺可视化触诊成像系统鉴别诊断乳腺良恶性病变：与临床触诊、X线及超声对照[J]. 中国医学影像技术， 2014,30(4): 527-530.
	SONG Y , LI J , ZHANG R Z , et al. Palpation imaging system in differentiating benign and malignant breast lesions:comparedwith physical examination, mammographyand ultrasonograph[J]. Chinese Journal of Medical Imaging Technology, 2014,30(4): 527-530.
[5]	盖小荣，王振捷，王健，等. 触诊成像系统在乳腺疾病健康筛查中的应用研究[J]. 中华健康管理学杂志， 2013,7(6): 398-401.
	GAI X R , WANG Z J , WANG J , et al. The application of palpation imaging in the screening of breast disease[J]. Chinese Journal of Health Management, 2013,7(6): 398-401.
[6]	于志勇，左文述，刘岩松，等. 乳腺触觉成像诊断系统的临床应用评估[J]. 中华肿瘤防治杂志， 2011(1): 50-53.
	YU Z Y , ZUO W S , LIU Y S , et al. Clinical evaluation of breast palpation imaging system[J]. Chinese Journal of Cancer Prevention and Treatment, 2011(1): 50-53.
[7]	方礼妮，郑春兰 . 女性乳腺癌患者主要照顾者抑郁状况及影响因素研究[J]. 中华健康管理学杂志， 2015,9(3): 209-211.
	FANG L N , ZHENG C L . Study on depression status and influencing factors of main caregivers of female breast cancer patients[J]. Chinese Journal of Health Management, 2015,9(3): 209-211.
[8]	王忠民，曹洪江，范琳 . 一种基于卷积神经网络深度学习的人体行为识别方法[J]. 计算机科学， 2016,43(11A): 56-58,87.
	WANG Z M , CAO H J , FAN L , et al. Method on human activity recognition based on convolutional neural networks[J]. Computer Science, 2016,43(11A): 56-58,87.
[9]	沈花玉，王兆霞，高成耀 . BP神经网络隐含层单元数的确定[J]. 天津理工大学学报， 2008,24(5): 13-15.
	SHEN H Y , WANG Z X , GAO C Y , et al. Determining the number of BP neural network hidden layer units[J]. Journal of Tianjin University of Technology, 2008,24(5): 13-15.
[10]	郑伟，马楠 . 一种改进的决策树后剪枝算法[J]. 计算机与数字工程， 2015(6): 960-966,971.
	ZHENG W , MA N . An improved postpruning algorithm for decision tree[J]. Computer and Digital Engineering, 2015(6): 960-966,971.
[11]	张琪，周琳，陈亮，等. 决策树模型用于结核病治疗方案的分类和预判[J]. 中华疾病控制杂志， 2015(5): 510-513.
	ZHANG Q , ZHOU L , CHEN L , et al. A decision tree model for classification and prediction of tuberculosis treatment[J]. Chinese Journal of Disease Control &Prevention, 2015(5): 510-513.
[12]	李雅秋，王旗 . 构建用于预测中药化学成分心脏毒性的定量构效关系模型[J]. 北京大学学报（医学版）， 2017,49(3): 551-556.
	LI Y Q , WANG Q . Quantitative structureactivity relationship model for prediction of cardiotoxicity of chemical components in traditional Chinese medicines[J]. Journal of Peking University Health Science, 2017,49(3): 551-556.
[13]	徐斌，苏一丹，黄山 . 基于km-smote和随机森林的不平衡数据分类[J]. 计算机技术与发展， 2015,25(9): 17-21.
	XU B , SU Y D , HUANG S . Classification of imbalance data based on km-smote algorithm and random forest[J]. Computer Technology and Development, 2015,25(9): 17-21.
[14]	李克文，杨磊，刘文英，等. 基于RSBoost算法的不平衡数据分类方法[J]. 计算机科学， 2015,42(9): 249-252，267.
	LI K W , YANG L , LIU W Y , et al. Classification of imbalance data based on KM-SMOTE algorithm and Random Forest[J]. Computer Technology and Development, 2015,42(9): 249-252，267.
[15]	肖魏娜，张为群，王玲玲 . 一种基于BP神经网络的软件需求分析风险评估模型的研究[J]. 计算机科学， 2011,38(4): 199-202.
	XIAO W N , ZHANG W Q , WANG L L . Study on evaluation model of software requirement analysis risk based on neural network[J]. Computer Science, 2011,38(4): 199-202.
[16]	张月，黄钢，章小雷，等. 贝叶斯网络在医学领域中的应用研究[J]. 中国医学创新， 2013(4): 145-146.
	ZHANG Y , HUANG G , ZHANG X L , et al. Study on application of bayesian networks in the medical field[J]. Medical Innovation of China, 2013(4): 145-146.

	数据集1	数据集2	数据集3
查重前阳性数据量/条	50	11	74
查重后阳性数据量/条	185	22	96

数据状况	使用SMOTE次数/次	阴性数据量/条	阳性数据量/条	准确率	精确率	召回率	F值
训练集A	1	15 487	606	99.21%	90.53%	88.03%	89.26%
训练集B	2	15 487	1 212	98.72%	89.80%	92.71%	91.23%
训练集C	3	15 487	2 424	98.75%	92.30%	98.99%	95.53%

数据名	数据状况	数据说明
训练集	16 120条数据	数据经过预处理、查重并处理、SMOTE阳性增量处理
特殊测试集1	1 549条数据	数据经过预处理
特殊测试集2	902条数据	数据经过预处理
随机测试集A~F	每个数据集有1 579条数据	数据经过预处理

算法名	准确率	精确率	召回率	F值
决策树	99.59%	89.66%	61.42%	72.90%
	99.75%	92.30%	98.99%	95.53%
SVM	98.67%	99.59%	99.59%	99.59%
	98.67%	98.75%	98.75%	98.75%
贝叶斯网络	93.21%	68.86%	90.55%	78.23%
	98.57%	95.41%	92.70%	94.63%
神经网络
	96.57%	94.89%	93.87%	93.99%
逻辑回归	93.98%	81.11%	72.06%	76.31%

数据状况	参数调整	模型调整	基分类器	准确率	精确率	召回率	F值
训练集	采用	不采用	神经网络	98.57%	95.41%	93.87%	94.63%
训练集	采用	采用	神经网络	99.15%	95.16%	98.75%	96.92%