利用二维格子复杂性挖掘肝癌CT图像预后信息  

Prognostic information extract from CT images of patients with liver cancer using two-dimensional lattice complexity

作　　者：武瑞霞 张子瑞[1] 陈宇彬 叶苏哲 郑明华[2] 柯大观[1] WU Ruixi;ZHANG Ziru;CHEN Yubin;YE Suzhe;ZHENG Minghua;KE Daguan(School of Biomedical Engineering,Wenzhou Medical University,Wenzhou,325035;Department of Hepatology,the First Affiliated Hospital of Wenzhou Medical University,Wenzhou,325015)

机构地区：[1]温州医科大学生物医学工程学院,浙江温州325035 [2]温州医科大学附属第一医院感染内科,浙江温州325015

出　　处：《温州医科大学学报》2018年第6期396-400,共5页Journal of Wenzhou Medical University

基　　金：国家自然科学基金资助项目(11005081)

摘　　要：目的:初步验证通过二维格子复杂性能否有效提取某些医学图像中隐含的预后信息。方法:将92例原发性肝癌患者手术前的原始腹部CT图像转为32像素×32像素的二值图像,利用二维希尔伯特曲线将图像转化为一维符号序列并计算其格子复杂性特征,利用支持向量机就全部病例进行十折交叉验证,基于46例患者的特征建立分类模型并检验对其余46位患者术后存活时间的模式识别效果。结果:在每位患者使用28幅图像的情况下,控制参数为19的格子复杂性在十折交叉验证中的平均准确率达到75.0%,在46例测试集上的识别准确率为69.6%。结论:二维格子复杂性算法能够挖掘出CT图像中人眼捕捉不到的预后信息。Objective： To verify whether the two-dimensional（2-D） lattice complexity algorithm can effectively extract hidden prognostic information in some medical images. Methods： The preoperative abdominal CT images of 92 patients with hepatocellular carcinoma（HCC） were converted into binary images with a size of 32×32 pixels and then into one-dimensional binary symbolic sequence by using two-dimensional Hilbert curves for calculating lattice complexity. All cases were scanned by support vector machine（SVM） for 10-fold crossvalidation to select the best characteristics. And then, based on the characteristics of 46 patients, a classification model was established to identify the pattern recognition of other 46 patients＇ survival period. Results： When using 28 images for each patient, the best average accuracy of classification in 10-fold cross-validation was 75.0% with the lattice complexity under the control parameter 19. For the 46 patients, the test accuracy was 69.6%. Conclusion： With 2-D lattice complexity, we could extract from CT images some prognostic information not previously captured.

关 键 词：肝癌CT图像 二维格子复杂性 支持向量机 模式识别 希尔伯特曲线 

分 类 号：R811[医药卫生—放射医学]