基于多模态超声建构甲状腺乳头状癌决策树模型及其应用价值

Construction of Multimodal Ultrasound-based Decision-making Tree Model for the Diagnosis of Papillary Thyroid Carcinoma and Its Application Value

目的分析基于多模态超声建构甲状腺乳头状癌决策树模型及其应用价值。方法选择甲状腺乳头状癌患者106例作为研究对象。多模态超声的甲状腺乳头状癌指标有位置、形态、大小、边界、内部回声、血流分布、淋巴结转移、分级、分期、累及位置共11项指标。在计算机上使用SPSSClementine12.0软件将106例患者的超声指标建模进行分析,包括logistic回归分析和决策树模型。在数据基础上,使用已经筛选的超声指标作为建立数学模型的输入指标,病理诊断结果作为输出,成为建模的金标准。采用SPSS 20.00随机抽取106例患者样本的50%作为训练集,另50%作为预测集验证模型优劣。建模过程分为训练集和测试集2个分支流。训练集先将随机抽样得到的70%的数据资料分别建立Logistic回归与决策树2个数学模型。结果决策树在训练集结果中正确率明显高于Logistic回归(P<0.05)。决策树在测试集结果中正确率明显高于Logistic回归(P<0.05)。决策树灵敏度、特异度和准确度均高于Logistic回归(P<0.05)。Logistic回归的AUC面积为0.554(95%CI:0.392~0.717);决策树的AUC面积为0.810(95%CI:0.676~0.943);决策树诊断的检验准确性更高。结论决策树模型用于多模态超声建构甲状腺乳头状癌诊断具有较好的价值,建议使用。

Objective To construct a multimodal ultrasound-based decision-making tree model for the diagnosis of papillary thyroid carcinoma,and to analyze its application value.Methods 106 patients with papillary thyroid carcinoma were selected.Multimodal ultrasound was applied to monitor various parameters including tumor location,morphology,size,boundaries,internal echogenicity,blood flow distribution,lymph node metastasis,grade,stage,and involvement.The ultrasound indicators of 106 patients were processed via SPSSClementine program,Logistic regression analysis and Decision Tree algorithms.During the model establishment,the selected ultrasound indexes were set as the input indexes,and the pathological diagnosis was used as the output indexes which was the gold standard of modeling.SPSS 20.00 was adopted to randomly select 50%of 106 patients as the training set and the other 50%as the validation set to verify the model.The modeling process was divided into two sets:training set and test set.In training set,70%of the randomly sampled data were used to build two mathematical models,a Logistic regression and a decision tree.Results The accuracy rate of decision tree model in training set was notably higher than that of Logistic regression model(P<0.05),and the accuracy rate of decision tree model in test set was also significantly higher than that of Logistic regression model(P<0.05).The sensitivity,specificity and accuracy of decision tree model were higher than those of Logistic regression model(P<0.05).The AUC of Logistic regression model was 0.554(95%CI:0.392~0.717),and the AUC of decision tree model was 0.810(95%CI:0.676~0.943),indicating that the decision tree model had the higher accuracy for the clinical diagnosis than Logistic regression model.Conclusion Multimodal ultrasound-based decision-making tree model is of good value in the clinical diagnosis of thyroid papillary carcinoma.