决策树技术在乳腺癌诊断中的应用

The use of decision tree in diagnosis of breast cancer

目的探讨决策树统计分析方法在乳腺癌诊断中的适用性，为发展乳腺癌的计算机辅助诊断提供依据。方法将699例乳腺肿块病例完全随机分成两部分，70％病例作为训练集，用来建立决策树模型；30％病例作为测试集，用来验证模型的有效性。以细针吸取细胞学的9项指标作为自变量，以手术后病理检查的诊断结果作为因变量，利用SPSS的TREE模块建立决策树分类规则。将测试集数据输入建立好的模型，将模型的诊断结果与术后病理诊断进行比较，以灵敏度、特异度、正确率来评价模型的有效性。结果该模型在乳腺肿块病例中诊断出恶性病例的灵敏度为92．7％（95％CI：86．2％～99．1％），特异度为92．5％（95％CI：87．9％～97．2％），诊断的正确率为92．6％（95％CI：88．8％～96．3％）。结论基于决策树技术的计算机诊断系统在目前技术条件下已达到了不逊于人工诊断的效果。随着相关技术的进步，该模型的诊断效果还有提高的空间。

Objective To explore the use of computer-aided decision tree method in diagnosis of breast cancer, and to provide evidence for the development of computer-assistant diagnosis of breast cancer. Methods Six hundred and ninety-nine cases of breast mass were randomized into two parts, including 70% of cases assigned as the training set for establishing decision tree models and the remaining 30% cases as test set for verification of the effectiveness of these models. The tree module of SPSS was used to establish the classification regulation of decision tree, with 9 indicators of fine needle aspiration cytology （FNAC）, as independent variables and the diagnostic results of postoperative pathological examination as dependent variables. Thereafter, the data of test set was input into the established models. The diagnostic results and postoperative pathological diagnosis were then compared for evaluation of the effectiveness of models according to sensitivity, specificity and accuracy rate. Results The sensitivity, specificity and accuracy rate were 92.7% （95%CI： 86.2%-99.1%）, 92.5% （95%CI： 87.9%-97.2%） and 92.6% （95%CI： 88.8%N 96.3% ） in the malignant cases diagnosed from cases with breast mass by models, respectively. Conclusion So far, the computer diagnostic system based on decision tree has achieved the effectiveness not worse than that of artificial diagnosis under current conditions. Along with the development of relative techniques, improvement of diagnostic effectiveness of models will go a long way.