结直肠癌患者肝脏CT图像分类中决策树模型的应用

目的:探讨数据挖掘中决策树模型在结直肠癌患者肝脏CT图像分类中的应用。方法:分别选取结直肠癌患者肝转移、单纯性肝囊肿以及正常肝脏的CT增强图像各20例。对该60例肝脏CT增强图像分别进行灰度直方图、灰度共生矩阵以及图像变换的纹理特征提取,然后采用朴素贝叶斯分类器和决策树归纳分类器对图像进行分类。最终分类结果与临床事实分类对照,利用十折交叉验证法验证两种分类模型的有效性。结果:基于数据挖掘的决策树模型对结直肠癌患者肝脏CT图像进行分类准确性较高。决策树归纳的分类准确性远高于朴素贝叶斯分类器(准确性96.7%vs 76.7%,Kappa值0.95 vs 0.65,P＜0.05)。结论:基于数据挖掘的决策树模型可以对结直肠癌患者肝脏CT图像进行分类,不仅能够判断肝脏有无相关病灶,而且仅依据图像的基本特性,可以自动识别肝脏乏血供转移瘤与单纯性肝囊肿,为未来计算机辅助诊疗疾病提供有效的参考信息及途径。

Liver CT Image Classification of Colorectal Cancer Patients Based on Decision Tree Model

Objective： To evaluate the application of decision tree model based on data mining in liver CT image classification in patients with colorectal cancer. Methods： 60 patients with colorectal cancer were enrolled in this study, including 20 cases with liver metastasis, 20 cases with simple hepatic cyst and 20 cases with normal liver respectively. All patients underwent CT contrast enhancement examination. The texture features of liver CT images of the 60 cases were extracted by using gray histogram, gray level co-occurrence matrix and image transform. Then use the naive Bias classifier and decision tree model to classify the images. Eventually compare the final classification results with clinical fact and verify the validity of the two classification models with ten-fold cross validation method. Results： For the evaluation of liver lesions, the classification of decision tree based on data mining had much higher accuracy than that of the naive Bias classifier （accuracy 96.7% vs 76.7% P 〈 0.05; Kappa 0.95 vs 0.65, P 〈 0.05）. Conclusion： The decision tree data based on mining model not only can judge whether the liver has related lesions in colorectal cancer patients, but also can automatically identify the liver metastasis and simple hepatic cysts based on the basic characteristics of the image, which may provide the reference information and effective way of computer aided diagnosis and treatment of diseases for the future.