CNN-OVA-SVM模型在结直肠腺癌多分化诊断中的价值

Value of CNN-OVA-SVM model in multi differentiation diagnosis of colorectal adenocarcinoma

目的探究CNN-OVA-SVM模型应用于多分化类型的结直肠腺癌鉴别诊断中的实用价值。方法本研究选取了2012年1月-2016年3月间新疆医科大学附属肿瘤医院病理科收治的高、中、低分化类型腺癌患者各20例进行回顾性研究。提出了用于多分化类型结直肠腺癌鉴别诊断的CNN-OVA-SVM模型,该模型使用预先训练的ResNet50卷积神经网络对患者的结直肠组织切片图像进行特征提取,使用子分类器为支持向量机(Support Vector Machine,SVM)的一对多集成分类器对结直肠癌进行诊断,并与主流的分类算法K近邻(K-nearest Neighbor,KNN)、随机森林(Random Forests,RF)等进行比较。通过绘制受试者工作特征曲线(Receiver Operating Characteristic,ROC)和混淆矩阵,计算模型的精确率、召回率、F1分数、准确率等指标,对模型的准确率和性能进行直观的评估。结果本研究所提出的CNN-OVA-SVM模型对3种分化类型结直肠腺癌的总体分类准确率为86.11%,AUC值均超过0.88。结论CNN-OVA-SVM模型对于不同分化类型的结直肠腺癌的鉴别具有一定的临床价值。

Objective To explore the practical value of CNN-OVA-SVM model in the differential diagnosis of multi differentiated colorectal adenocarcinoma.Methods Twenty patients with high,medium,and poorly differentiated adenocarcinoma admitted to the Department of Pathology,Affiliated Tumor Hospital of Xinjiang Medical University from January 2012 to March 2016 were selected for a retrospective study.A CNN-OVA-SVM model for the differential diagnosis of multi-differentiated colorectal adenocarcinoma was proposed,which was used by a pre-trained ResNet50 convolutional neural network to extract features from the patient′s colorectal tissue slice image,and by sub-classifiers as support vector machine(SVM)one-to-many ensemble classifier for diagnosis of colorectal cancer,and mainstream classification algorithms K-nearest neighbor(KNN),random forests(Random forests,RF),etc.Receiver operating characteristic(ROC)curve was drawn and confusion matrix,the accuracy rate,recall rate,F1 score,accuracy rate and other indicators of the model were calculated,and the accuracy and performance of the model would be intuitively evaluated.Results The CNN-OVA-SVM model proposed in this study had an overall classification accuracy of 86.11%for the three types of differentiated colorectal adenocarcinoma,and the AUC value exceeded 0.88.Conclusion The CNN-OVA-SVM model has certain clinical value in the identification of different types of colorectal adenocarcinoma.