基于CT图像的胰腺癌智能辅助诊断模型研究

Research on intelligent assistant model for pancreatic cancer diagnosis based on CT image

为了更好的为临床疗效评估提供参考,本研究使用胰腺癌患者的APT、DPT、NPT、VPT 4期CT图像数据集进行肿瘤良恶性分类。分别建立深度学习和机器学习模型,评价对胰腺肿瘤良恶性分类的准确率,比较了深度学习和机器学习模型在CT图像应用性能的优势与不足。深度学习模型采用VGG16,训练测试后评估识别率较高的是APT数据集和VPT数据集,其测试集准确率均达到了86%以上;机器学习模型采用支持向量机、逻辑回归、随机森林、K近邻和AdaBoost 5种模型,随机森林和AdaBoost两种模型效果良好,训练测试后识别率较高的是NPT数据集和VPT数据集。在使用AdaBoost模型时,VPT时期的准确率达到了77%。综合表现说明,胰腺癌在VPT时期利用计算机辅助诊断技术具有良好的应用效果。

In order to provide a better reference for the evaluation of clinical efficacy,the four-phase APT,DPT,NPT and VPT CT image datasets of pancreatic cancer patients are used to classify benign and malignant tumors.In this paper,deep learning and machine learning models are established respectively to evaluate the accuracy of benign and malignant pancreatic tumors classification,and the advantages and disadvantages of deep learning and machine learning models in CT image application are analyzed.The deep learning model adopts VGG16,and the APT dataset and VPT dataset have higher recognition rates after training,where the accuracy of the test set is above 86%.Support vector machine,logistic regression,random Forest,K-nearest neighbor and AdaBoost are used in the machine learning model.In these machine learning methods,random forest and AdaBoost have good effects,and NPT dataset and VPT dataset have high recognition rates.When AdaBoost model is used,the accuracy rate in VPT period reaches 77%.The comprehensive experiments show that the application of computer-aided diagnosis in VPT stage of pancreatic cancer has good effect.