术前精准预测胰腺囊性肿瘤的类型，对制定个体化诊疗方案具有重要的临床价值．针对胰腺浆液性和黏液性囊性肿瘤的分类鉴别问题，本文探讨了基于深度学习的多源特征分类模型在胰腺囊性肿瘤的术前辅助诊断中的应用．首先，通过深度学习和影像组学技术从分割图像中提取深度学习特征和影像组学特征，并对病例的临床特征进行评估和量化，然后采用最小绝对收缩选择算子（LASSO）及交叉验证的方法筛选特征，随之构建出两个多源特征模型，即影像组学联合深度学习（RAD_DL）模型、临床特征联合RAD_DL（Clinical_RAD_DL）模型，把传统的影像组学（RAD）模型和深度学习（DL）模型作为对照，最后选用支持向量机（SVM）、自适应提升算法（ADAboost）、随机森林（Random Forest）以及逻辑回归（Logistic）进行分类．采用准确率、召回率、精确率、曲线下面积（AUC）值以及精确率和召回率的调和平均数（F1值）作为评价指标，比较上述4种不同特征模型的分类效能，用校准曲线和决策曲线来评估其临床应用价值．结果显示Clinical_RAD_DL特征模型的分类效能表现最佳，准确率是0.923 1，召回率是0.882 4，精确率是0.882 0，F1是0.882 2，AUC是0.912 6，校准曲线和决策曲线显示出Clinical_RAD_DL特征模型的临床应用价值是最高的．实验表明基于深度学习的多源特征分类模型，对胰腺黏液性和浆液性囊性肿瘤具有较好的分类效果，可以为临床上精准诊疗提供帮助．

This study aims to classify and differentiate mucinous and serous cystic neoplasms of the pancreas using a multi-source feature classification model based on deep learning for preoperative auxiliary diagnosis. Deep learning features and radiomics features were extracted from segmented images using deep learning and radiomics technology, respectively. Clinical features were also evaluated and quantified. LASSO (least absolute shrinkage and selection operator) and cross-validation methods were applied to screen the features, and two multi-source feature models were constructed: the radiomics combined with deep learning (RAD_DL) model and the clinical feature combined with RAD_DL (Clinical_RAD_DL) model. Traditional radiomics (RAD) and deep learning (DL) models were used as controls. SVM (support vector machine), ADAboost (adaptive boosting), Random Forest, and Logistic were selected for classification. The Clinical_RAD_DL feature model shows the best classification performance, with the accuracy of 0.923 1, recall rate of 0.882 4, precision of 0.882 0, F1-score of 0.882 2, and AUC value of 0.912 6. The experimental results indicate that the multi-source feature classification model based on deep learning has good performance in classifying pancreatic serous cystic neoplasms and pancreatic mucinous cystic neoplasms, and can assist clinical accurate diagnosis and treatment.