【摘要】 目的：研究将口腔癌患者术前临床病理资料与CT影像组学特征相结合的口腔癌淋巴结转移（Lymph node metastasis,LNM）诊断的临床价值。方法：回顾性分析2016年1月至2023年3月确诊为口腔癌的295例患者的临床、病理和影像学资料。排除数据不全和影像不清晰病例后，最终纳入218例，随机分为训练组和验证组（训练组152例，验证组66例）。2名经验丰富的影像科医师使用ITK-SNAP软件标注出肿瘤和淋巴结转移的感兴趣区域（Regions of interest,ROI）。构建2个独立分支：一个使用多层感知器（Multilayer perceptron,MLP）对CT数据进行特征提取和降维，另一个提取临床病理特征。在多层感知器的第三层神经网络中，将临床病理特征和影像组学特征进行融合，最终构建一个基于临床病理特征和影像组学特征的口腔癌淋巴结转移诊断模型。结果：结合临床病理特征和影像组学特征的模型在诊断口腔癌淋巴结转移方面表现良好，训练组和验证组曲线下面积（Area under the curve,AUC）分别为0.997和0.961。临床病理-影像组学模型的校准曲线和决策曲线均优于单独的影像组学模型和临床病理模型。结论：结合术前临床病理特征和CT影像组学特征构建的模型在预测口腔癌淋巴结转移方面表现出较高的效能，有望在临床实践中广泛应用，以推动淋巴结转移诊断的精准化和个性化发展。更多还原

【Abstract】 Objective : To explore the clinical value of combining preoperative clinical pathological data and CT radiomic features for diagnosing lymph node metastasis(LNM) in patients with oral cancer. Methods : A retrospective analysis was conducted on the clinical, pathological, and imaging data of 295 patients diagnosed with oral cancer from January 2016 to March 2023. After excluding patients with incomplete data and unclear imaging, 218 cases were included in the study. They were randomly divided into training group and validation group(training group: 152 cases, validation group: 66 cases). Two experienced radiologists used ITK-SNAP software to delineate the regions of interest(ROI) for tumors and lymph node metastasis. Subsequently, two independent branches were constructed: one branch used a multilayer perceptron(MLP) for feature extraction and dimensionality reduction of the CT data, while the other branch extracted clinical pathological features. In the third layer of the MLP neural network, these two types of features were integrated to ultimately construct a diagnostic model for lymph node metastasis in oral cancer, based on clinical pathological features and radiomic features. Results : The model combining clinical pathological features and radiomic features performed well in diagnosing lymph node metastasis in oral cancer, with area under the curve(AUC) of 0.997 and 0.961 for the training and validation sets, respectively. The calibration curve and decision curve of the clinical pathological-radiomic model outperformed those of the radiomic model and the clinical pathological model alone. Conclusion: The model constructed by combining preoperative CT and clinical pathological features demonstrated high efficiency in predicting lymph node metastasis in oral cancer. It has the potential for widespread application in clinical practice, promoting the precision and personalization of lymph node metastasis diagnosis.更多还原