Abstract:

Objective　To construct and verify an intelligent prediction system for recurrence of patients with differentiated thyroid cancer (DTC) after endoscopic surgery based on depth algorithm. Methods　The clinical data of 189 patients with DTC who underwent endoscopic surgery at Shangluo Central Hospital from October 2020 to October 2023 were retrospectively analyzed, and were divided into a training set (126 cases) and a validation set (63 cases) by simple random sampling according to the ratio of 2∶1. In the training set, there were 39 males and 87 females who were (49.14±7.78) years old. In the verification set, there were 22 males and 41 females who were (50.38±8.12) years old. The patients in the training set were divided into a recurrence group (24 cases) and a non-recurrence group (102 cases) according to whether they had recurred or not. The Cox regression analysis was used to explore the factors affecting the patients' recurrence. Based on the obtained influencing factors, the convolutional neural network (CNN) depth algorithm was used to construct an intelligent prediction system for the patients' recurrence. The receiver operating characteristic curve (ROC) was used to evaluate the predictive efficacy of the intelligent prediction system. t and χ² tests were applied. Results　The postoperative recurrence rate of the patients undergoing endoscopic treatment was 18.52% (35/189). Maximum tumor diameter ≥2 cm [hazard ratio (HR)=1.660, 95% confidence interval (95%CI) 1.169-2.358], moderate differentiation (HR=1.484, 95%CI 1.081-2.039), lymph node metastasis (HR=1.876, 95%CI 1.258-2.798), subtotal thyroidectomy (HR=1.800, 95%CI 1.238-2.618), no adjuvant therapy (HR=1.737, 95%CI 1.213-2.486), and serum thyroid globulin antibody (TgAb) ≥40 IU/ml (HR=1.590, 95%CI 1.126-2.246) were risk factors for the patients' recurrence (all P<0.05). The accuracy of the DTC endoscopic postoperative recurrence intelligent prediction system based on CNN dept algorithm was 0.85; the recall rate was 0.88; the F1 value was 0.88. The ROC analysis results showed that the areas under the curves (AUC) of the training set and the validation set of the DTC postoperative recurrence intelligent prediction system constructed based on CNN were 0.901 (95%CI 0.835-0.947) and 0.872 (95%CI 0.763-0.943). Conclusion　The intelligent prediction system for recurrence of patients with DTC after endoscopic surgery based on depth algorithm has good prediction efficacy, and has good application value in the prediction of recurrence of patients with DTC after endoscopic surgery.

Key words:

Differentiated thyroid carcinoma, Depth algorithm, Endoscopic surgery, Recurrence, Intelligent prediction

摘要：

目的　构建基于深度算法的分化型甲状腺癌（differentiated thyroid cancer，DTC）患者腔镜术后复发智能预测系统并验证。方法　回顾性分析2020年10月至2023年10月于商洛市中心医院行腔镜手术治疗的189例DTC患者的临床资料。采用简单随机抽样法按照2∶1的比例将其分为训练集（126例）和验证集（63例）。训练集男39例，女87例，年龄（49.14±7.78）岁。验证集男22例，女41例，年龄（50.38±8.12）岁。根据是否复发将训练集患者分为复发组（24例）和未复发组（102例）。采用Cox回归分析探究影响DTC患者腔镜术后复发的因素。以得到的影响因素为基础，利用卷积神经网络（convolutional neural network，CNN）深度算法构建DTC患者腔镜术后复发的智能预测系统。采用受试者操作特征曲线（receiver operating characteristic curve，ROC）评价该智能预测系统的预测效能。采用t检验和χ²检验。结果　行腔镜术治疗的DTC患者术后复发率18.52%（35/189）。肿瘤长径≥2 cm［风险比（hazard ratio，HR）=1.660，95%置信区间（95% confidence interval，95%CI）1.169～2.358］、中分化（HR=1.484，95%CI 1.081～2.039）、淋巴结转移（HR=1.876，95%CI 1.258～2.798）、甲状腺切除范围次全切（HR=1.800，95%CI 1.238～2.618）、未进行辅助治疗（HR=1.737，95%CI 1.213～2.486）、血清甲状腺球蛋白抗体（thyroglobulin antibody，TgAb）≥40 IU/ml（HR=1.590，95%CI 1.126～2.246）均是DTC患者腔镜术后复发的危险因素（均P<0.05）。基于CNN深度算法构建的DTC腔镜术后复发智能预测系统的准确度为0.85，召回率为0.88，F1值为0.88。ROC分析结果显示，基于CNN构建的DTC患者腔镜术后复发智能预测系统的训练集和验证集曲线下面积（area under the curve，AUC）分别为0.901（95%CI 0.835～0.947）、0.872（95%CI 0.763～0.943）。结论　基于深度算法构建的DTC患者腔镜术后复发智能预测系统预测效能良好，在DTC患者腔镜术后复发的预测中具有良好的应用价值。

关键词:

分化型甲状腺癌, 深度算法, 腔镜术, 复发, 智能预测