基于机器学习预测机器人甲状腺全切联合颈侧区清扫手术难度
本研究回顾性纳入496例接受达·芬奇机器人辅助甲状腺全切联合颈侧区淋巴结清扫的分化型甲状腺癌患者,按7∶3分为训练集和验证集,利用LASSO筛选特征并构建8种机器学习模型。支持向量机模型表现最佳(AUC 0.815,F1 0.708,灵敏度0.727),SHAP分析明确了手术难度增加的危险因素。
要点
- 支持向量机模型预测手术难度效能最佳,AUC 0.815。
- 关键预测特征包括侧区阳性淋巴结最大短径、Ⅱ/Ⅲ区阳性等8项。
- SHAP分析可个体化可视化术前风险评估。
核心结论
- 基于术前多维特征构建的模型对机器人辅助甲状腺全切联合颈侧区淋巴结清扫手术难度具有良好预测效能。
- 结合SHAP可解释性分析可实现术前风险评估的个体化与可视化。
研究设计与样本
- 回顾性病例系列研究,纳入2014年1月至2024年7月496例分化型甲状腺癌患者。
- 按7∶3随机分为训练集(347例)和验证集(149例)。
- 通过LASSO回归筛选特征,构建8种机器学习模型。
- 采用AUC、F1分数等评价模型效能,并用SHAP进行可解释性分析。
主要发现
- LASSO筛选出8个关键特征:侧区阳性淋巴结最大短径、颈部Ⅱ/Ⅲ区淋巴结阳性、侧区阳性淋巴结、性别、超声弥漫性改变、体重、术者工作年限、TPOAb阳性。
- 支持向量机模型表现最佳,AUC 0.815,F1 0.708,灵敏度0.727。
- SHAP分析显示颈侧区淋巴结特征、超重、男性、超声弥漫性改变及TPOAb阳性为手术难度增加的危险因素。
关键数值与推荐
- 支持向量机模型AUC为0.815,F1分数为0.708,灵敏度为0.727。
临床怎么用
- 可用于术前个体化评估机器人甲状腺全切联合颈侧区清扫的手术难度。
- 为外科医师合理配置医疗资源及优化围手术期管理提供科学依据。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
Objectives: To construct a prediction model based on machine learning algorithms for the surgical difficulty of robot-assisted total thyroidectomy combined with lateral cervical lymph node dissection,and to validate its predictive performance and clinical utility. Methods: This retrospective case series study collected clinical data from 496 patients with differentiated thyroid cancer admitted to the Thyroid and Breast Surgery Department of the 960th Hospital of the People's Liberation Army Joint Logistics Support Force between January 2014 and July 2024, who underwent robot-assisted total thyroidectomy combined with lateral neck lymph node dissection using the Da Vinci system. There were 143 male cases and 353 female cases,aged (36.5±10.4)years (range: 18 to 70 years). Patients were randomly divided into a training set (n=347) and a validation set (n=149) at a ratio of 7∶3. Clinical characteristics,ultrasound features,and hematological parameters were collected. Feature selection was performed using the least absolute shrinkage and selection operator (LASSO) regression. Based on the selected features,eight machine learning algorithms-including logistic regression,support vector machine (SVM),gradient boosting machine,neural network,random forest,extreme gradient boosting (XGBoost),k-nearest neighbors and adaptive boosting-were utilized to construct prediction models. Model performance was evaluated using metrics such as the area under the receiver operating characteristic curve (AUC) and F1 score. The Shapley additive explanations (SHAP) method was applied to conduct an interpretable analysis of the optimal model. Results: A total of 8 key features for predicting surgical difficulty were selected through LASSO regression dimensionality reduction: the maximum short-axis diameter of positive lateral cervical lymph nodes,positive lymph nodes in cervical levels Ⅱ and Ⅲ,the number of positive lateral cervical lymph nodes,sex,diffuse thyroid changes on ultrasound,weight,surgeon's years of experience,and thyroid peroxidase antibody (TPOAb) positivity. The SVM model demonstrated the best performance,with the highest AUC (0.815),the highest F1 score (0.708), and a relatively high sensitivity (0.727). SHAP analysis confirmed that lateral cervical lymph node characteristics,higher body weight,male sex,diffuse thyroid changes on ultrasound,and TPOAb positivity were high-risk factors associated with increased surgical difficulty. Conclusions: The model constructed based on patients' preoperative multidimensional features exhibits good predictive performance for the surgical difficulty of Da Vinci robot-assisted total thyroidectomy combined with lateral cervical lymph node dissection. The integration of SHAP interpretable analysis enables an individualized and visualized preoperative risk assessment,which provides a scientific basis for surgeons to rationally allocate medical resources and optimize perioperative management. 目的: 构建基于机器学习算法的机器人辅助甲状腺全切联合颈侧区淋巴结清扫手术难度的预测模型,并验证其预测效能与临床实用性。 方法: 本研究为回顾性病例系列研究。收集2014年1月至2024年7月解放军联勤保障部队第九六〇医院甲状腺乳腺外科收治的接受达·芬奇机器人辅助甲状腺全切联合颈侧区淋巴结清扫的496例分化型甲状腺癌患者的临床资料。男性143例,女性353例;年龄(36.5±10.4)岁(范围:18~70岁)。按7∶3的比例将患者随机划分为训练集(347例)和验证集(149例)。收集患者的临床特征、超声特征及血液学指标,通过最小绝对收缩与选择算子(LASSO)回归进行特征筛选。纳入筛选出的特征,分别利用逻辑回归、支持向量机、梯度提升机、神经网络、随机森林、极端梯度提升、K-最近邻和自适应增强8种机器学习算法构建预测模型。通过受试者工作特征曲线的曲线下面积(AUC)、F1分数等评价模型的预测效能,并利用沙普利加性解释(SHAP)方法对最佳模型进行可解释性分析。 结果: 通过LASSO回归降维共筛选出8个预测手术难度的关键特征:侧区阳性淋巴结最大短径、颈部Ⅱ、Ⅲ区淋巴结阳性、侧区阳性淋巴结、性别、超声检查提示弥漫性改变、体重、术者工作年限和甲状腺过氧化物酶抗体(TPOAb)阳性。支持向量机模型表现出最高的AUC(0.815)、最高的F1值(0.708)和较高的灵敏度(0.727)。SHAP分析明确了颈侧区淋巴结特征、超重、男性、超声检查提示弥漫性改变及TPOAb阳性等为手术难度增加的危险因素。 结论: 基于患者术前多维特征构建的模型对达·芬奇机器人辅助全甲状腺切除联合颈侧区淋巴结清扫手术难度具有良好的预测效能。结合SHAP可解释性分析实现了术前风险评估的个体化与可视化,可为外科医师合理配置医疗资源及优化围手术期管理提供科学依据。.
引用
- Li WT, He QQ, Ma YH, et al. [A clinical study on predicting the surgical difficulty of robot-assisted total thyroidectomy combined with lateral cervical lymph node dissection based on machine learning algorithms]. Zhonghua Wai Ke Za Zhi. 2026. PMID: 42951849. doi:10.3760/cma.j.cn112139-20260402-00140
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。