基于机器学习的[⻊母]外翻术后复发预测模型

Machine learning-based prediction model for hallux valgus recurrence after surgery

目的本研究采用常见的机器学习算法,基于[⻊母]外翻(hallux valgus,HV)患者的术前负重位及术后即刻非负重位X线片测量指标,构建[⻊母]外翻术后复发预测模型,为临床早期识别[⻊母]外翻复发患者提供客观、精准的临床决策辅助系统。方法回顾性纳入天津医科大学总医院的172例HV患者,共包括230只接受远端截骨矫形及软组织手术的HV足。收集患足术前的负重位及术后非负重位X线片,并基于X线片获得足部各解剖参数。同时根据手术后6个月患足[⻊母]外翻角是否大于20°来判断其是否存在复发。之后将足部各解剖参数作为特征训练机器学习模型,并预测HV足是否出现复发,模型训练及测试采用10折交叉验证法以获得各模型的预测效能。为了进一步提高机器学习模型预测精度,在进行机器学习模型训练之前,首先采用K-means聚类将训练样本划分为两个亚型,对每个亚型分别训练机器学习模型。对于测试样本,首先根据其与两个亚型中心的距离判断其归属类别,再进一步采用相应模型进行预测,且同时采用10折交叉验证法来获得机器模型预测效能。结果机器学习能较好地预测[⻊母]外翻的术后复发,且通过K-means聚类划分亚型后构建机器学习模型,能够有效提高预测精度。结论采用机器学习模型能够准确预测[⻊母]外翻术后复发情况,为开发[⻊母]外翻术后复发的临床预测模型提供了新的思路。

Objective This study aims to use common machine learning algorithms based on preoperative weight-bearing position and immediate postoperative non-weight-bearing X-ray measurements in hallux valgus(HV)patients to construct a predictive model for HV recurrence after surgery.This model aims to provide an objective and precise clinical decision support system for early identification of HV recurrence in clinical practice.Methods This study retrospectively included 172 HV patients from Tianjin Medical University General Hospital,comprising a total of 230 HV feet that underwent distal osteotomy correction and soft tissue surgery.Weight-bearing and non-weight-bearing X-rays were collected before and after the surgery,and various anatomical parameters of the feet were obtained based on these X-rays.At 6 months post-surgery,HV recurrence was determined based on whether the hallux valgus angle exceeded 20°.The various anatomical parameters of the feet were used as features to input into machine learning models to predict the occurrence of HV recurrence.A 10-fold cross-validation was used to assess the predictive performance of each model.To further improve the accuracy of machine learning model predictions,K-means clustering was used to divide the training samples into two subtypes.Separate machine learning models were trained for each subtype.For testing samples,their category was first determined based on their distance from the cluster center,followed by prediction using the corresponding model.A 10-fold cross-validation was also used to evaluate the predictive performance of the machine models.Results Machine learning can effectively predict the population at risk of HV recurrence after surgery,and using K-means clustering to divide subtypes and training machine learning models significantly improves the predictive accuracy of each model.Conclusion This study demonstrates that using machine learning models can accurately predict HV recurrence after surgery,and the use of clustering algorithms to divide subtypes and input into machine learning models can effectively improve prediction accuracy.This research provides a new approach for developing clinical predictive models for HV recurrence after surgery.