| 引用本文： |
-
韩春霞,郑嘉祺,王焕,等.神经网络预测股骨颈骨折术后并发症的价值[J].同济大学学报（医学版）,2022,43(2):174-180. [点击复制]
- HAN Chunxia,ZHENG Jiaqi,WANG Huan,et al.Neural network models in predicting postoperative complications in patients with femoral neck fracture[J].同济大学学报（医学版）,2022,43(2):174-180. [点击复制]
|
|
| 摘要: |
| 目的应用神经网络建立股骨颈骨折术后股骨头坏死预测模型，并与Logistic回归预测模型进行比较，探讨其临床应用价值。方法收集了2013年3月—2017年1月在上海市3家医院行内固定治疗的378例新鲜股骨颈骨折患者，按4：1划分为训练集和测试集，使用SPSS 20.0分别建立Logistic回归模型、MLP神经网络模型以及RBF神经网络模型，比较3种模型的预测性能。结果多因素分析结果显示，VAS评分、Garden分型、完全负重时间、受伤至手术时长、BMI、术后错位程度、取不取内固定、Charlson合并症指数（Charlson comorbidity index, CCI）均与股骨颈骨折术后股骨头坏死预后相关。MLP神经网络模型训练集和测试集的AUC为0.940、0.923；Logistic回归模型训练集和测试集的AUC为0.850、0.834；RBF神经网络模型训练集和测试集的AUC为0.809、0.788。所有预测变量中，变量重要性排名依次为VAS评分、Garden分型、CCI、BMI、完全负重时间、取不取内固定、受伤至手术时长、术后错位程度。结论MLP神经网络在预测股骨颈骨折术后股骨头坏死方面的预测效能高于Logistic回归，具有较好的应用前景。 |
| 关键词: 股骨颈骨折 股骨头坏死 神经网络 预后 |
| DOI：10.12289/j.issn.1008-0392.21224 |
| 投稿时间：2021-06-01 |
| 基金项目:国家自然科学基金（81872718）；上海市卫生和计划生育委员会课题（201840041） |
|
| Neural network models in predicting postoperative complications in patients with femoral neck fracture |
| HAN Chunxia,ZHENG Jiaqi,WANG Huan,XU Linxiao,LIU Jun,AI Zisheng |
| (School of Medicine, Tongji University, Shanghai 200092, China) |
| Abstract: |
| ObjectiveTo establish prediction models of femoral head necrosis in patients with femoral neck fracture and to compare their clinical application value among different models. MethodsA total of 378 patients with fresh femoral neck fracture who received internal fixation treatment in 3 hospitals in Shanghai from March 2013 to January 2017 were enrolled in the study. Patients were randomly divided into training and testing cohorts in a 4：1 ratio. Logistic regression model, multilayer perception(MLP) neural network model and radial basis function(RBF) neural network model were established by SPSS 20.0, and the performance of the three models was compared. ResultsThe VAS pain score, Garden classification, full load time, time from injury to operation, BMI, postoperative dislocation degree, internal fixation and Charlson comorbidity index(CCI) were significantly associated with femoral head necrosis after femoral neck fracture. The area under ROC curve(AUC) of MLP neural network model in training cohort and test cohort were 0.940 and 0.923; that of Logistic regression model was 0.850 and 0.834; that of RBF neural network model was 0.809 and 0.788. Among all the predictive variables, the VAS pain score had the highest importance, followed by Garden classification, CCI, BMI, full load time, internal fixation, time from injury to operation, and postoperative dislocation degree. ConclusionMLP neural network is more effective than Logistic regression in predicting femoral head necrosis after femoral neck fracture, and has a better application prospect. |
| Key words: femoral neck fracture femoral head necrosis neural network prognosis |
