影像特征在膝关节炎功能评估中的预测价值研究

Research on the Prediction Value of Image Features Analysis in the Evaluation of Symptom Function of Knee Osteoarthritis

旨在探讨基于影像的膝关节定量结构特征是否可以预测膝关节的症状,并评估这些特征在膝关节症状功能评估中的预测价值。从公开数据库Osteoarthritis Initiative(OAI)中筛选了282个病例数据;然后,采用Pearson相关分析以及最大相关最小冗余方法进行降维,以选择与目标向量相关且非冗余的特征;最后,基于筛选特征构建了随机森林预测器模型,对膝关节炎症状功能进行了预测。经过两步特征筛选最终得到了20个影像特征。基于这些特征建立的随机森林预测器模型得到预测值和实际值之间表现出了良好的一致性和准确性:4个KOOS分数预测器的平均绝对误差(MAEs)和均方误差(RMSEs)在训练和测试集中均在5分以下。构建的随机森林模型具有良好的准确性和可行性。

This paper is to investigate whether image-based quantitative structural features of the knee joint can predict knee joint symptoms,and evaluate the predictive value of these features in the evaluation of knee symptom function.First,282 cases were selected from the open database——Osteoaarthritis Initiative(OAI).Then,Pearson correlation analysis and MRMR method were used to reduce the dimensions,so as to select the non-redundant features which are related to the target vectors.Finally,a random-forest predictor model was built based on the selected features to predict the symptom function of knee arthritis,and also.After two-step feature selection,20 image features were finally obtained.The random forest predictor model built based on these features shows good consistency and accuracy between the predicted value and the actual value:the mean absolute error(MAEs)and mean square error(RMSEs)of the four KOOS score predictors were all below 5 points in the training and test sets.The built random forest model has acceptable accuracy and feasibility.