采用机器学习算法中的支持向量机(SVM)对步态图数据集进行特征分类并建立识别模型，探索该模型区分膝骨关节炎或韧带损伤患者与健康人群的分类性能。

本研究从上海体育大学等单位纳入样本135例。其中健康人55例，确诊膝关节损伤患者80例(膝骨关节炎13例、单纯前交叉韧带损伤51例、多发韧带损伤16例)，排除合并有其他影响步行的组织、系统损伤患者。采用三维运动捕捉系统采集膝关节运动学数据，构建步态图数据集。通过数据降维等预处理方法后，使用SVM分类器将每例样本特征以高维空间位置和超平面的关系分类建立识别分类模型。采用单因素方差分析比较纳入患者的基本信息并使用准确率和精确度等指标来验证模型分类性能。

除年龄外，纳入样本的基本信息差异无统计学意义(均为P>0.05)。基于SVM所建立的识别分类模型对判断是否为健康人测量准确率为91.4% [95%CI(87.9，94.7)%] ；与健康人群相比，对于分类膝骨关节炎患者的准确率为98.5% [95%CI(93.0，99.1)%]，前交叉韧带损伤患者为90.6% [95%CI(81.9，99.4)%] ，多发韧带损伤患者为91.5% [95%CI(83.4，99.6)%] 。

采用机器学习算法对基于便携式运动捕捉设备所建立的健康人群、膝骨关节炎和韧带损伤患者步态图特征数据集进行分类具有较高的准确度，为基于步态分析的临床辅助诊断提供了一种新思路。

To use support vector machine (SVM) in machine learning algorithms to classify the features of gait map dataset and build a recognition model, and to explore the classification performance of the model in distinguishing between patients with osteoarthritis or ligament injuries of the knee and the healthy population.

A total of 135 cases of samples were included in this study from Shanghai University of Sports and other units. Among them, 55 cases were healthy people, and 80 cases were patients with confirmed knee injuries (13 cases of osteoarthritis of the knee, 51 cases of simple anterior cruciate ligament injuries, and 16 cases of multiple ligament injuries), and patients with other tissues and systems injuries that affect walking were excluded. A three-dimensional motion capture system was used to collect knee kinematic data and construct the gait map data set. After preprocessing methods such as data dimensionality reduction, an SVM classifier was used to classify the sample features of each case in terms of class- the relationship between high-dimensional spatial position and hyperplane to establish an identification ification model. One-way ANOVA was used to compare the basic information of the included patients and to validate the model classification performance using metrics such as accuracy and precision.

No significant differences were seen in the basic information of the included sample except for age (all P>0.05). The accuracy of the classification model based on SVM for determining whether a person was healthy or not was 91.4% [95% CI (87.9, 94.7)%]; the accuracy for classifying patients with knee osteoarthritis was 98.5% [95% CI (93.0, 99.1)%], patients with anterior cruciate ligament injuries was 90.6% [95% CI (81.9 , 99.4)%] and 91.5% [95% CI (83.4, 99.6)%] for patients with multiple ligament injuries.

The use of machine learning algorithms to classify the gait map feature datasets established based on portable motion capture devices for healthy people, patients with osteoarthritis of the knee and ligament injuries has a high accuracy and provides a new idea for clinical aid diagnosis based on gait analysis.