基于模式识别的高水平运动员RS-fMRI数据分类研究

Classification of RS-fMRI Data in Elite Athletes Based on Pattern Recognition

为弥补传统的广义线性模型(generalized linear model,GLM)方法的不足,并探索模式识别在运动脑科学当中的应用价值。使用支持向量机(support vector machine,SVM)模式识别算法,以低频振幅(fractional amplitude of low-frequency fluctuations,fALFF)、局部一致性(regional homogeneity,ReHo)和度中心度(degree centrality,DC)作为学习特征,对射击运动组和滑冰运动组(分类1)、射击运动组和对照组(分类2)以及速滑运动组和对照组(分类3)之间进行二分类,并计算每一个脑区在分类算法当中的权重。使用留一交叉验证法计算分类正确率,使用总的准确率、接受者操作特性曲线(receiver operating characteristic curve,ROC)、以及预测准确率来衡量机器分类算法的优劣性。结果表明:分类1中SVM算法的正确率较高且分类效果更稳定,总的准确率(total accuracy,tACC)可以维持在96.67%以上,曲线下面积(area under curve,AUC)均为1,说明SVM算法对区分不同项目运动员脑静息态功能特征时更有优势;在分类2和分类3中,SVM算法效果取决于使用的指标。其中,使用fALFF或者综合使用三个静息态指标的分类效果较稳定(tACC均在80%以上,AUC均在0.88以上);小脑在分类1算法中占较多的权重,提示不同运动项目运动员的脑功能活动之间差异最明显的部位主要在小脑上。而分类2和3中,除了小脑,还有一些与运动执行和控制及其他功能活动相关的脑区参与了算法的构成。通过SVM分类算法的应用获得较为理想的结果,展示了模式识别方法在运动科学领域的应用价值。研究成果有助于体育科学研究者从新的角度更加全面地理解运动与脑的关系。

To address the limitations of traditional generalized linear model(GLM)methods and explore the application value of pattern recognition in sports neuroscience.For binary classification between the shooting and skating groups(classification 1),the shooting and control groups(classification 2),and the speed skating and control groups(classification 3),the support vector machine(SVM)algorithm was employed using fractional amplitude of low-frequency fluctuations(fALFF),regional homogeneity(ReHo),and degree centrality(DC)as learning features.Additionally,the weight of each brain region in the classification algorithm was calculated.Leave-one-out cross-validation was utilized to calculate the classification accuracy.The pros and cons of the machine classification algorithm were evaluated based on overall accuracy,ROC curve,and prediction accuracy.Results show that higher accuracy and more stable classification performance are shown in Category 1 using the SVM algorithm,with a total accuracy(tACC)maintained above 96.67%and area under curve(AUC)of 1.This indicates that the functional characteristics of athletes’brain resting states across different tasks are excellently distinguished by the classification algorithm.In categories 2 and 3,the performance of the SVM algorithm is found to vary depending on the metrics used.Importantly,more stable classification performance is achieved by employing fALFF or the combined use of the three resting-state metrics,with tACC above 80%and AUC above 0.88 for both categories.A higher weight is assigned to the cerebellum by the classification 1 algorithm,indicating that the most significant difference in brain function activities among elite athletes of different sports lies primarily in the cerebellum.The application of SVM classification algorithm has achieved satisfactory results,demonstrating the practicality of pattern recognition methods in sports science.The research results offer a new perspective to enhance the understand of the relationship between exercise and the brain,which could prove valuable for researchers in the field of sports science.