Multivariate pattern analysis for MEG: A comparison of dissimilarity measures
Multivariate pattern analysis for MEG: A comparison of dissimilarity measures
复制标题
DOI：
10.1016/j.neuroimage.2018.02.044
复制
发表时间：
2018-06-01
期刊：
影响因子：
5.7
通讯作者：
Cichy, Radoslaw Martin
中科院分区：
文献类型：
--
作者：
Guggenmos, Matthias;Sterzer, Philipp;Cichy, Radoslaw Martin
多变量模式分析（MVPA）方法，如解码和代表性相似性分析（RSA）的分析脑磁图（MEG）的数据正在迅速增长的流行。然而，很少有人知道的相对性能和特性的具体相异性措施来描述诱发激活模式之间的差异。在这里，我们使用多段MEG数据集来定性表征一系列相异性测量，并定量比较它们的解码准确性（用于解码）和代表性相异性矩阵（用于RSA）的会话间可靠性。我们测试了一系列分类器（线性判别分析- LDA，支持向量机- SVM，加权鲁棒距离- WeiRD，高斯朴素贝叶斯- GNB）和距离（欧几里得距离，皮尔逊相关性）的相异性测量。此外，我们评估了三个关键处理选择：1）预处理（噪声归一化，去除模式均值），2）通过决策值加权解码精度，以及3）在三种不同的分区方案（非交叉验证，交叉验证，类内校正）中计算距离。从我们的研究结果中得出了四个主要结论。首先，适当的多元噪声归一化大大提高了解码精度和相异性测量的可靠性。第二，LDA，SVM和WeiRD产生高峰解码精度和几乎相同的时间过程。第三，虽然使用RSA的解码精度明显不如连续距离可靠，但通过解码精度的决策值加权来改善这一缺点。第四，交叉验证的欧几里得距离提供了无偏的距离估计和高度可复制的代表性相异性矩阵。总的来说，我们强烈建议使用多元噪声归一化作为一般的预处理步骤，推荐LDA，SVM和WeiRD作为解码的分类器，并强调交叉验证的欧几里得距离作为RSA的可靠和无偏的默认选择。
Multivariate pattern analysis (MVPA) methods such as decoding and representational similarity analysis (RSA) are growing rapidly in popularity for the analysis of magnetoencephalography (MEG) data. However, little is known about the relative performance and characteristics of the specific dissimilarity measures used to describe differences between evoked activation patterns. Here we used a multisession MEG data set to qualitatively characterize a range of dissimilarity measures and to quantitatively compare them with respect to decoding accuracy (for decoding) and between-session reliability of representational dissimilarity matrices (for RSA). We tested dissimilarity measures from a range of classifiers (Linear Discriminant Analysis - LDA, Support Vector Machine - SVM, Weighted Robust Distance - WeiRD, Gaussian Naive Bayes - GNB) and distances (Euclidean distance, Pearson correlation). In addition, we evaluated three key processing choices: 1) preprocessing (noise normalisation, removal of the pattern mean), 2) weighting decoding accuracies by decision values, and 3) computing distances in three different partitioning schemes (non-cross-validated, cross-validated, within-class-corrected). Four main conclusions emerged from our results. First, appropriate multivariate noise normalization substantially improved decoding accuracies and the reliability of dissimilarity measures. Second, LDA, SVM and WeiRD yielded high peak decoding accuracies and nearly identical time courses. Third, while using decoding accuracies for RSA was markedly less reliable than continuous distances, this disadvantage was ameliorated by decision-value-weighting of decoding accuracies. Fourth, the cross- validated Euclidean distance provided unbiased distance estimates and highly replicable representational dissimilarity matrices. Overall, we strongly advise the use of multivariate noise normalisation as a general preprocessing step, recommend LDA, SVM and WeiRD as classifiers for decoding and highlight the cross- validated Euclidean distance as a reliable and unbiased default choice for RSA.