Working on paper

Author: wmvanvliet

doc/rsa.svg

@@ -174,3 +174,51 @@ @article{Klimovich-Gray2021
   abstract = {Cortical circuits rely on the temporal regularities of speech to optimize signal parsing for sound-to-meaning mapping. Bottom-up speech analysis is accelerated by top--down predictions about upcoming words. In everyday communications, however, listeners are regularly presented with challenging input---fluctuations of speech rate or semantic content. In this study, we asked how reducing speech temporal regularity affects its processing---parsing, phonological analysis, and ability to generate context-based predictions. To ensure that spoken sentences were natural and approximated semantic constraints of spontaneous speech we built a neural network to select stimuli from large corpora. We analyzed brain activity recorded with magnetoencephalography during sentence listening using evoked responses, speech-to-brain synchronization and representational similarity analysis. For normal speech theta band (6.5--8~Hz) speech-to-brain synchronization was increased and the left fronto-temporal areas generated stronger contextual predictions. The reverse was true for temporally irregular speech---weaker theta synchronization and reduced top--down effects. Interestingly, delta-band (0.5 Hz) speech tracking was greater when contextual/semantic predictions were lower or if speech was temporally jittered. We conclude that speech temporal regularity is relevant for (theta) syllabic tracking and robust semantic predictions while the joint support of temporal and contextual predictability reduces word and phrase-level cortical tracking (delta).},
   file = {/home/vanvlm1/Zotero/storage/IQ64IVQT/Klimovich-Gray et al. - 2021 - One Way or Another Cortical Language Areas Flexibly Adapt Processing Strategies to Perceptual And C.pdf}
 }
+@article{Gramfort2013,
+  title = {{{MEG}} and {{EEG}} Data Analysis with {{MNE-Python}}},
+  author = {Gramfort, Alexandre and Luessi, Martin and Larson, Eric and Engemann, Denis A. and Strohmeier, Daniel and Brodbeck, Christian and Goj, Roman and Jas, Mainak and Brooks, Teon and Parkkonen, Lauri and H{\"a}m{\"a}l{\"a}inen, Matti S},
+  year = {2013},
+  journal = {Frontiers in Neuroscience},
+  volume = {7},
+  number = {December},
+  pages = {1--13},
+  issn = {1662-453X},
+  doi = {10.3389/fnins.2013.00267},
+  abstract = {Magnetoencephalography and electroencephalography (M/EEG) measure the weakelectromagnetic signals generated by neuronal activity in the brain. Using thesesignals to characterize and locate neural activation in the brain is achallenge that requires expertise in physics, signalprocessing, statistics, and numerical methods. As part of the MNE softwaresuite, MNE-Python is an open-sourcesoftware package that addresses this challenge by providingstate-of-the-art algorithms implemented in Python that cover multiple methods of data preprocessing, source localization, statistical analysis, and estimation offunctional connectivity between distributed brain regions.All algorithms and utility functions are implemented in a consistent manner with well-documented interfaces, enabling users to create M/EEG data analysispipelines by writing Python scripts.Moreover, MNE-Python is tightly integrated with the core Python libraries for scientificcomptutation (Numpy, Scipy) and visualization (matplotlib and Mayavi), as wellas the greater neuroimaging ecosystem in Python via the Nibabel package. The code is provided under the new BSD licenseallowing code reuse, even in commercial products. Although MNE-Python has onlybeen under heavy development for a couple of years, it has rapidly evolved withexpanded analysis capabilities and pedagogical tutorials because multiple labs have collaborated during code development to help share best practices.MNE-Python also gives easy access to preprocessed datasets,helping users to get started quickly and facilitating reproducibility ofmethods by other researchers. Full documentation, including dozens ofexamples, is available at http://martinos.org/mne.},
+  keywords = {eeg,electroencephalography,electroencephalography (EEG),magnetoencephalography,magnetoencephalography (MEG),meg,neuroimaging,open-source,python,software},
+  file = {S:\work\vanvlm1\papers\Gramfort et al. - 2013 - MEG and EEG data analysis with MNE-Python.pdf}
+}
+@article{Guggenmos2018,
+  title = {Multivariate Pattern Analysis for {{MEG}}: {{A}} Comparison of Dissimilarity Measures},
+  shorttitle = {Multivariate Pattern Analysis for {{MEG}}},
+  author = {Guggenmos, Matthias and Sterzer, Philipp and Cichy, Radoslaw Martin},
+  year = {2018},
+  month = jun,
+  journal = {NeuroImage},
+  volume = {173},
+  pages = {434--447},
+  issn = {1053-8119},
+  doi = {10.1016/j.neuroimage.2018.02.044},
+  urldate = {2019-06-25},
+  abstract = {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 Na{\"i}ve 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.},
+  keywords = {Cross-validation,Decoding,EEG,Machine learning,MEG,Multi-voxel pattern analysis,Noise normalisation,Representational similarity analysis},
+  file = {S\:\\work\\vanvlm1\\papers\\Guggenmos et al. - 2018 - Multivariate pattern analysis for MEG A compariso.pdf;C\:\\Users\\wmvan\\Zotero\\storage\\T6PGLTUL\\S1053811918301411.html}
+}
+@article{Hanke2009,
+  title = {{{PyMVPA}}: A Python Toolbox for Multivariate Pattern Analysis of fMRI Data},
+  shorttitle = {{{PyMVPA}}},
+  author = {Hanke, Michael and Halchenko, Yaroslav O. and Sederberg, Per B. and Hanson, Stephen Jos{\'e} and Haxby, James V. and Pollmann, Stefan},
+  year = {2009},
+  month = mar,
+  journal = {Neuroinformatics},
+  volume = {7},
+  number = {1},
+  pages = {37--53},
+  issn = {1559-0089},
+  doi = {10.1007/s12021-008-9041-y},
+  urldate = {2025-06-13},
+  abstract = {Decoding patterns of neural activity onto cognitive states is one of the central goals of functional brain imaging. Standard univariate fMRI analysis methods, which correlate cognitive and perceptual function with the blood oxygenation-level dependent (BOLD) signal, have proven successful in identifying anatomical regions based on signal increases during cognitive and perceptual tasks. Recently, researchers have begun to explore new multivariate techniques that have proven to be more flexible, more reliable, and more sensitive than standard univariate analysis. Drawing on the field of statistical learning theory, these new classifier-based analysis techniques possess explanatory power that could provide new insights into the functional properties of the brain. However, unlike the wealth of software packages for univariate analyses, there are few packages that facilitate multivariate pattern classification analyses of fMRI data. Here we introduce a Python-based, cross-platform, and open-source software toolbox, called PyMVPA, for the application of classifier-based analysis techniques to fMRI datasets. PyMVPA makes use of Python's ability to access libraries written in a large variety of programming languages and computing environments to interface with the wealth of existing machine learning packages. We present the framework in this paper and provide illustrative examples on its usage, features, and programmability.},
+  langid = {english},
+  keywords = {Brain Mapping,Computational Neuroscience,Functional magnetic resonance imaging,Image analysis,Machine learning,Machine Learning,Multivariate Analysis,MVPA,Neuroimaging software,Python,Scripting,Statistical Software},
+  file = {S:\work\vanvlm1\papers\Hanke et al. - 2009 - PyMVPA a Python Toolbox for Multivariate Pattern Analysis of fMRI Data.pdf}
+}

paper/paper.bib

@@ -22,10 +22,37 @@ bibliography: paper.bib
 
 # Summary
 
+MNE-RSA is a software package for performing representational similarity analysis (RSA) on non-invasive measurements of brain activity, namely electroencephalography (EEG), magnetoencephalography (MEG) and functional magnetic resonance imaging (fMRI).
+It serves as an extension to MNE-Python [@Gramfort2013] to provide a straightforward way to incorporate RSA in a bigger analysis pipeline that otherwise encompasses the many preprocessing steps required for this type of analysis.
+
+## About RSA
+RSA is a technique to compare information flows within complex systems [@Kriegeskorte2008].
+In the context of this software package, this mostly means comparing different representations of input stimuli to neural representations at different locations and times in the brain.
+Example representations of a stimulus would be the pixels of an image, or the semantic features of the object depicted in the image ("has a tail", "barks", "good boy"), or an embedding vector obtained with a convolutional neural network (CNN) or large language network (LLM) [@Diedrichsen2017].
+Example neural representations include the pattern of electric potentials across EEG sensors, or the magnetic field pattern across MEG sensors, or the pattern of source localized activity across the cortex, or the pattern of beta values across fMRI voxels.
+Whenever one can create multiple representations of the same stimuli, one can compare these representations using RSA to judge their "representational similarity" (\autoref{fig:rsa}).
+The key to this is the creation of a representational dissimilarity matrix (RDM) which is an all-to-all distance matrix between the representations of a set of stimuli, usually obtained by correlating the representation vectors of each pair of stimuli.
+Once an RDM is obtained for the different representation schemes (typically you have one obtained through some model and one obtained from brain activity) they can be compared (again using correlation) to yield an RSA score.
+When one does this in a "searchlight" pattern across the brain, the result is a map of RSA scores indicating where and when in the brain the neural representation corresponds to the model.
+
+![Schematic overview of representational similarity analysis (RSA).\label{fig:rsa}](RSA.pdf)
+
 # Statement of need
-Studies using MNE-RSA [@Hulten:2021; @Xu:2024; @Messi:2025; @Ghazaryan:2023; @Klimovich-Gray:2021]
+While the core computations behind RSA are simple, getting the details right is hard.
+Creating a "searchlight" patches across the cortex means using geodesic rather than Euclidean distance (\autoref{fig:geodesic}), combining MEG gradiometers and magnetometers requires signal whitening, creating proper evoked responses requires averaging across stimulus repetitions, and creating reliable brain RDMs requires cross-validated distance metrics [@Guggenmos2028].
+
+![Depiction of geodesic versus Euclidean distance between points along the cortex.\label{fig:geodesic}](geodesic.pdf)
+
+At the time of writing, MNE-RSA has been used in five studies, two of which involve the author [@Hulten2021; @Xu2024; @Messi2025; @Ghazaryan2023; @Klimovich-Gray2021].
+
+## Software ecosystem
 
-# Software ecosystem
+The original RSA-toolbox was implemented in MATLAB (https://github.com/rsagroup/rsatoolbox_matlab), with the third iteration now implemented in python (https://github.com/rsagroup/rsatoolbox). 
+While its focus is mostly on fMRI analysis, the RSA-toolbox aims for a broad implementation of everything related to RSA and its documentation includes an MEG demo.
+Another python package worth mentioning is PyMVPA (http://www.pymvpa.org), which implements a wide array of machine learning methods, including an RSA variant where RDMs are created using decoding performance as distance metric [@Hanke2009].
+While it is possible to use it for EEG and MEG analysis, it mostly focuses on fMRI.
+In contrast to these packages, the scope of MNE-RSA is more narrow, aiming to be an extention of MNE-Python.
+Hence its focus is mostly on MEG and EEG analysis, providing a streamlined user experience for the most common use cases in this domain.
 
 # Design