Supervised Hyperalignment for Multisubject fMRI Data Alignment

Abstract

Hyperalignment has been widely employed in Multivariate Pattern (MVP) analysis to discover the cognitive states in the human brains based on multi-subject functional Magnetic Resonance Imaging (fMRI) datasets. Most of the existing HA methods utilized unsupervised approaches, where they only maximized the correlation between the voxels with the same position in the time series. However, these unsupervised solutions may not be optimum for handling the functional alignment in the supervised MVP problems. This paper proposes a Supervised Hyperalignment (SHA) method to ensure better functional alignment for MVP analysis, where the proposed method provides a supervised shared space that can maximize the correlation among the stimuli belonging to the same category and minimize the correlation between distinct categories of stimuli. Further, SHA employs a generalized optimization solution, which generates the shared space and calculates the mapped features in a single iteration, hence with optimum time and space complexities for large datasets. Experiments on multi-subject datasets demonstrate that SHA method achieves up to 19% better performance for multiclass problems over the state-of-the-art HA algorithms.

Single-Trial MEG Data Can Be Denoised Through Cross-Subject Predictive Modeling

“…Our framework is close to hyperalignment (Haxby et al, 2011 ) and the Shared Response Model (SRM) (Chen et al, 2015 ) in that all approaches aggregate data from multiple subjects. However, these methods have been employed primarily in studies involving fMRI data (Xu et al, 2012 ; Baldassano et al, 2017 ; Yousefnezhad et al, 2020 ), and do not explicitly seek to denoise the data. MEG has much lower SNR, and to the best of our knowledge, our work is the first to show that single-trial naturalistic MEG data can be denoised successfully.…”

Shared Space Transfer Learning for analyzing multi-site fMRI data

“…The task-based functional magnetic resonance imaging (fMRI) is one of the prevalent tools in neuroscience to analyze how human brains work [1][2][3][4][5]. It can be used to visualize the neural activities when subjects are performing cognitive tasks -such as watching photos or making decisions [1].…”

“…As every human brain has a different connectome, each person will have a different neural response for the same stimulus [1]. Recent studies suggested applying functional alignment as an extra processing step before generating a prediction model for fMRI analysis [1,[3][4][5]. This functional alignment process extracts a set of common features from multi-subject fMRI data, which can be used to boost the prediction rate [3][4][5].…”

“…Recent studies suggested applying functional alignment as an extra processing step before generating a prediction model for fMRI analysis [1,[3][4][5]. This functional alignment process extracts a set of common features from multi-subject fMRI data, which can be used to boost the prediction rate [3][4][5]. However, functional alignment techniques need temporal alignment -i.e., the i-th time point for all subjects must involve the same type of cognitive task [4,5].…”

“…This functional alignment process extracts a set of common features from multi-subject fMRI data, which can be used to boost the prediction rate [3][4][5]. However, functional alignment techniques need temporal alignment -i.e., the i-th time point for all subjects must involve the same type of cognitive task [4,5]. Although applying temporal alignment to a single-site fMRI data is a relatively straightforward process, this approach cannot be directly used for any multi-site datasets with different schemes of experimental designs [3,4].…”