Automated simulation of fMRI experiments
Presented During:
Tuesday, June 27, 2017: 10:55 AM - 11:08 AM
Vancouver Convention Centre
Room:
Ballroom AB
Submission No:
1838
Submission Type:
Abstract Submission
On Display:
Monday, June 26 & Tuesday, June 27
Authors:
Leila Wehbe1, Alexander Huth1, Fatma Deniz1, Marie-Luise Kieseler1, Jack L Gallant1
Institutions:
1University of California, Berkeley, Berkeley, CA
First Author:
Introduction:
In a typical fMRI experiment responses are recorded under a few conditions (e.g. abstract words and concrete words) and then contrasts are performed between conditions. Locations of significant differences are reported, usually in a table listing peak locations in standardized space. However, statistical thresholds are usually not directly comparable across experiments because of differences in design and analysis. Thus, it is difficult to replicate experiments or synthesize results across them.
Naturalistic experiments and voxel-wise modeling provide one alternative to the contrast-based approach. These studies sample the stimulus space broadly and characterize the relationship between linearized stimulus features and brain activity in single voxels. Here, we provide a means to bridge between contrast-based studies and naturalistic studies. Specifically, we present a web-based replication engine that uses data derived from naturalistic voxel-wise modeling experiments to simulate any simple language contrast that can be expressed in terms of a list of words reflecting each of two conditions.
Naturalistic experiments and voxel-wise modeling provide one alternative to the contrast-based approach. These studies sample the stimulus space broadly and characterize the relationship between linearized stimulus features and brain activity in single voxels. Here, we provide a means to bridge between contrast-based studies and naturalistic studies. Specifically, we present a web-based replication engine that uses data derived from naturalistic voxel-wise modeling experiments to simulate any simple language contrast that can be expressed in terms of a list of words reflecting each of two conditions.
Methods:
The replication engine is based on data from Huth et al. (Nature, 2016). In that study seven subjects listened to hours of recorded narrative stories. Semantic features were extracted from the stories by projecting the stimulus words into a 985 word co-occurrence space computed over a large text corpus. Ridge regression was used to fit a linearized model to each voxel, predicting activity through time as a function of the semantic vectors.
Our replication engine uses these voxel-wise models to simulate the results that would be expected to occur given a new contrast-based language experiment. A contrast is first defined between two conditions by providing two lists of stimulus words. Brain activity for each stimulus word is then predicted by multiplying its semantic vector with the semantic model weights for each subject, and averaging predicted activity across words for each condition. The contrast is simulated by computing the difference between predicted activities and a contrast map is obtained for each subject, and for an MNI-normalized average of the subjects. A non-parametric two-sample test is used to establish the significance of the difference in activity at each voxel. Finally, we produce a series of visualizations of the contrasts, including interactive pycortex (Gao, et al. Frontiers in Neuroinformatics, 2015) brain viewers. For contrast words drawn from previous studies that are included in our database, we compare the replicated results against the original published results.
Our replication engine uses these voxel-wise models to simulate the results that would be expected to occur given a new contrast-based language experiment. A contrast is first defined between two conditions by providing two lists of stimulus words. Brain activity for each stimulus word is then predicted by multiplying its semantic vector with the semantic model weights for each subject, and averaging predicted activity across words for each condition. The contrast is simulated by computing the difference between predicted activities and a contrast map is obtained for each subject, and for an MNI-normalized average of the subjects. A non-parametric two-sample test is used to establish the significance of the difference in activity at each voxel. Finally, we produce a series of visualizations of the contrasts, including interactive pycortex (Gao, et al. Frontiers in Neuroinformatics, 2015) brain viewers. For contrast words drawn from previous studies that are included in our database, we compare the replicated results against the original published results.
Results:
The automated replication engine is available at https://boldpredictions.gallantlab.org/. It can be used to replicate experiments that are already published, or to simulate new contrasts by entering stimulus words for each condition. Thus, this tool can be used to predict responses to conditions of interest and to plan new experiments (see Figures 1-2).
The website includes several experiments published previously. Inspection of the published ROIs with our simulated replication results shows that the peaks of the simulated contrasts align with many of the reported ROIs. However, the replication fails to identify some of the reported ROIs. In most cases these ROIs appear to be associated with some task-related aspect of the original study that is not present in our narrative comprehension task.
The website includes several experiments published previously. Inspection of the published ROIs with our simulated replication results shows that the peaks of the simulated contrasts align with many of the reported ROIs. However, the replication fails to identify some of the reported ROIs. In most cases these ROIs appear to be associated with some task-related aspect of the original study that is not present in our narrative comprehension task.
Conclusions:
The online replication engine can be used to replicate existing experiments, and to simulate any language contrast that can be tested in a contrast-based design. We invite the community to use the engine and contribute to the database of published studies that can be replicated.
Higher Cognitive Functions:
Higher Cognitive Functions Other
Imaging Methods:
BOLD fMRI
Informatics:
Informatics Other
Language:
Language Comprehension and Semantics 2
Modeling and Analysis Methods:
Multivariate modeling 1
Keywords:
Cognition
Computational Neuroscience
Cortex
Data analysis
fMRI CONTRAST MECHANISMS
FUNCTIONAL MRI
Informatics
Language
Modeling
Statistical Methods
1|2Indicates the priority used for review
Would you accept an oral presentation if your abstract is selected for an oral session?
I would be willing to discuss my abstract with members of the press should my abstract be marked newsworthy:
Please indicate below if your study was a "resting state" or "task-activation” study.
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute the presentation in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels or other electronic media and on the OHBM website.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Internal Review Board (IRB) or Animal Use and Care Committee (AUCC) Approval. Please indicate approval below. Please note: Failure to have IRB or AUCC approval, if applicable will lead to automatic rejection of abstract.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references in author date format
Gao, James S., et al. "Pycortex: an interactive surface visualizer for fMRI." Frontiers in neuroinformatics 9 (2015).