Scene-selectivity in CA1/subicular complex: Multivoxel pattern analysis at 7T
Poster No:
2099
Submission Type:
Abstract Submission
Authors:
Marie-Lucie Read1, Samuel Berry2, Kim Graham3, Natalie Voets4, Jiaxiang Zhang5, John Aggleton1, Andrew Lawrence3, Carl Hodgetts2
Institutions:
1Cardiff University, Cardiff, United Kingdom, 2Royal Holloway University of London, London, United Kingdom, 3The University of Edinburgh, Edinburgh, United Kingdom, 4Wellcome Centre for Integrative Neuroimaging, FMRIB Building, Oxford, United Kingdom, 5Swansea University, Swansea, United Kingdom
First Author:
Co-Author(s):
Introduction:
Univariate functional magnetic resonance imaging (fMRI) studies suggest that the hippocampal anteromedial subicular complex is a hub for scene-based cognition (Hodgetts et al., 2017; Dalton & Maguire, 2017). However, univariate analyses may be less sensitive than multivariate approaches (which utilize activation patterns across voxels) to scene-related activity in additional hippocampal subfields implicated in spatial processing (e.g. CA1). Further, as hippocampal connectivity-based functional gradients do not respect anatomical subfield boundaries, scene-selective activity patterns may be distributed across subfields (Aggleton & Christiansen, 2015; Dalton et al., 2019), so anatomical region-of-interest approaches might conceal cross-boundary category selectivity. We hypothesised that, by applying searchlight multivariate pattern analysis (MVPA) to high resolution fMRI data, scene-selective information would be identified in hippocampal regions crossing subicular complex and CA1.
Methods:
Using FSL FEAT (Woolrich et al., 2001) and MVPA-light (Treder, 2020), we applied searchlight linear support vector machine classification to 7T fMRI data (voxel size=1.2x1.2x1.2mm) of 25 healthy adults (16f 18-35yrs) who undertook a visual odd-one-out discrimination task for scenes and non-scenes (faces, objects, shapes; Fig.1A,B).
1) Hippocampal searchlight classification was restricted to the hippocampal formation, designed to be sensitive, and accommodated inter-individual differences in category selectivity locations. Binomial tests were performed on individual-participant searchlight accuracy maps (for each 2-way classification between categories) against chance level. Each individual analysis resulted in 6 significant searchlight accuracy maps (scene v face, scene v size, scene v object, face v size, face v object, object v size), which were overlayed at group level (overlap maps), and then combined into scene and face 'hotspot' maps. The scene hotspot map was the sum of scene v face/size/object overlap maps, with face v size/object overlap maps subtracted.
2) Whole searchlight classification included the whole field of view (Fig.1C) and used conservative group-level statistics (FSL Randomise; threshold-free cluster enhancement; 5000 permutations, Winkler et al., 2014). T-maps were masked using corrected p-maps (α=0.0083; 0.05/6 tests). To isolate a 'scene-selective map', conjunction analysis was performed (the product of all the significance masks from searchlight contrasts that included scenes with the product of all the significance masks from the non-scene searchlights removed).
1) Hippocampal searchlight classification was restricted to the hippocampal formation, designed to be sensitive, and accommodated inter-individual differences in category selectivity locations. Binomial tests were performed on individual-participant searchlight accuracy maps (for each 2-way classification between categories) against chance level. Each individual analysis resulted in 6 significant searchlight accuracy maps (scene v face, scene v size, scene v object, face v size, face v object, object v size), which were overlayed at group level (overlap maps), and then combined into scene and face 'hotspot' maps. The scene hotspot map was the sum of scene v face/size/object overlap maps, with face v size/object overlap maps subtracted.
2) Whole searchlight classification included the whole field of view (Fig.1C) and used conservative group-level statistics (FSL Randomise; threshold-free cluster enhancement; 5000 permutations, Winkler et al., 2014). T-maps were masked using corrected p-maps (α=0.0083; 0.05/6 tests). To isolate a 'scene-selective map', conjunction analysis was performed (the product of all the significance masks from searchlight contrasts that included scenes with the product of all the significance masks from the non-scene searchlights removed).
·Figure 1. Methods.
Results:
1) Hippocampal searchlight classification: Contrasting scene and face hotspot maps revealed medial-lateral (distal-proximal) category selectivity gradients, with higher values for scenes medially (distally) and faces laterally (proximally), across the hippocampus, and within the subicular complex and CA1 (Fig1.A). 2) Whole searchlight classification: Scene-selective regions (Fig1.B) overlapped with anteromedial (distal) right subicular complex (max ROI probability: 55%, MNI x=19 y=-21 z=-21) and left posterolateral (proximal) CA1 (max ROI probability: 50%, x=-33 y=-34 z=-10).
·Figure 2. Results.
Conclusions:
Our work advances mapping of hippocampal scene-processing networks by using: a) MVPA to improve sensitivity to category-selectivity patterns of hippocampal BOLD, thereby providing evidence of scene selectivity in CA1 as well as subicular complex; b) a subfield-agnostic approach, revealing category selectivity regions that cross subfield boundaries; and c) subject-level, as well as group-based, analyses, which revealed scene selectivity gradients across the hippocampus, and individual variation in category-selective locations (also see Hodgetts et al., 2015). Our results align with a scene representation pathway described in non-human primates (Aggleton & Christiansen, 2015), including anteromedial (distal) subicular complex and lateral (proximal) CA1 regions.
Modeling and Analysis Methods:
Multivariate Approaches 2
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Anatomy and Functional Systems 1
Subcortical Structures
Novel Imaging Acquisition Methods:
BOLD fMRI
Perception, Attention and Motor Behavior:
Perception: Visual
Keywords:
Memory
Multivariate
Perception
Other - hippocampus; scene processing; face processing
1|2Indicates the priority used for review
Provide references using author date format
Dalton, M. A., & Maguire, E. A. (2017). The pre/parasubiculum: a hippocampal hub for scene-based cognition? Current Opinion in Behavioral Sciences, 17, 34-40. https://doi.org/10.1016/j.cobeha.2017.06.001
Dalton, M. A., McCormick, C., & Maguire, E. A. (2019). Differences in functional connectivity along the anterior-posterior axis of human hippocampal subfields. Neuroimage, 192, 38-51. https://doi.org/10.1016/j.neuroimage.2019.02.066
Hodgetts, C. J., Postans, M., Shine, J. P., Jones, D. K., Lawrence, A. D., & Graham, K. S. (2015). Dissociable roles of the inferior longitudinal fasciculus and fornix in face and place perception. Elife, 4. https://doi.org/10.7554/eLife.07902
Hodgetts, C. J., Voets, N. L., Thomas, A. G., Clare, S., Lawrence, A. D., & Graham, K. S. (2017). Ultra-High-Field fMRI Reveals a Role for the Subiculum in Scene Perceptual Discrimination. Journal of Neuroscience, 37(12), 3150-3159. https://doi.org/10.1523/jneurosci.3225-16.2017
Treder, M. S. (2020). MVPA-Light: A Classification and Regression Toolbox for Multi-Dimensional Data. Frontiers in Neuroscience, 14, 289. https://doi.org/10.3389/fnins.2020.00289
Winkler, A. M., Ridgway, G. R., Webster, M. A., Smith, S. M., & Nichols, T. E. (2014). Permutation inference for the general linear model. Neuroimage, 92(100), 381-397. https://doi.org/10.1016/j.neuroimage.2014.01.060
Woolrich, M. W., Ripley, B. D., Brady, M., & Smith, S. M. (2001). Temporal autocorrelation in univariate linear modeling of FMRI data. Neuroimage, 14(6), 1370-1386. https://doi.org/10.1006/nimg.2001.0931