Re-imagining fMRI for awake behaving infants

Abstract

Thousands of functional magnetic resonance imaging (fMRI) studies have provided important insight into the human brain. However, only a handful of these studies tested infants while they were awake, because of the significant and unique methodological challenges involved. We report our efforts to address these challenges, with the goal of creating methods for awake infant fMRI that can reveal the inner workings of the developing, preverbal mind. We use these methods to collect and analyze two fMRI datasets obtained from infants during cognitive tasks, released publicly with this paper. In these datasets, we explore and evaluate data quantity and quality, task-evoked activity, and preprocessing decisions. We disseminate these methods by sharing two software packages that integrate infant-friendly cognitive tasks and eye-gaze monitoring with fMRI acquisition and analysis. These resources make fMRI a feasible and accessible technique for cognitive neuroscience in awake and behaving human infants.

Fig. 1. Cumulative data retention for individual scanning sessions.

For a Cohort I and b Cohort II, the age of the child in months is shown on the x axis and the duration of the scan on the y axis, including a breakdown by color for different categories of data. Functional time-points were deemed usable when the translational motion was below 3 mm (the voxel resolution) and when the infant’s eyes were open and on-screen (determined by manual gaze coding of a video recording of the infant’s face). Epochs of data during a task (blocks in a block design or trials in an event-related design) were excluded if more than 50% of the time-points were excluded because of motion and/or eye-gaze. Runs were excluded if no blocks were usable within a run. If the infant fell asleep during functional scans, we occasionally continued collection and labeled it as resting data. Anatomical and scout scans are included if they were completed, although not distinguished based on quality. Individual infants who completed more than one session have been assigned a letter code, which is shown beneath the age of each of their sessions (no letter means the infant from that session only participated once).

Fig. 2. Comparison of signal-to-fluctuation-noise ratio (SFNR) in adults and infants.

Adults had both the top and bottom coils attached, while infants only had the bottom coil. SFNR was computed for each voxel and then the values from a random sample of 1000 voxels in each coronal slice of at least that size were averaged, spanning the posterior-anterior axis of the brain. Each gray line is one run from one participant (N = 64 for infants, N = 16 for adults) and the colored lines represent the average. The solid blue line is the mean of all infant runs, whereas the dashed blue line is the mean of five low-motion runs.

Fig. 3. Visual evoked activity for each run from Cohort I.

a Mean proportion of voxels showing significant visual responses within run (thresholded at p < 0.05) for V1, LOC, and A1 regions of interest (ROIs). Data are presented as mean values ± between-run standard error of the mean (SEM). Two-tailed bootstrap resampling compared against the chance proportion level (0.05): **p < 0.01, ***p < 0.001 for the N = 32 runs. Inset: change in proportion of significant voxels across the ROIs for each run for infants younger (orange) or older (magenta) than a year old. b t-value for voxels across the whole brain showing reliable responses across runs (two-tailed p < 0.005, uncorrected). V1, LOC, and A1 ROIs are outlined in green, blue, and gray, respectively.

Fig. 4. Proportion of significant voxels across translational motion thresholds.

Values are separately shown for V1 (green), LOC (blue), and A1 (gray). The dashed line indicates the motion threshold that was used as the default. None indicates the results when no motion threshold was applied. The lefthand axis reports the proportion of significant voxels (p < 0.05) for each ROI. The righthand y-axis reports the proportion of included TRs in mustard and included runs in purple (out of 38 runs with at least 2 blocks). Note that regardless of the motion threshold, some blocks/runs were excluded because the infant’s eyes were closed. Data are presented as mean values ± between-run SEM as shaded area.

Fig. 5. Proportion of significant voxels after various preprocessing decisions.

Results are shown for V1 (green), LOC (blue), and A1 (gray). The parameter setting we used for the other analyses is shown in bright blue. a Number of time-points into the future removed following above-threshold motion, with the percentage of retained time-points below (N = 32, 30, and 25 for 0, 1, and 2 time-points removed, respectively). b Full-width half maximum (FWHM) of the Gaussian spatial smoothing kernel (N = 32). c Minimum correlation threshold for excluding independent components based on their relationship to motion parameters with lower values leading to more components excluded (N = 32). d Turning on or off AFNI’s voxelwise despiking (N = 32). e Inclusion of temporal derivatives in the design matrix of the general linear model (N= 32). Significance of one-tailed Chi-square test for omnibus linear mixed model: *p < 0.05, **p < 0.01, ***p < 0.001. Significant two-tailed simple effects between our chosen parameter setting (in bright blue) and other settings are indicated by a bold line (p <0.05). Data are presented as mean values ± between-run SEM.

Fig. 6. Visual evoked activity for each run from Cohort II (akin to Fig. 3 for Cohort I).

a Proportion of voxels showing significant visual responses within run (thresholded at p < 0.05) for V1, LOC, and A1. Data are presented as mean values ± between-session SEM. Two-tailed bootstrap resampling compared against the chance proportion level (0.05): *=p < 0.05, **=p < 0.01 for N = 26 runs. Inset: change in proportion of significant voxels across the ROIs for each run (all infants younger than a year old). b t-value for voxels across the whole brain showing reliable responses across runs (two-tailed p < 0.005, uncorrected). V1, LOC, and A1 ROIs are outlined in green, blue, and gray, respectively.

Fig. 7. Schematic of the scanning environment.

a Overview of the setup and wiring diagram of the equipment and communications. b Key elements inside the scanner room as well as a view of the screen projected onto the ceiling of the bore. The camera is adjacent to the screen projection. The video feed is depicted on a screen on the wall but could be shown a monitor through the window. 3D rendering created using Sweet Home 3D from Sweet Home 3D assets shared under a Free Art 1.3 licence and CC-BY 3.0 license. Additional assets are from Trimble 3d Warehouse in accordance with their license. All rights reserved.