A large longitudinal neuroimaging investigation tracking infants across their first year of life has demonstrated that awake functional magnetic resonance imaging is feasible at scale for mapping visual representations in the developing brain. Researchers with the Foundations of Cognition Project recorded non-invasive brain activity from healthy full-term infants and infants admitted to neonatal intensive care at two critical developmental stages, approximately two months and nine months corrected gestational age. By developing specialized presentation protocols and carefully tailored visual tasks, the scientific team established that awake infants could participate in functional imaging sessions of substantial length while maintaining sufficient stillness and attention. The findings demonstrate that infants exhibit robust, stimulus-driven neural activity across the ventral visual stream, an anatomical hierarchy of cortical regions responsible for adult object and scene recognition, providing an experimental foundation for probing early cognitive development through computational and NeuroAI frameworks.
Understanding how structured visual representations emerge during early infancy has long challenged cognitive scientists due to the practical difficulties of measuring regional brain activity in conscious babies. While non-invasive scalp-based modalities such as electroencephalography and functional near-infrared spectroscopy are common in pediatric laboratories, their measurement sensitivity is largely confined to superficial cortical layers near the skull. Consequently, these surface methods struggle to record fine-grained, localized neural computations in deeper ventral occipitotemporal structures. Conversely, functional magnetic resonance imaging offers exceptional spatial resolution across the whole brain, but infant scanning has historically been conducted while infants sleep. Because sleeping paradigms largely preclude the systematic presentation of complex visual stimuli that require conscious gaze and active cognitive processing, researchers have had limited opportunities to map higher-level representational spaces in the youngest age groups.
To overcome these methodological constraints, the investigators created two distinct passive viewing paradigms designed to reconcile conflicting experimental demands: sustaining infant alertness, engaging ventral stream regions along semantic dimensions known to activate adult cortex, and limiting total experimental conditions to maximize repeated presentations within variable scanning durations. The first paradigm comprised a sequence of six naturalistic, dynamic 22.5-second video clips representing everyday infant experiences such as mealtime, the playground, and the beach. These continuous videos were selected to maintain engagement and naturally encourage motor stillness without imposing explicit task instructions. The second paradigm consisted of thirty-six segmented object images presented on a dark background, spanning twelve semantic categories evenly divided among animate entities, small inanimate items, and large inanimate objects. To capture attention and assist feature integration, each static object grew in size through a continuous, logarithmic looming animation.
Designing visual stimuli for awake infant imaging required systematic adjustments for the biological immaturities of early vision, including lower contrast sensitivity, developing color vision, and restricted visual acuity. Using computational simulations from the TinyEyes model, the investigators approximated the visual perception of two-month-old infants, accounting for screen dimensions, viewing distance, and opponent color channel blurring. Stimuli were projected directly onto an angled A4 display surface attached to the scanner bore approximately 27 centimeters above the infant’s eyes, yielding a bounded rectangular presentation window measuring 29 centimeters wide by 16.3 centimeters high. This restricted display window prevented salient visual content from appearing in distant peripheral visual angles, where two-month-old acuity drops four- to eight-fold compared to central gaze. Furthermore, by keeping maximum resolvable spatial frequency below 5.7 cycles per degree, the display ensured that presented scenes remained interpretable to both two-month-old and nine-month-old participants.
The study recorded high participation rates and functional scan durations across both target age cohorts. Among 134 enrolled infants who attended the two-month scanning session, awake functional datasets were successfully acquired from 130 infants, representing an acquisition success rate of ninety-seven percent. These two-month-old participants contributed a median duration of 18.63 minutes of awake task-based functional data across one or two scan sessions during a single visit, with 72 infants completing at least one awake functional run in each of two separate scanning sessions. At nine months, 104 infants returned for follow-up testing, and awake functional data were successfully obtained from 65 infants, yielding a participation rate of 62.5 percent and a median awake scanning duration of 13.02 minutes. In addition, when infants transitioned into natural sleep within the scanner, the team acquired anatomical T2-weighted and resting-state scans, including anatomical data from 103 two-month-olds and 32 nine-month-olds.
Infant motor compliance was quantified by calculating framewise displacement across functional volumes acquired with a multiband accelerated echo planar imaging sequence operating at a repetition time of 610 milliseconds and 3-millimeter isotropic voxel resolution. Head motion was controlled by applying a framewise displacement exclusion threshold of 1.5 millimeters, leading to the censoring of 17 to 18 percent of timepoints in retained two-month-old runs and 8 to 12 percent of timepoints in retained nine-month-old runs. Across usable functional runs, nine-month-olds exhibited significantly less overall motion than two-month-olds, with a median framewise displacement of 0.36 millimeters compared to 0.55 millimeters in the younger cohort. Within the two-month group, infants from the neonatal intensive care unit cohort displayed slightly higher median displacement than full-term control infants, though usable functional run retention remained high across both subgroups after censoring excessive movement.
Behavioral coding of synchronized in-bore video recordings demonstrated that infants remained visually attentive during stimulus presentation rather than averting their eyes or fussing out of the scanner. At two months of age, infants exhibited active scanning interest for a median proportion of 75.27 percent of total scan time, compared to 48.52 percent among nine-month-olds. While older infants looked away from the projection screen slightly more often than younger infants, the overall proportion of time spent looking away remained remarkably small, registering at a median of 1.13 percent at nine months and 0.0 percent at two months. Within a standardized sub-cohort of two-month-old infants who completed consecutive video and picture runs, infants displayed significantly greater visual interest and lower facial fussing during their initial video exposure compared to the initial pictures run, although fussing decreased substantially when returning to subsequent video runs.
To evaluate the scientific quality and reproducibility of the acquired functional signals, the researchers conducted representational similarity analyses and noise-ceiling evaluations across cytoarchitectonically defined regions of interest spanning primary visual cortex V1, intermediate area hV4, and higher visual area VO1. Group-level representational similarity matrices were computed by performing pairwise cross-subject run correlations, evaluating multivariate parameter estimates for the 36 objects and temporal response profiles across the six naturalistic videos. Non-parametric bootstrap resampling across one thousand iterations demonstrated that split-half reliability increased systematically as sample sizes expanded for both paradigms across all visual cortical regions. This scaling trajectory confirmed that awake infant functional imaging reliably captures stimulus-driven representational geometry when sufficiently large group cohorts are aggregated.
Despite the success of group-level representational modeling, the researchers noted key limitations regarding individual-level analysis and longitudinal comparisons. A variance partitioning analysis revealed substantial non-neural noise at the single-subject level, which restricted the ability to track individual representational shifts over time. Among the subset of 36 infants who contributed multiple high-quality picture runs at both two and nine months, individual representational similarity matrices were not significantly more correlated with an infant’s own longitudinal follow-up scan than with the matrices of other infants. The authors concluded that while awake infant functional imaging is fully viable for group-level investigations and computational comparisons at scale, isolating individual developmental trajectories will require larger per-subject run yields, refined infant-specific hemodynamic response models, and further advancements in prospective and deep learning-based motion correction.
Subject of Research: Feasibility and data quality of awake infant functional MRI for visual neuroscience
Article Title: Awake infant functional MRI for visual neuroscience: reflections from the Foundations of Cognition Project
Article References: Dineen, Á. T., O’Doherty, C., Truzzi, A., King, G., Zaadnoordijk, L., D’Arcy, E.-L., Caldinelli, C., Kravchenko, A., Holloway, T., White, J., Wade, A. R., De Burca, A., Healy, M., Burke, A., Campion, K., Murphy, C., Herbert, K., Ambre, C., Wadhwa, A., … Cusack, R. (2026). Awake infant functional MRI for visual neuroscience: reflections from the Foundations of Cognition Project. Frontiers in Neuroscience, 20, 1852835. https://doi.org/10.3389/fnins.2026.1852835
Image Credits: AI Generated
DOI: 10.3389/fnins.2026.1852835
Keywords: awake infant fMRI, ventral visual stream, visual neuroscience, infant neuroimaging, cognitive development, Foundations of Cognition Project, representational similarity analysis, visual cortex, pediatric neuroimaging, framewise displacement, multiband EPI, visual perception development, longitudinal neuroimaging, naturalistic stimuli, computational neuroscience, neurodevelopment
News Source: Harold Sullivan. (November 20, 2026). Awake Infant Brain Scans Successfully Map Visual Hierarchy Across Early Development. Scienmag.



