Poster No:
1225
Submission Type:
Abstract Submission
Authors:
Cliona O'Doherty1, Áine Dineen1, Anna Truzzi1, Graham King1, Keelin Harrison1, Lorijn Zaadnoordijk1, Enna-Louise D'Arcy1, Jessica White1, Tamrin Holloway1, Chiara Caldinelli1, Anna Kravchenko1, Sojo Joseph1, Eleanor Molloy2, Adrienne Foran3, Ailbhe Tarrant3, Angela Byrne2, Rhodri Cusack1
Institutions:
1Trinity College Dublin, Dublin, Ireland, 2The Coombe Hospital, Dublin, Ireland, 3The Rotunda Hospital, Dublin, Ireland
First Author:
Co-Author(s):
Introduction:
Object recognition and categorisation are foundational skills required to build conceptual knowledge. Although categorisation is well-studied in young cohorts there is still debate surrounding its origin in the ventral visual cortex (VVC). Initial infant neuroimaging shows that category representations appear early [Kosakowski et al., 2022] but the success of deep neural networks (DNNs) as models for the ventral stream suggests that categories are formed according to a hierarchy of feature complexity. In agreement with this, behavioural studies show very young infants can form mental groups according to perceptual features [Quinn et al., 1993] with grouping by conceptual features emerging later [Younger & Cohen 1993]. However, behavioural studies are limited as they test if an infant is acting upon knowledge and cannot provide evidence for the absence of a capacity for such knowledge. Neuroscience and computational modelling can provide unique insights richly complementing behavioral measures. To date, it has not been possible to probe infant brain representations underlying categorisation at scale. With pioneering advances in awake infant fMRI [Ellis et al., 2020] we have begun to tackle these questions by collecting the largest cohort of awake, behaving infants in an fMRI study to-date and quantifying their visual representations to a broad variety of categories.
Methods:
Infants (n=134) attended two scanning sessions at 2- and 9-months and were scanned in a Siemens 3T MAGNETOM Prisma. We measured brain responses to looming images: 12 categories of objects with 3 exemplars across viewpoints. The chosen categories had varying degrees of familiarity to young infants and spanned cortical organisation principles such as animacy and real-world size [Konkle & Caramazza, 2013]. Using multivariate pattern analysis, we quantified the distributed patterns of visual responses within the VVC and compared infant category organisation to adults. Representational similarity analysis with perceptual and categorical models was used to test the feature complexity in infant cortex and quantify the level of abstraction encoded across development. Infant brain activity was also compared to DNNs that were either untrained or trained to recognise 1000 categories of natural images. We measured the similarity to each layer of the DNN assuming that lower layers encode simpler features than later layers.
Results:
At 2-months we observed an impressive organisation by category in ventral visual stream with significant distinctions for within versus between category representations as well as organisation by animacy and real-world size. This initial representation becomes more adult-like by 9-months and interestingly, anterior ventral regions were found to become more similar to adults earlier than posterior visual regions. Analogously, comparison to a DNN trained on an object recognition task showed that anterior regions become closer to adults in their correlations to later layers of a DNN encoding complex features. In contrast, infants' early visual regions are more similar to an untrained network than adults as well as to earlier layers in the DNN.
Conclusions:
We have demonstrated that the neural basis of categorisation is present from as early as 2-months and that anterior ventral regions become more adult-like before posterior regions. The presence of this organisation is prior to its documented functional emergence from behavioural literature and contrary to the idea that simpler features in early visual cortex develop before complex category representations. For the first time we apply multivariate analysis methods to a large, innovative infant neuroimaging dataset and successfully define similarities between the infant brain and AI models. This provides novel insight into the workings of the youngest of humans and demonstrates the promise of awake infant fMRI for the future of understanding the brain and mind in early life.
Lifespan Development:
Early life, Adolescence, Aging 1
Modeling and Analysis Methods:
Multivariate Approaches
Novel Imaging Acquisition Methods:
BOLD fMRI 2
Perception, Attention and Motor Behavior:
Perception: Visual
Keywords:
Cognition
Computational Neuroscience
Development
Machine Learning
Multivariate
Vision
Other - Infancy
1|2Indicates the priority used for review
Provide references using author date format
Ellis, C. T., Skalaban, L. J., Yates, T. S., Bejjanki, V. R., Córdova, N. I., & Turk-Browne, N. B. (2020). Re-imagining fMRI for awake behaving infants. Nature Communications, 11(1), 4523.
Konkle, T., & Caramazza, A. (2013). Tripartite organization of the ventral stream by animacy and object size. Journal of Neuroscience, 33(25), 10235-10242.
Kosakowski, H. L., Cohen, M. A., Takahashi, A., Keil, B., Kanwisher, N., & Saxe, R. (2022). Selective responses to faces, scenes, and bodies in the ventral visual pathway of infants. Current Biology, 32(2), 265-274.
Quinn, P. C., Eimas, P. D., & Rosenkrantz, S. L. (1993). Evidence for representations of perceptually similar natural categories by 3-month-old and 4-month-old infants. Perception, 22(4), 463-475.
Younger, B. A., & Cohen, L. B. (1986). Developmental change in infants' perception of correlations among attributes. Child development, 803-815.