Plasticity in Face Perception
The following highlights show the multimodal approach I take to answer fundamental questions about face plasticity, including:
-
can a child, deprived of sight in the first few years of life, still learn to distinguish faces from non-faces, or to visually identify their peers?
-
what are the developmental mechanisms supporting this learning and can we use these schemes to facilitate learning later in life?
-
are the constituents of face recognition specific to faces per se, or rather to the experience we, as humans, have with faces?
Explore this recent talk I gave at the Bar Illan Vision Seminar about my work on plasticity in face recognition.
Key Findings
1
Plasticity in late visual pathways
Cataract-treatment for blindness since birth leads to structural changes in the brain’s white matter in late-, but not early-, visual pathways, showing selective plasticity to some parts of the visual cortex. These late visual pathways are believed to be involved in higher-order visual function, such as face perception. Indeed, we found that changes to these pathways directly modulate the children’s behavioral improvements in discriminating between face and nonface patterns, but not changes to their visual acuity.
This work was a close collaboration between Project Prakash and Bas Rokers lab, spearheaded by the talented Catarina Pedestini.
Tractography-generated white matter tract pathways in a cataract patient.
Read our paper:
Pedersini, C.A., Miller, N.P., Gandhi, T., Gilad-Gutnick, S., Mahajan, V., Sinha, P., Rokers, B., (2023). White Matter Plasticity Following Cataract Surgery in Congenitally Blind Patients. PNAS, 120(19) e2207025120
Read the MIT News coverage of this work:
Scientists discover anatomical changes in the brains of the newly sighted
2
Development of cortical specialization
Longitudinal fMRI dynamic face localizers reveal that despite not showing any face-specific brain activity prior to sight-providing surgery, sight onset leads to the emergence of stereotypical face-selective activations in both the ventral and dorsal pathways of the distributed face network. In fact, after a year of visual experience we even see some frontal “person knowledge” areas light up. Our data analysis is ongoing and our next goal is to determine which of these areas comes online first, and if one type of information (e.g. encoding of dynamics) drives the emergence of other types of information (e.g. static face/nonface discrimination).
Dynamic: Faces > Objects
Static: Faces > Objects
Ventral view of child who gained sight at 7 years old shows she starts off without any face selectivity in the ventral area most associated as representing faces, but within one year of treatment she gains an increasingly large and robust FFA, or face specific response, regardless of if faces are shown in motion or not.
Dorsal view shows initially widespread and non-specific activity, that becomes more localized to the stereotypical face-selective areas over the course of one year.
Overall, see all the stereotypical face-selective patches of the network emerging, including some of the frontal “person knowledge” areas.
Orange = face activation; Blue = deactivation.
3
Motion's a special role in learning to categorize facial expressions
This study explores the impact of two cues on facial expression recognition following treatment for early visual deprivation: initial visual acuity and motion cues.
To address the role of initial visual acuity, cataract-treated children were grouped by pretreatment visual acuity, comparing those with acuity comparable to a newborn to those with acuity worse than a newborn. We find that even prior to treatment, residual vision prior to treatment that is comparable to a newborn allows some facial expression recognition to emerge, whereas acuity worse than a newborn hinders this skill regardless of cue type (static or dynamic). Interestingly, some skill acquisition before treatment doesn't influence post-treatment trajectories; both groups follow similar developmental paths. However, years after sight onset, the most profoundly impaired group exhibits a 'dynamic advantage,' or significantly better recognition of motion-based facial expressions, compared to static expressions. This advantage is absent in controls or children with better pretreatment visual acuity. The dynamic advantage may stem from a shift toward using lower spatial frequencies, influenced by non-differential fixation patterns during static and dynamic stimulus viewing.
Watch poster presentation from Cognitive Neuroscience Society meeting, 2021:
Update: The manuscript for this work is under review. Check soon for updates...
4
The adaptive role of low visual acuity for the development of facial identification
In this study, we empirically test predictions derived from our computational work. We propose that the initial period of extremely poor visual acuity characteristic of a newborn serves not only as a sign of visual system immaturity but also plays a crucial adaptive role in learning. This hypothesis suggests that initial low visual acuity encourages the system to develop larger areas of spatial integration, known to be effective for robust facial-identification, but not necessary for object-identification.
Our computational models confirm that low-resolution input leads to larger receptive fields, enhancing facial identification. Testing these predictions in cataract-treated children, we compared improvements on face- versus object- identification tasks following late sight onset. As predicted, children who experience low- followed by high-visual acuity improve in face identification, while those who begin their visual journey with high acuity do not. This supports the idea that starting with low-resolution input before high-resolution input aids future face identification learning, while high-resolution input from the beginning can be detrimental to facial, but not object, identification.
The results are discussed in the context of typical and atypical theories of visual development, along with potential frameworks for visual rehabilitation.
Update: manuscript is under review so be on the lookout for publication updates soon…
Update: I will present this work at the Vision Science Society Annual Meeting in May 2024