Spatial Frequency Training Modulates Neural Face Processing: Learning Transfers from Low- to High-Level Visual Features

Abstract

Perception of visual stimuli improves with training, but improvements are specific for trained stimuli rendering the development of generic training programs challenging. It remains unknown to which extent training of low-level visual features transfers to high-level visual perception, and whether this is accompanied by neuroplastic changes. The current event-related potential (ERP) study showed that training-induced increased sensitivity to a low-level feature, namely low spatial frequency (LSF), alters neural processing of this feature in high-level visual stimuli. Specifically, neural activity related to face processing (N170), was decreased for low (trained) but not high (untrained) SF content in faces following LSF training. These novel results suggest that: (1) SF discrimination learning transfers from simple stimuli to complex objects; and that (2) training the use of specific SF information affects neural processing of facial information. These findings may open up a new avenue to improve face recognition skills in individuals with atypical SF processing, such as in cataract or Autism Spectrum Disorder (ASD).

Keywords: ASD; ERP; face processing; learning; neuroplasticity; spatial frequency.

Figure 1

Experimental design. (A) Timeline of the experimental protocol. After a pre-training (“baseline”) electroencephalogram (EEG) measurement on day 1, subjects participated in behavioral sessions on day 2, 4 or 5 and 7 in which they trained low spatial frequency (LSF) discrimination on grating stimuli. Finally, a post-training EEG was acquired on day 8, while subject performed an emotion categorization and oddball detection task identical to the pre-training EEG measurements. (B) Tasks in the LSF training (left) and EEG (right) sessions. Left: LSF discrimination skills were trained by detecting the odd-one-out target grating which SF was increasingly similar to the reference gratings as performance improved (i.e., staircase tracking 84% accuracy). In this example trial, the grating with a different SF than the fixed SF (2 cpa) of the reference gratings, is the second grating in the row. Therefore, the correct answer is “2”. In catch trials, where two target gratings were shown, participants pressed the spacebar (instead of the number corresponding to the position of the deviant grating). Gratings were always presented in the left hemifield, to allow comparisons between trained (left hemifield) and untrained (right hemifield) visual field locations. Right: in the EEG measurements, subjects performed an emotion categorization (left image series) and oddball detection (right) task on low-pass (LSF) and high-pass filtered (HSF) faces. Note that faces were presented at the same position as the gratings in the training task (trained hemifield) or at the mirror location in the opposite hemifield (untrained hemifield).

Figure 2

Average learning curve for the LSF training. The difference between the SF of the target and reference (as percentage of the SF reference grating with 2 cpa SF) as a function of concatenated trials of session 1, 2 and 3. The shaded area indicates standard error of the mean.

Figure 3

Grand average waveforms of LSF and HSF faces presented in the trained hemifield in the pre- and post-training session elicited at electrode PO8 in the (A) oddball detection and (B) emotion categorization task. (C) Differential mean N170 activation between the pre- minus post-training per hemifield stimulation (x-axis) and SF content in the emotion categorization task (arrows indicate the correspondence between activity shown in B,C). Note that the training-induced difference (*) is only present for LSF faces presented in the trained hemifield. Error bars indicate standard error of the mean.