Multi-Stage Cortical Plasticity Induced by Visual Contrast Learning

Abstract

Perceptual learning, the improved sensitivity via repetitive practice, is a universal phenomenon in vision and its neural mechanisms remain controversial. A central question is which stage of processing is changed after training. To answer this question, we measured the contrast response functions and electroencephalography (EEG) before and after ten daily sessions of contrast detection training. Behavioral results showed that training substantially improved visual acuity and contrast sensitivity. The learning effect was significant at the trained condition and partially transferred to control conditions. Event-related potential (ERP) results showed that training reduced the latency in both early and late ERPs at the trained condition. Specifically, contrast-gain-related changes were observed in the latency of P1, N1-P2 complex, and N2, which reflects neural changes across the early, middle, and high-level sensory stages. Meanwhile, response-gain-related changes were found in the latency of N2, which indicates stimulus-independent effect in higher-level stages. In sum, our findings indicate that learning leads to changes across different processing stages and the extent of learning and transfer may depend on the specific stage of information processing.

Keywords: ERP; contrast gain; latency; perceptual learning; response gain.

FIGURE 1

Model predictions, experimental stimuli and protocol. (A) Three different mechanisms in the sensory gain model that predict different pattern of contrast response function (CRF) changes following perceptual learning. From left to right: change in contrast gain, multiplicative response gain, or baseline shift. c₅₀: the stimulus contrast that evokes half of the neuron’s maximal response. R_max: maximal response to stimuli. b: baseline activity. (B) A typical trial procedure. Each trial started with an attention cue (500–800 ms). Stimulus was presented for 114 ms, and subjects were asked to report grating orientation within 1,500 ms. After response or 1,500 ms, a blank screen was presented for 600 ms and next trial started afterward. Training was performed in the upper left visual field location relative to the fixation dot. The dashed, white circles indicate two control locations: the upper right and the lower left visual field location. To ensure task compliance, subjects were asked to focus on the central fixation dot and press corresponding key when the black dot changed to “x” or “o” (with 5% probability each). (C) Experimental design. Participants were instructed to practice contrast detection tasks for ten consecutive days. Pre- and post-training psychophysical measurements covered contrast sensitivity function and visual acuity. ERP tests were conducted before and after contrast sensitivity training in different days to examine learning-induced changes in neural processing.

FIGURE 2

Behavioral results. (A) Post- versus pre-training logMAR visual acuity of trained (red circles) and untrained (gray triangles) eyes. Each symbol represents the data of one subject. The dashed line is the identity line (slope = 1), indicating no improvement. (B) Learning curve. Error bars represent standard errors across subjects. The first (blue triangles) and last data points (red circles) were derived from pre- and post-training CSF measurements in the trained condition, respectively. Black open circle: data from training phase. (C) Pre- (blue curves) and post-training (red curves) CSFs and the difference between the best fitting post- and pre-training CSFs (gray curves) measured in the trained location (the upper left), the upper right (Location change-contralateral) and the lower left (Location change-ipsilateral) visual field location in LE, and the upper left location in RE (Eye change). The enlarged symbols indicate the trained condition (spatial frequency: 5 cpd; location: the upper left; eye: LE) before (blue triangles) and after training (red circles). BW: the bandwidth of perceptual learning.

FIGURE 3

(A) Averaged ERP waveforms of the trained condition. The ERPs evoked by contralateral stimuli of 4.26, 8.90, 18.61, 38.90, and 81.13% Michelson contrast levels were subtracted by that evoked by contralateral 0%-contrast stimuli. Significant sensory ERP components, e.g., P1, N1, P2, and N2, were identified. Shaded regions denote standard errors across subjects. (B) Latency and amplitudes from early to late ERP components at each contrast levels of the trained condition in pre-training and post-training sessions. Statistical analysis showed that the latency and amplitude from early to late ERP components at each contrast levels were modified differently by training. Error bars represent standard errors across subjects. *: significant main effects of training; n.s.: non-significant. (C) The grand-mean topographical map series from 100 to 900 ms in steps of 80 ms evoked by stimuli of 81.13% contrast level of the trained condition in pre-training (upper part) and post-training (middle part) sessions. The difference topographical maps were also displayed (lower part). Four components occurred at this time window, from P1, N1, P2, to N2.

FIGURE 4

(A) Averaged ERP responses of control conditions. From left to right: Frequency change, Location change-contralateral (the upper right visual field location in LE), Location change-ipsilateral (the lower left location in LE) and Eye change condition (the upper left location in RE). Shaded regions denote standard errors across subjects. (B) Latency and amplitudes from early to late ERP components of the four control conditions in pre-training and post-training sessions. Statistical analysis showed that the amplitude and latency from early to late ERP components of the four control conditions were also modified differently by training. Error bars represent standard errors across subjects.

FIGURE 5

Effects of perceptual learning on the mean latency and amplitude of the P1, N1-P2 complex, and N2 components in the trained condition as function of contrasts (i.e., ERP-dependent CRF). (A,D) CRFs for P1 latency and amplitude. (B,E) CRFs for N1-P2 complex latency and amplitude. (C,F) CRFs for N2 latency and amplitude. For the latency CRF (A–C), training lead to c₅₀ improvement for both the P1 and N1-P2 complex, and c₅₀ and response increase for N2. For the amplitude CRFs (D–F), training led to c₅₀ improvement and multiplicative response increase for P1, and multiplicative response and baseline increase for N1-P2 complex.