Neural correlates of visual plasticity
Throughout life neural connectivity is continuously changing, correlating with for instance learning new skills, storing memories and improving perceptual performance. In other words, our brain adapts to the needs that are required to perform optimally in our environment and interact with it. Part of my research focuses on measuring neural responses associated with this so-called plasticity in an attempt to learn more about the involved mechanisms. We use an experimental technique called 'frequency tagging' in which different parts of a stimulus are flickering on the screen at different temporal frequencies. This results in peak responses at the flickering frequencies in the fast fourier transform (FFT) performed on the EEG signal. We are interested in how these so-called steady state visually evoked responses (SSVEPs) are modulated through visual experience.
Contrast adaptation modulates SSVEP responses
When stimuli are repetitively presented with one and the same stimulus the brain adapts to this stimulus, becoming less sensitive to the features of the adapted stimulus. In case of contrast adaptation, the repetitive presentation of a high-contrast oriented grating reduces the sensitivity to this orientation. We are measuring SSVEPs before and after adaptation as a neural marker of the effect of contrast adaptation.
Experiment 1
Methods
|
Experiment 2
Method
|
Results
Conclusions
|
Manuscript in preparation
Larger higher-order SSVEP responses after shape categorisation training
Relatively little is known about the neural mechanisms responsible for integrating parts into coherent wholes and about the learning mechanisms involved. In addition to the tagged frequencies and their harmonics, SSVEP responses can also be observed at so-called intermodulation frequencies (e.g., f1+f2; 2*f1-f2). Evoked activity at such intermodulation frequencies (IMs) is caused by nonlinear interactions between the neural signals carrying the two individual frequencies. Previous findings have suggested that intermodulation components reflect the neural activity involved in the integration of local elements, for instance, in face processing. Here we were interested in whether the strength of IM components are indicative of strengthening the integration of shape parts during categorisation learning.
Behavioural training
Stimuli were created by combining several radial frequency components (RFCs) into Fourier boundary descriptors (FBDs). Moreover, the contour of each shape was created as a sum of seven sine waves with different radial frequencies (2, 3, 4, 5, 6, 7, 8Hz). Two distinct families of shapes were created. Within one shape family, members differed in amplitude on 2 radial frequencies, while the amplitude was kept constant for the remaining 5 frequencies. The phases were kept constant for all 7 amplitudes. The corresponding 2-dimensional shape space for each family is displayed in the figure below. During the four-day training (in total 16 blocks of 160 trials), participants categorised shapes of one of both families into 2 categories without prior knowledge of the underlying linear categorisation rule. Trial-by-trial feedback was provided to allow participants to evaluate and adjust their performance.
Stimuli were created by combining several radial frequency components (RFCs) into Fourier boundary descriptors (FBDs). Moreover, the contour of each shape was created as a sum of seven sine waves with different radial frequencies (2, 3, 4, 5, 6, 7, 8Hz). Two distinct families of shapes were created. Within one shape family, members differed in amplitude on 2 radial frequencies, while the amplitude was kept constant for the remaining 5 frequencies. The phases were kept constant for all 7 amplitudes. The corresponding 2-dimensional shape space for each family is displayed in the figure below. During the four-day training (in total 16 blocks of 160 trials), participants categorised shapes of one of both families into 2 categories without prior knowledge of the underlying linear categorisation rule. Trial-by-trial feedback was provided to allow participants to evaluate and adjust their performance.
EEG recording and analysis After four days of training, participants completed an EEG recording session. In each trial of the EEG recording a shape from one of both families (either trained or untrained) was presented for 13 seconds while the left and right halves of the shapes were contrast modulated at different temporal frequencies (7.5Hz vs 5.45Hz). Our analysis focused on differences between conditions (trained and untrained exemplars and families) in frequency power for intermodulation (IM) frequencies (i.e., combinations of the 2 given frequencies), as these are argued to reflect holistic integration (see Figure 2). |
EEG results (SSVEPs)
|
Discussion
- We show behavioural effects of shape categorisation specific to the trained shape parameter space that do not transfer to an untrained family.
- This training does not lead to significant general changes in SNR for the trained shapes compared to untrained shapes, for any of the tested frequency types.
- For intermodulation frequencies, significant learning effects were found for higher-order IM frequency components in occipital areas.
- We speculate that this higher-order interaction between the tagged frequencies reflects the higher-order processing that is involved in learning successful shape categorisation, which requires global, holistic integration of stimulus elements into wholes.
Manuscript under revision as Vergeer, M., Kogo, N., Nikolaev, A.R., Alp, N., Loozen, V., Schraepen, B., & Wagemans, J. EEG frequency tagging provides a neural signature of holistic shape representations learned during shape categorization.