Frequency tagging with infants: The visual oddball paradigm

Abstract

Combining frequency tagging with electroencephalography (EEG) provides excellent opportunities for developmental research and is increasingly employed as a powerful tool in cognitive neuroscience within the last decade. In particular, the visual oddball paradigm has been employed to elucidate face and object categorization and intermodal influences on visual perception. Still, EEG research with infants poses special challenges that require consideration and adaptations of analyses. These challenges include limits to attentional capacity, variation in looking times, and presence of artefacts in the EEG signal. Moreover, potential differences between age-groups must be carefully evaluated. This manuscript evaluates challenges theoretically and empirically by (1) a systematic review of frequency tagging studies employing the oddball paradigm and (2) combining and re-analyzing data from seven-month-old infants (N = 124, 59 females) collected in a categorization task with artifical, unfamiliar stimuli. Specifically, different criteria for sequence retention and selection of harmonics, the influence of bins considered for baseline correction and the relation between fast periodic visual stimulation (FPVS) responses and looking time are analyzed. Overall, evidence indicates that analysis decisions should be tailored based on age-group to optimally capture the observed signal. Recommendations for infant frequency tagging studies are developed to aid researchers in selecting appropriate stimulation and analysis strategies in future work.

Keywords: analysis strategies; categorization; fast periodic visual stimulation; frequency tagging; infants; visual processing.

Figure 1

Flow diagram of study selection process for systematic review (Page et al., 2021).

Figure 2

Schematic illustration of the stimuli and experimental paradigms. Ten different angular red-orange and round blue-green stimuli were employed, with contrasting categories differing in the shape of all parts (angular vs. round outlines) and colour spectrum. Phase-scrambled control images were created from original stimuli. In 20-s-sequences, images were presented by sinusoidal contrast modulation at a rate of 6 cycles per second = 6 Hz (1 cycle ≈ 170 ms). Angular red-orange stimuli (A) were presented as standards, with round blue-green stimuli presented as deviants at every 5^th cycle (B; 6/5 Hz = 1.2 Hz).

Figure 3

Grand-average frequency-spectrum responses. Base (6 + 12 Hz) and categorization (1.2, 2.4, 3.6… 10.8 Hz) responses are clearly visible and highly significant. The effect of the number of bins employed for baseline correction was evaluated by comparing signal-to-noise-ratios (SNRs) was compared across 10 (2–6) and 20 (2–12) bins. While employing a larger number of bins did not seem to affect responses at higher frequencies (from 6 Hz onwards) detrimentally, response strength was stronger when employing fewer bins at lower frequencies (below 6 Hz). Crucially, the effect seemed strongest at 1.2 Hz, corresponding to the stimulation frequency.

Figure 4

SNR of grand-average base responses summed across harmonics 1–5 and compared across numbers of sequences averaged. The respective N contributing to each analysis is specified per graph. Highly significant base responses were obtained in all analyses.

Figure 5

SNR of grand-average categorization responses summed across harmonics 1–19 and compared across numbers of sequences averaged. The respective N contributing to each analysis is specified per graph. Highly significant categorization responses were obtained in all analyses.

Figure 6

SNR of grand-average base responses summed across harmonics 1–5 and compared across numbers of sequences averaged for a subsample of N = 20 who provided usable data for at least 8 sequences. Highly significant base responses were obtained in all analyses, with a tendency for increasing amplitude with number of sequences averaged.

Figure 7

SNR of grand-average categorization responses summed across harmonics 1–19 and compared across numbers of sequences averaged for a subsample of N = 20 who provided usable data for at least 8 sequences. The respective N contributing to each analysis is specified per graph. Significant categorization responses were obtained in all analyses. Averaging across larger numbers of sequences increased response strengths.

Figure 8

SNR of base grand-average responses summed across increasing numbers of harmonics (considering only the 1^st or up to harmonics 1–10). Highly significant responses were obtained in all analyses. The analysis on harmonics 1–5, marked by a black box, was indicated by the Z-score criterion.

Figure 9

SNR of categorization grand-average responses summed across increasing numbers of harmonics (considering only the 1^st or up to harmonics 1–29, always excluding harmonics corresponding to the frequency). Highly significant responses were obtained in all analyses. The analysis on harmonics 1–19, marked by a black box, was indicated by the Z-score criterion.