Effect of the Stimulus Timing in the fNIRS-based BCI using a RSVP paradigm
Poster No:
69
Submission Type:
Abstract Submission
Authors:
Seongyeon Lim1, Suh-Yeon Dong1
Institutions:
1Sookmyung Women's University, Seoul, Korea, Republic of
First Author:
Co-Author:
Introduction:
Functional near-infrared spectroscopy (fNIRS) has recently gained prominence in advancing brain-computer interfaces (BCIs) but faces limitations in rapid serial visual presentation (RSVP) tasks due to its temporal resolution issues[1,2]. Despite delays in hemodynamic responses, we observed notable differences in hemodynamic responses between target and non-target groups. These finding suggests the potential of fNIRS-BCI with quick stimuli.
Methods:
From the STL-10 dataset[3], we used vehicle images with 288×288 pixels such as airplanes, cars, ships, and trucks. A single airplane image was the target, while non-target images were randomly chosen from other vehicles. The background of each image was blurred to minimize its interference with object recognition. In total, we made 40 image sequences. Each image sequence comprised 20 randomly selected images, with target image sequence containing one target image along with 19 non-target images. Thirty healthy adult females (22.34 ± 2.48 years) participated. Participants completed 40 sets, each with 20 target and 20 non-target image sequences in a random order. Figure 1 shows the timeline of the experiment. The onset and offset of each set were signaled by 'Start' and 'End' messages for 1-second, respectively. Participants viewed a burst of 20 images for 2 seconds, presented at a rate of 10 Hz They were required to identify the presence of the target image by key press. A 30-second resting period accounted for response delays. The fNIRS were recorded using a 15-channel NIRSIT Lite device (OBELAB Inc., Seoul, Republic of Korea. Approval for this study was granted by the Institutional Review Board of Sookmyung Women's University (IRB No. SMWU- 2209-HR-083-01).
The target groups were categorized into early, middle, and last groups based on image presentation order. The 20 target image sequences were distributed into 7 for early group, 7 for middle group, and 6 for last group. We extracted ∆HbO mean values and normalized the data by scaling the range from 0 to 1. We performed normality and homoscedasticity of the data using Shapiro-Wilk and Levene tests, respectively. Games-Howell and Bonferroni post-hoc tests were applied to compare each group after conducting One-way ANOVA, Welch's ANOVA, and Kruskal-Wallis tests. If the data exhibit characteristics of normality and homogeneity of variance, we performed One-way ANOVA, and Bonferroni post-hoc tests. For non-parameter data, we conducted Kruskal-Wallis and Bonferroni post-hoc tests. We conducted Welch's ANOVA and the Games-Howell test on parametric data with non-homogeneity of variance. Significance was determined based on Bonferroni and Games-Howell adjusted p-values <0.05.
The target groups were categorized into early, middle, and last groups based on image presentation order. The 20 target image sequences were distributed into 7 for early group, 7 for middle group, and 6 for last group. We extracted ∆HbO mean values and normalized the data by scaling the range from 0 to 1. We performed normality and homoscedasticity of the data using Shapiro-Wilk and Levene tests, respectively. Games-Howell and Bonferroni post-hoc tests were applied to compare each group after conducting One-way ANOVA, Welch's ANOVA, and Kruskal-Wallis tests. If the data exhibit characteristics of normality and homogeneity of variance, we performed One-way ANOVA, and Bonferroni post-hoc tests. For non-parameter data, we conducted Kruskal-Wallis and Bonferroni post-hoc tests. We conducted Welch's ANOVA and the Games-Howell test on parametric data with non-homogeneity of variance. Significance was determined based on Bonferroni and Games-Howell adjusted p-values <0.05.
Results:
After excluding poor-quality data, we analyzed 28 data. We examined the concentration changes of oxygenated hemoglobin (∆HbO) variations in the prefrontal cortex (PFC) across four groups using mean values. Across all PFC channels, significant distinctions were observed between the target and non-target groups(p<0.001), as shown in Figure 2. Specially, Channel 3 showed noticeable differences among all groups (p<0.001). Channel 7 exhibited significant differences within target groups(p<0.001). Differences between early and middle groups were observed in Channels 4(p=0.002), 8(p=0.025), 1, 5, 9, and 11(p<0.001). Among the early and last groups, there were distinctions in Channels 4(p=0.006), 6(p=0.014), 1, 9, 10, 12 and 15(p<0.001). Channels 10, 11, 12, and 15 significantly differed between middle and last groups (p<0.001), with lower significance in Channel 8(p=0.046). However, there was no significant difference in Channel 13 among all groups.
Conclusions:
Our findings back the use of fNIRS-BCI in RSVP tasks, showing discernible differences in hemodynamic responses between target and non-target groups. These differences varied with the timing of the target image presentation. Our future work aims to develop a dependable and robust BCI system using fNIRS.
Brain Stimulation:
Non-Invasive Stimulation Methods Other 1
Modeling and Analysis Methods:
Activation (eg. BOLD task-fMRI) 2
Keywords:
Near Infra-Red Spectroscopy (NIRS)
Statistical Methods
Other - Rapid serial visual presentation (RSVP); Brain-Computer Interface (BCI)
1|2Indicates the priority used for review
Provide references using author date format
[2] Ahn, S. (2017), ‘Multi-modal integration of EEG-fNIRS for brain-computer interfaces–current limitations and future directions’, Frontiers in human neuroscience, vol. 11, pp. 503.
[3] Coates, N.A. (2011), 'An analysis of single-layer networks in unsupervised feature learning', In Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, pp. 215-223.
This work was supported by the Agency For Defense Development Grant Funded by the Korean Government(UI233005TD).