Predicting scholastic ability with resting-state functional near-Infrared spectroscopy
Poster No:
2430
Submission Type:
Abstract Submission
Authors:
Taehoon Kim1, Jihyun Cha1, Junggu Choi2, JongKwan Choi1, Sanghoon Han2
Institutions:
1OBELAB, Seoul, Korea, Republic of, 2Yonsei University, Seoul, Korea, Republic of
First Author:
Co-Author(s):
Introduction:
Recent studies employing connectome-based predictive modeling (CPM), a robust data-driven approach for constructing predictive models of brain-behavior relationships from neural connectivity (Shen, et al., 2017), have shown promise in elucidating individual differences. This method has successfully predicted diverse aspects of human cognition and personality traits including fluid intelligence (Finn et al., 2015), creative ability (Beaty et al., 2018), and an individual's level of anxiety (wang et al., 2021). The potential application of predicting real-life outcomes, such as professional aptitude or susceptibility to mental/physical disorders, is significant. Given the pivotal role the prefrontal cortex (PFC) plays in higher cognition, our study applies the CPM framework to resting-state functional near-infrared spectroscopy (fNIRS) signals acquired from the PFC. Specifically, we aim to examine the relationship between resting-state connectivity and academic ability, as represented by the scores on the Korean version of college scholastic assessment test (K-SAT).
Methods:
We analyzed resting-state fNIRS data from 100 participants (mean age = 19.17, sd = 0.55, 60 females) who had recently taken the K-SAT. Their resting-state fNIRS data were collected using NIRSIT LITE (OBELAB, Inc., Rep. of Korea), a portable 15-channel fNIRS system covering BA 10. Participants underwent a 5-minute resting-state scan with their eyes open.
The preprocessing involved converting raw light intensity to delta optical density, followed by motion artifact correction using spline interpolation (Scholkmann et al., 2010) and wavelet filtering (Molavi et al., 2012). The cleaned optical density data were then converted to the relative concentration changes in oxy- and deoxy-Hb based on the modified Beer-Lambert law (Delpy et al., 1988). A Butterworth filter (bandpass cutoff of 0.005 and 0.1 Hz, respectively) was applied to eliminate slow drift and physiological noise. In addition, short channel regression (Santosa et al, 2020) and an autogressive model (Lanka et al., 2020) were applied to remedy superficial physiological noise and serial correlation in the timeseries.
Functional connectivity between 15 channels was calculated, forming connectivity matrices with 105 edges. This resulted in 100 matrices reflecting intrinsic PFC connectivity patterns of 100 individuals (Figure 1). Using a leave-one-out cross-validation (LOOCV) approach, we divided the data into training (99 participants) and testing sets (1 participant). In the training set, we tested a linear relationship between each edge's strength and K-SAT score. Edges with significant positive or negative correlations were extracted and combined to generate positive and negative weights for individual score prediction. The resulting model was then applied to the testing data to yield prediction on the K-SAT score. The process was iterated through cross-validation, providing predictions for the K-SAT scores of the entire 100 participants.
The preprocessing involved converting raw light intensity to delta optical density, followed by motion artifact correction using spline interpolation (Scholkmann et al., 2010) and wavelet filtering (Molavi et al., 2012). The cleaned optical density data were then converted to the relative concentration changes in oxy- and deoxy-Hb based on the modified Beer-Lambert law (Delpy et al., 1988). A Butterworth filter (bandpass cutoff of 0.005 and 0.1 Hz, respectively) was applied to eliminate slow drift and physiological noise. In addition, short channel regression (Santosa et al, 2020) and an autogressive model (Lanka et al., 2020) were applied to remedy superficial physiological noise and serial correlation in the timeseries.
Functional connectivity between 15 channels was calculated, forming connectivity matrices with 105 edges. This resulted in 100 matrices reflecting intrinsic PFC connectivity patterns of 100 individuals (Figure 1). Using a leave-one-out cross-validation (LOOCV) approach, we divided the data into training (99 participants) and testing sets (1 participant). In the training set, we tested a linear relationship between each edge's strength and K-SAT score. Edges with significant positive or negative correlations were extracted and combined to generate positive and negative weights for individual score prediction. The resulting model was then applied to the testing data to yield prediction on the K-SAT score. The process was iterated through cross-validation, providing predictions for the K-SAT scores of the entire 100 participants.
·Overview of Connectome-based Predictive Modeling Framework with fNIRS
Results:
Out of discrete sections of K-SAT probing diverse domains, we primarily focus on the prediction results for the language comprehension section. To evaluate the model performance, we calculated the spearman's rank correlation between the predicted and their actual scores (Figure 2). A moderate correlation (r(98) = .37, p < .001), comparable to reported CPM results based on whole-brain fMRI was observed. A separate analysis indicated no significant correlation between the predicted scores and the participant's Raven's progressive matrices scores. Overall, the results suggest that the predicted score uniquely reflects an individual's language comprehension ability, independent of their general IQ.
·Relationship between the predicted and actual K-SAT scores and features participated in prediction
Conclusions:
Our study highlights the potential of resting-state fNIRS in predicting academic ability, demonstrating the relevance of CPM even in its simpler form. Further exploration of these neural connections promises valuable insights into cognitive abilities and educational outcomes.
Higher Cognitive Functions:
Higher Cognitive Functions Other
Lifespan Development:
Early life, Adolescence, Aging
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Novel Imaging Acquisition Methods:
NIRS 1
Keywords:
Modeling
Near Infra-Red Spectroscopy (NIRS)
Other - Scholastic Aptitude Test
1|2Indicates the priority used for review
Provide references using author date format
Delpy, D. T., Cope, M., Van Der Zee, P., Arridge, S., Wray, S., & Wyatt, J. (1988). Estimation of optical pathlength through tissue from direct time of flight measurement. Physics in Medicine and Biology, 33(12), 1433–1442.
Finn, E. S. (2015). Functional connectome fingerprinting: Identifying individuals using patterns of brain connectivity. Nature Neuroscience, 18(11), 1664–1671.
Lanka, P., Bortfeld, H., & Huppert, T. J. (2022). Correction of global physiology in resting-state functional near-infrared spectroscopy. Neurophotonics, 9(3), 035003-035003.
Molavi, B., & Dumont, G. A. (2012). Wavelet-based motion artifact removal for functional near-infrared spectroscopy. Physiological measurement, 33(2), 259.
Santosa, H., Zhai, X., Fishburn, F., Sparto, P. J., & Huppert, T. J. (2020). Quantitative comparison of correction techniques for removing systemic physiological signal in functional near-infrared spectroscopy studies. Neurophotonics, 7(3), 035009-035009.
Scholkmann, F., Spichtig, S., Muehlemann, T., & Wolf, M. (2010). How to detect and reduce movement artifacts in near-infrared imaging using moving standard deviation and spline interpolation. Physiological measurement, 31(5), 649.
Shen, X. (2017). Using connectome-based predictive modeling to predict individual behavior from brain connectivity. Nature Protocols, 12(3), 506–518.
Wang, Z., Goerlich, K. S., Ai, H., Aleman, A., Luo, Y. J., & Xu, P. (2021). Connectome-based predictive modeling of individual anxiety. Cerebral Cortex, 31(6), 3006-3020.