Data-driven optimization of preschoolers' hemodynamic response in a VR setup: advancing analytic methods for
Letizia Contini1, Rebecca Re1,2, Paola Pinti3,4
1Politecnico di Milano, Dipartimento di Fisica, Milan, Italy.
Insights
Analyzing fNIRS data in children requires new methods. A data-driven approach (AICopt) refines hemodynamic response functions, improving accuracy for overlapping events in naturalistic studies.
Area of Science:
- Neuroscience
- Developmental Psychology
- Biomedical Engineering
Background:
- Naturalistic functional near-infrared spectroscopy (fNIRS) studies in children offer insights into real-world behaviors.
- Standard analysis methods like block averaging and the general linear model (GLM) face challenges with overlapping events and atypical hemodynamic responses in children.
- These challenges can lead to misattribution of neural responses and reduced study sensitivity.
Purpose of the Study:
- To develop a data-driven framework to refine the shape and timing of hemodynamic response functions (HRFs) for each brain region and event type.
- To address the issue of overlapping responses in naturalistic paradigms, reducing the misattribution of neural signals to incorrect stimuli.
- To improve the accuracy and interpretability of fNIRS data analysis in pediatric populations.
Main Methods:
- Introduction of an automated HRF optimization procedure (AICopt) for GLM-based fNIRS analysis when HRFs are unknown.
- Evaluation of AICopt in 40 preschoolers (3-5 years) using a virtual-reality paradigm with immediate, overlapping emotional and neutral events.
- Comparison of AICopt performance against block-averaging and conventional GLM with canonical HRFs.
Main Results:
- AICopt produced activation patterns consistent with block-averaging for events.
- AICopt successfully avoided spurious activations related to choices that were present in the canonical GLM analysis.
- Data-driven HRFs enhanced sensitivity and reduced response misattribution compared to fixed canonical HRFs in an overlapping-event design.
Conclusions:
- Data-driven HRF modeling is crucial for analyzing fNIRS data in atypical populations like young children, especially in naturalistic settings.
- The AICopt method offers a viable approach to adapt GLM analyses for overlapping events and diverse populations, enhancing accuracy.
- AICopt provides a reusable workflow for analyzing child fNIRS data from non-standard experimental setups.
Significance:
Naturalistic fNIRS data acquired on children enable studying real-world behaviors but challenge standard analysis methods such as block averaging and general linear model (GLM). In naturalistic paradigms, events often overlap, whereas children's hemodynamic responses generally deviate from the adult canonical model, possibly leading to responses' misattribution and low sensitivity.
Aim:
We aim to reduce the risk of misattributing neural responses to stimuli by refining the shape and timing of the hemodynamic response function (HRF) for each brain region and event type in a data-driven framework, addressing cases where overlapping responses lead to neural responses being mistakenly assigned to the wrong stimulus, distorting results, and leading to misleading conclusions.
Approach:
We introduce a data-driven HRF optimization procedure (AICopt) that enables GLM-based analyses when the HRF is unknown. We evaluated the AICopt approach in 40 preschoolers (3 to 5 years) within a virtual-reality paradigm, featuring emotionally relevant and neutral events followed immediately by choices, without fixed inter-trial baselines. Then, we compare its performance with what is obtained using the block-averaging method and canonical HRF model-based GLM analysis.
Results:
AICopt yielded activation patterns that converged with block-averaging results for events while avoiding likely spurious choice-related activations seen with the canonical GLM. Overall, the use of data-driven HRFs improved sensitivity and reduced misattribution relative to the fixed canonical HRF in this overlapping-event design.
Conclusions:
Our results suggest that data-driven HRF modeling is a necessary step when analyzing fNIRS data from atypical populations such as young children, particularly in studies employing naturalistic setups. The presented AICopt method represents a possible approach to adapt GLM analyses to overlapping events and diverse populations, improving accuracy and interpretability of the obtained activation maps, and offering a reusable workflow for child fNIRS datasets collected in nonstandard setups.
