DrSVision: A Machine Learning Tool for Cortical Region-Specific fNIRS Calibration Based on Cadaveric Head MRI

Area of Science:

• Neuroscience
• Biomedical Engineering
• Medical Physics

Background:

• Functional Near-Infrared Spectroscopy (fNIRS) is a non-invasive neuroimaging method.
• A challenge in fNIRS is calibrating source-detector separation (SDS) for optimal sensitivity at depth (SAD).
• Accurate calibration is crucial for monitoring specific cortical regions in neuroscience and neuroimaging.

Purpose of the Study:

• To present DrSVision version 1.0, a standalone software tool for fNIRS calibration.
• To address the limitation of practical tools for optimizing SDS and SAD in fNIRS.
• To enable more precise application of fNIRS for targeted cortical region monitoring.

Main Methods:

• Monte Carlo (MC) simulations using segmented MRI data from cadaveric heads to model light attenuation.
• Computation of Sensitivity at Depth (SAD) for various source-detector separations (SDS).
• Training a Gaussian Process Regression (GPR) machine learning model to recommend optimal SDS for maximal sensitivity at target depths.

Main Results:

• DrSVision software was developed as a standalone tool, independent of third-party platforms.
• The software provides region-specific calibration outputs for experimental goals.
• The GPR model successfully recommends optimal SDS for achieving maximal SAD.

Conclusions:

• DrSVision software offers a practical solution for fNIRS calibration challenges.
• The tool supports more precise fNIRS applications by tailoring calibration to experimental needs.
• Future work includes subject-specific calibration and broader support for diverse experimental setups.