Advancing Neuroimaging Precision: AI-Driven Analysis of Multi-Frequency fNIRS for Enhanced Brain Function Mapping

Functional near-infrared spectroscopy (fNIRS) is increasingly recognized as a portable, non-invasive neuroimaging modality for studying brain health and activity [1]. Traditional fNIRS approaches typically use 2 or more wavelengths via a continuous wave (CW) approach: that is only measuring the attenuation of light as it samples the tissue, which limits applications to only measuring relative changes of the signal and only absorption-related changes. As a result, absolute quantification of haemoglobin concentrations is not possible, scattering effects are ignored, and depth sensitivity is reduced, which complicates inter-subject or longitudinal comparisons. This project aims to develop, evaluate, and test a framework that harnesses multi-frequency fNIRS signals for a transformative improvement in brain state decoding, artifact rejection, and biomarker discovery. Specifically, through the use of a novel multi-Frequency Domain (mFD) device we will measure both the attenuation and ‘time-of-flight’ information of light as sampled from tissue, hence providing a mechanism for not only absolute measures of brain state and activity, but also information about both tissue absorption and scatter. Through advanced machine learning (ML) and model/data-driven approaches, we will leverage the rich spectral content of multi-frequency datasets to achieve significantly greater accuracy in image reconstruction, neurocognitive state classification, and clinical insights, ultimately positioning fNIRS as a scalable alternative to higher-cost modalities such as fMRI.

The proposed PhD project centres around the development, testing and validation of model-based and data-driven optimization approaches that will allow the utilization of multi-frequency NIRS data for the recovery (and imaging) of brain function and health. Although we have a good understanding of mFD models [2] and have previously demonstrated that such an approach can improve both the resolution and depth-sampling of the NIRS signal within the brain [3], our work to date has been limited to numerical studies only. The mFD system developed by Brain Optics is the first and only multi-frequency fNIRS device worldwide, and through direct collaboration with the inventors, we will create and validate multi-frequency computational models, investigate the use of time-dependent data-types for absolute (rather than relative) parameter recovery, and demonstrate that use of mFD will provide stable data that is able to provide information at depth. Furthermore, the project will develop data-preprocessing techniques for mFD, to further investigate its robustness to noise and artifacts in line with the FRESH study [4]. The impacts are several folds: demonstrating the benefits of mFD-fNIRS that will outpace the widely used CW techniques will have a significant impact in clinical research settings. Furthermore, all models and optimization algorithms will form part of the systems processing tools which will contribute significantly to the refinement of the system as well as adoption and use of such approaches to ultimately benefit both academia and our industrial partners. We envisage that the ML and computational modelling and development to contribute significantly to the academic impact, while validation, testing and refinement will enhance the update of this system within the neuroimaging and clinical community to both enhance clinical and commercial impact.

References:

1. Hasan Ayaz et al, "Optical imaging and spectroscopy for the study of the human brain: status report," Neurophoton. 9(S2) S24001 (2022) https://doi.org/10.1117/1.NPh.9.S2.S24001

2. Hamid Dehghani et al, ”Near infrared optical tomography using NIRFAST: Algorithm for numerical model and image reconstruction. Commun. Numer. Meth. Engng., (2008) https://doi.org/10.1002/cnm.1162

3. Guy A. Perkins et al, "Quantitative evaluation of frequency domain measurements in high density diffuse optical tomography," J. Biomed. Opt. 26(5) 056001 (2021) https://doi.org/10.1117/1.JBO.26.5.056001

4. Meryem Ayşe Yücel et al, “The fNIRS Reproducibility Study Hub (FRESH): Exploring Variability and Enhancing Transparency in fNIRS Neuroimaging Research” Commun Biol 8, 1149 (2025). https://doi.org/10.1038/s42003-025-08412-1