Rapid estimation of photon measurement density functions using a deep convolutional neural network for functional near-infrared spectroscopy
- 1Beijing Institute for Brain Research, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, 102206, China; Chinese Institute for Brain Research, Beijing, 102206, China; Beijing Key Laboratory of Brain Science and Brain-Machine Interface, China
- 2State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University, Beijing, China; IDG/McGovern Institute for Brain Research, Beijing Normal University, China; Center for Collaboration and Innovation in Brain and Learning Sciences, Beijing Normal University, China
Abstract
The photon measurement density function (PMDF) plays a fundamental role in both pre-experimental optode arrangement and post-experimental data analysis in functional near-infrared spectroscopy (fNIRS). Conventionally, PMDFs are derived from structural MR images through tissue segmentation and photon propagation modeling (PPM), which are computationally demanding and time-consuming, thereby limiting their practical use. In this study, we propose a novel deep learning–based framework to estimate PMDFs directly from MR images and channel configurations. The proposed method supports flexible source–detector distances and eliminates the need for explicit tissue segmentation and repeated photon simulations. Specifically, a convolutional neural network is trained to predict photon fluence distributions, from which PMDFs are subsequently derived using the adjoint formulation. The trained model is evaluated on channels placed in both trained and unseen scalp regions across commonly used source–detector distances. The results demonstrate that the proposed method achieves PMDF estimations comparable to those obtained from PPM. Overall, this approach significantly reduces computational cost and has the potential to facilitate broader adoption of PMDF-based methods in the fNIRS community.
Related
- Interpretable attention-deficit hyperactivity disorder diagnosis from functional magnetic resonance imaging via hybrid convolutional neural network–transformer learning and multi-level graph-based comprehensibility
- Estimation and detection of rotor mass imbalance of wind turbines using convolutional neural network
- Three-dimensional measurement of RBC speed, flux, and density using optical projection functional angiography
- Spatiotemporal feature-based investigation of major depressive disorder and emotion recognition using functional near infrared spectroscopy.
- Functional assessment of experimental lymphedema: a focused mini-review of animal models and measurement methods
BibTeX
@article{zhao2026rapid,
title = {{Rapid estimation of photon measurement density functions using a deep convolutional neural network for functional near-infrared spectroscopy}},
author = {Yang Zhao and Xue-Ting Sun and Wen-Di Shi and Chao-Zhe Zhu and Lei Zhang},
journal = {NeuroImage: Reports},
year = {2026},
month = oct,
doi = {10.1016/j.ynirp.2026.100425},
url = {https://doi.org/10.1016/j.ynirp.2026.100425},
note = {CC BY-NC-ND},
}
RIS
TY - JOUR
TI - Rapid estimation of photon measurement density functions using a deep convolutional neural network for functional near-infrared spectroscopy
AU - Yang Zhao
AU - Xue-Ting Sun
AU - Wen-Di Shi
AU - Chao-Zhe Zhu
AU - Lei Zhang
PY - 2026
DA - 2026/10/24
JO - NeuroImage: Reports
DO - 10.1016/j.ynirp.2026.100425
UR - https://doi.org/10.1016/j.ynirp.2026.100425
AB - The photon measurement density function (PMDF) plays a fundamental role in both pre-experimental optode arrangement and post-experimental data analysis in functional near-infrared spectroscopy (fNIRS). Conventionally, PMDFs are derived from structural MR images through tissue segmentation and photon propagation modeling (PPM), which are computationally demanding and time-consuming, thereby limiting their practical use. In this study, we propose a novel deep learning–based framework to estimate PMDFs directly from MR images and channel configurations. The proposed method supports flexible source–detector distances and eliminates the need for explicit tissue segmentation and repeated photon simulations. Specifically, a convolutional neural network is trained to predict photon fluence distributions, from which PMDFs are subsequently derived using the adjoint formulation. The trained model is evaluated on channels placed in both trained and unseen scalp regions across commonly used source–detector distances. The results demonstrate that the proposed method achieves PMDF estimations comparable to those obtained from PPM. Overall, this approach significantly reduces computational cost and has the potential to facilitate broader adoption of PMDF-based methods in the fNIRS community.
SN - 2666-9560
ER -