fNIRS Single Trial Analysis and Classification using the General Linear Model

Key Researchers: Alexander von Lühmann, BU; Meryem Ayşe Yücel, BU; Antonio Ortega-Martinez, BU; David Boas, BU

Summary: fNIRS signal preprocessing and cleaning pipelines for single-trial classification often follow simple recipes. General Linear Model (GLM), when correctly applied in single trial analysis, e.g., in BCI, can significantly enhance contrast to noise ratio of the brain signal, improve feature separability and ultimately lead to better classification accuracy. In this manuscript, we provide a brief introduction into the GLM and show how to incorporate it into a typical BCI preprocessing pipeline and crossvalidation. We show that the GLM-based approach can provide better single trial estimates of brain activity as well as a new feature type, i.e., the weight of the individual and channel-specific hemodynamic response function (HRF) regressor. The improved estimates yield features with higher separability, that significantly enhance accuracy in a binary classification task when compared to conventional preprocessing—on average +7.4% across subjects and feature types.

Publication: von Lühmann et al., Frontiers, 2020, PMID: 31870944

Link: https://github.com/avolu/GLM-BCI

Funding: NIH R24NS104096