Deep learning-based motion artifact removal in functional near-infrared spectroscopy

Abstract

Functional near-infrared spectroscopy (fNIRS), a well-established neuroimaging technique, enables monitoring cortical activation while subjects are unconstrained. However, motion artifact is a common type of noise that can hamper the interpretation of fNIRS data. Current methods that have been proposed to mitigate motion artifacts in fNIRS data are still dependent on expert-based knowledge and the post hoc tuning of parameters. Here, we report a deep learning method that aims at motion artifact removal from raw fNIRS data while being assumption-and parameter selection-free. To the best of our knowledge, this is the first investigation to report on the use of a denoising autoencoder (DAE) architecture for motion artifact removal. To facilitate the training of this deep learning architecture, we (i) designed a specific loss function and (ii) generated data to mimic the properties of recorded fNIRS sequences. The DAE model was able to remove 93% of motion artifacts in the simulated dataset and 100% in the real dataset, outperforming current gold standard methods. In addition to that, the DAE model exhibited excellent computational efficiency. Overall, this work demonstrates the potential of deep learning models for parameter-free, accurate and fast motion artifact removal in fNIRS data.