Tutorial on Pre-processing of fNIRS Signals (NIRSport 2 System)

Manousos A. Klados
Jul 8
2 min read

Introduction to Functional Near-Infrared Spectroscopy (fNIRS)

Functional near-infrared spectroscopy (fNIRS) is a non-invasive neuroimaging technique that measures brain activity through the hemodynamic responses associated with neuronal behavior. It utilizes near-infrared light to detect changes in oxygenated (HbO) and deoxygenated (HbR) hemoglobin concentrations in cortical tissues. Due to its portability, ease of use, and high temporal resolution, fNIRS has become an increasingly popular tool in cognitive neuroscience, clinical diagnostics, and developmental studies, allowing researchers to investigate brain function in naturalistic environments and populations where traditional neuroimaging techniques like fMRI are challenging to implement.

Step-by-Step fNIRS Preprocessing Pipeline

Step 1: Loading SNIRF Data

First, we need to load our SNIRF data. Python’s mne library offers convenient support for reading SNIRF files.

import mne
# Load SNIRF 
dataraw_fnirs = mne.io.read_raw_snirf('your_file.snirf', preload=True)
print(raw_fnirs.info)

Step 2: Converting Raw Intensity to Optical Density

Raw intensity signals must be converted into optical density to reflect changes in absorbed light.

raw_od = mne.preprocessing.nirs.optical_density(raw_fnirs)

Step 3: Converting Optical Density to Hemoglobin Concentration

Next, convert optical density to oxy- (HbO) and deoxyhemoglobin (HbR) concentrations using the modified Beer-Lambert law.

raw_haemo = mne.preprocessing.nirs.beer_lambert_law(raw_od)

Step 4: Filtering

Apply a band-pass filter to remove physiological artifacts (e.g., heartbeat, respiration) and instrumental noise.

raw_haemo.filter(l_freq=0.01, h_freq=0.2, h_trans_bandwidth=0.05, verbose=True)

Step 5: Artifact Detection and Correction

Identify and correct motion artifacts, often prominent in fNIRS data.

from mne.preprocessing.nirs import scalp_coupling_index
# Calculate Scalp Coupling Index (SCI) to identify bad channels
sci = scalp_coupling_index(raw_od)
raw_haemo.info['bads'] = list(sci[sci < 0.5].index) # Mark bad channels  
# Interpolate bad channels
raw_haemo.interpolate_bads()

Step 6: Data Downsampling (Optional)

Reduce data size and computational load if necessary.

raw_haemo.resample(sfreq=2.0)  # Resample to 2 Hz

Step 7: Epoching

Segment continuous data into epochs aligned with experimental events.

events, event_dict = mne.events_from_annotations(raw_haemo)
epochs = mne.Epochs(raw_haemo, events, event_id=event_dict, tmin=-5, tmax=15,                    baseline=(None, 0), detrend=1, preload=True)

Final Thoughts

This preprocessing pipeline will give you clean, artifact-free fNIRS data ready for statistical analysis or machine learning applications. You can customize each step to better fit your research needs and data characteristics. Happy analyzing!