Classification of Tactile and Motor Velocity-Evoked Hemodynamic Response in Primary Somatosensory and Motor Cortices as Measured by Functional Near-Infrared Spectroscopy
Round 1
Reviewer 1 Report
The current paper titled “Classification of velocity-evoked hemodynamic response in primary somatosensory and motor cortices as measured by functional near-infrared spectroscopy” continues the work from the group of pneumotactile and sensorimotor stimuli response in the brain, first by MEG, fMRI and now with fNIRS. The authors use a regularized discriminant analysis to distinguish 3 velocity levels of tactile stimuli (active and passive) above the chance level. The discrimination between 3 velocity stimuli is novel and the sensorimotor area, a perfect target for fNIRS. However I missed justifications for choice of analysis and explanation of processing methods. In addition, I am missing the discussion regarding the usefulness for applicability of the information acquired here in the conclusion, which was eluded to in the introduction.
Specific comments:
Line 70 – Why use RDA among the machine learning algorithms? Please justify. How applicable to clinical settings is RDA and predictability etc. In addition, comparison to standard GLM would be appropriate as whole datasets are used anyway (preprocessing done on entire dataset before looking at samples) [1].
Line 93- Figure 1.a. Where are these channels located in regards to the 10-20 system or previous work from your group on velocity differences in MRI data? In results, where there specific channels particularly good in discriminating?
Line 98- Figure 1.e. Please state the range for time intervals between two consecutive nodes based on hand measurements. Is there a possibility that one person got less stimulation in the 20s window because his/her hand was bigger or is this completely neglectable?
Line 124 – What default modules were used? Please specify as there can be differences which affect the data [2].
Line 128 – Why was bandpass filter set to 0.01-0.18 Hz with the relevant frequency at 0.025Hz (every 40 seconds).
Line 133 -how many samples per trial of 20s were used? If I understand correctly, 5 second windows with 80% overlap of the last 10 second of every trial results in approximately 6 samples per trials. With 10 repetitions per velocity, 60 samples per velocity were used (not independent). Is that correct? An example of these samples or averaged over the samples of each velocity in 1 representative subject would be helpful to see if differences between velocities are visible on the timecourse or location. What was your hypothesis?
Line 143 – Why were the channels excluded, which had < 60% accuracy in differentiating between on and off stim? These were supposedly all healthy subjects and data quality should be determined beforehand. Please explain. In addition, how many channels were excluded per subject? - Figure 4 makes it really hard to discriminate between the >60% cutoff colors and below. Please revise.
Line 197 - Confusion matrix. Please add citation or give additional explanation.
Line 184 and Line 233. You expect a higher amplitude in the active paradigm as possible explanation for the higher accuracy in active vs passive. Did you see that in the data? You also mention motion as possible confounder, did you make sure the preprocessing steps resulted in adequate removal of the motion artifacts (were there more motion artifacts in the active in comparison to passive task?)?
- von Lühmann, Alexander; Ortega-Martinez, Antonio; Boas, David A.; Yücel, Meryem Ayşe (2020): Table_1_Using the General Linear Model to Improve Performance in fNIRS Single Trial Analysis and Classification: A Perspective.XLSX. Frontiers. Dataset. https://doi.org/10.3389/fnhum.2020.00030.s001
- Hocke, L.M.; Oni, I.K.; Duszynski, C.C.; Corrigan, A.V.; Frederick, B.D.; Dunn, J.F. Automated Processing of fNIRS Data—A Visual Guide to the Pitfalls and Consequences.Algorithms 2018, 11, 67.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This is an interesting paper with an objective to find a classifier based on RDA for encoding cerebral hemodynamic and velocity of stimulation. The NIRScout is used as a tool for measuring the respected region of interests that are the primary motor cortex (M1) and the primary somatosensory cortex (S1).
In general, the manuscript is well written however there are potentials for improvements.
Page 2, line 52, "in sensorimotor rehabilitation.......", the sentence is difficult to understand.
page 3. line 115: "The source wavelengths were 760nm and 850nm for HbR and Hbo, respectively". This is a wrong statement. The two selected wavelengths are applied to calculate HbR and Hbo from absorption spectra.
page 3. line 126: what does the calculation of optical density is applied for?
page 4, line 127: the input dataset to our algorithm were the HbO and HbR concentration....". The only way that these concentrations are calculated correctly is to measure the Hematocrite levels of the subjects. Otherwise, it is only an estimation for these concentrations.
Further, I have the following questions and comments with regards to the design of the experiment:
1) Previous research has reported a positive linear relationship the applied force and the activity in the contralateral primary motor cortex (M1) and primary somatosensory cortex (S1). How can the authors distinguish the velocity and force in this experiment? In particular, when the subjects are instructed to respond to an active task to repeat a pattern.
2) Did the authors consider the latency response time when doing the active part of stimuli with the motoric cortex in their velocity response? wouldn't that affect the analysis with the two-class dataset approach?
3) Did the authors used a standard montage (layout arrangement based on 10-10 or 10-20 system) or it is their own modified version? In the case of the latter, why did the authors modify the layout?
4)It is mentioned that the Polhemus Patriot digitizer is used, but how the data is mapped further is not elaborated.
5) The authors mention the sound from the TAC-machine which is shown in figure 1 and is located in a separate room. Did the authors measure the background sound levels? I cannot see any isolation in the hole where the tubes are passed through.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Round 2
Reviewer 1 Report
Comments are in the pdf below.
Comments for author File: Comments.pdf
Author Response
Please see the attachment. Thanks.
Author Response File: Author Response.docx
Reviewer 2 Report
No further comments. The authors have addressed all issues and questions accordingly.
Author Response
We thank the reviewer for their helpful comments in improving the paper.
