Recognition of the Mental Workloads of Pilots in the Cockpit Using EEG Signals^†

Round 1

Reviewer 1 Report

Authors presented a study in which applied a transfer learning models for workload recognition by EEG during a simulated flight task. Although the work is interesting, it seems to be many flaws to be addressed before the publication. Here below a list of the comments:

Line 82: The application fields fall into five...It is not clear why authors selected just 5 of the possible fields of applications. In particular, all passive BCI applications could employ deep learning, so, not only to measure workload, but even other mental states, such as attention, stress, fatigue, and so on. Please adjust this sentence. 

 

There are other studies in literature exploring this. E.g. https://ieeexplore.ieee.org/abstract/document/9630613, or  https://www.frontiersin.org/articles/10.3389/fnins.2016.00530/full, or https://www.mdpi.com/1424-8220/21/7/2369, or https://dr.ntu.edu.sg/bitstream/10356/145972/2/EEG-based%20Cross-subject%20Mental%20Fatigue%20Recognition.pdf. Please, refer to these works and enlarge this state of art part.

 

Figure 1: In the paper it is not mentioned any ethical committee approval, or any informed consent signed by the subjects. 

 

Line 221: If authors classify between rest and workload, conditions, the model would be trained to recognize every difference, including e.g. the number of movements. It could generate a big bias, since authors cannot say that a difference is associated just to a change in workload, but also to differences in movements for example.

 

Line 229: it is not clear why the authors used 8Hz of sampling rate, also because to analyzed frequencies in the theta band (e.g. 4-8Hz), they need at least the double of srate (i.e. 16 Hz)

 

Line 234: IQR filer to not assure that all the blinks were recognized and removed. It is needed a regression method (e.g. gratton and coles, or MWF, or blinker, to recognize a blink occurrence). Please address this bias. (https://ieeexplore.ieee.org/document/8166859, https://pubmed.ncbi.nlm.nih.gov/6187540/) 

 

Line 242: "No evidence about what are the most discriminating sensors". It is not true! The most of the literature demonstrated that the most discriminating sensors and frequency bands for workload are frontal theta and parietal alpha.  (e.g. https://www.sciencedirect.com/science/article/abs/pii/S0149763412001704)

 

Line 251: Authors seem to not consider the stationarity of EEG signal, that is at maximum of 4 seconds. So, 40 seconds should be not allowed to keep stable features. (https://pubmed.ncbi.nlm.nih.gov/5776646/) 

 

Line 255: Even in testing dataset authors used mean and STD? It could mean that the method is not applicable for online studies.

 

Line 312: Typo

 

Line 341: No statistics has been performed, making the results inconclusive. 

 

Line 344: Hyper link Error in the text

 

Line 388. There are different methods that can be used to delete artifacts contribution. E.g. https://www.sciencedirect.com/science/article/abs/pii/S0165027003003479?via%3Dihub

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

General remarks

In the manuscript, the problem of characterization of the mental workload of flying pilots in the cockpit based on EEG signal analysis is discussed. To solve this problem, a workload recognition task is formulated. The basic tool used in the study is convolutional neural networks.

In general, the work is well written and structured; however, there are some issues to be addressed.

1. In the experimental section, two variants of algorithms based on convolutional neural networks are used. The authors should add other methods for comparison purposes.
2. In the Contribution and Conclusions section, the authors mentioned that the presented solution is based on the fusion of EEG sensor signals. However, there is no formulation of the fusion signal problem.

Detailed remarks

1. Page 8, line 344. There are no proper references to the tables.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Authors addressed properly the comments, so I can endorse the publication.

Reviewer 2 Report

In the revised version of the manuscript, all the issues of the reviewer are addressed. In my opinion, the manuscript can be published after professional English correction.