We presented a tutorial to the Society for fNIRS educational webinar series on April 12, 2021 on using Homer and AtlasViewer to analyze high density fNIRS measurements of brain activity evoked by a finger tapping task. A video of that webinar can be viewed here. Here, we provide some additional documentation so that others can reproduce the results presented in webinar.
To start, make sure you have installed the latest version of Homer and AtlasViewer by going to https://openfnirs.org/software/homer/ and following the links to github to install the latest releases. The latest release of Homer at the time of the webinar (https://github.com/BUNPC/Homer3/releases/tag/v1.31.2) should work. But it is possible that the latest release of AtlasViewer (https://github.com/BUNPC/AtlasViewer/releases/tag/v2.11.3) had a bug that we fixed, in which case you should use the next release when it becomes available or until then use the developers branch (https://github.com/BUNPC/AtlasViewer/tree/development).
The data we use for this demonstration was acquired with a NIRx NIRSport2. We designed a high-density hexagonal arrangement of 7 sources and 16 detectors as shown in the figure above. This provided for overlapping measurements with a short separation of 18 mm and a long separation measurement of 30 mm. Data was collected for a run of 5 minutes. The subject tapped their fingers for 10 seconds, repeating every 30 seconds such that 9 trials of finger tapping happened during the 5-minute run. The run was repeated 3 times. The data and the probe geometry file can be downloaded from here.
The time-series data were processed with Homer3 using the processing stream shown in the figure. Short separation regression was performed in the GLM estimate of the hemodynamic response function (HRF) by setting the threshold for short separation measurements (rhoSD_ssThresh) to 20 mm and using flagNuisanceRMethod = 2 which takes the average of all short separation measurements to obtain the regression vector. We encourage the user to. Compare these results with that obtained with no short separation regression obtained by setting rhoSD_ssThresh = 0.
The group results with and without the short separation regression are shown in the figure below.
We then open AtlasViewer and produce the measurement sensitivity matrix using the MCXlab plugin for AtlasViewer. We then perform the image reconstruction on the brain and compare it with the tomographic image reconstruction simultaneously on the brain and scalp. The image results are shown below for the reconstructed HbO images. Note that in latter case we only show the image reconstructed on the brain, not what is reconstructed on the scalp. We see that the brain and scalp reconstruction produces an image of brain activation that is far more localized than the brain only reconstruction. This happens because the short separation regression performed in Homer3 removes a great deal of the interfering scalp signal but does not remove all of the scalp interference. When the image reconstruction allows for absorption changes to be reconstructed in the scalp as well as the brain, this residual scalp interference can then be reconstructed in the scalp further reducing it from interfering with the estimate of the brain activity. When we just do a brain image reconstruction, that residual scalp interference is forced to reconstruct as changes in brain activity. These results very nicely demonstrate the benefit of doing high density tomographic image reconstruction in addition to short separation regression as was very nicely pointed out in 2010 by [Gregg2010].
Gregg NM, White BR, Zeff BW, Berger AJ, Culver JP. Brain specificity of diffuse optical imaging: improvements from superficial signal regression and tomography. Frontiers in neuroenergetics. 2010;2. PMCID: PMC2914577