The story of Grey Level Index: forty years of misleading consistency?

6. How justifiable would be using results of 2D GLI profiling in modern multi-modal 3D atlases of human brain?

Despite almost infinite chain of re-publishing of the same method from one paper to another, and persistent claims regarding novelty of GLI approach it is 40 years old technology.

Given the availability of uninterrupted serial sections of the whole brain from several individuals, the only reasonable sampling method is sampling in 3D data space. In this case, the whole volume of the cortex will be covered by correctly oriented 3D pyramid-shaped regions, and valid borders between areas will be obtained along the whole “circumference” of the boundary of any particular area. As a second benefit, in such a case the cellular density estimate will be available in 3D as well, so ambiguous and vague term “grey level index” will become obsolete. Instead, the estimated profile will be called clear and simple “volume fraction profile”, as it should be all this time from the very beginning.

If 3D sampling would be impossible due to lack of correctly registered sections, then, in the absence of better choice, 2D GLI profile could be used. However, regardless of the specific way of how position of the border between two areas is obtained, two features should be included in addition to its coordinate: “width” and “origin”.

First, we cannot assume that borders between areas are “razor-sharp”. All published data regarding differences of the “block size” systematically indicate that position of the boundary is “distributed” across rather wide region of the cortical section. Block width variability in different cortical areas evidently has biological significance and should not be ignored. Accordingly, the method should produce the boundary with clear indication of the “width” of the border, expressed in some agreed upon unit measure.

Second, the “origin” of the boundary has to be indicated as well. The border of each cortical area should have clear and distinctly visible indication of how a particular “section” of the border is obtained: by statistical testing or by pure guessing (or “visual interpolation”).

7. Conclusion: Grey Level Index profiling is 40-years old and questionable methodology.

Paraphrasing Lady Bracknell from famous play by Oscar Wilde one could say: “Publishing quite similar papers once looks like misfortune. Publishing quite similar papers many times seems like carelessness”. However, let’s set aside the issue of novelty and related ethically problematic re-publishing essentially the same technology as “new” numerous times.

Let’s consider much more important problem. If computerized, observer-independent method of parcellation of the human brain cortex was available for the last 20 years to say the least, why “new”, ultra-high-resolution cortical cytoarchitectonic map of human brain still does not exists or, in a best-case scenario, is still incomplete? Why downloadable “probability map” of human neocortex posted by INM-1 is still a series of quite low-resolution images (152 by 154 by 188 pixels) with numerous blank areas, especially in frontal and temporal lobes?

Maybe GLI-based parcellation is not as “observer-independent” and automated, as authors would like to present? Maybe in reality the cytoarchitectonic mapping is still a manual process, heavily dependent upon visual analysis of infinite sequence of serial sections, and tedious manual point-by-point tracing of the section image? Admittedly, nowadays this tracing is done on the computer screen instead of a sheet of paper, but does it essentially make a difference?

So, it would not be unreasonable to conclude that modern approaches to cortical areas mapping require much more sophisticated and more robust methods. To say the least, traditional GLI approach has to be extended in 3D and become truly stereological and observer-independent.

Copyright © 2017-2019, Brain Imaging Informatics, LLC

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.