Meta

Perusing reference 1, about extracting macro-scale geometry of the brain from micro-scale scanner data, I was pointed to reference 2, about extracting geometry of the brain from meta-analysis of papers in journals, now around 15,000 of them – orders of magnitude more than could be consumed by hand by even a large research team. This geometry, together with still more meta-data, is to be found at reference 3.

You can supply a term, perhaps ‘fear’ or ‘anxiety’, to the database which underpins reference 3 and it will draw you a map of the bits of the brain which appear to be involved in that particular term; that is to say the bits of the brain which fear or anxiety seem to particularly activate or deactivate. I think you can do it the other way around: give the database a position and it will tell you what terms are reported to have hit that position. All this by getting a computer to read all the papers, to count up usage of terms and to extract positions from tables – that is to say likely looking numbers from tables.

It does not say, but maybe the database also provides an index into the papers. You supply some terms or some positions and it tells you what papers there are out there, papers you might otherwise have missed. Maybe more reliable than Bing or Google.

All of which involves a lot of computational and statistical trickery. But the results seem to look good, a lot better than you might expect. The maps you get for a term agree pretty well with those you might construct, rather more laboriously, by hand.

As far as I can tell, terms are extracted from papers and positions are extracted from papers quite independently. Within paper, there is no attempt to link particular terms with particular positions, with particular tables. This sort of thing comes out in the statistical wash, a bit later on.

Other limitations

17 journals were included in this work. But not, Science, Nature or Nature Neuroscience, apparently as these did not focus particularly on functional neuroimaging studies of the brain. Presumably the authors did not think it worth their while to load it all in.

I did not notice anything about paper quality, perhaps along the lines of Google saying that a webpage is important if lots of already important webpages link to it. Again, presumably not thought worthwhile.

Although I could turn up a list of 1,334 terms at reference 3, a list which starts ‘abilities, ability, abstract, abuse, acc [anterior cingulate cortex], accumbens, accuracy, accurate, accurately, acoustic’, I did not turn up anything about how those terms came to be selected or about the sort of matching used.

Positions are logged without regard to the frame of reference being used in the paper in question. Presumably these frames do not differ enough to be a problem, at least in this context.

I did not turn up anything about the particular set of 200,000 or so voxels used.

I did not notice anything about mapping voxels onto any of the popular parcellations of the brain, the sort of parcellations which get down from 200,000 voxels to 200 regions or areas. But it would be easy enough to add in, if it were absent. I associate to the terms ‘Brodmann’ (a famous parcellation) and Broca (a famous area).

Other matters

Reference 1 includes the intriguing idea that the connectivity of two places in the brain is related in a systematic way to the shortest distance between them, as measured by the length of the geodesic running in the grey matter of the cortical sheet, that is to say respecting the myriad folds. This despite the variety and complexity of the connections available to two such places.

Conclusions

It was surprising to me that such a technique works as well as it appears to. Meta-analysis coming of age!

PS: maybe it is all down to the law of large numbers: a big enough sample will smooth out all kinds of noise.

References

Reference 1: Situating the default-mode network along a principal gradient of macroscale cortical organization - Margulies, D.S., Ghosh, S.S., Goulas, A., Falkiewicz, M., Huntenburg, J.M., Langs, G., Bezgin, G., Eickhoff, S.B., Castellanos, F.X., Petrides, M., et al – 2016.

Reference 2: Large-scale automated synthesis of human functional neuroimaging data - Tal Yarkoni, Russell A Poldrack, Thomas E Nichols, David C Van Essen & Tor D Wager – 2011. 

Reference 3: https://www.neurosynth.org/.