Wednesday, 17 June 2015

A critique of functional brain imaging (1995)

*
This is extracted from an un-published essay I wrote in 1995 named A manifesto for research in cognitive neuropsychiatry, with additional commentary from 1996.

I have corrected a few phrases and added a few sub-heading, but made no substantive changes.

19 years later, I can only observe "I was right".

*

A Critique of Functional Brain Imaging, 1995
by Bruce G Charlton
School of Psychology, Newcastle University, England 

Structure-Function correlations
One of the most fruitful constraints in biology has been the assumption of correlations between structure and function. This assumption is valid only when structure is examined at the appropriate level; and may be obscured by inter-subject differences in developmental history, age, coincident disease etc. Furthermore, the necessary scale of analysis to reveal an association is not always obvious.

Nonetheless, a useful degree of functional localization has been described for many neuroanatomical scales in the brain ranging from the basic divisions between primitive fore-, mid-, and hind-brain to the tendency of individual cells to respond to specific classes of visual information (Zeki, 1993).

The existence of structure-function correlations is not surprising given an understanding of natural selection. Newly-evolved functions usually require appropriately-modified structures: so that a cognitive adaptation usually needs a specifically constructed neural circuitry to perform novel computations.

And because natural selection must produce a reproductive advantage for each evolutionary change (Dawkins, 1987) , this limits evolution to an incremental, stepwise modification of the parameters of existing neural circuitry. Redesigning the brain from scratch is not an option, and ‘rewiring’ of the brain is therefore piecemeal and additive, rather than radical and substitutive.  

Hence the neocortex has evolved many additional stages of complexity to its hierarchy of processing throughout evolutionary history. New structural/ functional specializations have typically been accommodated by lateral expansion of the surface of the cortex (Barton & Dunbar, 1997). Each distinct intermediate cognitive process tends to be functionally localized to a cortical area; the integrity of which area is necessary for performance of a particular class of inferential reasoning tasks; and disruption to which area is sufficient to impair performance of this class of tasks.  

The proper purpose of functional brain imaging
The purpose of imaging in cognitive research is therefore to fractionate function by displaying the location and sequence of activation in he brain structures which are necessary for performance of specific stages in cognitive processing (Jacobs & Carr, 1995). A method is therefore required that discriminates neural activity at an adequately precise level of both spatial and temporal resolution (Kosslyn, 1994).

Such an imaging technology could potentially demonstrate both the modularity of intermediate level processes, and the number of steps of cognitive processing within each of these pathways.

Functional imaging might then constrain theories of cognitive functioning, yielding a template of the number of modules, the number of discrete processing steps within modules and the nature of the temporal inter-relationship between these processes.

Brain imaging would, in a sense, provide a flow diagram comprising ‘boxes’ with ‘arrows’ between them: it would remain for cognitive psychology to fill the boxes by defining the nature of processing at each locus.

Inadequate spatial and temporal resolution of functional brain imaging
It is uncertain exactly what level of spatial resolution is required of an imaging technique to accomplish such fractionation, because the necessary size of neural structures needed for intermediate processing is not yet properly established (probably, the more complex the computational task, the larger the necessary size of neural circuitry).

However, many specialized cortical areas are some millimetres in diameter, so a resolution of at least tenths of a millimetre would seem to be necessary.

And temporal recording of neural activity would need millisecond resolution in order to discriminate the steps of neural activation involved in cognitive processing.

Unfortunately, none of the currently available techniques for functional imaging of the brain are able to provide the minimum necessary structural and functional discrimination of neural activation.

Magneto-encephalography (MEG) can so far attain this temporal resolution only for restricted cortical regions, and in situations where (probably) thousands of orientated neurons are operating synchronously (Naatanen et al, 1994).

Positron Emission Tomography (PET) has low spatial resolution and very low temporal resolution, and visualizes blood flow changes – which are a crude, and probably unreliable, proxy measure of neural activation (Posner & Raichle, 1994).

Functional Magnetic Resonance Imaging (fMRI) probably has adequate spatial resolution, but cannot form images rapidly enough to resolve cognitive processes; and also uses proxy measures of neural activity (Kosslyn, 1994).

PET and fMRI should, therefore, be considered as yielding proxy measures of brain anatomy averaged over time, rather than revealing sequences of neural activity corresponding to cognitive function.

Electroencephalography (EEG), by contrast, can demonstrate brain activation events in real-time; but even in its modern ‘high resolution’ form, the technique provides only a very blurred, two-dimensional structural image with a centimetre level of spatial resolution (Gevins at al, 1995).

Functional Brain Imaging - an enumeration of misleading artefacts
The present practice of disregarding the systematic limitations of existing functional imaging, and using techniques such as PET to explore cognitive functioning, is a dubious practice.

The intrinsic tendency of low-resolution techniques of imaging to mislead is exacerbated by the practice of constructing images using ‘subtraction’ techniques, averaging of repeated trials, and the pooling of patients in (often diagnostic) group studies (Posner & Raichle, 1994). Presumably, investigators believe (but without evidence or testing the assumption) that such procedures simply remove ‘noise’ and thereby enhance precision and sensitivity.

However, given the complexity and dynamic, non-repeating nature of brain states, it is almost certain that investigators are pooling systematically-different instances. A comparison of single case, real time data from MEG with averaged images from groups reveals highly significant averaging effects with time and across groups (personal communication, Andreas A Ioannides, 6. 12. 95).

Despite the proliferation of high status publications in the field of functional brain imaging; it therefore remains entirely possible that the results reported so far represent little more than an enumeration of artefacts.

*

  • Excerpt from note added in 1996:
[Functional brain] imaging has not ‘yet’ contributed anything of significance to the understanding of human psychology – despite the billions of pounds pent on it and the numbers of papers published into journals.

And I mean nothing at all.

Under the hype, the published work is merely fourth-rate, stamp-collecting, pre-science; pumped-up to high prestige by the cost of the technology and the attractiveness of its pictures.

  • Further note added today, 18 years later:
Nothing to add to, or subtract from, the above. Functional brain imaging was a successful scam for extracting funding; a quarter century  of which contributed nothing substantive to science, and grossly misallocated prestige away from real science and into an assortment of deluded and dishonest careerist confidence tricksters.


References

Barton R& Dunbar R. (1997). Evolution of the social brain. In RW Byrne, A Whiten (Editors) Machiavellian Intelligence II. Cambridge University Press: Cambridge, UK.

Dawkins R. (1986) The Blind Watchmaker. Longmans: London.

Gevins A et al. (1995). Mapping brain function with modern high-resolution electroencephalography. Trends in Neurosciences (TINS). 18: 429-436.

Kosslyn SM. (1994). Image and Brain. MIT Press: Cambridge, MA, USA.

Naatanen R et al. (1994). Magnetoencephalography in studies of human cognitive brain function. Trends in Neurosciences (TINS). 17: 389-395.

Posner MI. (1994). Images of Mind. Scientific American Library: New York.

Zeki S. (1993). A Vision of the Brain. Blackwell: Oxford.

3 comments:

Nicholas Fulford said...

I question then is: Are there any technological advancements or methods of application that you see on the horizon that would have the necessary spatial and temporal resolution to allow the type of research into brain studies that would be fruitful?

Also, with respect to temporal limits, are there methods akin to digital oversampling see http://www.analog.com/media/en/training-seminars/tutorials/MT-017.pdf for a discussion of oversampling interpolation with regard to analog streams (in this case sound). Can this be combined with fractal interpolation methods? (Yes, interpolation does not add data samples, but it does allow a smoothing that is consistent with natural forms below a certain frequency.)

No argument with what you are saying here, but surely some creative engineer out there has some ideas about how to get as much as is possible out of what is available? Without knowing what the limiting factors are with such technologies as fMRI - though I would expect magnetic hysteresis is a contributing factor - I would not begin to know where to look. I just cannot help but think that there are improvements which can but have not as yet been realized. I suppose the focus of research must shift to the tools before anything more substantial can be done.

Bruce Charlton said...

@NF - My main idea was to study specifically those brain functions where the problems may not rule-out validity - a possible example I mentioned was 'working memory' - which is supposed to deploy long-acting 'delay neurons' that have a timescale of firing in seconds (rather than in tens of milliseconds). But nobody in functional brain imaging was doing real science - they were pursuing careers, and they were mostly extremely successful at that, via producing hyped-up nonsense.

sykes.1 said...

A very detailed critique from a neurologist is given by Raymond Tallis, "Aping Mankind: Neuromania, Darwinitis, and the Misrepresentation of Humanity," Acumen, 2011. Amazon also has a paper back edition.

In sum, the people doing brain imaging claim too much, and the published images are not quite what is claimed. Most are averages over several images, and individual images don't always closely match one another.