[page 6 & 7 of handout]
Savoy, R. L. (2001). History and future directions of human brain mapping and functional neuroimaging. Acta Psychologica, 107(1-3), 9-42. (Another “wide-ranging reference”.)

It has long been known that there is some degree of localisation of function in the human brain, as indicated by the effects of traumatic head injury. Work in the middle of the 20th century, notably the direct cortical stimulation of patients during neurosurgery, suggested that the degree and specificity of such localisation of function were far greater than had earlier been imagined. One problem with the data based on lesions and direct stimulation was that the work depended on the study of what were, by definition, damaged brains. During the second half of the 20th century, a collection of relatively non-invasive tools for assessing and localising human brain function in healthy volunteers has led to an explosion of research in what is often termed “Brain Mapping”. The present article reviews some of the history associated with these tools, but emphasises the current state of development with speculation about the future. (C) 2001 Elsevier Science B.V. All rights reserved.

Savoy, 2001 page 28
There is no shortage of things to be worried about in the domain of functional brain mapping. The theme of this section will be a collection of related concerns that all stem from the limitations of reporting data as a collection of activated voxels. The variations on this theme concern "thresholds", the consequences of increasing statistical power of the tools, and the interpretation of the "most active" voxels across a small set of stimulus classes.

Consider, first, the question of thresholds. In a typical block design fMRI or PET experiment the data collected during one type of block are compared to the data collected during another type of block. A statistical test is applied at each voxel in space to decide when the difference between the distributions collected during the two types of blocks are statistically significant. A "map" is presented, typically showing anatomy in the background, with a coloured overlay indicating those voxels for which the statistic exceeds some threshold. How is that threshold determined?

Savoy, 2001 top of page 30

This problem is likened to the classic question raised by Paul Meehl in the context of general experimental questions in psychology. Meehl's observation was simply that if psychological questions take the form "Will Group A differ from Group B by scoring statistically differently (either higher or lower) on some behavioural measure X"?, then as we increase the power of our statistical tests, the answer will almost certainly be Yes, for any A, B, and X! That is, while correlation between two presumptively equivalent groups on an unrelated task (e.g., Group A ¯ persons with red hair; Group B ¯ persons with brown hair; and the task X is IQ score) is likely to be insignificant when the number of subjects is modest (say, N=20 in each group), the story changes when N=55,000. The reason is simply that even hair colour has some association with ethnicity, which might be associated with religious orientation, which might be associated with emphasis on education, etc. The associations are weak enough that they do not yield a significant relationship between the groups and the task when N is small, but they do when N is large. As a dramatic demonstration of this point, Meehl cites a study of 55,000 Minnesota high school seniors, for whom statistically significant correlations were found in 91% of pairwise comparisons among miscellaneous measures such as sex, college choice, club membership, mother's education, dancing, interest in woodworking, birth order, religious preference, number of siblings, etc. These were not "spurious" correlations, but real associations that were detected because of the form of the question and the power of the test made possible by a very large N.

Savoy 2001, bottom of page 30

  • It is believed that there is a highly analogous problem lurking in the context of functional neuroimaging.

  • The analogous concern applies to questions of the form, “Will area A in the cortex show a change in activity (an increase or decrease) in response to task X, compared to its response to task Y?”;

  • If our imaging system is powerful enough;

  • if we design more effective receiver coils for the MR scanner;

  • if we use more subjects;

  • if we collect data for a longer period of time),

  • then the answer will almost always be Yes, for any A, X, and Y.

  • Once we get past the peripheral sensory systems (i.e., once we have reached the thalamus), there are only about 5 synapses between any two neurones in the brain.

  • It is likely that the activity in any one neurone (or collection of neurones, given the spatial resolution of our non-invasive imaging techniques) is going to influence almost any other neurone, albeit weakly.

  • Data collected across multiple subjects on the same task give a hint of such a conclusion, in that the apparent area of activation increases as the number of subjects increases, as long as the threshold for statistical significance is held constant.

Savoy, 2001, page 10
It is important to note that the enterprise of brain mapping did not begin with fMRI or any other non-invasive imaging tool. The understanding that localisation of function is pervasive in the human brain has been well established for more than 50 years. Consider, for example, the summary of this knowledge represented by Fig. 1, reproduced from a book published in 1957 (Polyak, 1957). There are at least two kinds of questions that should be asked about this figure. The first questions are methodological: What is the basis for this figure? Where did the data come from? What were the technologies that gave rise to this data? The answer to these first questions is that the figure is based upon two techniques: study of people with lesions (caused, for example, by stroke, disease or traumatic wounds) and the direct electrical stimulation of the cortex of patients undergoing brain surgery. More about these techniques will be written below.

Fig. 1. This figure schematically summarises the state of knowledge of localisation of human functional brain in 1957. It is based on data from lesions and studies using direct cortical stimulation during neurosurgery. (Reproduced with permission from the publisher from Fig. #275, p. 456 of "The Vertebrate Visual System", by Stephen Polyak).

Thus, Fig. 1 teaches us that we were far from ignorant or misguided about localisation of brain function in 1957. So, what is all the current excitement about? The primary answer1 is that today there are a host of technologies that can be used to give us information non-invasively that address the same issue. The study of patients with lesions, or those who are undergoing direct cortical stimulation during surgery, has substantial limitations. For ethical reasons, neither lesions (obviously) nor direct electrical stimulation of the brain via surgery (for reasons of general risk associated with exposing the brain) may be used in the study of healthy human subjects

Meehl, P.E., 1967. Theory-testing in psychology and physics: a methodological paradox. Philosophy of Science 34 2, pp. 103¯115.

Polyak, S., 1957. The vetebrate visual system, The University of Chicago Press, Chicago.