Hudspeth’s Method of Coherence Analysis

by admin | August 25th, 2005

In Hudspeth’s method, coherence values derived from 19-electrode QEEG data collection are used to generate a graphical display which consists of a three-dimensional depiction of a cube-like space in which little circles float, something like fish in a fish tank. Upon closer look one sees that there are 19 such circles, each labeled as one of the 19 QEEG electrode sites. In general, the circles tend to cluster fairly close to one another; but at times a particular circle stands out because it floats at a distance from the others. As Hudspeth explained, when a circle stands out from the others, it identifies an electrode site that has a problematic pattern of coherence relationship with the other sites.

This display is far from being the usual kind of QEEG coherence display which highlights paired electrode sites when they have unusually large or small coherence values. Hudspeth’s display highlights individual sites only, doing so on the basis of how well the pattern of obtained coherence values for a site agrees with the global patterning of coherence values of all of the sites taken together. The method identifies a site as problematic when its pattern of values does not fit well with the overall pattern. It identifies problematic locations, not problematic coherence values.

Before describing how Hudspeth generates his coherence display, some key concepts will be developed by discussing intelligence testing work which uses a method that is similar to Hudspeth’s but is simpler, easier to understand and probably somewhat familiar already. The method in question is factor analysis.

In this area of work, it has long been known that how a person scores on different measures of intelligence largely depends on the strengths of his or her performance in two major areas: Verbal ability and Spatial ability. A simple, concrete example of how this has been established would be as follows. Suppose that a group of people is given a battery of four tests: 1.) Vocabulary; 2.) Verbal Analogies; 3.) Embedded Figures; 4.) Object Rotation (1. is a test of word knowledge; 2, of verbal thinking; 3, of facility in visually analyzing a complex design; 4, of facility in internally manipulating a mental image). In a situation such as this (i.e. 1 and 2 particularly involve Verbal ability while 3 and 4 particularly involve Spatial), the inter-correlations of the group’s results on each of the tests would typically show the following pattern. All of the tests would have positive correlations with one another; but at the same time the correlation between the Verbal tests would be stronger than the correlations of those tests with the Spatial tests and similarly for the Spatial tests, the correlation would be stronger with each other than their correlations with the Verbal tests.

Thus, in addition to the overall pattern of positive inter-correlations, two sub-tendencies exist. It is not easy to detect such sub-tendencies in much larger datasets without the help of a tool such as factor analysis. If, say, there were 30 tests, factor analysis would involve putting the inter-correlations among them into a 30 x 30 table and then performing complex computations on this matrix to “extract” a mathematically defined “factor” for each sub-tendency that was present. For present purposes it is not necessary to know anything specific about how that definition is mathematically expressed except that the computations also generates a percentage number that measures the strength of the factor. That number measures how much of the total amount of correlation information in the matrix has been captured by the extracted factor. In technical terms, the factor is said to account for a percentage of the total variance in the matrix. If the total number of extracted factors is relatively small compared to the size of the correlation matrix and if the sum of the percentages of all of the factors is high, say 80% or 90%, this would mean that only 10% or 20% of the matrix information has been left out of account and that the factor analysis had thus achieved an efficient grasp of the major underlying factors in the starting data.

At this point in the process, all operations have been mathematical. The factors have been identified mathematically but have not been given verbally expressed meaning. To accomplish that, the “loadings” for each of the tests in the dataset on each of the factors are computed. These numbers measure how strongly each test is involved with each of the factors. In the present example, the Vocabulary test would thus have a high loading on the Verbal factor and a low loading on the Spatial factor. By studying the nature of the tests that load highly on a given factor, one is finally able to verbally define what the factor is about and thus apply labels such as “Verbal” or “Spatial” to them.

That a dominant Verbal-Spatial pattern is typically found in factor analyses of intelligence tests should be no surprise given that there appears to be no larger form of differentiation of function in the human brain than that between the left and right hemispheres. Verbal functioning is well known to be primarily left hemisphere and spatial functioning, right. These two major factors do not come out of the blue; instead they exist in the test data because of a fundamental way that the brains of the people who generated the test data are organized. This important point deserves re-statement: Whenever it is possible to capture with relatively few factors most of the information in a matrix of measures of co-relationship, it is reasonable to conclude that one has captured in simpler mathematical form something about the fundamental organization of the actual data source.

In essentially the same way that factor analysis has just been described, Hudspeth uses a somewhat different matrix algebra technique called principal component analysis to extract not “factors, but rather “principal components from a 19 x 19 matrix of QEEG coherence values. Note that both correlation coefficients and coherence values are measures of co-relationship. In practice, he finds that extraction of just four principal components usually yields results that closely approach the ideal previously described of accounting for most of the information in the matrix. This means that he thus gets a mathematical handle on the major functional patterns of connectivity in the person’s brain. These four components measure the dominant, major patterns in the ways that the 19 electrode sites simultaneously relate to one another.

His method then uses the “loadings” (described earlier) that are computed for each electrode site on each of the four components to plot displays. Because we live in a three- dimensional world, one display is devoted to one of the sets of component loadings while loadings on the other three components are used for the three dimensional display described earlier. This main display is plotted in much the same way that Compressed Spectral Arrays are generated using measures of time, frequency and amplitude along the three axes of the display. Here the axes of the cubical display are based on the sets of loadings for three of the components.

The distribution of electrode sites in the main display does not depict the strength of their coherence relationships with one another. When sites cluster closely together in the display this does not mean that the sites have high coherence values with one another. Indeed, the coherence values that each site might have with the other sites in a cluster might variously be small, large or medium. Instead the clustering exists because the sites that belong to it are those that work well together to produce the global patterning of connectivity that the principal component analysis has grasped. Similarly, an electrode site that stands apart from the clustering is one that shows a dysfunctional relationship with the overall organization of connectivity.

This organization exists in any person except when extreme pathology is present. In most cases, the problems addressed by any form of therapy exist within the otherwise, more-or-less-healthy organized functioning of the person. Moreover, part of that healthy functioning consists of those self-healing processes without which the therapy is impotent. The clustering of electrode sites in Hudspeth’s display represents the integrated organization of connectivity in the individual’s brain. Without the context of that organization, it would not be possible to identify sites that are not well related to it. In an analogous way, the factor analytic identification of specific types of intelligence depends upon the existence of an overall pattern of positive inter-correlation among all types of tests.

In the conventional QEEG analysis of coherence values, the detection of a problematic paired-site value says little about its source location. Is it at one of the electrodes sites, or the other, or both? Moreover, the actual value obtained is not simply a reflection of the functional relationship of only the two sites in question. Rather, the way each site functions individually is influenced by its relationship with many other sites and these multiple influences will affect in unknown ways the strength of any of the single, paired-site coherences that are measured. The conventional approach looks at paired sites in a piecemeal fashion and ignores much of the information that exists about those multiple influences in the full set of coherence values. While eyeballing the separate problematic coherences in a conventional Q display will sometimes indicate that a particular single site needs attention, Hudspeth’s results often differ from such conclusions, sometimes even identifying sites that the usual approach finds no problem with at all.

Both the uniqueness and strength of Hudspeth’s approach is that it utilizes in a single analysis much of the information that is present in the 171 coherence values (19 times 18, divided by 2) that QEEG data collection generates. This gives the analysis great power as shown in the Jon Walker’s work with learning disabilities. This is an area in which neurofeedback has not been very successful in the past. When working with a particular disability, Walker first makes a list of the brain locations that might be involved in the person’s problem, basing this list on established neurological knowledge of the brain’s modular functioning. Then, if any one of the locations on the client’s list is also a Hudspeth “stand out”, he trains there and finds that he often gets surprisingly swift, good results. Hudspeth also reported much the same with other case material. This radically new method appears to have enabled a new level of clinical effectiveness. Surely it deserves to be more widely known.

NOTE: In his workshop at the upcoming iSNR conference, Hudspeth will present new slides said that will help understanding of his technique.

Ruminations on the Highly Sensitive Person

A Fish Story

Our Recent Publications

Leave a Reply