ANALYSIS AND COMMENTARY Limitations of Brain Imaging in Forensic Psychiatry Donald Reeves, MD, Mark J. Mills, JD, MD, Stephen B. Billick, MD, and Jonathan D. Brodie, PhD, MD J Am Acad Psychiatry Law 31:89–96, 2003 Over the several decades of its existence, brain imag- the complexity of the technology and must make ing has moved from the laboratory to the bedside. statements cautiously to avoid saying more than the Brain imaging now plays a routine role in the diag- science warrants. nosis of many central nervous system disorders. Given the steps in brain imaging—from the gen- However, the appeal of a “snapshot” of the brain eration of an image to the conclusion drawn by the extends beyond medicine. Brain imaging, now avail- expert—the psychiatrist must consider several ques- able in color, with its simplicity and vividness, has tions. What is to be measured? Is the technique sen- sometimes proven irresistible to defense attorneys sitive, accurate, precise, and reproducible? What do seeking to exonerate their clients of responsibility for the measurements mean? Each of these questions is their crimes. Putting aside the question of what brain discussed in this article. imaging means for a justice system predicated on the A brain image is the vivid representation of anat- assumption of free will, misrepresentation of brain omy or physiology through a pictorial or graphic imaging may mislead a judge and jury. display of data. The data are some property in or of the brain (e.g., attenuation of x-rays, magnetic mo- The Technology of the Machines ments or dipoles, electrical signals, radioactive events) that the imaging technique detects, often Brain imaging, with its many technological vari- without actually invading the brain. The image may ables and requirement for clinical inference, has not be structural, chemical, electrical, psychological, or advanced to the point that it can be introduced in physiological. Techniques include computed tomog- court without real and significant caveats. These raphy (CT), magnetic resonance imaging (MRI; in- same variables and inferences lend themselves to po- cluding functional MRI (fMRI) techniques such as tential manipulation and distortion within the adver- blood oxygenation level-dependent (BOLD) fMRI, sarial system of the court. Therefore, in using brain and diffusion tensor imaging (DTI)), magnetic reso- imaging, the forensic psychiatrist needs education in nance spectroscopy (MRS), quantitative electroen- cephalography (qEEG), positron emission tomogra- Dr. Reeves is Assistant Professor of Psychiatry, University of Medicine and Dentistry of New Jersey, Newark, NJ. Dr. Mills is a principal, phy (PET), single-photon emission computed Forensic Sciences Medical Group, Washington, DC. Dr. Billick is tomography (SPECT), and magnetoencephalogra- Clinical Professor of Psychiatry, New York Medical College, New York, NY. Dr. Brodie is the Marvin Stern Professor of Psychiatry, New phy (MEG). York University School of Medicine, New York, NY. This paper was Each imaging technique produces a detectable sig- presented in part by Dr. Brodie at the scientific session of the 25th annual meeting of the Tri-State Chapter of the AAPL, New York, NY, nal with a characteristic sensitivity, precision, accu- January 22, 2000. Address correspondence to: Donald Reeves, MD, racy, and fidelity for the physiological process being Department of Psychiatry, University of Medicine and Dentistry of New Jersey, 185 South Orange Avenue, Newark, NJ 07103. E-mail: measured. To interpret these signals, the clinician [email protected] must apply a model. That is, the data must be recon- Volume 31, Number 1, 2003 89 Brain Imaging in Forensic Psychiatry structed by using a set of mathematical assumptions comprise the norm? Are these people screened for use that are framed by an experimental context or design. of psychoactive drugs? The norms may even be pro- Such a reconstruction involves statistical analysis and prietary. Furthermore, to achieve a norm, a statistical comparison, from which experimental conclusions manipulation must be made to compensate for dif- or clinical inferences may be drawn. Although the ferences in brain size. In other words, each head is forensic psychiatrist need not possess the under- mathematically “squeezed” to look the same size. By standing of a physicist to use brain imaging, the psy- de-emphasizing natural variability in favor of a single chiatrist should not ignore the technology. The gen- average image, this procedure may give the false im- eration of an image involves many assumptions, pression that an individual is abnormal when actually corrections, and compromises and various levels of the person is merely not average. analysis (see “Statistical Parametric Mapping” Web 1 The definition of normal may be ambiguous in site). These steps are not standardized from one brain imaging. Normal may mean the rigorous ex- technology to the next or from one machine or lab- clusion of disease or it may mean the selection of a oratory to the next. Thus, the image the psychiatrist sample that has been matched for several parameters reads varies, depending on the signal threshold, that may (or may not) be relevant but that are not the color, contrast, or ordinates the technician chooses or variable under study, such as age, gender, or educa- even the brand of machine available in a particular tion. Even if a finding is abnormal, it may not be laboratory. Furthermore, the conditions under dysfunctional. A simple analogy is height. One man’s which the scan was obtained may be idiosyncratic height is six feet, six inches and another’s is five feet, and impossible to compare in a meaningful way with two inches. Both men are abnormal in that they have data obtained at other centers. Not only are the steps not standardized, they are heights far removed from the average. However, it easily manipulated by a person with knowledge of would be ridiculous to say that their abnormal the technology. Color coding, for example, can be heights necessarily make them dysfunctional. In ad- arbitrary and may present the illusion of huge differ- dition, even assuming a simple Gaussian distribution ences in some aspect of brain activity, when little of any variable (e.g., laboratory value), five percent of actually exists. The signal-to-noise ratio (the thresh- the normal population may have an abnormal value old for a signal) may be changed, with the inevitable and yet may be defined as normal. tradeoff in the type of information obtained. The The definition of normal is still more slippery in analogy is that of mountaintops to valleys. If a large forensic evaluations in which the brain activity being scale is used, the result shows only mountaintops. If evaluated is purported to relate to function or behav- the scale is decreased, then the buildings, or even the ior. In this situation, the psychiatrist should be aware people in the valleys between the mountaintops, may of the period for which the definition of normal become visible. The problem is that as more detail is properly holds. Because the brain exhibits elastic- visualized, the data become more confounded.2 ity—a dynamic state of continuous adjustment to external and internal stimuli—an image taken at one What Is Normal? time, may not resemble the image taken in the same The variables do not end with the machine itself. subject at a different time. Yet both images could Statistical maps of brain activity are a common and reflect normal function. popular way to illustrate how an individual brain Furthermore, because the brain is a nearly closed compares with an average brain. A statistical map is a system at equilibrium, considerable normal variation probability function compared with a defined in components can be expected. What may be more norm.3–7 In brain imaging, the norm is obtained by characteristic of brain function, and thus more im- pooling and averaging the brain images of normal portant for evaluation of function, are patterns of people. The question, then, is how useful are the component relationships. For example, in verbal flu- norms? The manner in which the norm is obtained is ency tests in males, BOLD activation of fMRI is seen not always stated. Are these so-called normal people in the left prefrontal cortex and right cerebellum, friends of the researcher, college students, employees and deactivation is seen in the posterior cingulate of the hospital, or random persons from the commu- gyrus and in parietal and superior temporal cortices nity? How many people have provided the data that (Fig. 1).8 90 The Journal of the American Academy of Psychiatry and the Law Reeves, Mills, Billick, et al. Figure 1. Verbal fluency: grouped male images. Activation is shown in the left prefrontal cortex and right cerebellum. Deactivation is shown in the posterior cingulate and the parietal and superior temporal cortices. This figure was reprinted with permission from the BMJ Publishing Group (J Neurol Neurosurg Psychiatry 64:492–8, 1998). What Brain Imaging Can Properly to GABA neurons. The relatively quiescent GABA Accomplish neurons in turn exert a diminished inhibition of do- Properly conducted, brain imaging effectively pre- pamine-containing neurons. Thus, in the presence of sents an enormous amount of data. Imaging can lo- ketamine, more dopamine is released at the caudate nucleus than in the normal state and can be visual- calize lesions, show subject or group properties and 11 relationships, enhance signal over noise by limiting ized using the labeled dopamine antagonist, C raclopride. In the presence of ketamine, and the in- the data to be acquired, and show functional associ- 11 ations. Thus, in clinical medicine, brain imaging is creased dopamine in the caudate nucleus, C raclo- pride occupies fewer caudate dopamine receptors used in diagnosis and in assessment of treatment re- 10 sponse. For example, a suggestive clinical examina- than in the normal state.