What is fMRI?

Functional neuroimaging (as opposed to structural imaging, as in CT scans or structural MRIs) can track the workings of the brain in real time. Functional MRI does this by tracking changes in local blood flow and blood oxygenation level (BOLD) that are associated with neuronal activity. When neurons are active, they demand energy, which is supplied by ATP, which in turn is produced by oxygen and glucose in the blood. Because more oxygen is supplied to the active brain region than is consumed (since most glucose is metabolised into ATP nonoxidatively), the ratio of oxygenated to deoxygenated blood in the active region increases. This results in changes in magnetic resonance (MR) signal intensity, as measured by an MR scanner, because oxygenated and deoxygenated blood have different magnetic susceptibilities (e.g., Huettel, Song and McCarthy, Reference Huettel, Song and McCarthy2004).

Functional MRI has two major advantages over PET and SPECT. First, fMRI is capable of greater spatial and temporal resolution, on the order of millimetres and seconds, respectively (Huettel et al., Reference Huettel, Song and McCarthy2004; Illes and Racine, Reference Illes and Racine2005).Footnote ¹ Second, fMRI is noninvasive, unlike PET and SPECT, which require the injection of a radioactive tracer. This enables fMRI researchers to include participants who might not want to undergo the injection and to test for people’s performances more often and under many more tasks or conditions (because fMRI does not involve injections, it is not subject to limits on the number of injections that may be given to any given person within a given time frame), thus enhancing the potential reliability and validity of the results (Constable, Reference Constable2006; Parry and Matthews, Reference Parry and Matthews2002).

Many thousands of investigative and exploratory studies indicate that fMRI has great promise in a wide range of in neuroscientific, medical and psychological applications. Its more precise functional brain mapping may facilitate planning for neurosurgery and allow for more effective development and administration of pain therapies (Columbia University functional MRI, 2005). Functional MRI can detect abnormalities in brain function in persons at risk for Alzheimer’s before the clinical presentation of symptoms characteristic of the disease (Parry and Matthews, Reference Parry and Matthews2002). And fMRI has already yielded unprecedented insights into the physiological bases for human cognitions, perceptions and judgments (e.g., Greene, Reference Greene2003; Greene et al., Reference Greene, Sommerville, Nystrom, Darley and Cohen2001).

Possible legal applications of fMRI

The wide range of possible legal applications of fMRI is suggested in part by the previous and ongoing uses of PET, SPECT and other neuroimaging technologies in various legal proceedings. First, fMRI evidence may be offered to prove adjudicative facts (Davis, Reference Davis1942), i.e., facts that are germane and limited to the particular case at hand. These include:

• • determining the presence of brain injury due to trauma (e.g., Green v. K-Mart, 2003; cf. Penney v. Praxair, 1997);

• • diagnosing brain injury, including ruling in or out possible causes of that brain injury (e.g., Elam v. Lincoln Electric Co., 2005; Hose v. Chicago Northwestern Transp. Co., 1995; Rhilinger v. Jancsics, 1998);

• • determining mental capacity for various purposes, including: (a) the insanity defence (e.g., United States v. Hinckley, 1981; People v. Weinstein, 1992; see also President’s Council on Bioethics, 2004); (b) mens rea, i.e., the specific intent required for proof of most crimes (e.g., People v. Williams, 2004; cf. Jackson v. Calderon, 2000); and (c) competency, whether to stand trial (e.g., United States v. Gigante, 1997), to receive the death penalty or to be put to death (e.g., State of Delaware v. Red Dog, 1993), or to execute a contract or a will or to give informed consent (Greely, Reference Greely and Garland2004; Tancredi, Reference Tancredi and Garland2004);

• • declaring brain death, especially in cases where non-functional testing, such as CAT scans, yield ambiguous results (In re Guardianship of Schiavo, 2005; Tancredi, Reference Tancredi and Garland2004);

• • determining racial or other bias (Greely, Reference Greely and Garland2004; see Eberhardt, Reference Eberhardt2005); and

• • detecting deception by witnesses (see, e.g., Langleben et al., Reference Langleben, Schroeder, Maldjian, Gur, McDonald, Ragland, O’Brien and Childress2002; Lee et al., Reference Lee, Liu, Li-hai, Chetwyn, Mahankali, Feng, Hou, Fox and Gao2002, Spence et al., Reference Spence, Hunter, Farrow, Green, Leung, Hughes and Ganesan2004).Footnote ²

Second, fMRI may also be offered to prove legislative facts (Davis, Reference Davis1942), i.e., general facts relevant to legal reasoning and the lawmaking process, including the formulation of legal rules or principles by judges. For instance, fMRI studies showing differences between adolescents’ and adults’ brain activity while participants were engaged in moral appraisal and reasoning tasks were cited in the briefs to the United States Supreme Court that successfully argued that executing juveniles is unconstitutional (Roper v. Simmons, 2005; Roper v. Simmons, Briefs for Respondent and Amicus Curiae ABA, 2004; for a review of the research, see Beckman, Reference Beckman2004).Footnote ³ And in the cases alluded to earlier, fMRI (and other) studies purporting to show connections between exposure to media violence and aggressive ideation and behaviour have been introduced in defence of state laws restricting children’s access to violent video games (Entertainment Software Association v. Blagojevich, 2005a; Entertainment Software Association v. Granholm, 2005, 2006).Footnote ⁴

Evidentiary framework

Testimony based on fMRI studies is, of course, scientific evidence, and is therefore subject to the special rules governing that kind of evidence. In American federal courts and a majority of state courts, the law governing this subject derives from Federal Rule of Evidence 702 (or the state law versions thereof) and the leading Supreme Court decision, Daubert v. Merrell Dow Pharmaceuticals (1993). Under Rule 702, if ‘scientific, technical, or other specialised knowledge will help the trier of fact . . . a witness qualified as an expert . . . may testify thereto in the form of an opinion or otherwise, if[:] (1) the testimony is based upon sufficient facts or data, (2) the testimony is the product of reliable principles and methods, and (3) the witness has applied the principles and methods reliably to the facts of the case’ (Federal Rules of Evidence, 2006). Both Rule 702 and Daubert require the proponent of the evidence to persuade the court that the evidence is sufficiently reliable and relevant to the matter to be proven for the trier of fact to consider it. Daubert proposes four factors for trial judges to use in gauging the reliability of proffered scientific evidence: whether the theory or technique on which the testimony rests can be or has been tested; whether the theory or technique has been subjected to peer review and publication; the known or potential error rate of the theory or technique when applied; and whether the theory or technique has been generally accepted in the relevant scientific community (Daubert, 1993, pp. 593–95). Many courts apply these factors with varying degrees of rigour (Merlino, Murray and Richardson, Reference Merlino, Murray and Richardson2005), while other courts have turned to additional or alternative tests (see Advisory Committee Note to Rule 702 (Federal Rules of Evidence, 2006); Daubert v. Merrell Dow Pharmaceuticals, 1995). In states that do not follow Daubert, the admissibility of scientific evidence continues to be governed by the older Frye test, which asks whether the proffered evidence is based on a theory or technique that is ‘generally accepted’ in the relevant scientific community (Frye v. United States, 1923). In any event, the decision whether to admit scientific evidence (like any other evidence) is within the trial judge’s discretion and may be overturned only if the appellate court finds that the trial judge abused that discretion (see Joiner v. General Electric, 1997).

The admissibility of the fMRI images implicates a few additional rules, although here, too, the focus is on the reliability of the (visual) evidence. Most importantly, admissibility may depend on whether the images are considered substantive or demonstrative evidence. Substantive evidence is evidence offered to prove a fact in the case; demonstrative evidence is offered merely to illustrate testimony rather than as independent proof (Mauet and Wolfson, Reference Mauet and Wolfson2005).Footnote ⁵ Substantive evidence is generally held to a more rigorous standard. Images considered as substantive evidence must be authenticated (under Federal Rule 901(b)(9) or the state law equivalent) by testimony ‘describing [the] process or system used to produce [the image] and showing that the process or system produces an accurate result’ (Federal Rules of Evidence, 2006). Merely illustrative evidence, by contrast, is ordinarily admissible if, in the opinion of the trial court, it will help the jury to understand some fact of consequence in the case (Mauet and Wolfson, Reference Mauet and Wolfson2005).Footnote ⁶

Finally, all evidence, oral testimony or tangible (real or demonstrative) evidence, is subject to Federal Rule of Evidence 403 (or the state law equivalent), which provides that even relevant evidence ‘may be excluded if its probative value is substantially outweighed by the danger of unfair prejudice, confusion of the issues, or misleading the jury’ (Federal Rules of Evidence, 2006).

Legal treatment of PET and SPECT

To date there have been roughly 130 reported opinions involving PET and/or SPECT evidence.Footnote ⁸ Several general observations are worth making. First, on more than two-fifths (85 of 133, 63.9%) of the occasions on which this expert evidence has been at issue, its proponents have presented or sought to present it not to juries but to judges alone (e.g., proceedings before administrative law judges or relating to non-jury issues such as requests for injunctive relief or determinations of competency to stand trial). This is noteworthy because concerns about the use of fMRI data (and perhaps especially fMRI images; see below) in court usually focus on their use by jurors; more often than not, however, judges, not jurors, will be the decision makers on the issues for which neuroimaging evidence is offered. Second, PET and/or SPECT evidence has been admitted in more than four-fifths (73 of 89, 82.0%) of cases in which it has either been admitted or excluded.Footnote ⁹ This perhaps surprisingly high admissibility rate may be due in part to the large proportion of bench proceedings: when a judge sits without a jury, there is no jury to protect from evidence deemed insufficiently reliable; instead, the judge typically hears the evidence and then decides how much weight to give it. Indeed, in cases in which PET and/or SPECT evidence has been admitted or excluded, it has been admitted at a significantly higher rate in proceedings before judges (44 of 46, 95.7%) than in those before juries (29 of 43, 67.5%), Χ² = 15.95, p <.01.

Third, there do not (yet) appear to be any clear patterns in the PET and SPECT case law regarding the admissibility of this evidence to prove particular kinds of facts (e.g., to prove brain injury due to workplace exposure to toxins). To a limited extent the decisions may break down by state: For instance, California seems to have been relatively unreceptive to both PET and SPECT evidence to diagnose brain trauma and other conditions (see, e.g., People v. Protsman, 2001 (affirming exclusion of PET-based testimony to diagnose prior brain trauma); People v. Yum, 2003 (affirming exclusion of SPECT-based evidence to prove defendant’s PTSD)), whereas Louisiana has tended to admit neuroimaging evidence for similar purposes (e.g., Green v. K-Mart Corp., 2003 (PET scan evidence admitted to prove traumatic brain injury)). There have been too few cases, however, to support valid inferences about this or any other sort of pattern or trend in the PET and SPECT cases.

If there are indeed no such general patterns, the explanation may well be traced to the fourth (and in some ways the most important) observation: The admissibility and persuasiveness of PET and SPECT evidence very much depend on the case-specific context. Courts do occasionally speak broadly about the general reliability or relevance of functional neuroimaging. For example, a California appellate court in a 2001 case approvingly referred to expert testimony indicating that PET scans were generally accepted for use only in cases of stroke, Alzheimer’s and epilepsy, and were not generally used to diagnose or evaluate head trauma; consequently, the court, applying California’s Frye-like standard for admissibility of scientific evidence, affirmed the trial court’s exclusion of the defendant’s scans to prove head trauma (People v. Protsman, 2001).Footnote ¹⁰ Similarly, a New York trial court recently ruled that PET scan studies did not meet the Frye standard when offered to prove traumatic brain injury, but that neither was such evidence ‘junk science’ justifying an award of costs and attorneys’ fees to the adversary (LaMasa v. Bachman, 2005).Footnote ¹¹

More often, however, courts make finer-grained determinations, taking into account the claimed relevance and probative value of the evidence, the strength of the testimony supporting or opposing admission, the strength of other evidence going to the same point, and, occasionally, the qualifications of the expert. Thus, where proponents of neuroimaging expert evidence are careful to offer that evidence not to ‘prove’ a diagnosis of, say, toxic encephalopathy, but rather as merely ‘consistent’ with such a diagnosis, courts have admitted the evidence (e.g., Elam v. Lincoln Electric Co., 2005; Hose v. Chicago Northwestern Transp. Co., 1995). More generally, PET and/or SPECT evidence has most often been admitted (when it is admitted) along with other neurological and clinical evidence of the party’s mental condition. As a consequence, neuroimaging expert evidence rarely seems in itself to be dispositive (in contrast, say, to the well-known CT scans in the Hinckley (1981) case); typically, it is admitted but does not seem to affect the outcome (although in the absence of controlled experimental data, this observation is necessarily speculative).

When courts have excluded PET and/or SPECT evidence, they have done so for several kinds of reasons. Proffers may be rejected where the evidence is not shown to be relevant (e.g., Feria v. Dynagraphics Co., 2004 (SPECT offered to prove plaintiff’s toxic encephalopathy, but insufficient evidence that condition could have been caused by exposure to the defendant’s chemicals); United States v. Mezvinksy, 2002 (PET scan evidence offered to prove defendant’s diminished frontal lobe activity ruled inadmissible because irrelevant to legal point of dispute, which was whether defendant could have formed an intent to deceive, the mens rea required to commit the charged crime of fraud)).Footnote ¹² Alternatively, the evidence in the particular case may not be sufficiently reliable to be heard by jurors. Under Federal Rule of Evidence 702(3) and Daubert, the expert must have reliably applied neuroimaging science to the facts of the case; if, however, the subject and control group in a neuroimaging study are not properly matched, so that a significant possibility of confounds remains, testimony based on the study is properly excluded (e.g., Penney v. Praxair, 1997 (PET) (subject and controls not adequately matched for age and subject but not controls under medication that might have affected scan results); United States v. Gigante, 1999 (same)). When the judge sits as trier of fact, expert testimony based on PET and/or SPECT evidence may be admitted but given very little weight because it is unsupported or even contradicted by other evidence (e.g., Falksen v. Secretary of Dept of Health and Human Servs., 2004 (in Vaccine Act case, court finds that claimant did not suffer encephalitis where SPECT scan showed abnormalities but no board certified neurologist diagnosed encephalitis and the use of SPECT to make such a diagnosis was not generally accepted); Prince v. Thomas, 1997 (in admiralty case, PET scan may have indicated that plaintiff suffered permanent brain damage from boating accident but court ‘d[id] not believe that [this evidence] offset the essentially negative MRI and the absence of objective neurological symptoms’). And some proffers of PET or SPECT evidence have been rejected because the witnesses who sought to introduce that evidence were found unqualified to do so (e.g., Commonwealth v. Frangipane, 2001 (PET); Craig v. Orkin Exterminating Co., 2000 (SPECT)).

Fifth and finally, in only two cases have the initial decisions to admit or exclude the evidence been reversed on appeal (Commonwealth v. Frangipane, 2001 (admission of PET evidence reversed); Berry v. CSX Transportation, Inc., 1998 (exclusion of SPECT evidence reversed)). This is consistent with the law requiring deference to trial judges’ determinations regarding the admission or exclusion of scientific evidence (Joiner v. General Electric, 1997)Footnote ¹³ and with the low rate of reversal of evidentiary rulings generally (Berger, Reference Berger1992).

In short, the courts’ treatment of PET and SPECT evidence offers some clues about how fMRI evidence is likely to be received. Broad declarations regarding the reliability of the underlying science may be infrequent, and in any event, of limited precedential value as fMRI science and its applications, especially in non-legal settings, develop. Admissibility and probative value are likely to depend instead on relevant features of the particular case in which the evidence is offered. Finally, trial judges’ decisions to admit or exclude fMRI evidence will rarely be overturned.

fMRI: Conceptual and methodological issues

The analysis and interpretation of fMRIs for purposes of neuroscience and psychology, and hence for law, can be understood as based on a chain of inferences running from the data to the psychological function or construct of interest:Footnote ¹⁴

Each step of this model raises potential questions about the reliability (and, in the case of inference (3), the applicability or relevance) of fMRI evidence for law, both of which are crucial considerations, as explained above, under either Federal Rule of Evidence 702 and Daubert or the alternative Frye test. Inference (1), from the fMRI data to the BOLD data in the regions of interest in the brain(s) of subject(s) in the task or experimental condition, is affected by researchers’ decisions regarding and assumptions underlying data processing methods, most of which have not yet converged on the kind of consensus that would allow the basic technology to be ‘black boxed’ (Halfon, 1998; Latour and Woolgar, Reference Latour and Woolgar1986), obviating further doubts about reliability. Inference (2), from the BOLD data to neuronal activity, is less problematic but still fraught with difficulty; although most researchers believe that local blood flow in the brain is related to neural activity, the precise relationship is not yet completely understood (Wolpe, Foster and Langleben, Reference Wolpe, Foster and Langleben2005). And inference (3), from neuronal activity to psychological function, raises fundamental questions about the theories and concepts relied upon in the design of fMRI studies and the associations drawn between fMRI data and the cognitive or emotional function of interest, and hence about the application of the data to the legal issue at hand.

Severally or cumulatively, these issues may affect the perceived reliability and hence admissibility of fMRI research on two levels. First, any aspect of fMRI methods or principles that makes any given finding less robust, less replicable, may make it harder for that finding to acquire general support and acceptance in the relevant scientific community, and this in turn will make courts less likely to regard that finding as a reliable enough basis for legal decision making. Second, in the aggregate, the various concerns raised by fMRI methods and principles may lead courts to find more generally that fMRI studies do not (yet) provide a sufficiently reliable basis for legal decision making.

From fMRI data → BOLD data in brain(s) of subject(s) in task condition

Whether represented in numbers or the well-known ‘pretty coloured brain pictures’, fMRI data are the outputs of an exceptionally complex technology. The nature and meaning of those data are shaped by a series of methodological and conceptual choices by the fMRI scientists. While the same could be said of many if not all types of scientific data, the decisions implicit in the fMRI data may be especially problematic for law: functional MRI technology is still relatively new, so the experts themselves have not yet reached consensus on methods for producing and interpreting data, and laypeople (including judges and jurors) are less likely to be or to feel that they are familiar with the ramifications of those methodological and conceptual choices, possibly undermining the perceived reliability of the data. The inference from the fMRI data obtained in properly conducted brain studies to BOLD levels in brain regions of interest – the foundation for all subsequent inferences to neuronal activity and hence to psychological function – raises at least half a dozen potential reliability issues.

First, fMRI scans present differences in brain activation between control (or rest) and experimental (or task) states, obtained by subtracting the BOLD signal of one from the other. That is, fMRI yields a relative, not an absolute, measure of brain activity. What follows from this? Typically, most brain regions are active in both the control and experimental conditions, but the differences in BOLD levels between the conditions (which is what the scans show) may be very small – in the typical reported fMRI study, only a few percent – or nonexistent (Wexler, Reference Wexler, Shulman and Rothman2004). If the brain region is more or less equally active in the two conditions, the fMRI will not show anything. The presentation of only these small differences (what appears depends in part on the thresholding level; this is discussed next) thus ignores the majority of brain activity that is occurring in both conditions. The fMRI data, therefore, may not provide a very firm basis for inferences about ‘active’ versus ‘inactive’ brain regions during a given task – which is how fMRI findings often tend to be described (especially to lay audiences) – and the supposed psychological implications of those different brain activity levels.Footnote ¹⁵

Second, fMRI data are thresholded for display at an arbitrary level of significance (often but not always p < .05) (Wolpe et al., Reference Wolpe, Foster and Langleben2005). The lower the bar for statistical significance of reported differences in brain activity between baseline and experimental conditions, the more ‘active’ regions appear and the more widely distributed in the brain they appear to be. Because at present there seems to be no consensus among fMRI researchers about the proper significance cutoff, researchers can obtain different brain activity patterns, and hence find support for different inferences about the structural distribution of psychological functions of interest, based on the identical underlying data.

Third, most fMRI images reported in the literature represent an average of data derived by subtracting an average of experimental scans from an average of control scans (e.g., Huettel et al., Reference Huettel, Song and McCarthy2004; Illes and Racine, Reference Illes and Racine2005). ‘Averaging images across brains can . . . produce apparent patterns of data that are spatially misleading, with group activity appearing in regions not found in any individual or eliminating aggregate patterns because of structural differences across individuals’ (Cacioppo et al., Reference Cacioppo, Bernston, Lorig, Norris, Rickett and Nusbaum2003, p. 652).Footnote ¹⁶

Fourth, and relatedly, variability in brain anatomy between individuals creates uncertainty in localisation (Brett et al., Reference Brett, Johnsrude and Owen2002). In any multiperson study, many steps are involved in normalising the data by reconciling individual anatomical variation to some standardised reference brain map (Brett et al., Reference Brett, Johnsrude and Owen2002; Huettel et al., Reference Huettel, Song and McCarthy2004). And every brain registration algorithm embeds assumptions that may not be shared by all or most other researchers (Constable, Reference Constable2006). These differences and uncertainties may not matter if the brain region of interest is as large as, say, the dorso-lateral prefrontal cortex, but the smaller the structural unit(s) of interest, the more important the decisions regarding the reconciliation of anatomical variability become.Footnote ¹⁷

Fifth, although it might seem that the preceding two concerns would be avoided in a single-subject study, fMRI data and hence the inference to BOLD signals for even a single person may be affected by all sorts of intrasubject variables that may muddy the data. These include the subject’s level of expertise in the experimental task (which may raise or lower BOLD signals and hence inferred brain activity levels; see below), the amount of attention the subject devotes to the task, and physiological variations in brain blood flow (on which the BOLD levels depend) due to the subject’s absorption of caffeine, nicotine and other substances that may affect baseline (resting) brain activation patterns and hence the measured differences in brain activity between baseline and experimental conditions (Constable, Reference Constable2006; Huettel et al., Reference Huettel, Song and McCarthy2004).Footnote ¹⁸ At least one leading brain researcher has gone so far as to state that, at present, ‘[t]here is no single-subject reliability’ (Robinson, Reference Robinson2004). Since many if not most likely courtroom uses of fMRI data will be to help prove adjudicative facts about individuals – mental impairment as evidence of injury, incompetence to stand trial, reduced brain function as a mitigating factor in sentencing, and so on (see above) – these caveats regarding single-subject data are of great practical importance.

Sixth, the complexity and variability of the experimental designs and methods used in fMRI research, and especially the variability from one lab to another, may undermine the generalisability of results (see, e.g., Illes and Racine, Reference Illes and Racine2005). For instance, MR signal changes increase with increasing magnetic field strength. But clinical fMRI scanners vary in strength – many are 1.5 Tesla, but an increasing number are 3.0 Tesla or greater – so the brain activation patterns detected may vary from lab to lab, even holding everything else constant, making it more difficult for specific findings to be replicated.

None of these issues has prevented the tremendous growth and widespread acceptance of fMRI as a brain research tool, as attested to by the thousands of fMRI studies now appearing annually in dozens of specialised and general scientific journals (Illes and Racine, Reference Illes and Racine2005). Even though peer reviewers routinely recognise (properly conducted) fMRI research as ‘good enough’ science, however, courts may well hesitate before admitting expert testimony based on fMRI. The inference from fMRI data to BOLD data may well be strong enough to support the always provisional findings of experimental science when addressed to audiences of fellow scientists, but not the conclusive determination of legal facts by non-scientists. Authors of peer-reviewed fMRI studies can satisfy their professional obligations in this regard simply by reporting their methodological and interpretive choices, trusting that their readers will understand the theoretical and practical contexts for those choices and appreciate the limitations on what may be validly inferred their findings. Judges applying rules of evidence to proffers of fMRI data, in contrast, ought not to be as sanguine, and might well decline to consider (or to allow jurors to consider) that data unless persuaded that the data are trustworthy enough to be relied upon by a decision maker whose judgment might deprive a party of property, liberty, or even life.

Although astounding progress has been made in functional brain imaging over the last decade, fMRI science is still in its infancy. Some if not all of the concerns raised above regarding the reliability with which any given fMRI data indicate relative BOLD levels in brain regions of interest may be minimised or eliminated as fMRI technology advances and brain researchers converge on protocols and methods for the processing and interpretation of data. Until enough of these and other conceptual and technological issues are resolved, however, the reliability of the fMRI data (even from properly designed and conducted studies) on which courtroom experts will rely in providing opinions should remain open to question.

From BOLD data in regions of interest in brain(s) of subject(s) in task condition → neuronal activity in those regions in brain(s) of subject(s) in task condition

The assumption underlying all fMRI work is that the measured BOLD signal is strongly correlated with neuronal activity; that is, that increases in the MR signal due to increases in the ratio of oxygenated to deoxygenated hemoglobin in particular brain areas reflects increases in neuronal activity (spiking or local field potentials or both) in those areas (Heeger et al., Reference Heeger, Huk, Geisler and Albrecht2000). But while neuroscientists generally agree that local blood flow in the brain is related to neuronal activity, and recent research has shown a strong correlation between neuronal spiking, local field potentials and BOLD signals (Mukamel et al., Reference Mukamel, Gelbard, Arieli, Hasson, Fried and Malach2005), ‘the relationship remains incompletely understood’ (Wolpe et al., Reference Wolpe, Foster and Langleben2005, p. 5). The BOLD signal ‘is a complex function of blood flow, oxygenation level and volume’ (Henson, Reference Henson2005, p. 5), and other physiological events besides neuronal spiking rates, such as the transitory supply of the neurotransmitter dopamine (Huettel et al., Reference Huettel, Song and McCarthy2004), may independently affect local blood flow and hence the BOLD signal. The inference from BOLD signals to neuronal activity levels may well be the least problematic step in the inferential chain and the most likely step to be black-boxed (cf. Halfon, 1998; Latour and Woolgar, Reference Latour and Woolgar1986) in the foreseeable future. Even so, the science is not yet settled, and therefore, even if this inference is generally accepted in the scientific community for purposes of the Frye standard, expert testimony based on fMRI may still be susceptible to objection and exclusion on this ground by judges taking a sufficiently strict view under Daubert.

From neuronal activity in brain(s) of subject(s) in task condition → psychological function of interest

All brain researchers (and many, but not all, psychologists) presume that there must be some kind of systematic mapping of psychological function onto brain structure; otherwise, measuring brain activity wouldn’t tell us anything about the mind (Henson, Reference Henson2005). But the nature and conceptual precision of this mapping, which should be critical to almost any reliance on fMRI studies in courts, raise a number of questions.

First, many brain imaging studies simply assume that the differential activation of a particular region of the brain during a specific mental operation means that that region is the neural substrate for that operation, and conversely, that the absence of (relative) activity in a region during a specific operation means that that region is not a neural substrate for that operation (i.e., a one-to-one mapping). But neither assumption is necessarily true (Cacioppo et al., Reference Cacioppo, Bernston, Lorig, Norris, Rickett and Nusbaum2003; Sarter, Bernston and Cacioppo, Reference Sarter, Bernston and Cacioppo1996). The degree to which the neural correlates of cognitive and emotional functions are localised versus distributed in the brain is a topic of ongoing (and maybe irresolvable) debate, but there is at least substantial support for a many-to-many relationship between regions of (relative) brain activity and cognitive function or behaviour.Footnote ¹⁹ So ‘activation [in a localised brain region] may reflect not the locus in which a particular function originates but rather may reflect a region that is part of a more distributed network of processing mechanisms that work together to perform the task’ (Cacioppo et al., Reference Cacioppo, Bernston, Lorig, Norris, Rickett and Nusbaum2003, p. 654). In addition, it could well be that ‘in a different context or when using a different comparison task, what appears to be the same information-processing operation could be related to activation in a different region of the cortex’ (Cacioppo et al., Reference Cacioppo, Bernston, Lorig, Norris, Rickett and Nusbaum2003, p. 654). Moreover, the more complex the mental function of interest, the more widely distributed the structural activity patterns, because ‘more complex’ when used to describe functions means (among other things) that more discrete processes are involved (Cacioppo et al., Reference Cacioppo, Bernston, Lorig, Norris, Rickett and Nusbaum2003). The complexities of widely distributed structure-function relationships should make it harder to establish sufficiently reliable support for generalisable associations between brain region activation and cognitive functions of interest, because the more multifarious the findings, the more difficult it will be (all things being equal) to replicate the results adequately.

Second, the interpretation of the psychological significance of fMRI data (i.e., significant differences in measured brain activity between experimental and control conditions) always depends on the specification of the psychological construct of interest and the care with which the experimental manipulations are designed. ‘Unless the information-processing operations that differ between comparison tasks are clearly and narrowly specified, the fact that the data were collected using fMRI and can be rendered in dramatic fashion on anatomically precise [images] may not help address the psychological or theoretical significance of the data’ (Cacioppo et al., Reference Cacioppo, Bernston, Lorig, Norris, Rickett and Nusbaum2003, p. 658; see also Willingham and Dunn, Reference Willingham and Dunn2003). Thus, a psychologist may want to do an fMRI study to learn more about memory, but what is memory? If (as most psychologists believe) memory consists of many distinct but sometimes interrelated functions, what are those functions, and how are they related? One must have some theory about this to draw any inferences about what is happening in the mind when certain patterns of activation appear in the brain. So any inferences from brain to mind are necessarily provisional because they depend on the existing (and often contested) state of psychological theory. Relatedly, the experimental and baseline tasks must also be matched for all dimensions not being deliberately varied, including difficulty and interest, to minimise confounds (Phelps, Reference Phelps and Sternberg1999), and this is more difficult to do the less well defined the psychological function or construct of interest (see Wexler, Reference Wexler, Shulman and Rothman2004).

It could be argued that the validity of any experimental psychological research depends on the clarity and coherence of the underlying theoretical constructs and the precision of the experimental design, and that fMRI research is no different in these respects, but merely uses a different dependent variable (see Henson, Reference Henson2005; Phelps, Reference Phelps and Sternberg1999). The exceptional sensitivity and complexity of that dependent variable (see generally Huettel et al., Reference Huettel, Song and McCarthy2004), however, makes any flaws resulting from imperfections in conceptualisation and design (not to mention other factors like individual variability in response; see above) that much more salient.

Third, even to the extent that scientists generally support an inference from relative activity levels in the brain region(s) of interest to psychological function, there may still be a gap between that function and the concept that is material to legal judgments of responsibility or guilt, raising concerns under Rule 702(3) (Federal Rules of Evidence, 2006 and Daubert, 1993). For instance, a criminal defendant who wants to use neuroimaging evidence to support an argument that he was incapable of having the mens rea required to commit the crime must do more than show a brain abnormality or even a decrease in cognitive function. He ‘must introduce evidence of a cognitive or behavioral impairment that directly relates to his capacity to form criminal intent or to control his own behavior’ (Kulynych, Reference Kulynych1997, p. 1257; see also Dumit, Reference Dumit2004; Kevles, Reference Kevles1997). It was on just this ground that the Ninth Circuit Court of Appeals in Jackson v. Calderon (2000), for instance, rejected the defendant’s proffer of PET scan evidence to diagnose his PCP abuse. The court reasoned that the PET scan evidence could at most establish only some PCP-induced brain abnormality but not the effect of that abnormality on the defendant’s capacity for higher-level thought, including the specific intent required for conviction for the crime charged.Footnote ²⁰ Given the current state of forensic psychological neuroimaging, fMRI evidence offered to prove similar matters may face the same sort of challenge.Footnote ²¹

fMRI goes to court: The Entertainment Software Association litigation

How much any of the issues discussed above will affect the admissibility or perceived probative value of fMRI evidence will depend on the circumstances of the particular case in which the evidence is offered. The very first reported cases, however, illustrate some of the difficulties that proponents of fMRI evidence may face (Entertainment Software Association v. Blagojevich, 2005a; Entertainment Software v. Granholm, 2005, 2006). The plaintiffs in both cases were Entertainment Software Association, a trade organisation consisting of video game publishers and console manufacturers, and others who sued in federal court to enjoin enforcement of anti-violent video game legislation in Illinois, Michigan and other states.Footnote ²² The most important fact at issue is the legislative finding that (in the words of the Illinois legislature) minors who play violent video games are more likely to ‘experience a reduction of activity in the frontal lobes of the brain which is responsible for controlling behavior’ (Entertainment Software Association v. Blagojevich, 2005a, p. 1073). To support this finding at the hearing on the plaintiffs’ motion for a preliminary and permanent injunction in the Illinois case, the state offered the testimony of several expert witnesses, including William Kronenberger, a clinical psychologist who, relying in part on fMRI data, sought to show that adolescents exposed to a substantial amount of media violence exhibited deficits (lower activity levels) in particular brain locations (dorso-lateral prefrontal cortex and anterior cingulate cortex), and that these patterns resembled those of the brains of other adolescents clinically diagnosed as aggressive or violent.

In response, the plaintiffs offered the testimony of Howard Nusbaum, a cognitive psychologist. Nusbaum criticised the defendant’s fMRI studies and their proposed use in court on many grounds, three of which are particularly noteworthy here because they clearly illustrate concerns raised by the third link in the inferential chain discussed above. First, Nusbaum called attention to a problem of experimental design. In one of the defendant’s studies, researchers used fMRI to measure the effects of violent versus nonviolent video game play. Since participants could not actually play the games while in the scanner, however, they were instead shown videotapes of game play and asked to press a fake joystick button whenever they would have done so during actual play. But, as Nusbaum pointed out, brain activity would very likely be different during actual play, which requires interacting with the (mediated) world, as opposed to mere viewing. So the study was not a valid measure of the effects of violent versus nonviolent game play (Entertainment Software Association v. Blagojevich, 2005b, pp. 384–5).

A second and more fundamental criticism attacked the inference that Kronenberger drew from the observed difference in brain activity (in the task versus rest condition) to the claimed difference in mental function. The fMRI study just described showed reduced activity in participants’ frontal lobes during (simulated) game play. Kronenberger testified that this reduced level of activity meant a reduced capacity for impulse control, self-regulation, and so on, and thus a greater susceptibility to aggressive or violent behaviour. Nusbaum, however, explained that the frontal lobes are involved in many brain functions, and that many areas besides the frontal lobes (e.g., parietal lobes, thalamus, etc.) are also involved in attention (and hence self-regulation), so that it is not possible to draw the kind of one-to-one connection between frontal lobe activity and level of self-regulation on which the defendant’s expert’s testimony relied. In addition, reduced activity in the frontal lobes may indicate not a mental deficit, but rather greater expertise in some task-relevant function(s), because the more expert one is at a task, the less attention one needs to devote to it (Entertainment Software Association v. Blagojevich, 2005b, pp. 361–5). Thus, the fMRIs of adolescents who play violent (or any) video games may show less frontal lobe activity not because those adolescents are aggressive or violent but because they are experts at playing those sorts of games.

Third, during his cross-examination, Nusbaum indicated the gap between the psychological constructs measured by well-designed fMRI studies and those on which legal judgments depend. For example, the legislative finding supporting the Illinois statute speaks of behavioural ‘control,’ and the defendant’s attorney asked Nusbaum whether certain brain functions were associated with ‘self-control’ and ‘concentration’. Nusbaum responded that fMRI cognitive research might (for instance) associate a given part of the brain with ‘task-switching’, but would not purport to reach findings about functions as broadly (and colloquially) labeled as ‘self-control’ (Entertainment Software Association v. Blagojevich, 2005b, pp. 406–7).

The federal trial courts in both Illinois and Michigan found these and other criticismsFootnote ²³ of the fMRI research on which the defendant relied to be convincing, and enjoined enforcement of the anti-violent video game laws (Entertainment Software Association v. Blagojevich, 2005a; Entertainment Software Association v. Granholm, 2005, 2006).Footnote ²⁴ It is possible that, in contrast to the very ambitious offer of fMRI (and other) evidence in the Entertainment Software Association cases to establish an association between ill-defined stimuli (‘violent video games’) and a complex set of higher-order cognitive/emotional functions (e.g., self-control) to prove a broad legislative fact, other potential uses of fMRI may be less prone to at least some of the basic conceptual and methodological concerns outlined above. For instance, reliability and relevance problems posed by averaging and normalising multiperson data, specifying the psychological construct of interest, and translating neuroimaging findings into concepts both relevant and material to legal judgments may be mitigated or avoided when fMRI evidence is used to prove case-specific adjudicative facts such as whether a defendant displays mental function x that is characteristic of, say, toxic solvent encephalopathy (see, e.g., Hose v. Chicago Northwestern Transp. Co., 1995).Footnote ²⁵ On the other hand, the opposite may also be true: precisely because of difficulties in interpreting individual scans (due to insufficient single-subject reliability; Robinson, Reference Robinson2004), inferences from a scan to the classification of legal interest may be less reliable for individuals than for the aggregates that legislative facts purport to describe. In sum, the extent to which proffers of fMRI evidence will be successfully challenged on the basis of the fundamental methodological and conceptual questions discussed above is simply not yet clear.Footnote ²⁶

Functional MRI images may be inappropriately persuasive

Functional neuroimages may indeed be highly persuasive evidence of the facts they are offered to prove, but for reasons that go beyond the reasons why brain scientists and other experts might consider them persuasive. First, as one critic of the courtroom use of neuroimages has written, fMRIs ‘may appear to be deceptively similar to photographs of a person’s brain’ (Kulynych, Reference Kulynych1997, p. 1249). This remark nicely captures two facets of what media theorists call the appeal of transparency (Bolter and Grusin, Reference Bolter and Grusin1999), or what cognitive and social psychologists call naïve realism: the convincing notion that one has unmediated access to objective reality (Robinson et al., Reference Robinson, Keltner, Ward and Ross1995). First, fMRIs may offer convincing evidence because they seem to allow viewers to look directly at the brain. Writing of (and to) professional experimental psychologists, i.e., the experts themselves, Henson (Reference Henson2005, p. 42) states that ‘[t]here is a real danger that pictures of blobs on brains seduce one into thinking that we can now directly observe psychological constructs’. The risk that non-scientists (including jurors and judges) will be prone to similar epistemic confusion must be even greater.Footnote ²⁸ At the same time, fMRIs may offer convincing evidence because they look like what people expect images of the brain to look like, because they have seen brain images in magazines, on television and on the Internet. That is to say, these images have been assimilated (or remediated, to use Bolter and Grusin’s (Reference Bolter and Grusin1999) terminology) to a familiar medium (photography) which people are already habituated to ‘see through’ directly onto reality.

Second, fMRI pictures are likely to be especially credible to lay audiences because of their status as prototypical scientific images. The fMRI visualises brain function for the non-scientist audience as an instance of mechanised, computerised, and therefore objective (and, consequently, highly reliable), scientific knowledge (see Daston and Galison, Reference Daston and Galison1992). In addition, the fMRI is a mathematised object, the visual representation of a series of (computerised) measurements and computations (see Lynch, 1988). In both respects, the persuasiveness of fMRI images derives from their resonance with other representations of scientific truth in the culture at large.

Third, the presumptive credibility of fMRIs may trigger in lay viewers’ minds networks of folk psychological inferences that have consequences for legal judgments. Brain images tend to naturalise the psychological construct of interest (e.g., mental competence) by appearing to ground them in visible reality and to locate them in the person – specifically, in the brain (Dumit, Reference Dumit2004). This reduces psychosocial complexity to supposed features of the brain; it confuses the part (brain) for the whole (person-in-situation) (cf. Bennett and Hacker, Reference Bennett and Hacker2003). This could exacerbate people’s general tendency to commit the fundamental attribution error – to overattribute others’ behaviour to the kinds of people they are rather than to the circumstances in which they find themselves (Nisbett and Ross, Reference Nisbett and Ross1980).Footnote ²⁹ Moreover, neuroimages tend to convert (sometimes subtle) differences of degree into differences in kind. Statistically significant but still relatively small distinctions in BOLD levels between two sets of data become contrasts of bright colour versus no colour. The juxtaposed images ‘just look different,’ and therefore may be thought to be different in psychologically and legally relevant ways (cf. Eberhardt, Reference Eberhardt2005). This may encourage in triers of fact a false sense of ease in making legally required classifications (e.g., whether the defendant was or was not mentally impaired so as to diminish or avoid legal responsibility for his or her actions). More invidiously, neuroimages of difference may exacerbate the effects of unconscious prejudice against outgroups (although they may also mitigate prejudice by showing people how their race-based perceptions are situation-dependent and amenable to change; Eberhardt, Reference Eberhardt2005).

The probative value of fMRI images should outweigh their judgmental risks

Even though fMRI images may be unduly or improperly persuasive for the reasons just discussed, the dangers they may pose to good legal judgment do not warrant their per se exclusion from the courtroom. In the language of Federal Rule of Evidence 403 (Federal Rules of Evidence, 2006) and similar state law rules, the risks of unfair prejudice, confusion, or misleading the jury that fMRIs pose need not (substantially) outweigh their probative value in illustrating and clarifying expert testimony. Reliable legal knowledge about matters to which fMRI-based expert testimony is relevant is best sought, not by excluding fMRIs, but rather by admitting them and allowing expert witnesses and lawyers to educate jurors (and judges) about scientists’ own interpretations of the images.

Four arguments support this position: (1) contra Dumit (Reference Dumit2004), there is no ‘paradox’ of expert images. That judges and jurors need expert testimony to help them interpret fMRIs is not a reason to exclude the pictures; rather, in court as in the everyday practice of science, the scientist’s words and images can explain each other, increasing the likelihood that legal judgments will be grounded in acceptable science. (2) fMRIs may indeed resonate with jurors’ familiarity with other brain images and thus produce ‘unauthorised’ meanings – but all images are capable of doing this, yet the law does not for this reason exclude all demonstrative (or real) evidence, and fMRIs are no different from other images in this respect. (3) Although excluding fMRI images may avoid some kinds of potential bias, showing the pictures may help to correct for (or at least not exacerbate) other biases. (4) Finally, in addition to hearing expert witnesses’ explanations of how fMRI images are produced and what they mean, the best way for lay audiences to understand how fMRIs (or any images) represent knowledge is to be shown more pictures.

fMRIs and expert testimony can explain each other

The first argument is that Dumit’s ‘paradox’ of expert images is illusory. Judges and jurors can be taught to know what scientists already know: that fMRIs are not ‘snapshots’ of the brain, not direct read-offs from reality, but rather highly mediated outputs of scientific investigation, and that their probative value depends on the nature of the mediations – the validity of the underlying theories and concepts and of the principles that guide the multiple translations from experimental intervention to final visual representation (see, e.g., Galison, Reference Galison1997; Hacking, Reference Hacking1983; Latour, Reference Latour1987, Reference Latour1999). Judges and jurors can also be instructed to interpret fMRIs in light of their context within relevant scientific discourse. In routine scientific practice, including conference presentations and publications, scientists use images along with words and notations to construct arguments about the reliability, validity and importance of their work. Words and illustrations back each other up (Bastide, Reference Bastide, Lynch and Woolgar1988); each explains the other. The same should be true in court. Here, the relative opacity of the fMRI – ‘If it’s not a snapshot of the brain, then just what are we looking at?’ – enlists the expert as the interpreter of the image and (as Goodwin (1996) has explained) provides an opportunity for the expert to perform science by modeling informed looking at the image. Thus, when an expert’s words explain the image in court, each makes the other more intelligible and persuasive. Allowing jurors (and judges) to hear expert testimony and see expert images would therefore actually bring courtroom scientific discourse into closer alignment with scientific practice.Footnote ³⁰

There are reasons to believe that the fMRI pictures will not necessarily trump the testimony (as Dumit (Reference Dumit2004), Mobbs (Reference Mobbs2006) and others have suggested they might). First, expert testimony will frame the fMRI, from the authentication to the description to the interpretation of the image. The foundation needed to authenticate (see Federal Rule of Evidence 901(b)(9)), for instance, will prompt jurors to focus on the image’s mediated aspects, and competent cross-examination should help to ferret out the limitations of the data to prove the fact at issue. Second, expert witnesses are capable of setting forth the science clearly enough for laypeople to understand (as exemplified by Nusbaum’s testimony in the Entertainment Software hearing, discussed above). Research shows that when expert scientific evidence is clearly presented, jurors generally do a satisfactory job of understanding it and using it properly to inform their judgments (e.g., Vidmar and Diamond, Reference Vidmar and Diamond2001). More specifically, although jurors of course come to court with preconceived notions of every sort which may sometimes diverge from legal norms, when those preconceptions are explicitly identified and addressed, jurors are likelier to decide in accordance with the law rather than with the preconceptions (see, e.g., Diamond and Casper, Reference Diamond and Casper1992; Smith, Reference Smith1991, Reference Smith1993). Similarly, we might expect that if jurors’ naïve notions about the ‘pretty coloured brain pictures’ are acknowledged and then addressed through expert testimony, jurors’ judgments are less likely to be driven by their preconceptions about what the pictures mean.

The problem of polysemy is not confined to brain images

Although it is true that the polysemy of fMRIs may prompt ‘unauthorised’ readings, this is true of all images, expert or non-expert, displayed in court. Recognising that visual displays may prompt decision makers to reach judgments on inappropriate grounds (e.g., animosity toward a party), the law of evidence subjects them (as it does all other evidence) to a balancing test, admitting them unless any risk of undue prejudice or confusion they pose substantially outweighs their probative value (Federal Rule of Evidence 403). Some visual displays pass this test; others do not. There is no reason to treat neuroimages in general any differently.Footnote ³¹

fMRIs may on balance reduce, not aggravate, judgmental biases

Even if not showing the fMRI scans in court would avoid certain judgmental biases (as described above), showing the scans could help to avoid others. For example, many lawyers and commentators claim to have observed what has been called ‘the CSI effect’ (Baar, Reference Baar2006; Stockwell, Reference Stockwell2005; but cf. Podlas, Reference Podlas2006): jurors expect prosecutors to produce forensic DNA evidence in every case, and tend to judge the prosecution more harshly when the proof disappoints these expectations.Footnote ³² More generally, exposure to television programmes like CSI has shaped jurors’ ideas of what scientific truth looks like. In the case of United States v. Old Chief (1997), the United States Supreme Court recognised that parties have a right to present evidence in the form that they deem best suited to meet jurors’ expectations of what convincing proof should be, even if that evidence is not logically necessary to the determination of guilt or innocence. Thus, if the fMRI-based expert testimony itself is admissible, the proponent of the fMRI evidence ought to be permitted to show the pictures in order to avoid being unfairly prejudiced by having disappointed the jurors’ expectations about what scientific proof looks like.Footnote ³³

The cure for naïve realism is more, not fewer, images

To the extent that those opposing the courtroom display of fMRIs are concerned that judges and jurors will mistakenly believe that the images offer direct, infallible access to objective reality (and thus accord the fMRIs too much weight), the solution is not to keep the images out. Instead, the law should confront decision makers’ possible naïve realism head-on and refocus the search for legal truth where it should be, on the ways in which legal knowledge is mediated. And the way to do that is to show more images – to multiply the styles of representation. As Hacking (Reference Hacking1983) argued, when representations are undifferentiated, the problem of realism does not arise; in other words, we are all naïve realists. ‘But as soon as representations begin to compete, we ha[ve] to wonder what is real’ (p. 139). A variety of modes of representation makes it possible to ask how each mode mediates reality, and with what consequences.

If all viewers see are the standard axial views of the functional data superimposed on the structural brain image, highlighted in the usual fashion, they may well remain captivated by the notion that they are ‘looking at the brain’. But if they also see fMRI data represented on rotated 3-D views of a highly articulated cortex (e.g., Brett et al., Reference Brett, Johnsrude and Owen2002, figure 2), on ‘smoothed’ brain images (e.g., Spence et al., Reference Spence, Hunter, Farrow, Green, Leung, Hughes and Ganesan2004, figure 4), and/or ‘glass brain’ views, outlines of brains mapped onto grids and shown from three views (axial, coronal [from behind] and sagittal [from the side]) (e.g., Huettel et al., Reference Huettel, Song and McCarthy2004, figure 12.11; Spence et al., Reference Spence, Hunter, Farrow, Green, Leung, Hughes and Ganesan2004, figure 3), they are much more likely to understand that none of these is a ‘snapshot’ of the brain, and that each is no more (and no less) than a particular visual representation of the scientific data, the product of deliberate and often debatable choices by the research team (see Constable, Reference Constable2006).Footnote ³⁴ It is the variety of representational styles itself that makes the point that different modes of imaging can convey the ‘same’ information in very different ways, offering different signal-to-noise ratios, methods of visualising statistical significance, and implications regarding neuronal activity in areas other than the regions of interest quantification,Footnote ³⁵ as well as other explicit and implicit differences in visual meaning.Footnote ³⁶ Finally, images can help explain how fMRIs mediate scientific knowledge of the brain in yet another way: by explicitly diagramming the stages of data generation (e.g., Huettel et al., Reference Huettel, Song and McCarthy2004, box 8.1) and/or image processing (e.g., Brett et al., Reference Brett, Johnsrude and Owen2002, figure 1). The courtroom display of such images should not only greatly assist triers of fact in understanding the fMRI expert testimony but also disabuse them of the tendency to view the data representations naïvely and hence uncritically.