Gary Smith lay in an fMRI scanner as a series of pre-programmed questions appear on the screen in front of him, at five second intervals.
"Did you share an apartment with Mike McQueen?"
"Did you shoot Mike McQueen with a revolver?"
"Did you kill Mike McQueen?"
In response to each question, Smith clicked a button to indicate "yes" or "no," but as the scientists who designed the test later explained to me, his manual answers were largely irrelevant. The real objective was to monitor the processes taking place in his brain.
Smith, a 33-year old former US Army sergeant who served on two tours of Afghanistan, is now at the center of one of Maryland's most complex and prolonged murder cases in recent history. In September 2006, Smith's roommate, former army ranger Michael McQueen, was found shot dead in their shared apartment.
Two separate juries have already convicted Smith of second-degree murder, initially in 2008 and then again in 2012, with sentences of 35 and 28 years respectively. Both have since been overturned on appeal; later this year, Smith will stand trial for a third time.
With little concrete evidence to support either the defense or the prosecution, the objective of the fMRI scans is to determine whether Smith's claim of innocence is the truth, or a lie.
During both trials, Smith claimed he spent the night drinking, smoking marijuana, and playing pool with McQueen. After dropping him at home, he left the apartment; on returning, he found McQueen's body in an armchair—an apparent suicide. McQueen's girlfriend since high school had recently left him, and after a drunk driving conviction, he was concerned about his job prospects.
But it was Smith's gun that fired the fatal shot. During questioning, he gave three versions of events to police officers, initially suggesting that McQueen may have been murdered by local Hispanic men. Most incriminatingly, in the eyes of the prosecution, he disposed of the gun before police arrived and only confessed to doing so after ten hours of interrogation. However, for nine years, Smith has insisted he did not kill McQueen. He says he initially tried to make the crime scene appear like a murder, because he didn't want his friend to be remembered as someone who killed himself.
The prosecution and McQueen's family believe Smith is lying. Prosecutors have pointed out that he was heavily traumatized by what he witnessed in the Middle East. During his first trial, Smith was diagnosed with PTSD.
Two trials have yielded little hard evidence to prove Smith's guilt or innocence, but in the mind of Joel Huizenga, CEO of a company called Truthful Brain, there is no doubt. He is convinced Smith is completely innocent.
Huizenga's company and a team of independent experts conducted the fMRI scans on Smith's brain while questioning him about the events of that night. They were looking for certain patterns of activity that would indicate he might be lying.
"Deliberately lying is hard work," Huizenga told VICE. "When you're telling the truth, you're just retrieving a memory. But when you lie you have to bring back the truth first and then manipulate it and doing that requires much more of the brain to be active. This means blood rushes to specific areas which are never really used when you're being truthful, and the fMRI allows us to detect these relative changes of blood flow."
Huizenga initially asked Smith to intentionally lie in response to simple questions about his age and place of birth to get an idea of what his brain looks like when he's lying. Then he compared those scans to the ones taken while he was questioned about McQueen.
"In my opinion, they show beyond a shadow of a doubt that he is innocent. Everything about the scans is completely clean," Huizenga says. "None of the brain areas associated with lying light up whatsoever when he's asked about killing his room-mate."
But these scans will not be shown to the jury when Smith is tried again later this year. Last month, his attorney requested a new Frye hearing, a proceeding which determines whether new technology can be used as evidence in a court of law. The request was denied.
The previous Frye hearing, which took place ahead Smith's 2012 trial, was the most comprehensive legal review of fMRI 'lie detection' in the legal system so far. Ultimately, it ended in rejection.
"The legal decision is based on whether fMRI lie detection has been completely accepted in the scientific community and that's not the case right now," explained Gary Gibson, a professor at California Western School of Law, in an interview with VICE.
"You have scientists saying it's great, but there are others saying the brain is too variable and it isn't reliable. If an analyst runs a DNA sample, you'll get the same result one thousand times in a row. Is that the case for fMRI? We don't know and judges are going to be conservative as we've let a boatload of unreliable science in courts before. Look at the FBI's retractions of bullet lead and hair sample analysis. Those technologies were admitted despite thousands of inconsistent results and now they've had to be pulled. So people are going to be cautious."
Frank Haist is a cognitive neuroscientist in the Psychiatry Department at the University of California, who has testified as an expert witness in Smith's case, arguing for the use of brain imaging data. He believes that the prosecution resisted particularly strongly as it was a trial by jury.
"If the defense entered this kind of information, I think jurors would find it particularly compelling," Haist explained. "It would make a whole lot of sense to them. While the prosecution could put out counter-evidence, it's very approachable science so that's why they want to keep it out."
Haist believes the fMRI data is far more valid than much of the scientific evidence which has been presented during the two trials so far.
"There was an expert on the prosecution side who argued that the pattern of blood flow splatter from the victim was consistent with a gunshot at a particular angle from a particular distance, while an expert for the defense said it was more consistent with a suicide. Personally I would suggest that the forensic validity of blood splatter analysis is highly questionable. So the decision to allow something like that in, and exclude the fMRI analysis seemed rather arbitrary."
The opponents of brain scan lie detection argue that it remains too vulnerable to countermeasures, which can beat the system. One study done by the Harvard Medical School found that participants could obscure the ability to detect deception by making subtle motor movements during key points in testing.
"The countermeasure reduced the ability to detect deception, but they didn't eliminate it completely," Haist says. "It was still possible to tell they were lying. The field is still maturing and with improvements in pattern recognition, it will get to the point where we can exclude all the possibility of any counter-measure working. It's also important to look at particular personality styles such as sociopaths. Can they mask lying better than the typical person because they're delusional? That's an open question. We haven't done research in the deception field to decide conclusively whether the technology could be more than 90 percent accurate in those cases."
Huizenga believes the technology is up to the challenge but the problem is finding the funds to conduct such studies. "What would be the best for the field is if a government puts up the money for a study involving 1,000 people. That's only thing that people can hold against it at the moment, the study size work is small."
Outside of the courtroom, the demand for brain scan detectors is steadily increasing, with Florida State police signing up to use 'brain fingerprinting' technology last year. Many believe it will soon replace polygraphs as the lie detection method of choice.
"Polygraphs simply look at whether you're anxious or nervous and they're pretty good at telling that," Huizenga says. "But the leap of faith is from being anxious to telling a lie and that's why they're only 60 percent accurate. There are people who are anxious when they're telling the truth. And they're easy to manipulate. You can beat a polygraph through simple techniques like taking a beta-blocker."
Currently, polygraphs are used hundreds of thousands of times a year across the US but given their low accuracy rate, Gibson says the final judgement often comes down to the polygraph operator rather than the data itself. He believes that if the CIA and national security agencies switch to using fMRI instead, that could start the ball rolling for the technology to be accepted in court.
"There has been interest and that would be the vindication that it's passed beyond theoretical. In any case, sooner or later you're going to find a judge that's going to admit it. It just takes one judge in one state to start the ball rolling and that could happen in six months or six years. Frankly, I think the way it's going to get into court is going to be through a civil application, not a criminal application because the stakes are [lower]. You're only talking about money, not freedom."
But such a decision will undoubtedly come too late for Smith, who will discover his fate for a third time, by the end of 2015.
"I believe that if admitted, the fMRI evidence would have tipped it in his favor, since he's been consistent with the 'I didn't shoot him,' story," Gibson says.
Haist feels the evidence should have been presented at trial but he is agnostic as to whether Smith is really innocent or guilty.
"He says he's not lying, that he did not kill his room-mate and I believe the data is consistent with that. But I can't say that evidence alone should exonerate him. I don't think that the science is there yet to allow that kind of determination by itself. It still has to be taken in the context of all the other data and there was a lot of unusual behavior that night, which is difficult to explain. In any court, it has to be the weight of evidence, rather than any single piece of evidence, which ultimately counts."
Follow David Cox on Twitter.