Skip to main content

Can We Measure Delusions?

There might be a way—and if so, we could use it to detect and treat them in the earliest stages, before they become debilitating

This article was published in Scientific American’s former blog network and reflects the views of the author, not necessarily those of Scientific American


It was midday when an ambulance brought Rose to the Emergency Department.

The triage nurses, with their characteristic knack for brevity, had written “50 year old schizophrenic woman hearing/seeing dead boyfriend.” The medical team had done the standard workup—temperature, blood pressure, EKG, labs to screen for an electrolyte imbalance, drug or toxin that might explain Rose’s condition. Everything seemed normal, making Rose (whose name and narrative details have been changed to protect her privacy) a psychiatry patient.

So I made my way to the B wing of the E.D., which serves as a Limbo of sorts between the medical and psychiatric services.


On supporting science journalism

If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.


The B wing invariably bustles with activity. A long concierge-style counter with three computers faces the center of the room, which is essentially a large rectangle. When seated at one of these computers, you can see into each of the nine patient rooms that wrap around the three outer walls. From this vantage, the B wing becomes an amphitheater, with patients in gurney-sized niches showcasing some emergent medical concern: B7, chest pain; B9, acute shortness of breath.

Rose was medically cleared, so her gurney had been downgraded to stage right, to the end of the counter. I entered at stage left and noticed her across the room, feet at the head of the gurney propped on a pillow; her head was at the foot, neck slightly bent over the edge. Her hands were neatly resting on her belly while her thick hair formed a graying river that reached towards the linoleum.

In a firm yet conversational tone, Rose said towards the ceiling, “Why would Steven say he isn’t dead? How could anyone be so cruel?” Her mouth moved widely like a Claymation character as she slowly enunciated every syllable, chopping cruel into CRU-EL. She was smiling.

I stood quietly, observing the scene as Rose stared intensely upwards. A few seconds later, it occurred to me that she was waiting for the ceiling to reply.

“I see,” I muttered, recalling the triage nurses’ note, and went in search of a stool for my interview.

When I returned, Rose was sitting right-side up in the gurney with an entirely different expression. Her eyes narrowed in existential pain as her mouth and eyebrows curled in a grimace. I extended an unaccepted handshake and asked how I could help.

“Well, someone is being very CRU-EL,” she announced, vigilantly glaring upwards. “I hear Steven’s voice saying he isn’t dead. I’ve seen him. But he died in March.”

Then smiling, she turned to me and asked, “Can anyone be so CRU-EL to pretend he’s not dead? Why trick me? Why, DOC-TOR?” Her gaze returned to the ceiling. After a few unanswered questions, I dismissed myself and retreated down the hallway to the Psychiatric E.D. to review her chart.

Rose had a PhD in Economics, had taught at local universities. About 20 years before, she had been diagnosed with schizophrenia. Her delusions responded well to medications but she didn’t like taking them because they made her feel “out of sorts.” She would stop them, her delusions would return, and after a few days in the hospital, medications would quiet her mind and she would return home to begin the cycle again. Then she met Steven.

Steven helped Rose remember her medications, stabilized her life. Without Steven and—more importantly—without medications, she was a ship without keel. Eventually, she capsized.

What struck me was that at each admission, she would arrive with dogged delusions that would melt after a few days of medication. The neuroscientist in me was deeply frustrated: it was as if she came to us without warning, only after she had suffered a cognitive “heart attack.”

As a medicine intern, I’d measured patients’ blood pressure and cholesterol and, in combination with data like the patients’ age and ethnicity, I’d estimated their risk of a heart attack. It was routine for me to trace a patient’s blood pressure and cholesterol over time and prescribe (and increase) medications if they reached troubling levels. For heart attacks have known risk factors that physicians take seriously.

There is no risk calculator for delusions. In Rose’s case, there is only her arrival at the emergency department, when upside down on her gurney, she asks the ceiling about her dead boyfriend’s apparition. Yet heart attacks happen after years of atherosclerosis and hypertension; psychotic breaks must also have measurable warning signs. But what to measure?

KARL FRISTON AND THE FREE ENERGY PRINCIPLE

Schizophrenia is a severely disabling illness that warps the fabric of someone’s reality, often tearing it apart. I imagine that, on interviewing their first “Rose,” every clinician takes pause to consider the connection between reality and consciousness, between sensation and perception. In the 1980s, one such young clinician was Karl Friston.

Friston—who is now considered one of the most influential neuroscientists for his serial landmark contributions over decades—read medicine at Cambridge University’s Gonville and Caius College and at London’s King’s College Hospital. He completed his psychiatry residency at Oxford University. Friston spent most of his clinical time treating people with schizophrenia and puzzled over what was going on in their brains.

Positron Emission Tomography (PET) had recently been applied to the brain, allowing researchers to indirectly measure brain function based on how much radiotracer active regions absorb from the blood. And so, in the early 1990s, Friston turned to PET to study the brains of people with schizophrenia. At the time, each research group had their own in-house software to measure brain activity with PET.

Some would visually outline a region in each subject (i.e. that part looks like the motor cortex) and see how activity within that region changed with different conditions (e.g. pushing a button versus rest). Improvements in how to line up the anatomy of people’s different-shaped brains (this work by PhD mentor, Peter Fox, remains strikingly elegant) allowed researchers to subtract the entire PET images and see where and how much, on average, activity throughout the brain changed with different conditions.

But Friston reasoned that since each value in each subject’s PET image (a three-dimensional string of numbers) corresponded to similar anatomy, he might as well compare each corresponding value with rigorous statistical tests. So Friston hunkered down and created Statistical Parametric Mapping, a software program that allowed neuroscientists to compare PET scans with statistical tests.

Then Friston did something no other group had: he shared it with the neuroscience community, for free. The software was almost instantly adopted by the community and ~90% of neuroscientists across the globe still use it to analyze their neuroimaging data. Many studies showed that patients suffering from schizophrenia had abnormal brain activity in their prefrontal cortex, anterior cingulate cortex and thalamus, regions involved in managing decisions.

A few years later when Magnetic Resonance Imaging (MRI) became a useful method, Friston developed (free) tools to study the brain’s structure and function in greater detail than what PET could provide—tools he (and the field) then applied to understand schizophrenia. These studies showed that schizophrenia involves a network of brain regions that is progressively damaged over time. Crucially, this damage appeared to be a pernicious disconnection of individual regions within the larger network, an incarnation of 19th-century notions of a disintegrating psyche.

Friston then turned his sights on a unified theory of brain function to better understand what happens in the schizophrenic brain. He wanted this theory to provide a way to measure not simply the brain’s structure and function, but also how that structure and function ties in to the brain’s overall decision-making process, to how we perceive and behave. A central part of this theory is the free-energy principle.

The free-energy principle is something like a mission statement for the brain: minimize surprise. The sensory data gathered by our eyes, ears, etcetera is an incomplete sampling of the environment. The eye, for example, detects only a very narrow visible spectrum of light (400–700 nanometers) and yet the retina alone sends the brain a massive approximately 1010 bits (1.25 gigabytes) each second. In the larger universe, the brain is an underdog, a hapless organ meant to sift through an incomplete chaos of raw pixels and sound bites and (in real time) to navigate an uncertain and potentially hostile terrain. To do this effectively, the brain creates models.

HOW THE BRAIN MODELS THE WORLD

Imagine a forest. You’re probably not thinking of a specific forest, but your mental model of a forest. Within your “forest” model is the expectation that you’ll perceive trees (something for which you also have a mental model). Say now you’re tromping through this forest, and your brain is expecting sensory data indicative of tree trunks—stationary, vertically-oriented objects of specific colors in your visual stream.

Because you’re tromping through this forest, should you sense a series of brown pixels in a vertical pattern, your brain would generate the percept “tree trunk” and change course to avoid an unpleasant collision. (Of note, if you were sitting in your living room, you’d have a different “living room” model and a similar series of brown pixels could generate the percept “grandfather clock,” something that would be most surprising in a forest.) 

Experience has taught you that your “forest” model is so reliably stable and that you are so likely to perceive “tree trunks” while tromping in a forest-scenario, that you probably don’t even notice the trees, perceiving them subconsciously without thinking “that’s a trunk … over there’s a trunk … etcetera.” This “trunks in forest” model frees up your brain to talk to your hiking buddy or to ruminate over your day.

But what if, after you pass a “trunk,” you sense a rustling, then a growl. Then those stationary brown pixels begin to lean slightly towards you. Surprised, you look closer and notice a large shadow moving your way. Uh-oh.

Surprises are bad. A surprise means that your “trunks in forest” model failed to represent and predict reality. It means that you are tailspinning towards a state of disorder and possibly death. To minimize surprise, the brain must change something. It can update the percept (e.g. “Oh, that’s a bear!”) or change the sensory data through action (e.g. “Get away from that bear-looking thing!”)—in this case, it would probably do both.

The free-energy principle describes the brain as a machine whose goal is to predict what’s happening in the environment. Because it is predicting (and these predictions can be wrong), it is prone to statistical errors like false positives and false negatives. Perceiving a tree trunk when there isn’t one is a false positive. Not perceiving a bear is a false negative. These surprising, aberrant inferences carry different levels of cost: the cost of a “bear” false negative is quite high, while the cost of a “trunk” false positive is low. To stay alive, the brain needs to minimize surprise, or the difference between what is perceived and what is actually out there in the (potentially hungry) world.

ROSE’S FREE ENERGY

Still puzzled by Rose, I emailed Friston, who kindly accepted my trans-Atlantic consult via Google phone. With his sophisticated British accent, Friston sounded precisely as I imagined him: enlightened, brilliant. He spoke in punctuated sentences that flowed into well-formed paragraphs.

Friston is perpetually intrigued by schizophrenia: “Always, underneath all the imaging and theoretical neuroscience that I’ve done,” Friston explained, “there’s been this imperative to make it speak to psychiatry and, in particular, to schizophrenia research.” Yet despite all his work, he said only now were we starting to develop the tools to measure and understand the disorder.

I described Rose’s case, her history and our exchange in the emergency room, how she saw and spoke with her boyfriend whom she knew was dead, her quicksilver emotions that knew only extremes.

“So what was going on in her brain?” I asked.

We discussed his free-energy principle and how we form models of the world to predict how to behave in our environment. In a healthy brain, sensations are fed into a model that—based on learned, contextual expectations—generates percepts of the environment; brown vertically-oriented pixels are sensed and, in the context of a “trunk in forest” model, are perceived as tree trunks.  Should someone encounter a bear or a grandfather clock whilst tromping through the forest, they would be surprised and, in turn, update their “forest” model to include bears and grandfather clocks.

In schizophrenia, Friston suggested, the brain struggles to balance what is sensed with prior beliefs about what is expected. Because of this imbalance, the brain generates inaccurate and inappropriately stable models of its environment in the sense that it is less likely to update its models based on what is actually seen, heard and felt. In schizophrenia, the brain is so overpoweringly convinced that it should perceive trees or bears that it wills them into existence. This is the quintessential false positive that psychiatrists call a hallucination.

FREE ENERGY AND SCHIZOPHRENIA

A study recently published in Science brings experimental methods to bear on Friston’s free-energy principle. The project was led by two neuroscientists at Yale University, Phil Corlett and Al Powers (who is also a psychiatrist) in collaboration with Chris Mathys, a cognitive neuroscientist at the International School for Advanced Studies in Italy. The goal of their experiment was to train someone to hallucinate and then to measure what happened when they did.

In their experiment, participants were repeatedly shown a checkerboard paired with a tone. When they heard the tone, participants were asked to press a button; so checkerboard, tone, button; checkerboard, tone, button. After training people on this “checkerboard then tone” model, the researchers stopped playing the tone after the checkerboard. They wanted to see how different individuals would respond: checkerboard, tone, button; checkerboard, [silence], button?

As expected, healthy participants made a few mistakes; even a healthy mind is imperfect. But Powers and Corlett discovered that people with a diagnosable psychotic disorder inappropriately stuck to the trained model and were less likely to update their beliefs, even after multiple checkerboards were followed by multiple silences.

They then used Friston’s Statistical Parametric Mapping to look at participants’ brain activity, measured with functional MRI. They discovered that, in the unexpected silence following a checkerboard, people without psychosis had a surge of activity in their cerebellum, a region thought to process error signals or mismatches between what is expected and what is sensed. It appeared that the cerebellum was hard at work sorting out why the “checkerboard then tone” model had failed and why they hadn’t heard a tone.

But in people with psychosis, the activity in the cerebellum decreased the more stable a participant felt the model was. It was as if people with psychosis were so doggedly convinced of the “checkerboard then tone” model that their runaway model turned off the brain’s ability to accurately assess the environment, leading to persistent false positives.

Powers and Corlett also discovered that people who hallucinate voices (as part of a psychosis or as self-described “clairaudient psychics”) were five times more likely to “hear” nonexistent tones. They were also more certain they heard nonexistent tones. In fact, as a participant’s certainty increased, so too did the activity in brain regions responsible for auditory perception. A percept in the absence of sensation; the runaway model had gone rogue.

Applying this study to Rose, Friston explained that before Steven’s death, Rose had formed a stable model that allowed her to navigate her environment. “From this point of view,” he continued, “one would imagine that through medication and a fairly predictable, stable way of conducting her life, [Rose] was in a prodromal state,” meaning she had schizophrenia but she wasn’t experiencing symptoms. Even though Rose’s illness predisposed her to unsubstantiated percepts (i.e. hearing inexistent tones), medication had freed her brain to compare and update her model with what was actually sensed, so she couldn’t sustain delusions. With Steven and her medications, “her life was stable, it was predictable, which is to say, her generative model could predict it.”

Friston felt that Rose’s cognitive tipping point was Steven’s death. “A change or a stressful life event puts great demands on the brain in terms of modeling the world,” he said, pausing empathically. “A grief reaction is a profound remodeling of nearly every hierarchical level of your brain; of the generative model that transcends every modality—who you expect to see in terms of faces, who you touch, the food you eat.” Without Steven, Rose’s entire generative model—her routine, her expectations for every moment of every day—was upended, forcing her brain to revise how she believed her day would unfold. Without her routine, she stopped taking her medications. Without her medications, her brain lost touch with reality as her illness caused her to favor beliefs about what her environment should be like, ignoring what was actually there. And so, unsurprisingly, she began hearing Steven’s voice—because she expected it. If Rose expects to hear Stephen, then (from her point of view) hearing Stephen is the least surprising and optimal solution for her brain.

ROSE’S UNRAVELING AS A GRIEF REACTION

I arranged to speak with Corlett and Powers at a local pub to learn how their results could apply to Rose and to discuss Friston’s thoughts on grief. We met on the street one evening, so, three neuroscientists walked into a bar….

At a small, circular table in an otherwise empty backroom, we discussed Rose’s story. By any reasonable standard, one might assume that after five years of close collaboration, Corlett and Powers had coordinated their personas: both framed their stubbed faces with rectangular, plastic glasses. They both relied on their left hands and eyebrows to emphasize words and ideas when they spoke. Both were invariably eloquent. The lights were dim and jazz music played in the background as we discussed grief.

“Seeing a deceased loved one is probably the most common hallucination in the general population, across all cultures,” Powers told me. “This happens in grief—and in stress responses to all sorts of stuff. Patients have a terrible exacerbation of their symptoms in the setting of social stressors as well—losing housing, having to drop out of school, etcetera.”

Corlett—who hails from the U.K.—chimed in, building on the idea: “What one might say is: as you increase uncertainty, people start to rely on their priors more. We know this from psychology: you take people who are about to jump out of a plane for the first time and give them snowy images to look at and ask them what they see. There’s not anything there, but people say, ‘Ah, I can see a deer.’ As people do more and more jumps, their perception of deer [or whatever] decreases. Your perceptual priors change as a function of how stressed out you are and how uncertain you are about your future.”

So Rose actually had cognitively capsized. Steven’s death had rendered her model of the world no longer fit for purpose—and, in the grief and stress of her solitude (and bereft of her medication), her brain failed to learn an “after Steven” model of the world. Rose’s brain had built a model wherein Steven played a central part of her life and, based on the rigidity of her disease, she wasn’t able to relinquish or relearn that key part.

When I saw Rose in the Emergency Department, she was seeing and hearing and emotionally responding to a nonexistent Steven. At the time, I wanted to measure her delusions, but this is like checking someone’s cholesterol level after they’ve already had a heart attack. By the time Rose arrived in the emergency department, she was already psychotic, in the throes of a cognitive “heart attack.”

What would have been useful is multiple measures—similar to cholesterol levels and blood pressures over time—of how her brain interacts with reality, of whether she updates her mental models when contradicted by events and when her expectations near her perceptual tipping point, just before hallucinations appear. This data could have avoided the crisis, allowing us to increase her medications just enough to suppress those expectations, allowing her to accurately perceive—and reconnect with—her sensorium. In the end, two weeks of medication softened her mind, allowing it to learn from its surroundings. She was again well enough to return home.

As I think about her now, I wonder whether Rose became gradually less certain of Steven’s presence or if, like the dropping of a curtain, Steven suddenly disappeared from her mind.

I’m unsure which is worse.