AI can spot the pain from a disease some doctors still think is fake – By Olivia Goldhill – August 9, 2018
In the 16 years since Liptan had her illness [fibromyalgia] so summarily dismissed in 2002, there are still those who believe fibromyalgia isn’t “real.”
There’s no tissue damage that explains the pain fibromyalgia patients experience all over their body, and contemporary medicine struggles to treat and even accept an illness where pain seems to be rooted in the mind or brain, rather than a bodily injury.
Artificial intelligence, though, has the potential to make a diagnosis in minutes. Last year, researchers used machine learning to distinguish the brain scans of those with fibromyalgia from those without—with 93% accuracy.
The implications are immense:
Decoding the brain signature for fibromyalgia could hold the key to understanding the disease and which treatments work for which patients. But it’s also a definitive, objective sign that fibromyalgia really does exist.
There’s no accepted criteria for diagnosing fibromyalgia. There is no known biological malfunction, nor is there any biomarker that can be uncovered in a lab test.
Patients experience pain all over their body, fatigue, insomnia, difficulty focusing, depression, and 18 “tender points”—including the back of the neck, elbows, and knees—that are sore to the touch. Antidepressants, painkillers, physical therapy, acupuncture, massage, counseling, and exercise are all used to treat the condition, with varying effects.
Both patients and researchers believe skepticism around the illness partly reflects gender discrimination: Close to 90% of fibromyalgia patients are women.
Every week, Liptan argues with insurance companies about whether her patients really are in pain and if fibromyalgia is a legitimate condition. “For most of my patients, that’s the hardest thing for them. It’s not the illness but the judgment that they get from other people,” she says.
“People with fibromyalgia look fine; our lab tests are fine. Maybe we’re just lazy [and] we should suck it up?” The millions of people (an estimated 3% to 6% of the population) worldwide who experience fibromyalgia know that their pain is real.
Proving this to others, though, is a perpetual challenge.
The researchers who successfully used machine learning to identify fibromyalgia patients started by using fMRI machines to capture images of the brain signals of 37 fibromyalgia patients and 35 healthy people used as a control group.
All the participants had pressure applied to their right thumbnail to create “severe but tolerable pain,” explained the researchers in their paper, published in the journal Pain last year.
Those with fibromyalgia experienced more pain compared to the healthy controls, according to a neurological signature of physical pain, as well as different activity in the insula area of the brain, related to sensory integration, and the medial prefrontal cortex, which is important for emotional regulation.
Collectively, these different neurological responses created a brain signature for fibromyalgia patients.
A machine-learning algorithm that was programmed to recognize this neurological signature was able to use it to predict which brain scans were indicative of fibromyalgia and which were not.
As such, neuroimaging combined with artificial intelligence was able to create an objective snapshot of what, to date, has been characterized as a subjective sensation. It made perceptible an experience that was previously unknowable to anyone but the patient.
This study was small, and it will take years, likely at least a decade, before such techniques can be used in a clinical setting. The findings from this small dataset cannot be extrapolated or applied to other patients, and so researchers will need to repeat the process with thousands more.
I also immediately noticed how small the dataset was and I wonder if the results of this study can be generalized.
The lack of clear biomarkers can make it difficult for doctors to definitively diagnose and treat fibromyalgia, but López-Solà’s work provides hope that researchers are on the right path.
There are still providers who don’t understand it or think it’s real.”
By now, this attitude is inexcusable.
Any patient who has a doctor who still believes it’s unreal should hurry to find another doctor. Who knows what other outdated beliefs this doctor may harbor?
Refusing to accept newer medical and research findings could lead to dangerous treatments.
Some old-fashioned physicians may never change their preconceptions that conditions such as fibromyalgia exist, says Clauw. “The classic biomedical model is there’s something broken and you find it, you do surgery and repair it and the person’s better. A lot of illnesses don’t follow that model.
And this is the complexity of modern medicine: if it were only about diagnosing and treating by standard recipes, it could be done by computer algorithms. But this complexity and ambiguity is exactly why we still need educated and experienced doctors.
Machine learning was an essential tool in distinguishing fibromyalgia patients from healthy controls.
“If we hadn’t used machine learning, we wouldn’t have been able to identify the pattern of brain activity that was most predictive of patient status, and we would not have been able to say anything about an individual patient,” says López-Solà. Machine learning was the only way to do such deep and wide-raging statistical analysis on brain activity.
It’s not yet clear whether the brain signals symptomatic of fibromyalgia are reflecting
- the experience of pain,
- the brain malfunctions that cause the pain, or
- some combination of the two.
In general, patterns in the physical brain both cause and represent the sensations that play out in the mind. And so a biomarker that can definitively identify fibromyalgia should help patients get the serious consideration and treatment they need.
Every mental experience is “the consequence of something that’s happening at another level, the physiological processes,” says López-Solà. “The emerging properties of those underlying processes are what we refer to as the mind.”
Some fibromyalgia patients experience stronger depressive symptoms whereas others struggle more with the pain or difficulty concentrating. It’s likely different brain mechanisms cause the various symptoms of the disease, and López-Solà hopes machine learning will eventually be able to tell which patients are suffering which symptoms.
Why not just ask patients what symptoms they are experiencing?
Answer: because the symptoms of physical and mental distress and suffering are invisible.
Though neuroimaging provides new information, it still doesn’t offer a perfectly accurate picture.
Analyzing the neurological responses of thousands more patients would certainly increase accuracy, but reading any given individual brain is a complicated and messy business, and brain signatures provide guidance at best, and certainly not a definitive checklists of symptoms.
So, subjective experiences like pain cannot be objectively determined, no matter how close we get. What the human reports should still always be considered, no matter what the computer says.
Researchers are also adamant that machine learning should not be used to claim that any one patient is free from pain.
This is extremely important, but will likely be overlooked.
“In no way could this be used for not giving people treatment. We have false negatives—we have people who have pain but do not show expression of the marker,” says López-Solà. “It would be deeply wrong, unethical and technically incorrect to use a negative response in these markers to not give treatment to patients.”
But as we know from our recent experiences, the “hard data” of numbers or “objective” computer classifications will absolutely be used to deny treatment.
If the computer finds no patterns of pain, then no matter what you say differently, your words will be disregarded and only the computer’s decisions will be believed.
This is the posture medical science already takes towards other difficult-to-diagnose diseases: An ultrasound might show signs of endometriosis lesions, for example, but doctors know that a clear scan doesn’t rule out the condition, either. The point is that even with all our tools, it’s difficult to accurately look inside the body.
But it’s still easier to believe (and document) computer results than a patient’s story.
Though plenty can be learned from objective data, medical researchers must not forget the value of subjective experience. Liptan points to earlier research showing that patients’ self-reports of pain are the most accurate means of classifying patients with fibromyalgia
Nowadays with all the focus, analysis, employment decisions, and even DEA prosecutions dependent on data entered into EHRs (Electronic Health Records), “hard data” will always be given a higher priority than a patient’s words.
The value of doctors listening to their patients is a cornerstone of any medical treatment. But many doctors fail to do so and, even in cases where they are willing, such conversations can be impossible.
And there are plenty of other diseases that, like fibromyalgia, don’t necessarily come with clear-cut physical symptoms. Those who suffer from conditions such as Lyme disease, back pain, and chronic fatigue syndrome know what it’s like to be in terrible pain without being able to point to the physical cause.
Patients who fall in any of these categories are likely to benefit from artificial intelligence’s ability to perceive their suffering.
But it cannot. No external scan or analysis can determine what a human is feeling, physically or mentally.
The effort to objectively capture a physical image of pain is, in some ways, inherently at odds with with the amorphous, subjective experience of feeling pain.
But the way patients know their symptoms is necessarily different from doctors’ knowledge of their condition. And allowing machines to peer inside mental experiences means that, for the first time, patients whose pain is largely inside the mind can know that their suffering is truly seen.
As you may have guessed, I highly doubt this.