Artificial intelligence meets human intelligence to collaborate on addressing brain health
Researchers agree that early detection is the key to optimising the potential success of drug and device therapies – and that these therapies would work most effectively if they were administered when a patient first begins to develop a disease rather than waiting until when they are displaying obvious symptoms. However, it is extremely difficult to observe exactly when this process starts to occur.
Neurodegenerative diseases, such as Parkinson’s, for example, can currently only be detected by the presence of substantial and bothersome motor difficulties – but by the time a patient is exhibiting such symptoms, in some cases almost 70% of brain function may have already been lost. In contrast, subtle, often unseen motor symptoms manifest much earlier and offer the chance to detect patients at a more opportune time.
Likewise, in diseases such as Alzheimer’s, cognitive decline in certain subtypes can begin as early as age 45 and it is often ignored and/or labeled “Old-Timers Disease” until such time as the person becomes incapacitated in later stages.
As with many other unmet needs in healthcare, there are now several emerging digital innovations that hope to drive breakthroughs in neurological disease detection.
Digital health tools are not uncommon in other therapeutic areas. The app BiliScreen, for example, uses a phone camera to measure the bilirubin level in a person’s eyes for early diagnoses of pancreatic cancer. Meanwhile, ResApp Health uses the phone’s microphone to analyse coughing sounds to accurately diagnose pneumonia and other upper respiratory diseases.
Nevertheless, neurology remains one of the more difficult disease areas in which to develop any kind of tool.
“One difficulty in neurology is that even severe ongoing pathology in the brain may initially cause only subtle patient symptoms that are not outwardly obvious,” says Rahul Mahajan, MD, PhD, neurologist at Massachusetts General Hospital. “There is a definite need for reliable biomarkers for brain maladies. Tools or solutions that can precisely measure and uncover more about disease onset or disease progression will be critical in developing new cures and better managing patient treatments.”
Currently, HCPs and researchers mostly rely on clinical exams and medical imaging to diagnose neurodegenerative diseases with very few other diagnostic tests or markers available to establish diagnosis early, quickly, and accurately. This is contrary to many diseases in other areas where specific biomarkers allow us to diagnose with more certainty the condition of the patient.
“Neurology is a very attractive field to work in because there is so much potential for digital solutions to address these huge unmet needs,” says Teresa Arroyo-Gallego, a researcher in the biomedical engineering lab of the Massachusetts Institute of Technology.
Digital steps up
As one might expect of any challenging field, researchers have come up with some unconventional but exciting solutions that show promise.
For example, researchers from Universidad Politécnica de Madrid, Massachusetts Institute of Technology (MIT), and Johns Hopkins University have been testing voice recognition as a potential way to identify Parkinson’s disease at an early stage.3
Algorithms may be able to detect specific variations in sound vibrations linked to vocal tremors, breathlessness, and weakness, all of which are associated with neurological degeneration.
Their machine learning models learn the common differences in typing patterns between patients with neurodegenerative disorders and healthy controls.
“When your motor function is not affected by Parkinson’s, you tend to be more rhythmic in your typing. We see that the distribution of the metrics we’re evaluating are very stable over time. In Parkinson’s patients, distributions tend to be more spread out. They also show some unusual characteristics – for example Parkinson’s disease tends to affect one side of the body more than the other, and this asymmetry can be picked up in their typing.
The challenge, of course, is that most voice recognition software in consumer products isn’t of high enough quality to compare with voice recognition performed in a lab – and is likely to pick up more ambient noises in everyday life than one would find in a clinical study.
Meanwhile, research by the neuroQWERTY team at MIT, where Arroyo-Gallego developed her PhD research, took a similar but perhaps even simpler approach – monitoring people’s typing on their phones or computers to detect the early signs of motor decline in diseases such as Parkinson’s4.
“We’re analysing keystroke dynamics – information on how you are pressing and releasing keys or touchscreen touches,” she explains. “Sometimes Parkinson’s disease patients are unable to release the key quickly even when the fingers are getting the signal from the brain to do so. This is one of the very clear manifestations of PD-related motor decline in the typing patterns.”
The technology spun out from MIT in 2016, into a company founded as nQ Medical. But rather than rushing the computational biomarker to market, nQ spent two years meeting and interviewing the various healthcare stakeholders in neurological care – providers, payors, clinicians, patients, caregivers, and industry – to best understand the value proposition for the technology.
“We found that it was perfectly applicable to Parkinson’s, where there is an underserved population who might only see a clinician once or twice a year and where patient outcomes could be improved,” says the company’s CEO, Richie Bavasso. “The gold standard for assessment is patient self-reporting during that limited clinician intervention. nQ can show clinicians, quantitatively, what is happening to patients between those visits, thus optimising the opportunity to provide the patient better and more precise care.”
Details
Not every digital solution can rise above the hundreds of thousands of apps on the market and make a true impact on outcomes. Bavasso has his own thoughts on the key factors for making a successful digital solution and how this might apply to neurology patient care.
He says it’s important that the technology only uses existing devices, like the person’s own smartphone, laptop, desktop, and/or tablet.
“There are so many devices out there, with more being introduced every day, that it’s overwhelming for all of us,” he says. “You can easily enable health solutions on hardware that is already used by people. Requiring patient (or clinicians) to buy a proprietary device gets in the way of them participating and benefiting from technology.”
Secondly, everyone in healthcare knows that patient adherence is a huge issue in delivering quality care and outcomes. Requiring patients to perform tasks each day (or in some case throughout the day) is a recipe for non-adherence.
“I am 100% compliant in my own daily ritual and routines,” Bavasso explains. “If you want to have me as a participant in your digital healthcare solution, you have to adhere to me. I’m not going to adhere to you. The devices we touch every day can all be medical devices.
He adds that passive technologies like this can also be useful for clinical research.
“Most diagnostics and measuring tools are episodic. A reading is achieved only when the patient is in front of the clinician. Digital technologies can record clinical readings many times a day, every day. In nQ’s case, we record a score every 90 seconds for the several hours a day a person’s personal devices are used.
“nQ can be used for trial recruitment, determining if someone is eligible for a clinical trial from their home by having them type something on their device from anywhere. It provides real-world data during the course of the trial, and immediately measures the impact of the drug or device that is being tested and how it changes over time.”
Using passive data collection, one can detect presence of disease at the closest date of phenoconversion, remotely track disease progression over time, and measure the impact of therapy, drug, device, or other.
Bavasso stresses the importance of ensuring digital solutions are useful for clinicians.
“You need to ask if you have a tool that the clinician needs and will use. As part of our SaMD application for clearance, the FDA required us to articulate and demonstrate how clinicians will use our tool in everyday patient care.”
“While use of passive versus active patient participation addresses issues with adherence, it is also important that the measurement data is meaningful to clinicians and presented in a way that helps them efficiently and accurately interpret results,” adds Dr Mahajan. “Integration into the EMR which physicians are already using and ability to contextualise with other clinical data would be preferable.
“Like any other clinical tool, digital clinical applications should be appropriately validated with clinical testing and real evidence. This important bar separates clinical digital tools used to inform important healthcare decisions in a doctor’s office or a clinical trial from more recreational health applications.”
Bavasso adds: “At nQ, we have chosen the longer, harder path of seeking clinical validation via true trials supervised in appropriate academic environments, regulatory validation via the appropriate government authorities, and market validation to ensure clinician and patient adoption. I think it’s the right path to choose for technologies like this.”
Bavasso believes that the pharma industry has matured in its evaluation of digital technologies and is now more thoughtful and demanding about what these technologies can deliver. Referring to the Digital Health Alliance, he states that the industry itself has come together to better define and establish standards for digital technologies.
He expresses hope that simpler digital health tools will help the general public embrace health monitoring to ultimately improve outcomes.
“I often start my presentations by asking how many people in the audience know their blood pressure. Everybody raises their hands. Then I ask how many know their cholesterol level. A large number of people raise their hands. Finally, I ask how many people know the state of their brain health, and no one raises their hand. We hope to change that.
Customer comments
No comments were found for Artificial intelligence meets human intelligence to collaborate on addressing brain health. Be the first to comment!