人妻少妇专区

Machine learning lets Rochester researchers accurately identify signs of the neurological disease by analyzing facial muscles.

Every day, millions of people take selfies with their smartphones or webcams to share online. And they almost invariably smile when they do so.

To and his collaborators at the 人妻少妇专区, those pictures are worth far more than the proverbial 鈥渢housand words.鈥� Computer vision software鈥攂ased on algorithms that the computer scientist and his lab have developed鈥攃an analyze the brief videos, including the short clips created while taking selfies, detecting subtle movements of facial muscles that are invisible to the naked eye.

The software can then predict with remarkable accuracy whether a person who takes a selfie is likely to develop Parkinson鈥檚 disease鈥攁s reliably as expensive, wearable digital biomarkers that monitor motor symptoms. The researchers鈥� technology is described in .*

鈥淧arkinson鈥檚 is the fastest growing neurological disorder,鈥� says Hoque, an associate professor of . 鈥淲hat if, with people鈥檚 permission, we could analyze those selfies and give them a referral in case they are showing early signs?鈥�

Though ethical and technological considerations still need to be addressed, the has agreed to fund this novel research through a $500,000 grant, effective November of 2021.

鈥淭he foundation wants us to validate the feedback that we would give people if they did, indeed, show early signs of Parkinson鈥檚鈥攅specially if they are performing the test at home,鈥� Hoque says. 鈥淭he challenge is not only validating the accuracy of our algorithms but also translating the raw machine-generated output in a language that is humane, assuring, understandable, and empowering to the patients.鈥�

Software analyzes facial expressions, hand movements

Smiles are not the only behaviors that Hoque and his lab can analyze for early symptoms of Parkinson鈥檚 disease or related disorders.

In collaboration with 鈥攁 leading expert in Parkinson鈥檚 disease and the David M. Levy Professor of Neurology at Rochester鈥攁nd the University鈥檚 , the researchers have developed a five-pronged test that neurologists could administer to patients sitting in front of their computer webcams hundreds of miles away.

This could be transformative for patients who are quarantined, immobile, or living in underdeveloped areas where access to a neurologist is limited, Hoque says.

In addition to making the biggest smile, and alternating it with a neutral expression three times, patients taking the test are also asked to:

• Read aloud a complex written sentence
• Touch their index finger to their thumb 10 times as quickly as possible
• Make the most disgusted look possible, alternating with a neutral expression, three times
• Raise their eyebrows as high as possible, then lower them as far as they can, three times slowly

Using machine learning algorithms, the computer program shows鈥攚ithin minutes鈥攁 percentage likelihood from each of the tests whether the patient is showing symptoms of Parkinson鈥檚 disease or related disorders.

What exactly does the program look for? When patients are making a smile, the software can detect whether they show less control over their facial muscles while doing so, a symptom of Parkinson鈥檚 that clinicians refer to as 鈥渕odularity.鈥�

鈥淥ne thing about Parkinson鈥檚 is that you don鈥檛 show all the symptoms all the time, and not every symptom is shown in every part of your body,鈥� says Rafayet Ali 鈥�20, lead author of the paper. 鈥淔or example, you may not have hand tremors, but you may show a significant level of deviation in your smile.鈥�

Hence the importance of testing other expressions and movements, according to Ali, a former postdoctoral associate in Hoque鈥檚 lab who now is an associate data scientist at Sysco.

From pen-and-paper evaluations to 鈥榦bjective, digital assessments鈥�

Both Hoque and Ali have personal stakes in helping people with Parkinson鈥檚. Their mothers both have suffered from the disorder. Hoque鈥檚 late mother in Bangladesh, for example, was put on leovodopa, a leading medication for the disorder, after finally finding one of the country鈥檚 few neurologists. The tremors went away. 鈥淲e were so happy,鈥� Hoque says. Unfortunately, it was difficult to make follow-up appointments, and the tremors eventually returned.

That prompted Hoque to email Dorsey, 鈥渏ust to casually chat.鈥� When they finally met in 2016, Hoque recalls, Dorsey 鈥渢ook a big document and just threw it on the table.鈥� The pamphlet contained forms physicians need to fill out as part of the Movement Disorders Society鈥揢nified Parkinson鈥檚 Disease Rating Scale (MDS-UPDRS).

鈥淭hat鈥檚 the gold standard for measuring Parkinson鈥檚,鈥� Dorsey told him. 鈥淓verything we do is pen and paper. Any automation, any data analytics that you can bring into this would be a contribution. And he immediately helped us see how we could do that,鈥� Hoque says.

鈥淥bjective, digital assessments of Parkinson鈥檚 disease can help us diagnose people with the condition and evaluate new therapies for the condition faster,鈥� says Dorsey, an author of (2020).

Progress toward FDA approval

It will be a while yet, however, before Hoque and his researchers can start seeking permission to analyze people鈥檚 selfies, or even before neurologists can deploy the five-pronged test that the researchers have developed.

鈥淎n algorithm will never be 100 percent accurate,鈥� Hoque says. 鈥淲hat if it makes a mistake? We want to be very careful and follow guidance from the FDA if we want anybody from any part of the world to try this and get an assessment.鈥�

Moreover, there is a whole family of movement disorders that are closely related to Parkinson鈥檚 disease, including ataxia, Huntington鈥檚 disease, progressive supranuclear palsy, and multiple dystrophy.

鈥淭hey all share similar symptoms of tremor, but the tremors are very different in nature,鈥� Hoque says. 鈥淗owever, even expert neurologists find it very, very difficult to distinguish among them.鈥�

The researchers have made great progress in detecting Parkinson鈥檚 disease by automatically analyzing expressions, voice and motor movements. Yet further work is needed to develop algorithms to differentiate how these involuntary tremors differ across other movement disorders, including Ataxia and Huntington鈥檚.

鈥淲e can鈥檛 tell that just yet,鈥� Hoque says. 鈥淏ut we are in a pursuit of differentiating those tremors using AI to prevent the potential harm of misdiagnosis while maximizing benefit.鈥�

And speaking of timely Parkinson鈥檚 disease diagnosis . . .

Turns out it鈥檚 not just selfies and videos that can help with diagnosing Parkinson鈥檚 disease.

More and more, people are using speech-activated smart devices, such as Alexa, Apple Watch, and Google Voice Assistant, to accomplish everyday tasks. Could these devices analyze our speech and voices to alert us if we show early warning signs of Parkinson鈥檚 disease?

Recent work by Rochester researchers suggests it鈥檚 entirely possible. Wasifur Rahman, Sangwu Lee, Md. Saiful Islam, and other students in Hoque鈥檚 lab published findings in the Journal of Medical Internet Research that show how an online tool can be used to help screen almost anyone anywhere for Parkinson鈥檚 disease remotely using video- or audio-enabled speech tasks.

Taken together, the Rochester researchers鈥� efforts are contributing to a future in which equity and access to neurological care is as ubiquitous as owning a smart phone or other internet-enabled device.

*Editor鈥檚 note: As of February 2023, the authors have retracted the article published in聽npj Digital Medicine聽that established聽the relationship between smiles and the early onset of Parkinson鈥檚 disease because the classification described in the paper was performed on an inaccurate use of a data pre-processing tool. Please see the authors鈥櫬� for more information. In June 2025, a was published in NEJM AI聽demonstrating that smiling videos can effectively differentiate between individuals with and without Parkinson鈥檚 disease.