AI

AI Could Help Enable Accurate Remote Monitoring of Parkinson’s Patients

Share this post:

In a paper recently published in Nature Scientific Reports, IBM Research and scientists from several other medical institutions developed a new way to estimate the severity of a person’s Parkinson’s disease (PD) symptoms by remotely measuring and analyzing physical activity as motor impairment increased. Using data captured by wrist-worn accelerometers, we created statistical representations of PD patients’ motor movement that could be objectively evaluated using AI either in-clinic or from a more natural setting, such as a patient’s home.

The human motor system typically relies on a series of stereotyped unit of movement to perform a given task, such as arm swigging while walking. These discreet movements and the transitions linking them create a pattern of physical activity that can be measured and analyzed for signs of PD, an incurable neurodegenerative disease estimated to affect nearly one million people this year in the U.S. alone1. Measurements taken from PD patients deviate from those found in non-patients, and growth in those deviations marks the disease’s progression over time.

Existing approaches for evaluating patients for PD are limited. Doctors typically evaluate patients once or twice a year in a supervised clinical setting2 and make subjective assessments based on a standardized rating scale, known as the Movement Disorder Society’s Unified Parkinson’s Disease Rating Scale (or MDS-UPDRS)3. Such examinations tend to rely on patient-reported information that, combined with physicians’ interpretations of motor impairment (evaluated in MDS-UPDRS – Part III), could potentially lead to biased results.

In our study—conducted jointly with the Pfizer Innovation Research Lab, Boston University’s Spivack Center for Clinical and Translational Neuroscience and Tufts Medical Center’s Department of Neurology—we demonstrated an unsupervised technique that could be used on PD patients in their homes or in a doctor’s office to generate objective measurements of movement quality. Continuous signals from wearables were transformed in a sequence of “syllables” (see Figure 1). Those sequences – common across healthy subjects – are part of our learned motor repertoire with subsequences shared across different actions. The derived action-independent statistical distribution of ordered transition between syllables was a signature of healthy behavior.

Figure 1 – Arm swigging during walking can be converted to a sequence of discrete movements recorded at the wrist.

Disorganized sequences of symbols were observed in Parkinsonian patients (Figure 2). By means of AI, we were able to estimate both the gait impairment (PIGD, see Figure 2 – color coded in blue/green) and the overall severity of Parkinson’s symptoms (MDS-UPDRS Part-III, see Figure 2 – color coded in pink/yellow) by capturing increasingly disorganized transitions between movements as motor impairment increases.

Figure 2 – Discreet movements in a mild participant (ON drug and OFF drug) and in a severe participant. Upper raw sequence of movements over time during a walking tasks. Lower raw, estimation of postural instability and gait disturbance (PIGD) and overall disease state (total MDS-UPDRS Part-III) of the model vs. neurologist assessment.

Our Nature Scientific Reports study was part of IBM Research’s Bluesky Project with Pfizer, launched in 2016 which aimed to develop a system to improve clinical trials conducted for PD drugs in development. IBM Research’s specific role has been developing new algorithms that enable AI to analyze data collected from study participants. Bluesky’s basic premise was to digitize clinical trials to create a more accurate way of assessing patients than the traditional approach of having them self-report. Self-reporting can be especially problematic for PD patients experiencing cognitive impairment.

For the study, we applied our AI algorithms to data from three Bluesky studies that collected sensor data from people in three categories – 1) individuals diagnosed with PD undergoing the standard neurological exam, 2) healthy participants undergoing the same protocol and 3) people with PD in unconstrained behavior at home. We developed a new mathematical model to extract value out of the sensor data, which provides a way to objectively monitor and measure disease progression and movement quality in Parkinson’s patients in an unsupervised setting. This can allow a neurologist to compare patient evaluations both in a clinical setting and at home. Our approach also proved highly efficient—it required data from less than 10 minutes of activity on average, to create stable estimates: it allows a continuous 24/7 evaluation of the neurological state. This is particularly important when patients have fluctuations of their pathological state during the day.

At a time when there is increasing interest in expanding telemedicine capabilities to enable patients especially vulnerable to COVID-19 to remain at home, our research demonstrates how a neurologist could accurately evaluate PD patients remotely. The added benefit to such a scenario is that telemedicine checkups could be performed more frequently than is possible when patients are required to visit a doctor’s office.

Our current research is part of a larger body of work at IBM Research to study how data science and technology can potentially help improve the study of neurodegenerative diseases, including Huntington’s disease, which shares a lot of similarities with Parkinson’s. For instance, another recent publication by our team in Nature Partner Journals – Parkinson’s Disease, also part of the Bluesky project, demonstrated the ability to determine whether patients’ ingestion of levodopa, a palliative drug to substitute dopamine, has taken effect by measuring acoustic and content properties of simple and short speech samples.

Looking ahead, although we explicitly tested our proposed model on movement differences associated with PD, we believe this could be the basis for potentially helping to detect other neurological states with characteristic movement signatures.

Sources:

  • Prevalence of Parkinson’s disease across North America, npj Parkinson’s Disease 4, 21 (2018).
  • Standard strategies for diagnosis and treatment of patients with newly diagnosed Parkinson disease, Neurology: Clinical Practice,
  • The MDS-sponsored Revision of the Unified Parkinson’s Disease Rating Scale, International Parkinson and Movement Disorder Society, August 2019

Program Manager, Emerging Technology Experiences (ETX)

Avner Abrami

IBM Research

More AI stories

AI goes anonymous during training to boost privacy protection

Our team of researchers from IBM Haifa and Dublin has developed software to help assess privacy risk of AI as well as reduce the amount of personal data in AI training. This software could be of use for fintech, healthcare, insurance, security – or any other industry relying on sensitive data for training.

Continue reading

Moving beyond the self-reported scale: Objectively measuring chronic pain with AI

Together with Boston Scientific, we are presenting research that details the feasibility and progress towards our new pain measurement method at the 2021 North American Neuromodulation Society Annual Meeting.

Continue reading

Who. What. Why. New IBM algorithm models how the order of prior actions impacts events

To address the problem of ordinal impacts, our team at IBM T. J. Watson Research Center has developed OGEMs – or Ordinal Graphical Event Models – new dynamic, probabilistic graphical models for events. These models are part of the broader family of statistical and causal models called graphical event models (GEMs) that represent temporal relations where the dynamics are governed by a multivariate point process.

Continue reading