Healthcare

Understanding Parkinson’s Disease with Machine Learning and The Michael J. Fox Foundation

Share this post:

Parkinson’s disease (PD) is a chronic, degenerative neurological disorder that affects as many as one in a 100 people over the age of 60. It is estimated that more than five million individuals have PD worldwide, and the number is growing with today’s aging population. [1]

While primarily characterized by motor disturbances — including involuntary tremors and impaired movement — it is not uncommon for PD patients to also experience cognitive declines, behavioral issues, and sleep disorders.

The Michael J. Fox Foundation for Parkinson’s Research has provided a grant to IBM to try to better understand PD and the route it can take in patients, with the goal of paving the way for more effective treatments. Through this partnership, The Michael J. Fox Foundation is making available its data from the Parkinson’s Progression Markers Initiative (PPMI), an observational study that has collected a vast amount of anonymous longitudinal data across cohorts of PD patients.

IBM’s tiny fingernail sensor prototype uses AI to help clinicians better manage Parkinson’s disease.

This is IBM’s next step in our quest to better comprehend and track neurological conditions such as PD. A few weeks ago, we announced a new fingernail sensor prototype, which could one day help clinicians to continuously track, monitor and more accurately diagnose movement and neurodegenerative disorders with the help of data and AI.

Additionally, throughout this past year, we’ve published research which uses AI and machine learning to better detect and comprehend changes in patient’s speech, which can indicate markers of PD progression. This includes:

  • The progression of PD is marked by an impairment in processing action verbs, verb generation and comprehension of complex sentences. We’ve recently published a study about experiments with machine learning methods to analyze these changes and to automatically distinguish Parkinsonian patients from healthy controls. Our system could identify PD patients with an accuracy of 81 percent and may aid professionals in treatment planning.
  • The effectiveness of PD medications can induce subtle changes in a patient’s speech. We’ve worked to build machine learning algorithms that can detect and analyze these fluctuating changes in a patient’s speech on a long-term and objective basis and predict with a high accuracy the effect of medications in treating the symptoms of the disease.

IBM’s collaboration with The Michael J. Fox Foundation aims to provide a comprehensive view of the disease through longitudinal, clinical, behavioral and imaging assessments observed from patients in clinical settings, as well as genomic and biological samples such as genetic and epigenome data. The Michael J. Fox Foundation was instrumental in collecting and making this data available in a responsible way, and the broad nature of this information provides an unprecedented opportunity to gain insights into PD and its progression.

Though the potential of applying AI and machine learning techniques to learn from this data is vast, there are still challenges. Specifically, the heterogeneity of disease progression across patients, the multi-faceted nature of PD symptoms (motor, cognitive and behavioral), and the confounding factors of symptom-alleviating medications make this a formidable challenge.

To overcome these obstacles, we plan to design machine learning methods that explicitly account for the effects of medication and other confounding factors by drawing on our previous work in disease progression modeling to analyze the progression of PD.

Our hypothesis is that observed PD symptoms are a manifestation of an underlying disease process. Statistical models that account for latent variables – or hidden representations that can be inferred from statistical regularities in the data – are particularly attractive instruments for capturing such assumptions [2]. Hidden representations are learned by maximizing the likelihood of a model and can provide a succinct summary of PD symptoms that may correspond to distinct disease stages. Additionally, our models will make it possible to estimate the probability of transitioning between these representations over the course of the disease.

Thus, once learned, these statistical models will allow us to both stage PD patient profiles and provide predictions of their future progression. Moreover, because the models provide a quantitative description of progression, they may allow us to discover sub-groups of PD patient profiles that share a common progression pathway through the disease and may also help us in the identification of biomarkers that are predictive of progression.

More broadly, insights arising from an improved understanding of PD progression may have the potential to transform the care of PD patients. For instance, accurate staging can aid in the recruitment of subjects to clinical trials for new drugs, while the discovery of PD sub-groups can help inform more personalized treatments and may hopefully improve quality of life and outcomes.

Ultimately, all of our work will ladder up to the hope and promise of the possibility that one day we may accurately be able to predict the onset and progression of PD. The ability to accurately predict the progression of PD can help with early detection and, through appropriately timed interventions, allow PD patients and their care providers to better manage the disease.


  1. The Michael J. Fox Foundation: https://www.michaeljfox.org/understanding-parkinsons/living-with-pd/topic.php?causes
  2. “Unsupervised learning with contrastive latent variable models”, Kristen Serverson, Soumya Ghosh, Kenney Ng, to appear in AAAI 2019, Jan. 27 – Feb. 1, 2019

Research Staff Member, IBM Research

More Healthcare stories

Dark Matter Matters: AI Makes DNA Dark Matter Useful

What is the minimal description that captures a space? Asking a mathematician’s basic question of a  biological dataset reveals interesting answers about biology itself. This summarizes our underlying approach to subtyping hematological cancer. Disease subtyping is a central tenet of precision medicine, and is the challenging task of identifying and classifying patients with similar presentations […]

Continue reading

Helping to Untangle Cancer Drug Resistance with Data

Why do targeted cancer therapies often fail? We have acquired so much more understanding about cancer in the last fifty years than in the last five thousand years. Approaches to patient treatments have dramatically changed, and statistics show significant improvement in patient response and outcomes to therapy in the last half a century [1]. Yet […]

Continue reading

Novel AI tools to accelerate cancer research

At the 18th European Conference on Computational Biology and the 27th Conference on Intelligent Systems for Molecular Biology, IBM will present significant, novel research that led to the implementation of three machine learning solutions aimed at accelerating and guiding cancer research.

Continue reading