CHDI Foundation and IBM tracking the progress of Huntington's disease using AI

The organisations have developed a predictive model that can help assess whether there's a change in a patient's motor and cognitive performance.

US-based biomedical research organisation CHDI Foundation and IBM Research have released a joint research paper revealing the development of a new artificial intelligence-based predictive model that helps determine when patients will begin to experience symptoms of Huntington's disease (HD), and how quickly these symptoms will progress.  

HD is a neurodegenerative disease that causes the progressive breakdown of brain nerve cells. But unlike other neurodegenerative diseases, HD is caused by a single gene mutation "with a striking correlation to age of motor symptom onset", according to CHDI Foundation chief clinical officer Cristina Sampaio.

"People with HD may be identified and tracked from an early age long before the onset of manifest symptoms. As a result, HD may also be a good entry point for gaining insight into the mechanisms of and the development of treatments for other neurodegenerative disorders, including Alzheimer's disease, Parkinson's disease, and amyotrophic lateral sclerosis," she told ZDNet.

The published paper, titled Resting-state connectivity stratifies premanifest Huntington's disease by longitudinal cognitive decline rate, was released after three years of joint work between the two organisations and is the second paper to be published by the pair.

IBM Watson researcher Guillermo Cecchi said the paper focused on identifying how existing functional MRI (fMRI) data can be used to train artificial intelligence (AI) models to assess whether there is a change in a patient's motor and cognitive performance.

"Part of what we're trying to do is pinpoint with more accuracy what determines a particular patient with certain genetics will experience symptoms early in life or later in life," he said.

See also: IBM unveils new AI model to predict potentially harmful drug-to-drug interactions (TechRepublic)

He continued to explain how there was also the goal of uncovering people that were at higher risk of developing symptoms earlier, so they could eventually be a target for early intervention and monitoring in order to better understand the effects and outcomes of any new drug.

"Having that in mind, what we did was show that we can take a single brain scan and have accuracy about whether that particular patient belongs in the rapidly declining population or it belongs in the slowly declining population," Cecchi said.

"The way you know whether someone is slowly or rapidly declining is by looking at them over several years, so three, four, five years, and then you measure their motor symptoms and you can see over the course of five years whether the motor symptoms were changing slowly or changing very rapidly.

"But then you would need those five years to determine whether someone is deteriorating fast or slow, so what we're showing here is all you need is a single scan -- a functional MRI -- to have very good accuracy to determine whether that particular patient belongs to the fast declining or slow declining group."

Sampaio agreed that functional MRI can provide a "rich source of information", but noted its "technical complexity, until recently, has limited its broad application".

"In our study, we show that a single cross-sectional fMRI data point can predict future progression of cognitive and motor signs and symptoms of HD. Prognostic biomarkers that predict future events, like the fMRI in our study, are used to enrich for clinical-trial participants with certain pathological features to maximise the likelihood of success," she said.

"Our study results are a first step for HD clinical trials. We now need to further validate to develop fMRI as a robust prognostic biomarker in premanifest HD."

Read: Intel and GE Healthcare's X-ray machine uses embedded AI to prioritize scans (TechRepublic)

For the research, Cecchi said based on a "couple of hundred" scans, the AI model produced around an 80% accuracy output rate.

Moving forward, IBM Research and CHDI plan to replicate the study in other hospitals.

"We show that we can take data from one hospital, learn about it, and apply it to data acquired in another hospital, and still be robust and obtain the same results," Cecchi said.

Cecchi said the goal would be to eventually have the model approved by medical bodies globally and for it to be used as a standard in the field when it comes to not only HD, but other neurodegenerative diseases as well.

Similarly, a joint study by the Epilepsy Centre at Kuopio University Hospital, the University of Eastern Finland, and Neuro Event Labs resulted in the group successfully developing an AI algorithm to help quickly and automatically assess the severity of myoclonus jerks from video footage.

The model can be used to identify and track key points in the human body of myoclonus -- brief, involuntary muscle twitching -- which is the most progressive drug-resistant symptom in patients with myoclonus epilepsy type 1.

As part of the study, 10 clinical video-recorded test panels were used and it showed that the automatic method using the model correlated with the clinical evaluation. It was also able to quantify the smoothness of movement and detect small-amplitude and high-frequency myoclonic jerks by detecting and tracking predefined key points in the human body during movement. 

Updated 13 February 2020, 9:37AM (AEDT): Correction it is CHDI Foundation. 

Related Coverage