Israeli machine learning contest breaks new ground in Parkinson’s disease research

One of the most intriguing findings from the study was a newly discovered relationship between FOG frequency and the time of day.

By Pesach Benson, TPS

A machine learning contest initiated by Israeli researchers produced remarkable results with wearable sensors that could significantly advance the monitoring and treatment of Parkinson’s disease.

The project, spearheaded by researchers at Tel Aviv University’s Faculty of Medical & Health Sciences, aimed to aid neurologists by creating a continuous and automated system for tracking freezing of gait (FOG) episodes — a debilitating symptom affecting many Parkinson’s patients.

FOG, which affects 38-65% of Parkinson’s sufferers, is a phenomenon where patients experience sudden, temporary inability to walk. These episodes can last from a few seconds to over a minute, during which the patient’s feet seem “glued” to the floor.

“FOG can seriously impair the mobility, independence, and quality of life of people with Parkinson’s disease, causing great frustration, and frequently leading to falls and injuries,” explained Prof. Jeff Hausdorff, who led the study.

Currently, the diagnosis and tracking of FOG largely depends on self-reported questionnaires, visual observation by clinicians, or frame-by-frame video analysis of patients in motion.

While video analysis is reliable and accurate, it is also time-consuming, requires multiple experts, and is impractical for long-term monitoring.

“Researchers worldwide are trying to use wearable sensors to track and quantify patients’ daily functioning,” explained Amit Salomon, one of the study’s co-authors. “So far, however, successful trials have all relied on a very small number of subjects.”

To overcome these limitations, the TAU team collected data from various existing studies, compiling information on over 100 patients and approximately 5,000 FOG episodes.

This data was uploaded to Kaggle, a Google-owned platform known for hosting international machine learning competitions. Researchers worldwide were then invited to develop models to be incorporated into wearable sensors that could quantify FOG episodes in terms of duration, frequency, and severity.

The response to the competition was extraordinary: 1,379 teams from 83 countries submitted nearly 25,000 solutions. The prize for the best solutions, totaling $100,000, was funded by Kaggle and the Michael J. Fox Foundation for Parkinson’s Research.

The contest “brought together capable, dynamic teams all over the world, who enjoyed a friendly atmosphere of learning and competition for a good cause,” said Hausdorff.

The models developed in the contest performed impressively, with results closely matching those obtained through video analysis—currently the gold standard. In fact, the machine learning models outperformed previous trials relying on single wearable sensors.

One of the most intriguing findings from the study was a newly discovered relationship between FOG frequency and the time of day.

Eran Gazit, another co-author, noted that “we observed, for the first time, a recurring daily pattern, with peaks of FOG episodes at certain hours of the day.”

This pattern may be linked to clinical factors such as fatigue or the effects of medications, offering new insights into the treatment and management of FOG.

The findings were recently published in the peer-reviewed Nature Communications.

“Wearable sensors supported by machine learning models can continuously monitor and quantify FOG episodes, as well as the patient’s general functioning in daily life. This gives the clinician an accurate picture of the patient’s condition at all times,” Hausdorff explained.

Hausdorff said the competition’s results will lay the foundation for long-term, 24/7 FOG monitoring in patients’ homes and real-world environments, with data collected through the technology paving the way for developing new treatments.

FOGParkinson’s diseasewearable sensors