New Video-based Algorithm May Help Predict Tremor in Parkinson’s

Tool found to ID tremors in hands, legs, with up to 90% accuracy

Lindsey Shapiro, PhD avatar

by Lindsey Shapiro, PhD |

Share this article:

Share article via email
This illustration is a banner with the words

Researchers have developed a video-based algorithm to predict the severity of tremor in Parkinson’s disease patients based on the criteria of the Movement Disorder Society-Unified Parkinson’s Disease Rating Scale (MDS-UPDRS).

The new algorithm was found to identify tremors, both resting and postural, with high accuracy — ranging from 85% to 90% for various body parts, including the hands, legs, and jaw.

According to the team, their method offers a way of more easily monitoring this common Parkinson’s symptom, helping to ease the burden on doctors and patients alike.

“Our study demonstrated the effectiveness of computer-assisted assessment for PD [Parkinson’s disease] tremors based on video analysis,” the researchers wrote. “With this method, doctors can easily and quickly measure the severity of the patient’s symptoms to formulate better treatment plans.”

The study, “Vision-based estimation of MDS-UPDRS scores for quantifying Parkinson’s disease tremor severity,” was published in Medical Image Analysis

Recommended Reading
An illustration of a person presenting data to an audience.

Algorithm Able to Predict Initial 5-year Rate of Parkinson’s Progression

Developing new ways to predict Parkinson’s tremor

Tremor — an involuntary muscle contraction that often manifests as shaking — is a common motor symptom of Parkinson’s disease, occurring in the early stages of disease for about 70% of patients.

Such tremors can be classified as resting, meaning they occur when the muscles are relaxed and still (like when a person is lying in bed), or action, which happen when a person is performing a voluntary movement. A postural tremor is a type of action tremor that occurs when a person maintains a position against gravity, such as holding an arm up in the air.

Because they’re one of the main types of Parkinson’s symptoms, assessing tremors is a critical part of diagnosing Parkinson’s and monitoring its severity. However, these muscle contractions can be difficult to assess, given the variability of their occurrence from patient-to-patient and in different contexts.

Tremors usually are diagnosed in the clinic using a validated measure called the MDS-UPDRS Part III that contains a clinical evaluation for both resting and postural tremors.

The resting tremor test requires patients to sit still in a chair for 10 seconds, while their hands are resting on the chair’s armrests and their feet are placed comfortably on the floor. The specialist then scores tremors in the hands, legs, and jaw.

In the postural tremor test, patients are asked to hold their arm straight out in front of the body with separated fingers, and the hand tremors are assessed and scored.

However, such evaluation is time-consuming for doctors and requires sufficient clinical training. Overall, such testing poses a significant burden to the healthcare system.

“Clinical practice and health services urgently need an automated and objective method to assess [Parkinson’s] tremors” to help alleviate this burden, the researchers wrote.

While various approaches have been proposed, most of them require wearable devices that use sensors to capture movement data. Such devices “might affect the patient’s tremor performance during wearing due to the weight of the device, which could cause the results to be less accurate,” the researchers added.

Now, a team of researchers in China developed a video-based method to assess resting and postural tremors without the need for wearable equipment. Essentially, they created an algorithm that could quantify Parkinson’s tremor based on a video analysis of the patient while performing the MDS-UPDRS tremor-related tests.

To build the approach, a total of 130 Parkinson’s patients were recruited from 2019 to 2021. Each underwent an MDS-UPDRS-based tremor evaluation at the clinic that was recorded on video and scored by trained physicians.

The algorithm was “trained” to spot resting tremors by examining data from these videos, focusing specifically on the hands, legs, and jaws. For capturing postural tremors, the hands were the specific focus.

Compared with other video-based approaches, the new method contained a number of features to improve the accuracy of tremor identification. These include video magnification to capture more subtle tremors that could otherwise be missed, and an ability to analyze multiple body parts and tremor types.

When testing their method, the team found that it could predict MDS-UPDRS scores for both rest and postural tremors with a high level of accuracy.

[The algorithm is] a novel and reliable way to the assessment of PD tremors.

Particularly, it achieved a 90.6% accuracy for predicting resting tremors scores in the hands, 85.9% for those in the legs, and 89% for those in the jaw. For postural tremor scores, the overall accuracy was 84.9%.

Moreover, the accuracy of the new method was higher than “current state-of-the-art methods in the assessment of PD tremor videos,” the researchers wrote.

These findings highlight that the algorithm is “a novel and reliable way to the assessment of PD tremors, which was suitable for assessing rest and postural tremors,” the team wrote, noting that it has “excellent potential for future clinical assessment and remote monitoring of PD patients.”