Neuromuscular disorders such as muscular dystrophy and spinal muscular atrophy affect an estimated 1 million children worldwide. Many of these disorders progress, sometimes rapidly, significantly affecting movement and gait.

They can even be fatal.

Pharmaceutical companies are developing and testing medicines to treat these conditions, but gauging their effectiveness is difficult.

“Knowing when people, both those in clinical trials and others, are improving or becoming worse is critical,” says Stevens professor Damiano Zanotto. “However, the current methods of assessing mobility are not always accurate.”

Now Zanotto, working closely with Columbia University Irving Medical Center (CUIMC) professor Jacqueline Montes, is working to develop new AI techniques that analyze in-shoe sensor data to better monitor those disorders.

The National Institutes of Health (NIH) has just awarded a $2 million multi-principal investigator R01 award to support the work — with three of the nation’s leading medical centers involved as well.

“While these disorders are typically childhood-onset, they affect individuals across the lifespan,” notes Montes. Recent therapeutic approaches have resulted in extended survival and clinically meaningful improvements in motor function, with the best responses seen with early treatment.” “However, there is still no cure.”

Better data, to capture changes in function

In their study, Zanotto and Montes will focus on Duchenne muscular dystrophy (DMD) — an inherited disorder that affects mostly young boys — as well as spinal muscular atrophy (SMA), the world’s leading genetic cause of infant mortality.

“One of the problems here that interested me, from an engineering perspective,” adds Zanotto, “is that patients report that new medications really help them. However, when it’s time for those individuals to perform clinical tests to verify this, those benefits in motor improvements do not always show up."

Most of the standard assessment tests, he says, were first introduced decades ago, when wearable technology was still at its infancy and not widely available — and essentially rely on a stopwatch and clinical observation.

Zanotto developed in-shoe sensors over several generations, including this latest iteration“What you see in ten minutes in a doctor’s office doesn’t necessarily reflect real-world performance, nor the improvement in or progression of a disease,” he continues. “Existing sensor technologies such as activity trackers, which can be used to monitor performance continuously in everyday settings, are mostly limited to basic “gait quantity” metrics, such as step count or distances walked.”

“But they cannot accurately capture measures of ‘gait quality’ — such as stride length, time, and velocity — nor kinetic metrics that evaluate a person’s dynamic stability.”

Zanotto’s lab previously engineered a series of novel AI-enhanced, in-shoe sensors with support from partners including the National Science Foundation, Muscular Dystrophy Association and New Jersey Department of Health.

These sensors are especially useful for assessing motor function because they capture not only measure of gait quality but also unique kinetic data — such as the force of a foot striking the ground — something other sensors (such as ankle bracelets or smartwatches) simply can’t do.

The AI-enhanced in-shoe sensors were developed and tested in controlled environments with more than 600 volunteers, including healthy controls and individuals from various clinical populations. However, real-world gait monitoring was limited to a small group of healthy individuals.

This NIH study will be the first time the technology is deployed in patients' everyday environments for month-long gait monitoring periods.

“There’s a need to develop more sensitive quantitative assessments of real-world ambulatory function in individuals with neuromuscular disorders,” notes Zanotto. “That’s what we’re trying to do here.”

This project will fill that need, via clinical sites at CUIMC, Boston Children’s Hospital (which is affiliated with Harvard Medical School) and Stanford Medical School.

“We plan to enroll up to 100 individuals — 33 with SMA, 33 with DMD and 40 healthy controls — across the 3 clinical sites,” explains Montes.

“We will enroll ambulatory individuals at least 5 years old, and we will ensure there is a balance of children, adolescents and adult participants.”

Evaluations over time; AI for prediction

The plan is to evaluate participants’ gait with the Stevens-developed sensors during two one-month periods, one year apart.

By comparing metrics collected by the sensors in everyday environments with benchmark clinical assessments that will also be performed, and analyzing the results, the duo hopes to extract and pinpoint digital biomarkers of disease severity, portions of the motion data that appear most useful for accurately assessing patients — narrowing their focus for future work.

From a drug-development perspective, the work is also valuable.

“Having more sensitive outcome metrics can help reduce both the duration and size of future clinical trials,” points out Zanotto. “There’s growing interest from pharmaceutical companies in these types of technologies, to help them bring new treatments to patients sooner and in a more cost-effective way.”

As a secondary objective, the team will also use the collected data to train new machine learning models that could one day help clinicians forecast disease progression or improvement.

“Can we leverage AI methods to predict a patient’s functional status 12 months from now?” asks Zanotto. “It’s a more challenging, longer-term goal, but AI-based methods informed by wearable sensor data have recently shown encouraging results in predicting disease trajectories in other populations, so we are very hopeful.”

The project, “Establishing Walking-related Digital Biomarkers in Rare Childhood Onset Progressive Neuromuscular Disorders,” is funded through 2028.