AI-Assisted Diagnosis of SMA Hypotonia
—Using AI technology, researchers in France taught a computer how to recognize signs of SMA hypotonia at a younger age than when diagnosis is normally made, opening up the possibility for earlier treatment for this rare disease.
“Early detection and treatment of infants born with spinal muscular atrophy (SMA) is key to improved long-term outcomes,” Natalie Katz, MD, PhD, a pediatric neurologist at the Lenox Baker Children’s Hospital in Durham, North Carolina, told Bվ.
As of 2024, the U.S. has adopted SMA screening in all 50 states as part of the newborn screening program,1 although about 5% of SMA cases can be missed.2 It can be difficult to determine if a particularly young infant is displaying symptoms of SMA, noted Dr. Katz, who is also Assistant Professor, Department of Pediatrics, Duke University School of Medicine, in Durham.
“Use of video-assisted technology and artificial intelligence (AI) to identify changes in movement can provide valuable insight on whether an infant is displaying symptoms of hypotonia, which might suggest SMA as a possible diagnosis,” she added. “Use of AI technology is becoming increasing important in medicine to improve diagnostic efficacy among other applications.”
In this context, researchers in France developed and assessed the utility of an SMA screening tool that incorporated computer vision and artificial intelligence (AI) and found that this technology is able to identify abnormal motricity in SMA “as a first procedural step to diagnosis.” Their findings were reported in a recent issue of JAMA Pediatrics.3
Teaching computers to detect SMA
One-minute, 2-dimensional videos were made to record motor patterns of 25 pediatric patients admitted to the intensive care unit during 2020-2022 at Hôpital Raymond-Poincaré in Garches in France. Data for pose estimation (the computer vision process of determining position and orientation) were calculated using the authors’ contrast modification method4 that was applied to a pose estimator called AlphaPose.
The researchers first obtained the infants’ motor skills dataset based on articulations, limbs, and limb angles; calculated different linear, angular, and depth of movements; and then trained various models to identify SMA. XGBoost, a supervised machine learning model, was used to label data as either control motricity or SMA motricity.
AI identified SMA hypotonia with 97% accuracy
Five infants (mean age 29.2 weeks, median age 34 weeks) presented with SMA type 1; genetic analysis confirmed 2 copies of the SMN2 gene. The 20 remaining infants (mean age 15.6 weeks, median age 12.5 weeks) had normal neurologic examinations and comprised the control group.
“A limitation of the study was the small number of infants presenting with neurodevelopmental deficits,” the authors noted in their research letter.
The analysis revealed that depth of movements gave the best discrimination between the SMA group and control group. Depth of movements corresponded to the possibility that the infant had to move in the frontal plane, which is an SMA feature, the researchers explained. Mean (standard deviation) area under the curve for this discrimination was 0.97 (0.02). In other words, the researchers were able to teach a computer, via computer vision and AI, to identify abnormal mobility with an accuracy of 97% among infants who presented with SMA (versus a control group) and, as the authors remarked, who were “at a younger age than usual.”
The Shapley Additive Explanations model (a unified framework to interpret model predictions) demonstrated that the lower limbs were more discriminant than the upper limbs in classifying motricity as SMA. The AI decision to classify infants primarily according to lower limb motricity was “consistent with the clinical diagnosis of SMA,” the authors wrote.
“Though the study is small,” commented Dr. Katz, who was not involved in the study, “the results show that use of AI technology may improve the clinician’s ability to detect subtle signs/symptoms of hypotonia and reduce the diagnostic delay in SMA, leading to more rapid treatment and improved long-term outcomes.”
“With new therapies for SMA on the horizon, and given the need to counsel families properly,” the authors wrote, “the ability to make a clinically difficult diagnosis early becomes increasingly important for mobilizing limited resources and expertise.”
Published:
References