LOS ANGELES ― Artificial intelligence (AI) analysis of home videos could accelerate the diagnosis of infantile spasms, a severe form of epilepsy in babies, suggests a study presented at the American Epilepsy Society Annual Meeting.
Infantile spasms are seizures that disrupt normal brain development. If not identified and treated early, they can cause delays in the development of motor and social skills and language, as well as intellectual disabilities, permanent epilepsy and even death. Diagnosis is often delayed because the spasms are subtle, including sudden jerks, bending, stiffening of the arms and legs or the head dropping forward; parents and pediatricians often mistake them for typical infant behavior such as the startle reflex in response to a sudden stimulus like a loud noise or movement. Treatment typically includes hormonal therapy and or vigabatrin, an antiseizure medicine. Delaying treatment even for a week can be devastating.
“Early identification of infantile spasms is crucial because it enables timely treatment, ideally within one week of the onset of the spasms, which may significantly improve outcomes for these infants,” said Gadi Miron, MD, the lead author of the study and researcher at Charité-Universitätsmedizin Berlin, Germany. “Our video-based seizure AI detection model shows high performance on smartphone videos, including those with various camera resolutions and video capabilities. We believe this technique could empower parents by capturing critical information about their child’s symptoms.”
The researchers collected videos of infantile spasms in children younger than two years old that had been posted online, including on social media, as well as from videos collected on hospital-based cameras. The researchers fine-tuned an AI model designed for general video analysis by training it to detect infantile spasms using labeled video segments. They analyzed seizures that were clearly visible and could be recognized on video by an epilepsy specialist. They compared five-second video segments of seizures to non-seizure video segments from a dataset that included 168 infants with infantile spasms who had 1,106 seizures and 216 infants without seizures for a total of 13,869 segments featuring normal movements.
They determined the AI model was highly accurate at distinguishing between movements that were seizures and those that were not. The model correctly identified 83% of movements that were actually infantile spasms (true positive rate) and was 95% accurate at identifying movements that were not infantile spasms (true negative rate). Overall, it was 85% accurate at detecting infantile spasms.
As a follow-up to this study, the researchers have developed a digital tool that parents could use to upload videos of their infants’ suspicious movements to a secure platform. Those videos are then screened by the AI model, which flags concerning videos for review by specialists. If an infantile spasm is identified, the parent is notified and provided with a recommendation for further diagnostic evaluation. The researchers are conducting a study using this platform to see if it can be used by pediatricians and other clinicians who treat infants.
“This technique could also aid doctors by offering additional data for patient evaluations and potentially expedite referrals for diagnostic tests, leading to quicker treatment decisions,” said Christian Meisel, MD, PhD, the principal investigator of the study and a neurologist at Charité-Universitätsmedizin Berlin.
The research presented at AES 2024 coincides with Infantile Spasms Awareness Week, Dec. 1-7. AES is one of the entities in the Infantile Spasms Network collaboration, which is dedicated to raising awareness of infantile spasms.
*** AES 2024 news releases may contain updated data that does not match what is reported in the abstract.
For more information, visit the American Epilepsy Society online at aesnet.org. Join the AES social conversation today by following @AmEpilepsySoc on X and use the hashtag #AES2024.