By Denise Baez
VIRTUAL -- October 19, 2021 -- A deep learning model is useful in detecting nystagmus from video recording, and can be useful when evaluating patients with dizziness remotely, according to a study presented at the Virtual 146th Annual Meeting of the American Neurological Association (ANA).
“Identification and interpretation of nystagmus is challenging for non-expert neuro-otologists and neuro-ophthalmologists,” said Kemar E. Green, MD, Johns Hopkins University, Baltimore, Maryland. “This challenge is magnified when this task must be performed via telemedicine.”
“Nystagmus can be detected from low quality videos using deep-learning methods and can be useful for remote diagnosis of dizzy patients in a pandemic, as well as becoming a permanent feature of healthcare delivery,” he said.
The researchers developed, trained, and validated a deep learning system to classify 435 60 Hz recordings as videos with nystagmus or video without nystagmus. The dataset was created from monocular video-oculography recordings collected retrospectively used in the AVERT clinical trial, which collected data from vertiginous patients in 5 emergency rooms across the United States.
“Nystagmus precedes MRI changes by 48 to 72 hours in patients who have experienced a stroke presenting with isolated dizziness or vertigo,” noted Dr. Green.
Patients with no nystagmus and patients with nystagmus were split into training and test sets in a ratio of approximately 3:2. Nine different variations of the model were tested.
Using the test test, the model performed well in detecting nystagmus, which was calculated using the receiver-operating characteristic curve. The model had a sensitivity of 0.87, specificity of 0.88, a negative predictive value of 0.85, and a positive predictive value 0.84.
“Deep learning is useful in detecting nystagmus in 60 Hz video recordings, making it a useful screening tool for the vertiginous patient, and potentially applicable for future automated smartphone ocular motor diagnosis,” said Dr. Green.
[Presentation title: Deep Learning Model Detects Nystagmus From Video Recording. Abstract 267]
To read more Conference Dispatch articles, click here.