aEYE: A deep learning system for video nystagmus detection.
artificial intelligence
deep learning
dizziness
eye movements
machine learning
nystagmus
telemedicine
vertigo
Journal
Frontiers in neurology
ISSN: 1664-2295
Titre abrégé: Front Neurol
Pays: Switzerland
ID NLM: 101546899
Informations de publication
Date de publication:
2022
2022
Historique:
received:
08
06
2022
accepted:
20
07
2022
entrez:
29
8
2022
pubmed:
30
8
2022
medline:
30
8
2022
Statut:
epublish
Résumé
Nystagmus identification and interpretation is challenging for non-experts who lack specific training in neuro-ophthalmology or neuro-otology. This challenge is magnified when the task is performed We developed, trained, and validated a deep-learning system (aEYE) to classify video recordings as normal or bearing at least two consecutive beats of nystagmus. The videos were retrospectively collected from a subset of the monocular (right eye) video-oculography (VOG) recording used in the Acute Video-oculography for Vertigo in Emergency Rooms for Rapid Triage (AVERT) clinical trial (#NCT02483429). Our model was derived from a preliminary dataset representing about 10% of the total AVERT videos ( An ensemble between the ResNet-soft voting and the VGG-hard voting models had the best performing metrics. The AUROC, sensitivity, specificity, and accuracy were 0.86, 88.4, 74.2, and 82.7%, respectively. Our validated folds had an average AUROC, sensitivity, specificity, and accuracy of 0.86, 80.3, 80.9, and 80.4%, respectively. Models created from the compressed videos decreased in accuracy as image sampling rate decreased from 60 Hz to 15 Hz. There was only minimal change in the accuracy of nystagmus detection when decreasing image resolution and keeping sampling rate constant. Deep learning is useful in detecting nystagmus in 60 Hz video recordings as well as videos with lower image resolutions and sampling rates, making it a potentially useful tool to aid future automated eye-movement enabled neurologic diagnosis.
Sections du résumé
Background
UNASSIGNED
Nystagmus identification and interpretation is challenging for non-experts who lack specific training in neuro-ophthalmology or neuro-otology. This challenge is magnified when the task is performed
Methods
UNASSIGNED
We developed, trained, and validated a deep-learning system (aEYE) to classify video recordings as normal or bearing at least two consecutive beats of nystagmus. The videos were retrospectively collected from a subset of the monocular (right eye) video-oculography (VOG) recording used in the Acute Video-oculography for Vertigo in Emergency Rooms for Rapid Triage (AVERT) clinical trial (#NCT02483429). Our model was derived from a preliminary dataset representing about 10% of the total AVERT videos (
Results
UNASSIGNED
An ensemble between the ResNet-soft voting and the VGG-hard voting models had the best performing metrics. The AUROC, sensitivity, specificity, and accuracy were 0.86, 88.4, 74.2, and 82.7%, respectively. Our validated folds had an average AUROC, sensitivity, specificity, and accuracy of 0.86, 80.3, 80.9, and 80.4%, respectively. Models created from the compressed videos decreased in accuracy as image sampling rate decreased from 60 Hz to 15 Hz. There was only minimal change in the accuracy of nystagmus detection when decreasing image resolution and keeping sampling rate constant.
Conclusion
UNASSIGNED
Deep learning is useful in detecting nystagmus in 60 Hz video recordings as well as videos with lower image resolutions and sampling rates, making it a potentially useful tool to aid future automated eye-movement enabled neurologic diagnosis.
Identifiants
pubmed: 36034311
doi: 10.3389/fneur.2022.963968
pmc: PMC9403604
doi:
Types de publication
Journal Article
Langues
eng
Pagination
963968Subventions
Organisme : NIDCD NIH HHS
ID : U01 DC013778
Pays : United States
Informations de copyright
Copyright © 2022 Wagle, Morkos, Liu, Reith, Greenstein, Gong, Gangan, Pakhomov, Hira, Komogortsev, Newman-Toker, Winslow, Zee, Otero-Millan and Green.
Déclaration de conflit d'intérêts
Author DN-T conducts research related to diagnosis of dizziness and stroke, as well as diagnostic error. He serves as the principal investigator for multiple grants and contracts on these topics, including the NIH-sponsored AVERT clinical trial (NIDCD U01 DC013778, ClinicalTrials.gov #NCT02483429). Johns Hopkins has been loaned research equipment [video-oculography (VOG) systems] by two companies for use in DN-T's research; one of these companies has also provided funding for research on diagnostic algorithm development related to dizziness, inner ear diseases, and stroke. DN-T has no other financial interest in these or any other companies. DN-T is an inventor on a provisional patent (US No. 62/883,373) for smartphone-based stroke diagnosis in patients with dizziness. He gives frequent academic lectures on these topics and occasionally serves as a medico-legal consultant for both plaintiff and defense in cases related to dizziness, stroke, and diagnostic error. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Références
Sci Rep. 2019 Jul 18;9(1):10452
pubmed: 31320726
Biomed Signal Process Control. 2021 Jul;68:102676
pubmed: 33936249
Neurology. 2021 Jan 5;96(1):34-38
pubmed: 33004609
Stroke. 2009 Nov;40(11):3504-10
pubmed: 19762709
Digit Biomark. 2020 Dec 30;5(1):1-8
pubmed: 33615116
IEEE J Biomed Health Inform. 2021 May;25(5):1814-1823
pubmed: 32956068
Philos Trans A Math Phys Eng Sci. 2020 Mar 6;378(2166):20190054
pubmed: 31955675
Entropy (Basel). 2020 Dec 25;23(1):
pubmed: 33375658
Biomed Signal Process Control. 2021 Jul;68:102708
pubmed: 34276807
J Vis. 2015;15(14):11
pubmed: 26587699
Acad Emerg Med. 2013 Oct;20(10):986-96
pubmed: 24127701
Phys Med Biol. 2021 Sep 14;66(18):
pubmed: 34450599
J Clin Med. 2019 May 08;8(5):
pubmed: 31072056
Neural Comput. 1997 Nov 15;9(8):1735-80
pubmed: 9377276