Abstract Details

Title Human vs. Machine: The Brain and Optic Nerve Study with Artificial Intelligence (BONSAI)

Topic Neuro-ophthalmology/Neuro-otology

Presentation(s) S32 - Neuro-ophthalmology/Neuro-otology (3:42 PM-3:54 PM)

Poster/Presentation
Number 002

Objective To compare the diagnostic performance of an artificial intelligence deep learning system (AI-DLS) versus expert neuro-ophthalmologists in classifying optic nerves as “normal”, “papilledema” (optic nerve edema from proven intracranial hypertension), and “other optic nerve abnormalities” on ocular fundus photographs.

Background We developed and validated an AI-DLS to automatically classify optic discs as “normal” or “abnormal”, and specifically detect “papilledema”, but direct comparison of the diagnostic accuracy of a DLS versus expert neuro-ophthalmologists using the same sample is warranted.

Design/Methods Eight-hundred standard digital fundus photographs (400 “normal optic discs”, 201 “papilledema”, 199 “other optic disc abnormalities”) taken using multiple cameras with various fields were selected from the BONSAI multi-ethnic population. The image-naïve AI-DLS and expert neuro-ophthalmologists (VB, NJN)independently classified the optic discs on the randomly presented photographs as “normal”, “papilledema”, or “other optic disc abnormalities”, without any additional clinical information.

Results Classification by the machine took 25 seconds, compared to 61-74 minutes by humans for each of the tasks. The AI-DLS successfully classified 678/800 (84.7%) images, compared with 675/800 (84.4%) for human-1 and 641/800 (80.1%) for human-2. Agreement between the two humans was 656/800 (82%); between machine and human-1, 660/800 (82.5%); and between machine and human-2, 624/800 (78%). The machine’s performance in identifying papilledema was similar to humans’ (accuracy 91.5% vs. 89.0% and 89.6%, sensitivity 83.1% vs. 85.1% and 76.6%, specificity 94.3% vs. 90.3% and 94.0% for machine vs. human-1 and human-2, respectively). Similar results were observed for identification of normal optic discs (accuracy 92.1% vs. 93.1% and 88.6%, sensitivity 91.0% vs. 94.5% and 86.3%, specificity 93.3% vs. 91.8% and 91.0%).

Conclusions The classification performance of our trained AI-DLS was at least as good as two expert neuro-ophthalmologists lacking additional clinical information. Ongoing studies of machine versus non-ophthalmologic personnel will likely confirm the obvious superiority of AI-DLS over ophthalmoscopy in many clinical settings.

Authors/Disclosures
Nancy J. Newman, MD, FAAN (Emory University School of Medicine) PRESENTER	Dr. Newman has received personal compensation in the range of $10,000-$49,999 for serving as a Consultant for GenSight. Dr. Newman has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Chiesi. Dr. Newman has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Stoke. Dr. Newman has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Neurophth. Dr. Newman has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Lilly. The institution of Dr. Newman has received research support from GenSight. The institution of Dr. Newman has received research support from Santhera. The institution of Dr. Newman has received research support from NINDS/NIH. Dr. Newman has received publishing royalties from a publication relating to health care. Dr. Newman has received publishing royalties from a publication relating to health care. Dr. Newman has received publishing royalties from a publication relating to health care.
	No disclosure on file
	No disclosure on file
	No disclosure on file
	No disclosure on file
	No disclosure on file
Valerie Biousse, MD	Dr. Biousse has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Gensights Biologic. Dr. Biousse has received personal compensation in the range of $5,000-$9,999 for serving as a Consultant for Topcon. Dr. Biousse has received publishing royalties from a publication relating to health care. Dr. Biousse has received publishing royalties from a publication relating to health care.
	No disclosure on file
	No disclosure on file

��ɫ��

��ɫ��