Abstract Details

Title Identification of large vessel occlusion on CTA head using artificial intelligence

Topic Cerebrovascular Disease and Interventional Neurology

Presentation(s) P12 - Poster Session 12 (5:30 PM-6:30 PM)

Poster/Presentation
Number 13-007

Background

Endovascular thrombectomy has been shown to improve outcomes for patients with large vessel occlusions. Artificial intelligence provides the opportunity to expedite this treatment pathway and improve outcomes through automatically detecting large vessel occlusions.

Objective

To develop and assess a deep learning algorithm (“CTA LVO”) for detection, lateralization and localization of large vessel occlusions on CTA head imaging.

Design/Methods

We used a training dataset of 2,764 CTA head cases including 1,206 (43.6%) that were positive for large vessel occlusion. We trained a deep learning algorithm, which consisted of an ensemble of five DenseNet models, to detect large vessel occlusions in the anterior circulation. The algorithm also provided lateralization (left or right) and localization (ICA, M1 or M2) information. We assessed its performance on imaging from 6-months of consecutive stroke presentations across two hospitals.

Results The CTA LVO algorithm performed with area under the receiver operating characteristic curve 0.943, sensitivity 91.2% and specificity 82.1%. The algorithm identified 16 out of 16 (100%) ICA occlusions, 29 out of 29 (100%) M1 occlusions, and 7 out of 12 (58.3%) proximal M2 occlusions. It correctly identified the laterality for 51 out of 52 (98.1%) of these occlusions and the localization for 44 out of 52 (84.6%) of these occlusions.

Conclusions

The CTA LVO algorithm can detect, lateralize and localize large vessel occlusions on CTA head. It has the potential to improve patient outcomes through expediting the stroke treatment pathway.

Authors/Disclosures
James M. Hillis, MD (Massachusetts General Hospital) PRESENTER	Dr. Hillis has stock in Elly Health. The institution of Dr. Hillis has received research support from GE Healthcare. The institution of Dr. Hillis has received research support from Annalise.ai. The institution of Dr. Hillis has received research support from Viz.ai. Dr. Hillis has received intellectual property interests from a discovery or technology relating to health care.
	No disclosure on file
	No disclosure on file
	No disclosure on file
	No disclosure on file
	No disclosure on file
	No disclosure on file
	No disclosure on file
Michael H. Lev, MD (Mass General Hospital)	Michael H. Lev, MD has received personal compensation in the range of $500-$4,999 for serving as a Consultant for Takeda, Roche-Genetech. The institution of Michael H. Lev, MD has received research support from GE. Michael H. Lev, MD has received publishing royalties from a publication relating to health care.
	No disclosure on file