Scientific Reports (Jan 2023)
Head CT deep learning model is highly accurate for early infarct estimation
- Romane Gauriau,
- Bernardo C. Bizzo,
- Donnella S. Comeau,
- James M. Hillis,
- Christopher P. Bridge,
- John K. Chin,
- Jayashri Pawar,
- Ali Pourvaziri,
- Ivana Sesic,
- Elshaimaa Sharaf,
- Jinjin Cao,
- Flavia T. C. Noro,
- Walter F. Wiggins,
- M. Travis Caton,
- Felipe Kitamura,
- Keith J. Dreyer,
- John F. Kalafut,
- Katherine P. Andriole,
- Stuart R. Pomerantz,
- Ramon G. Gonzalez,
- Michael H. Lev
Affiliations
- Romane Gauriau
- Data Science Office, Mass General Brigham
- Bernardo C. Bizzo
- Data Science Office, Mass General Brigham
- Donnella S. Comeau
- Data Science Office, Mass General Brigham
- James M. Hillis
- Data Science Office, Mass General Brigham
- Christopher P. Bridge
- Data Science Office, Mass General Brigham
- John K. Chin
- Data Science Office, Mass General Brigham
- Jayashri Pawar
- Data Science Office, Mass General Brigham
- Ali Pourvaziri
- Data Science Office, Mass General Brigham
- Ivana Sesic
- Data Science Office, Mass General Brigham
- Elshaimaa Sharaf
- Data Science Office, Mass General Brigham
- Jinjin Cao
- Data Science Office, Mass General Brigham
- Flavia T. C. Noro
- Data Science Office, Mass General Brigham
- Walter F. Wiggins
- Data Science Office, Mass General Brigham
- M. Travis Caton
- Data Science Office, Mass General Brigham
- Felipe Kitamura
- Diagnosticos da America SA (Dasa)
- Keith J. Dreyer
- Data Science Office, Mass General Brigham
- John F. Kalafut
- GE Healthcare
- Katherine P. Andriole
- Data Science Office, Mass General Brigham
- Stuart R. Pomerantz
- Data Science Office, Mass General Brigham
- Ramon G. Gonzalez
- Data Science Office, Mass General Brigham
- Michael H. Lev
- Data Science Office, Mass General Brigham
- DOI
- https://doi.org/10.1038/s41598-023-27496-5
- Journal volume & issue
-
Vol. 13,
no. 1
pp. 1 – 11
Abstract
Abstract Non-contrast head CT (NCCT) is extremely insensitive for early ( 0.98). When this 150 CT test set was expanded to include a total of 364 CT scans with a more heterogeneous distribution of infarct locations (94 stroke-negative, 270 stroke-positive mixed territory infarcts), model sensitivity was 97%, specificity 99%, for detection of infarcts larger than the 70 mL volume threshold used for patient selection in several major randomized controlled trials of thrombectomy treatment.
