AI Ability to Predict Preterm Birth Improves with Retraining
Tuesday, April 9, 2024
12:06pm – 12:13pm
Location: 412
Authors: Neil Patel, University of Kentucky John O'Brien, University of Kentucky Robert Bunn, Ultrasound AI John Bauer, Kentucky Childrens Hospital Brandon Schanbacher, Kentucky Childrren's Hospital, University of Kentucky Garrett Lam, University of Kentucky ,
Artificial intelligence (AI) based image analyses for prediction of preterm birth (PTB) improves with retraining. Sensitivity of PTB prediction improved from 39% to 40% and specificity of 93% to 95%. For prediction of PTB >30 days, sensitivity rose from 51% to 59%, and specificity rose from 94% vs 96%. Continuous training of an AI improved its performance and accuracy to identify those individuals who deliver preterm and from an US exam >30 days from delivery. Through retraining, the AI “learns” to better recognize characteristics in the data and improves its ability to make unbiased predictions in individuals for preterm birth as the technology can for other applications.