Deep Learning Based Prediction of Pulmonary Hypertension in Newborns Using Echocardiograms (2024) Hanna Ragnarsdottir

Description

Pulmonary hypertension (PH) in newborns is a critical condition requiring early and accurate detection. This work presents an explainable multi-view video-based deep learning approach to predict and classify the severity of PH in a cohort of 270 newborns using echocardiograms. The model showed strong performance (e.g., F1-score of 0.84 for severity prediction via cross-validation) and uses saliency maps to confirm focus on clinically relevant cardiac structures, offering the first automated assessment method for PH in this population.