Enhanced Quality Assessment of Echocardiographic Images for Pulmonary Hypertension Using Convolutional Neural Networks

Pulmonary Hypertension (PH) is associated with cardiopulmonary disease and carries strong prognostic information.Its accurate diagnosis is based on invasive procedures such as right-sided heart catheterization.Alternative non-invasive approaches rely on Doppler imaging, where signal quality may impede correct readings potentially limiting its clinical value.In this study, we propose an automated approach for the quality assessment of Doppler signals using a convolutional neural network (CNN).The CNN was trained on echocardiographic Doppler images and their quality was assessed by 2 independent expert readers.The dataset was subjected to preprocessing and augmentation techniques to enhance model resilience and generalization.Leveraging the VGG-16 architecture, the CNN demonstrated an accuracy of 86%, sensitivity of 86%, precision of 88%, and F1-Score of 86% on the test set.The CNN showed improved accuracy, recall, and F1-score as compared to an unseen clinical reader assessment.The results emphasize the variability in clinical reader assessment, such that automated assessment may prove highly clinically useful in the future.In this way, deep learning-driven image quality assessment could enhance diagnostic accuracy, reduce practitioner variability, and streamline patient care in PH management.