Echocardiographic aortic insufficiency detection using 3-dimensional convolutional neural network from apical 4-chamber views

Abstract

This work proposes to use deep learning to automatically detect cases of aortic insufficiency from echocardiographic videos. Specifically, the use of a 3D CNN (Convolutional Neural Network) was proposed. First, we built our dataset from raw and unstructured data. A database was created containing all phenotypic parameters and echocardiographic measurements of the patients. In addition, all echocardiograms were labeled with the type of view to which they belonged, using a convolutional network. From these structured data, we were able to select 117 patients to form the dataset to be used for classification of aortic insufficiency. We developed a classifier based on R(2+1)D, which accepts video as input and provides in outuput the diagnosis of aortic insufficiency with an overall accuracy of 87.1%.

Objective

Use of AI for identification of aortic valve insufficiency in echocardiography

Methodologies

Bibliographic and experimental research

Conclusions

The developed model achieved an overall accuracy of 87.1% and was able to correctly detect 80% of cases of patients with aortic insufficiency and 90% of cases of patients without aortic insufficiency. This study thus demonstrated how the use of a 3D CNN network was effective in identifying this pathology from echocardiographic videos showing the A4C view.

Future developments

Extend the dataset, improve model performance and increase generalization of aortic regurgitation detection regardless of data source, identify other pathologies.