Multi-label 4-chamber segmentation of echocardiograms using Fully Convolutional Network

Arghavan Arafati, Daisuke Morisawa, Ramin Assadi, Reza Amini, Hamid Jafarkhani, Arash Kheradvar

Abstract: Precise segmentation of the heart is crucial for reliable calculation of clinical indices such as chambers’ volumes and ejection fraction (EF). Currently, in echocardiography, cardiac segmentation is manually performed through a tedious and time-consuming process. More importantly, this process is prone to inter- and intra-user variability. Motivated by current methods limitations, we have employed Fully Convolutional Networks to semantically segment heart chambers. Our model was trained with more than 900 images from 70 subjects and tested on 449 images from 30 patients. Further, it was evaluated using multiple evaluation metrics including Dice similarity coefficient and Hausdorff distance. The proposed model has shown robust and good accuracy.

Keywords: deep learning, segmentation, Echocardiography, fully convolutional neural network

Author Affiliation: University of California Irvine, University of California Los Angeles, Loma Linda University