Authors:
Xiaochen Wang
and
Natalia Khuri
Affiliation:
Department of Computer Science, Wake Forest University, 1834 Wake Forest Road, Winston-Salem, U.S.A.
Keyword(s):
COVID-19, Deep Learning, Generative Adversarial Network, Image Segmentation.
Abstract:
The coronavirus disease 2019 is a global pandemic that threatens lives of many people and poses a significant burden for healthcare systems worldwide. Computerized Tomography can detect lung infections, especially in asymptomatic cases, and the detection process can be aided by deep learning. Most of the recent research focused on the segmentation of the entire infected region in a lung. To automate a more fine-grained analysis, a generative adversarial network, comprising two convolutional neural networks, was developed for the segmentation of ground glass opacities and consolidations from tomographic images. The first convolutional neural network acts as a generator of segmented masks, and the second as a discriminator of real and artificially segmented objects, respectively. Experimental results demonstrate that the proposed network outperforms the baseline U-Net segmentation model on the benchmark data set of 929 publicly available images. The dice similarity coefficients of segm
enting ground glass opacities and consolidations are 0.664 and 0.625, respectively.
(More)