JMIR Publications

ABSTRACT

Background:

Generative Adversarial Network (GAN)-based synthetic images can be viable solutions to current supervised deep learning challenges. However, generating highly realistic images is a prerequisite for these approaches.

Objective:

We investigated and validated the unsupervised synthesis of highly realistic body CT images using a progressive growing GAN (PGGAN) trained to learn the probability distribution of normal data.

Methods:

We trained the PGGAN using 11 755 body CT scans. Ten radiologists then evaluated the results in a binary approach using an independent validation set of 300 images (150 real, 150 synthetic) to judge the authenticity of each image.

Results:

Mean accuracy for the entire image set was low (59.4%), and accuracy among three reader groups with different experience levels was not significantly different (58.0% - 60.5%, P = 0.36). Inter-reader agreements were poor (κ = 0.11) for the entire image set. In subgroup analysis, the discrepancies between real and synthetic CT images occurred mainly in the thoracoabdominal junction and in anatomical details.

Conclusions:

The GAN can synthesize highly realistic high-resolution body CT images, which are indistinguishable from real images; however, it has limitations in generating body images in the thoracoabdominal junction and lacks accuracy in anatomical details.