This paper presents a deep-learning method to synthesize polarization-sensitive optical coherence tomography (PS-OCT) images from standard OCT intensity images. A generative adversarial network (GAN) is trained on paired OCT intensity and PS-OCT images. The accuracy of the synthesized PS-OCT images is evaluated using structural similarity index (SSIM) and validated through cancer/normal classification using real and synthetic PS-OCT images. The results show that the synthetic PS-OCT images can be used interchangeably with real images for cancer diagnosis. The method is also applied to OCT images from a separate system, demonstrating its potential to reduce the cost and complexity of PS-OCT imaging.

Yi Sun, Jianfeng Wang, Jindou Shi, Stephen A. Boppart