Abstract
Purpose :
Long-term prediction for diabetic macular edema (DME) is important and this can be achieved using generative adversarial networks (GANs) with data from a randomized controlled trial. We evaluated post-treatment optical coherence tomography (OCT) images generated by GANs trained on data from a randomized controlled trial of anti-vascular endothelial growth factor (VEGF) treatment (q4w x 12) for DME (CRTH258B2305, KINGFISHER).
Methods :
Three-hundred twenty-seven DME eyes of 327 patients with Spectralis OCT images were included in this analysis. OCT B-scan images through the foveal center at week-0, -4, -12, and -52, fundus photography, and retinal thickness maps from each patient were collected. Datasets were divided into training and validation (n = 297), and test (n = 30) sets. Input images for each model comprised either baseline B-scan alone or in combination with others. Predictive post-treatment OCT B-scan images were generated using GAN models and compared with real week-52 images.
Results :
For 30 test images for GAN models trained with baseline OCT B-scans, 28, 29, 15, 20, and 30 acceptable OCT images were generated by cycleGAN, unitGAN, Pix2Pix, Pix2PixHD, and RegGAN, respectively (P<0.001). In comparison with real week-52 images, generated images showed sensitivity, specificity, and positive predictive values (PPV) for residual fluid ranging 0.500-0.917, 0.778-0.944, and 0.667-0.917 and those for hard exudate (HE) ranging 0.545-0.818, 0.895-1.000, and 0.750-1.000. RegGAN exhibited the highest values. RegGAN trained with multi-input images showed improved performance with sensitivity, specificity, and PPV for fluid and HE (0.818-0.909, 0.947-1.000, and 0.900-1.000; 0.818-0.909, 0.947, and 0.900-0.909, respectively).
Conclusions :
OCT images generated by GAN models could predict the presence of residual fluid and hard exudate after long-term treatment of DME using continuous anti-VEGF therapy. Implementation of this tool may help predict which eyes will remain refractory after long-term treatment, thereby facilitating the establishment of a proper treatment regimen for these eyes.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.