Training Deep Learning Models to Work on Multiple Devices by Cross-Domain Learning with No Additional Annotations.

PURPOSE: To create an unsupervised cross-domain segmentation algorithm for segmenting intraretinal fluid and retinal layers on normal and pathologic macular OCT images from different manufacturers and camera devices. DESIGN: We sought to use generative adversarial networks (GANs) to generalize a segmentation model trained on one OCT device to segment B-scans obtained from a different OCT device manufacturer in a fully unsupervised approach without labeled data from the latter manufacturer. PARTICIPANTS: A total of 732 OCT B-scans from 4 different OCT devices (Heidelberg Spectralis, Topcon 1000, Maestro2, and Zeiss Plex Elite 9000). METHODS: We developed an unsupervised GAN model, GANSeg, to segment 7 retinal layers and intraretinal fluid in Topcon 1000 OCT images (domain B) that had access only to labeled data on Heidelberg Spectralis images (domain A). GANSeg was unsupervised because it had access only to 110 Heidelberg labeled OCTs and 556 raw and unlabeled Topcon 1000 OCTs. To validate GANSeg segmentations, 3 masked graders manually segmented 60 OCTs from an external Topcon 1000 test dataset independently. To test the limits of GANSeg, graders also manually segmented 3 OCTs from Zeiss Plex Elite 9000 and Topcon Maestro2. A U-Net was trained on the same labeled Heidelberg images as baseline. The GANSeg repository with labeled annotations is at https://github.com/uw-biomedical-ml/ganseg. MAIN OUTCOME MEASURES: Dice scores comparing segmentation results from GANSeg and the U-Net model with the manual segmented images. RESULTS: Although GANSeg and U-Net achieved comparable Dice scores performance as human experts on the labeled Heidelberg test dataset, only GANSeg achieved comparable Dice scores with the best performance for the ganglion cell layer plus inner plexiform layer (90%; 95% confidence interval [CI], 68%-96%) and the worst performance for intraretinal fluid (58%; 95% CI, 18%-89%), which was statistically similar to human graders (79%; 95% CI, 43%-94%). GANSeg significantly outperformed the U-Net model. Moreover, GANSeg generalized to both Zeiss and Topcon Maestro2 swept-source OCT domains, which it had never encountered before. CONCLUSIONS: GANSeg enables the transfer of supervised deep learning algorithms across OCT devices without labeled data, thereby greatly expanding the applicability of deep learning algorithms.

Cataract Surgery in Very Old Patients: A Case-Control Study.

Advancements in surgical techniques and increased life expectancy have made cataract surgery more common among very old patients. However, surgical outcomes seem impaired in patients older than 90 years, especially with ocular comorbidities. A retrospective case-control study of 53 eyes of 53 very old patients (mean 92.6 ± 3.0) and 140 eyes of 140 matched patients (mean 75.2 ± 7.6) was undertaken. Groups were matched in terms of gender and systemic and ocular comorbidities. In very old patients, higher phacoemulsification energy (cumulative dissipated energy [CDE], 25.0 ± 22.4 vs. 16.1 ± 10.7, p = 0.01) and rate of intraoperative floppy iris syndrome (IFIS, 9.4% vs. 1.4%, p = 0.02) were observed compared to controls. Uncorrected (UCVA) and best-corrected distance visual acuity (BCVA) gains were significantly poorer among the very old patients than among the control at postoperative day 30 (0.20 ± 0.70 vs. 0.56 ± 0.61 logMAR, p < 0.001 and 0.27 ± 0.64 vs. 0.55 ± 0.62 logMAR, p = 0.006, respectively). Even after including CDE and IFIS as covariates, age remained an independent factor for poor visual gain at 30 days (p < 0.001). Cataract surgery in very old patients may demand more experienced surgeons due to higher nuclear density and the rates of IFIS. Expectations in visual acuity gains should be aligned with the patient's age.

Infantile cataract: comparison of two surgical approaches.

OBJECTIVE: To compare the results of 2 cataract extraction techniques with primary intraocular lens (IOL) implantation in children. DESIGN: Retrospective comparative case series study. METHODS: This study included children with congenital or developmental cataract. In all cases, anterior capsulorhexis, lens aspiration, posterior continuous curvilinear capsulorhexis, and primary IOL implantation were performed. We compared 2 surgical approaches. In the first approach, after IOL implantation in the bag, posterior optic capture (OC) was performed without anterior vitrectomy (OC technique); in the second approach, anterior vitrectomy was performed without OC of the IOL (AV technique). Patient demographic data as well as ethnic origin, child's age at cataract diagnosis, child's age at surgery, axial length, IOL power in diopters, visual acuity, visual axis opacification, and complications were assessed. RESULTS: One hundred twenty-three eyes were included for surgical approach outcomes comparison; 21 eyes underwent the OC surgical approach and 102 the AV surgical approach. The mean patient age at surgery was 57.3 ± 47.1 months. The mean follow-up was 63.13 months (range 12-202 months). Epithelial lens reproliferation was the major adverse event in our series, affecting 21.1% of patients' eyes; the mean time to epithelial lens reproliferation development was 90 ± 9.70 months. There was no statistically significant difference between the two groups in best spectacle-corrected visual acuity or epithelial lens reproliferation incidence. CONCLUSIONS: In our case series we did not find any difference between surgical techniques.