Discriminating Healthy Optic Discs and Visible Optic Disc Drusen on Fundus Autofluorescence and Color Fundus Photography Using Deep Learning-A Pilot Study.

The aim of this study was to use deep learning based on a deep convolutional neural network (DCNN) for automated image classification of healthy optic discs (OD) and visible optic disc drusen (ODD) on fundus autofluorescence (FAF) and color fundus photography (CFP). In this study, a total of 400 FAF and CFP images of patients with ODD and healthy controls were used. A pre-trained multi-layer Deep Convolutional Neural Network (DCNN) was trained and validated independently on FAF and CFP images. Training and validation accuracy and cross-entropy were recorded. Both generated DCNN classifiers were tested with 40 FAF and CFP images (20 ODD and 20 controls). After the repetition of 1000 training cycles, the training accuracy was 100%, the validation accuracy was 92% (CFP) and 96% (FAF), respectively. The cross-entropy was 0.04 (CFP) and 0.15 (FAF). The sensitivity, specificity, and accuracy of the DCNN for classification of FAF images was 100%. For the DCNN used to identify ODD on color fundus photographs, sensitivity was 85%, specificity 100%, and accuracy 92.5%. Differentiation between healthy controls and ODD on CFP and FAF images was possible with high specificity and sensitivity using a deep learning approach.

Optimizing intraocular lens power calculation using adjusted conventional keratometry for cataract surgery combined with Descemet membrane endothelial keratoplasty.

PURPOSE: To evaluate the utility of intraocular lens (IOL) power calculation using adjusted conventional keratometry (K) according to postoperative posterior to preoperative anterior corneal curvature radii (PPPA) ratio for eyes with Fuch's dystrophy undergoing cataract surgery combined with Descemet membrane endothelial keratoplasty (triple DMEK). METHODS: A fictitious refractive index (FRI) was determined (Pentacam HR®) based on the PPPA ratio in 50 eyes undergoing triple DMEK. Adjusted corneal power was calculated in every eye using adjusted K values: K values determined by the IOLMaster were converted to adjusted anterior corneal radius using the mean FRI. Posterior corneal radius was calculated using the mean PPPA ratio. Adjusted corneal power was determined based on the calculated corneal radii and thick lens formula. Refractive errors calculated using the Haigis, SRK/T, and HofferQ formulae based on the adjusted corneal power were compared with those based on conventional K measurements. RESULTS: Calculated PPPA ratio and FRI were 0.801 and 1.3271. Mean prediction error based on conventional K was in the hyperopic direction (Haigis: 0.84D; SRK/T: 0.74D; HofferQ: 0.74D) and significantly higher (P < 0.001) than that based on adjusted corneal power (0.18D, 0.22D, and 15D, respectively). When calculated according to adjusted corneal power, the percentage of eyes with a hyperopic shift > 0.5D fell significantly from 64 to 30% (Haigis), 62 to 36% (SRK/T), and 58 to 26% (HofferQ), respectively. CONCLUSION: IOL power calculation based on adjusted corneal power can be used to reduce the risk of a hyperopic shift after triple DMEK and provides a more accurate refractive outcome than IOL power calculation using conventional K.

Assessing the validity of corneal power estimation using conventional keratometry for intraocular lens power calculation in eyes with Fuch's dystrophy undergoing Descemet membrane endothelial keratoplasty.

PURPOSE: The present retrospective study was designed to test the hypothesis that the postoperative posterior to preoperative anterior corneal curvature radii (PPPA) ratio in eyes with Fuch's dystrophy undergoing Descemet membrane endothelial keratoplasty (DMEK) is significantly different to the posterior to anterior corneal curvature radii (PA) ratio in virgin eyes and therefore renders conventional keratometry (K) and the corneal power derived by it invalid for intraocular lens (IOL) power calculation. METHODS: Measurement of corneal parameters was performed using Scheimpflug imaging (Pentacam HR, Oculus, Germany). In 125 eyes with Fuch's dystrophy undergoing DMEK, a fictitious keratometer index was calculated based on the PPPA ratio. The preoperative and postoperative keratometer indices and PA ratios were also determined. Results were compared to those obtained in a control group consisting of 125 eyes without corneal pathologies. Calculated mean ratios and keratometer indices were then used to convert the anterior corneal radius in each eye before DMEK to postoperative posterior and total corneal power. To assess the most appropriate ratio and keratometer index, predicted and measured powers were compared using Bland-Altman plots. RESULTS: The PPPA ratio determined in eyes with Fuch's dystrophy undergoing DMEK was significantly different (P < 0.001) to the PA ratio in eyes without corneal pathologies. Using the mean PA ratio (0.822) and keratometer index (1.3283), calculated with the control group data to convert the anterior corneal radius before DMEK to power, leads to a significant (P < 0.001) underestimation of postoperative posterior negative corneal power (mean difference (∆ = - 0.14D ± 0.30) and overestimation of total corneal power (∆ = - 0.45D ± 1.08). The lowest prediction errors were found using the geometric mean PPPA ratio (0.806) and corresponding keratometer index (1.3273) to predict the postoperative posterior (∆ = - 0.01 ± 0.30) and total corneal powers (∆ = - 0.32D ± 1.08). CONCLUSIONS: Corneal power estimation using conventional K for IOL power calculation is invalid in eyes with Fuch's dystrophy undergoing DMEK. To avoid an overestimation of corneal power and minimize the risk of a postoperative hyperopic shift, conventional K for IOL power calculation should be adjusted in eyes with Fuch's dystrophy undergoing cataract surgery combined with DMEK. The fictitious PPPA ratio and keratometer index may guide further IOL power calculation methods to achieve this.

[Physical Therapy of Non-Exudative Age-Related Macular Degeneration]. / Physikalische Therapie der nicht exsudativen altersabhängigen Makuladegeneration.

Despite the success of anti-VEGF therapy (VEGF: vascular endothelial growth factor) in neovascular age-related macular degeneration (AMD) in the last decade, many unmet needs in AMD management remain. In order to improve patient eye health and relieve the burden on health systems, the development of new intervention options appears to be of great importance if they can delay or even prevent the progression of an early form into a late form. In the field of physical treatment for non-exudative AMD, there is no recognised therapy procedure to date. It now appears appropriate to pursue further research efforts in the field of intraocular blue filter lenses and subthreshold laser treatment in prospective studies. The following article provides an overview of the current strategies of physical therapy for non-exudative AMD.

[Expression of Motion Artifacts in OCT-Angiography Imaging in Healthy Subjects Using Two Different Devices]. / Vergleich der Bildqualität zweier unterschiedlicher OCT-Angiografie-Systeme mit Fokus auf Bewegungsartefakten bei gesunden Probanden.

BACKGROUND: To compare the expression of motion artifacts in optical coherence tomography angiography (OCT-A) in healthy subjects using two different devices. METHODS: In this study, 25 eyes of 25 healthy volunteers with no history of any ocular disease or ocular surgery were included. OCT-A imaging was performed using the RTVue XR Avanti (Optovue Inc., Fremont, California, USA) and the Spectralis OCT-A (Heidelberg Engineering, Heidelberg, Deutschland). The macula was imaged twice in each proband with active eye tracking (ET) using a 3 × 3 mm2 or a 10 × 10° scan, respectively. The expression of motion artifact was analyzed by two independent readers in the superficial OCT-angiogram using the Motion Artifact Score (MAS). RESULTS: The signal strength index (SSI) was 73.0 ± 7.8 (Optovue) and 39.6 ± 3.6 (Heidelberg), which is equivalent to 73.0% (Optovue SSImax = 100 = 100%) and 79.2% (SSImax = 50 = 100%) of the maximum quality score. Both devices showed a very good image quality (mean MAS Optovue: 1.32 ± 0.551, mean MAS Heidelberg: 1.7 ± 0.789, p = 0.006). Of all measurements, quilting/banding was found in 20% of Optovue patients (10/50) and 6% of Heidelberg patients (3/50). Stretching was found in 4% of Optovue patients (2/50) and in 6% of Heidelberg patients (3/50). Vessel doubling was only seen in one Optovue angiogram (2%) as well as a displacement (2%). Blink lines only existed in three Heidelberg angiograms (6%). CONCLUSION: Despite different software and hardware approaches, both devices were able to take high-quality images with a very low prevalence of motion artifacts. Nevertheless, these artifacts still also occur in healthy subjects with good fixation. With regards to MAS, there was a high agreement between the two readers. However, the analysis of artifacts remains complex and requires experience as well as a precise assessment in evaluating OCT-A images.

Using Deep Learning in Automated Detection of Graft Detachment in Descemet Membrane Endothelial Keratoplasty: A Pilot Study.

PURPOSE: To evaluate a deep learning-based method to automatically detect graft detachment (GD) after Descemet membrane endothelial keratoplasty (DMEK) in anterior segment optical coherence tomography (AS-OCT). METHODS: In this study, a total of 1172 AS-OCT images (609: attached graft; 563: detached graft) were used to train and test a deep convolutional neural network to automatically detect GD after DMEK surgery in AS-OCT images. GD was defined as a not completely attached graft. After training with 1072 of these images (559: attached graft; 513: detached graft), the created classifier was tested with the remaining 100 AS-OCT scans (50: attached graft; 50 detached: graft). Hereby, a probability score for GD (GD score) was determined for each of the tested OCT images. RESULTS: The mean GD score was 0.88 ± 0.2 in the GD group and 0.08 ± 0.13 in the group with an attached graft. The differences between both groups were highly significant (P < 0.001). The sensitivity of the classifier was 98%, the specificity 94%, and the accuracy 96%. The coefficient of variation was 3.28 ± 6.90% for the GD group and 2.82 ± 3.81% for the graft attachment group. CONCLUSIONS: With the presented deep learning-based classifier, reliable automated detection of GD after DMEK is possible. Further work is needed to incorporate information about the size and position of GD and to develop a standardized approach regarding when rebubbling may be needed.

Deep learning-based detection and classification of geographic atrophy using a deep convolutional neural network classifier.

PURPOSE: To automatically detect and classify geographic atrophy (GA) in fundus autofluorescence (FAF) images using a deep learning algorithm. METHODS: In this study, FAF images of patients with GA, a healthy comparable group and a comparable group with other retinal diseases (ORDs) were used to train a multi-layer deep convolutional neural network (DCNN) (1) to detect GA and (2) to differentiate in GA between a diffuse-trickling pattern (dt-GA) and other GA FAF patterns (ndt-GA) in FAF images. 1. For the automated detection of GA in FAF images, two classifiers were built (GA vs. healthy/GA vs. ORD). The DCNN was trained and validated with 400 FAF images in each case (GA 200, healthy 200, or ORD 200). For the subsequent testing, the built classifiers were then tested with 60 untrained FAF images in each case (AMD 30, healthy 30, or ORD 30). Hereby, both classifiers automatically determined a GA probability score and a normal FAF probability score or an ORD probability score. 2. To automatically differentiate between dt-GA and ndt-GA, the DCNN was trained and validated with 200 FAF images (dt-GA 72; ndt-GA 138). Afterwards, the built classifier was tested with 20 untrained FAF images (dt-GA 10; ndt-GA 10) and a dt-GA probability score and an ndt-GA probability score was calculated. For both classifiers, the performance of the training and validation procedure after 500 training steps was measured by determining training accuracy, validation accuracy, and cross entropy. RESULTS: For the GA classifiers (GA vs. healthy/GA vs. ORD), the achieved training accuracy was 99/98%, the validation accuracy 96/91%, and the cross entropy 0.062/0.100. For the dt-GA classifier, the training accuracy was 99%, the validation accuracy 77%, and the cross entropy 0.166. The mean GA probability score was 0.981 ± 0.048 (GA vs. healthy)/0.972 ± 0.439 (GA vs. ORD) in the GA image group and 0.01 ± 0.016 (healthy)/0.061 ± 0.072 (ORD) in the comparison groups (p < 0.001). The mean dt-GA probability score was 0.807 ± 0.116 in the dt-GA image group and 0.180 ± 0.100 in the ndt-GA image group (p < 0.001). CONCLUSION: For the first time, this study describes the use of a deep learning-based algorithm to automatically detect and classify GA in FAF. Hereby, the created classifiers showed excellent results. With further developments, this model may be a tool to predict the individual progression risk of GA and give relevant information for future therapeutic approaches.

Automated detection of exudative age-related macular degeneration in spectral domain optical coherence tomography using deep learning.

PURPOSE: Our purpose was to use deep learning for the automated detection of age-related macular degeneration (AMD) in spectral domain optical coherence tomography (SD-OCT). METHODS: A total of 1112 cross-section SD-OCT images of patients with exudative AMD and a healthy control group were used for this study. In the first step, an open-source multi-layer deep convolutional neural network (DCNN), which was pretrained with 1.2 million images from ImageNet, was trained and validated with 1012 cross-section SD-OCT scans (AMD: 701; healthy: 311). During this procedure training accuracy, validation accuracy and cross-entropy were computed. The open-source deep learning framework TensorFlow™ (Google Inc., Mountain View, CA, USA) was used to accelerate the deep learning process. In the last step, a created DCNN classifier, using the information of the above mentioned deep learning process, was tested in detecting 100 untrained cross-section SD-OCT images (AMD: 50; healthy: 50). Therefore, an AMD testing score was computed: 0.98 or higher was presumed for AMD. RESULTS: After an iteration of 500 training steps, the training accuracy and validation accuracies were 100%, and the cross-entropy was 0.005. The average AMD scores were 0.997 ± 0.003 in the AMD testing group and 0.9203 ± 0.085 in the healthy comparison group. The difference between the two groups was highly significant (p < 0.001). CONCLUSIONS: With a deep learning-based approach using TensorFlow™, it is possible to detect AMD in SD-OCT with high sensitivity and specificity. With more image data, an expansion of this classifier for other macular diseases or further details in AMD is possible, suggesting an application for this model as a support in clinical decisions. Another possible future application would involve the individual prediction of the progress and success of therapy for different diseases by automatically detecting hidden image information.

Quantitative changes in flow density in patients with adult-onset foveomacular vitelliform dystrophy: an OCT angiography study.

PURPOSE: To quantitatively compare the flow density, the retinal thickness, and the area of the foveal avascular zone (FAZ) between patients with adult-onset foveomacular vitelliform dystrophy (AOFVD) and a healthy controls. METHODS: Thirteen eyes (eight patients) with AOFVD and 13 matched eyes (13 patients) without any ocular pathology were included in this study. A 6 × 6 mm optical coherence tomography angiography (OCTA) scan was performed for every included eye. The flow density (superficial retinal vascular layer, deep retinal vascular layer and choriocapillary layer), retinal thickness and FAZ (superficial retinal vascular layer and deep retinal vascular layer) were subsequently analyzed. RESULTS: The mean flow density was decreased in the AOFVD patients in all measured vascular layers. The difference from the control group was statistically significant in the parafoveal sector of the deep retinal vascular layer (P = 0.02), and a clear trend was found in the superficial retinal vascular layer (P = 0.05). Both groups had comparable FAZs in the superficial and deep retinal vascular layers. The retinal thickness values were higher in the fovea (P = 0.840) and lower in the parafoveal sectors (P = 0.125). The difference was significant in the superior parafoveal sector (P = 0.034). CONCLUSIONS: Flow densities as measured by OCTA are decreased in the superficial retinal vascular layer and the deep retinal vascular layer in patients with AOFVD. These findings could be helpful for diagnosing and understanding the pathogenesis of this disease.