A Deep Feature Fusion of Improved Suspected Keratoconus Detection with Deep Learning.

Detection of early clinical keratoconus (KCN) is a challenging task, even for expert clinicians. In this study, we propose a deep learning (DL) model to address this challenge. We first used Xception and InceptionResNetV2 DL architectures to extract features from three different corneal maps collected from 1371 eyes examined in an eye clinic in Egypt. We then fused features using Xception and InceptionResNetV2 to detect subclinical forms of KCN more accurately and robustly. We obtained an area under the receiver operating characteristic curves (AUC) of 0.99 and an accuracy range of 97-100% to distinguish normal eyes from eyes with subclinical and established KCN. We further validated the model based on an independent dataset with 213 eyes examined in Iraq and obtained AUCs of 0.91-0.92 and an accuracy range of 88-92%. The proposed model is a step toward improving the detection of clinical and subclinical forms of KCN.

The effect of accelerated pulsed high-fluence corneal cross-linking on corneal endothelium; a prospective specular microscopy study.

BACKGROUND: Corneal collagen cross-linking (CXL) is a procedure utilized for halting keratoconus progression with different approved protocols. The current study aimed to assess the corneal endothelial changes following the relatively new accelerated pulsed high-fluence protocol of epithelium-off corneal cross-linking for the treatment of mild to moderate keratoconus. METHODS: This prospective case series study enrolled 45 eyes of 27 patients with mild to moderate progressive keratoconus who underwent accelerated pulsed high-fluence CXL (pl-ACXL, 30 mW/ cm2 UVA at 365 nm wavelength, 8 min pulsed mode 1 s on / 1 s off with a total energy of 7.2 J/ cm2). The main outcome measures were corneal endothelial changes assessed by specular microscopy at 3 and 6 months postoperatively including endothelial cell density (ECD), coefficient of variation, percentage of hexagonal cells, average, minimum and maximum endothelial cell sizes. Demarcation line depth was assessed 1 month following surgery. RESULTS: The mean age of the studied sample was 24.89 ± 7.21. The mean preoperative ECD (2944.6 ± 247.41 cell/mm2) showed non-significant reduction at 3 and 6 months postoperatively (2931.03 ± 253.82 and 2924.7 ± 224.88 cell/mm2, respectively, P-value = 0.361). There were no significant changes in the mean coefficient of variation, percentage of hexagonal cells, average, minimum and maximum endothelial cell sizes at 3 and 6 months following pl-ACXL (P-value > 0.05). The mean demarcation line depth 1 month after pl-ACXL was 214 ± 17.43 µm. CONCLUSIONS: Corneal endothelial changes following accelerated pulsed high-fluence CXL were minimal with stability of endothelial cell count and non-significant morphological changes. TRIAL REGISTRATION: Clinicaltrials.gov: NCT04160338 (13/11/2019).

Detecting dry eye from ocular surface videos based on deep learning.

OBJECTIVE: To assess the performance of convolutional neural networks (CNNs) for automated diagnosis of dry eye (DE) in patients undergoing video keratoscopy based on single ocular surface video frames. DESIGN: This retrospective cohort study included 244 ocular surface videos from 244 eyes of 244 subjects based on corneal topography. A total of 116 eyes were normal while 128 eyes had DE based on clinical evaluations. METHODS: We developed a deep transfer learning model to directly identify DE from ocular surface videos. We evaluated the performance of the CNN model based on objective accuracy metrics. We assessed the clinical relevance of the findings by evaluating class activations maps. MAIN OUTCOME MEASURE: Area under the receiver operating characteristics curve (AUC), accuracy, specificity, and sensitivity. RESULTS: The AUC of the model for discriminating normal eyes from eyes with DE was 0.98. Network activation maps suggested that the lower paracentral cornea was the most important region for detection of DE by the CNN model. CONCLUSIONS: Deep transfer learning achieved a high diagnostic accuracy in detecting DE based on non-invasive ocular surface videos at levels that may prove useful in clinical practice.

Q-value customized versus wavefront-optimized ablation in femtosecond laser-assisted LASIK for myopia and myopic astigmatism: a prospective contralateral comparative study.

BACKGROUND: Corneal refractive surgery for myopia results in an oblate shift with increased postoperative aberrations inversely affecting the quality of vision. Aspheric ablation profiles have been introduced to minimize such a problem. The aim of this study was to compare changes in corneal asphericity, central and mid-peripheral pachymetry between the Q-value customized and the wavefront-optimized (WFO) ablation profiles. METHODS: A prospective, comparative non-randomized fellow eye study was conducted. Eighty eyes of 40 eligible patients underwent femtosecond laser-assisted laser in situ keratomileusis for myopia and myopic astigmatism. In each patient, the more myopic eye was included in the custom-Q ablation experimental group and the other less myopic eye was included in the WFO control group. For the custom-Q group, the target asphericity was set to the preoperative Q-value. Corneal asphericity, central and mid-peripheral pachymetric changes and the root mean square of corneal higher-order aberrations (RMSh) were assessed 6 months following surgery. Visual and refractive outcomes were also evaluated in both platforms 6 months postoperatively. RESULTS: The mean preoperative refractive spherical equivalent was significantly more myopic in the custom-Q group than in the WFO group (P = 0.001). The mean Q-value changed from - 0.2 ± 0.1 to 0.6 ± 0.7 and from - 0.2 ± 0.1 to 0.4 ± 0.5 in the custom-Q and WFO groups, respectively. The oblate shift in corneal asphericity was not significantly different between both treatment groups (P = 0.094). The mean ablation depth at the pupillary center was significantly greater in the custom-Q group (P = 0.011), while there was no significant difference at the mid-peripheral pachymetry (P = 0.256). The RMSh significantly increased in both treatment profiles (P < 0.001) with no significant difference between the two groups (P = 0.06). The uncorrected distance visual acuity (UDVA) and the manifest refraction spherical equivalents (MRSE) significantly improved in both treatment groups (P < 0.001). CONCLUSIONS: The custom-Q treatment profile with target asphericity set at the preoperative Q-value achieved comparable outcomes vs. the WFO profile in terms of postoperative corneal asphericity and mid-peripheral pachymetry despite the greater amount of ablation, the smaller optical zone, and the resulting increase in postoperative corneal flattening in the custom-Q group. Trial registration (Clinicaltrials.gov): NCT04738903, 4 February 2021- Retrospectively registered, https://clinicaltrials.gov/ct2/show/NCT04738903.

Bridging the resources gap: deep learning for fluorescein angiography and optical coherence tomography macular thickness map image translation.

BACKGROUND: To assess the ability of the pix2pix generative adversarial network (pix2pix GAN) to synthesize clinically useful optical coherence tomography (OCT) color-coded macular thickness maps based on a modest-sized original fluorescein angiography (FA) dataset and the reverse, to be used as a plausible alternative to either imaging technique in patients with diabetic macular edema (DME). METHODS: Original images of 1,195 eyes of 708 nonconsecutive diabetic patients with or without DME were retrospectively analyzed. OCT macular thickness maps and corresponding FA images were preprocessed for use in training and testing the proposed pix2pix GAN. The best quality synthesized images using the test set were selected based on the Fréchet inception distance score, and their quality was studied subjectively by image readers and objectively by calculating the peak signal-to-noise ratio, structural similarity index, and Hamming distance. We also used original and synthesized images in a trained deep convolutional neural network (DCNN) to plot the difference between synthesized images and their ground-truth analogues and calculate the learned perceptual image patch similarity metric. RESULTS: The pix2pix GAN-synthesized images showed plausible subjectively and objectively assessed quality, which can provide a clinically useful alternative to either image modality. CONCLUSION: Using the pix2pix GAN to synthesize mutually dependent OCT color-coded macular thickness maps or FA images can overcome issues related to machine unavailability or clinical situations that preclude the performance of either imaging technique. TRIAL REGISTRATION: ClinicalTrials.gov Identifier: NCT05105620, November 2021. "Retrospectively registered".

Thin-Flap Laser In Situ Keratomileusis-Associated Dry Eye: A Comparative Study Between Femtosecond Laser and Mechanical Microkeratome-Assisted Laser In Situ Keratomileusis.

OBJECTIVE: To assess and compare postlaser in situ keratomileusis (LASIK) dry eye after LASIK with planned thin flaps created by a femtosecond laser (FS) and mechanical microkeratome (MK). METHODS: Patients were stratified according to the flap creation technique into FS and MK groups with planned 100 µm flap thickness in all eyes. Dry eye parameters including tear film break-up time (TBUT), Schirmer I test, ocular surface disease index (OSDI), and lower tear meniscus height and area (tear meniscus height [TMH] and tear meniscus area [TMA]) were assessed before surgery and at 3 and 6 months after surgery. RESULTS: The study included 102 eyes of 55 patients (52 eyes underwent FS-LASIK/50 eyes underwent MK-LASIK). The preoperative characteristics including age, gender, and spherical equivalents were similar in both groups (P>0.05). The difference in postoperative flap thickness was statistically significant (P<0.05). In each group, dry eye parameters showed a statistically significant worsening (P value<0.001) at 3 months with later significant improvement at 6 months but did not return to baseline. The mean TBUT, Schirmer I, and lower TMH and TMA were statistically higher, and the mean OSDI was statistically lower in FS-LASIK than the MK-LASIK group at both 3 and 6 months (P value<0.001). The depth of ablation had a statistically significant correlation with Schirmer I at 6 months and OSDI at both 3 and 6 months post-LASIK (P<0.05). CONCLUSION: The use of femtosecond laser technology in performing thin-Flap LASIK eye surgeries has improved the outcome of post-LASIK dry eye when compared with the mechanical microkeratome.

Pix2pix Conditional Generative Adversarial Networks for Scheimpflug Camera Color-Coded Corneal Tomography Image Generation.

Purpose: To assess the ability of pix2pix conditional generative adversarial network (pix2pix cGAN) to create plausible synthesized Scheimpflug camera color-coded corneal tomography images based upon a modest-sized original dataset to be used for image augmentation during training a deep convolutional neural network (DCNN) for classification of keratoconus and normal corneal images. Methods: Original images of 1778 eyes of 923 nonconsecutive patients with or without keratoconus were retrospectively analyzed. Images were labeled and preprocessed for use in training the proposed pix2pix cGAN. The best quality synthesized images were selected based on the Fréchet inception distance score, and their quality was studied by calculating the mean square error, structural similarity index, and the peak signal-to-noise ratio. We used original, traditionally augmented original and synthesized images to train a DCNN for image classification and compared classification performance metrics. Results: The pix2pix cGAN synthesized images showed plausible subjectively and objectively assessed quality. Training the DCNN with a combination of real and synthesized images allowed better classification performance compared with training using original images only or with traditional augmentation. Conclusions: Using the pix2pix cGAN to synthesize corneal tomography images can overcome issues related to small datasets and class imbalance when training computer-aided diagnostic models. Translational Relevance: Pix2pix cGAN can provide an unlimited supply of plausible synthetic Scheimpflug camera color-coded corneal tomography images at levels useful for experimental and clinical applications.

Comparison of the phaco chop and drill-and-crack techniques for phacoemulsification of hard cataracts: A fellow eye study.

PURPOSE: To quantify the intraoperative parameters and postoperative outcomes after using the phaco chop technique in one eye and drill-and-crack technique in the other eye in patients with bilateral dense brunescent cataract. METHODS: The Lens Opacities Classification System III grading system was used to select 132 eyes of 66 patients with bilateral nuclear opalescence (NO) grade NO4 or grade NO5. One eye in each patient with bilateral dense brunescent cataract was subjected to phacoemulsification using the phaco chop technique, while the other eye was subjected to phacoemulsification with the drill-and-crack technique for nucleus disassembly. The intraoperative parameters were quantified. Surgical outcome was assessed preoperatively and 1 day, 4 weeks and 12 weeks postoperatively, and the outcomes of the two techniques were compared. RESULTS: There was no significant difference between the techniques in operative parameters [cumulative dissipated energy (p = 0.74), surgical time (p = 0.68) or surgical difficulty during nucleus disassembly (p = 0.80)]. There was no significant difference in the postoperative change in central corneal thickness between the techniques at day 1, 4 weeks and 12 weeks or in corneal endothelial cell density loss at 4 and 12 weeks (p > 0.05). There was no significant difference between the techniques in the mean corrected distance visual acuity (logarithm of the minimum angle resolution) at 4 weeks postoperatively (p = 0.25). CONCLUSION: The phaco chop and drill-and-crack techniques are equally effective for disassembly of hard NO4 and NO5 cataracts.

Correlation between corneal demarcation line depth in epithelium-off and trans-epithelium accelerated corneal cross linking and keratoconus progression.

AIM: To compare corneal demarcation line (DL) depth in both accelerated epithelium-off and trans-epithelium cross linking (CXL) using anterior segment optical coherence tomography (AS-OCT) and its relation to maximum keratometry (Kmax) progression in both techniques. METHODS: A prospective comparative interventional study where patients with mild to moderate keratoconus (KC) were classified into two groups: accelerated epithelium-off and trans-epithelium CXL based on corneal pachymetry. Assessment of corneal DL depth was carried out after 3mo by AS-OCT. Kmax readings were evaluated after one year follow up using the Scheimpflug imaging system. RESULTS: Study included 74 eyes of 44 patients. Group A underwent epithelium-off CXL (41 eyes), while Group B underwent trans-epithelium CXL (33 eyes). At 3mo follow up, mean corneal DL depth in Group A was 219.9±58.4 µm while in Group B was 127.2±7.8 µm (P<0.05). The mean Kmax changed from 51.9±3.9 to 51.3±4.2 diopters in Group A and from 53.1±4.1 to 53.6±5 diopters in Group B with insignificant difference in Kmax changes in either group (P>0.05). In addition, no significant change in corneal pachymetry was found in both groups (mean change at 1y: 6.4±4.7 and -10.1±2.3 µm in Groups A and B respectively). CONCLUSION: Despite a significantly deeper corneal DL depth created by accelerated epithelium-off CXL technique compared to accelerated trans-epithelium CXL, there is no significant impact on keratoconus progression.

Flap Sliding Technique for Managing Flap Striae following Laser In Situ Keratomileusis.

PURPOSE: To assess the efficacy and safety of a simple, noninvasive, "flap-sliding" technique for managing flap striae following laser in situ keratomileusis (LASIK). METHODS: This prospective, interventional study included eyes with post-LASIK flap striae. All eyes underwent flap sliding 1-2 days after surgery. Following flap edge epithelialisation, a cellulose sponge was used to gently slide the flap perpendicular to the striae direction. This technique allows for flap striae treatment without flap lifting, avoiding any associated lifting complications. Uncorrected distance visual acuity (UDVA), corrected distance visual acuity (CDVA), and refractive error were monitored one day after the flap-sliding procedure. RESULTS: Fifteen eyes (15 patients) with post-LASIK flap striae were managed using the flap-sliding technique. The procedure did not successfully relocate the flap striae in 1 eye, and flap elevation and floating (using a balanced salt solution) were required. Therefore, 14 eyes were included in post-flap-sliding analyses. The UDVA improved in all patients the first day after the flap-sliding procedure was performed, with 11 of 14 eyes (78.57%) reaching an UDVA of 20/25 or better. Complications following flap sliding occurred in 2 eyes (14.29%). One eye had intraoperative epithelial abrasion, and 1 eye had residual postoperative striae outside of the optical zone. CONCLUSION: The flap-sliding technique is a simple, noninvasive, efficient, and safe technique for managing post-LASIK flap striae that develop after epithelial healing in the early post-LASIK period. This trial is registered with NCT04055337.

Causes and Clinical Impact of Loss to Follow-Up in Patients with Proliferative Diabetic Retinopathy.

PURPOSE: This study determined the clinical impact and causes of loss to follow-up (LTFU) from the patients' perspective in individuals with proliferative diabetic retinopathy (PDR) who received panretinal photocoagulation (PRP) and/or intravitreal injections (IVIs) of antivascular endothelial growth factor (VEGF). METHODS: This prospective cohort study included 467 patients with PDR who received PRP and/or IVIs of anti-VEGF between May 2013 and June 2018. LTFU was defined as missing any follow-up visit for any interval exceeding 6 months, provided that patients eventually resumed care. Main outcome measures include rates and causes of LTFU. RESULTS: A total of 391 patients (83.7%) were followed up, and 76 patients (16.3%) were LTFU over the study period. Rates of LTFU decreased with age (P=0.005). Questionnaire analysis conducted for patients' LTFU showed a significant positive correlation between best corrected visual activity (BCVA) loss and patient's lack of trust and satisfaction with treatment (rs = 0.458, P=0.005). Questionnaire analysis conducted for patients' LTFU showed a significant positive correlation between best corrected visual activity (BCVA) loss and patient's lack of trust and satisfaction with treatment (rs = 0.458, P=0.005). Questionnaire analysis conducted for patients' LTFU showed a significant positive correlation between best corrected visual activity (BCVA) loss and patient's lack of trust and satisfaction with treatment (rs = 0.458, P=0.005). Questionnaire analysis conducted for patients' LTFU showed a significant positive correlation between best corrected visual activity (BCVA) loss and patient's lack of trust and satisfaction with treatment (rs = 0.458. CONCLUSIONS: LTFU threatens vision in PDR patients receiving PRP and/or IVIs of anti-VEGF. Possibly, patient-specific LTFU causes should be addressed before treatment in order to minimize the risk of LTFU. The clinical trial is registered with NCT04018326 (trial registration: ClinicalTrials.gov Identifier: NCT04018326, 10th of July 2019 "Retrospectively registered").

Classification of Color-Coded Scheimpflug Camera Corneal Tomography Images Using Deep Learning.

Purpose: To assess the use of deep learning for high-performance image classification of color-coded corneal maps obtained using a Scheimpflug camera. Methods: We used a domain-specific convolutional neural network (CNN) to implement deep learning. CNN performance was assessed using standard metrics and detailed error analyses, including network activation maps. Results: The CNN classified four map-selectable display images with average accuracies of 0.983 and 0.958 for the training and test sets, respectively. Network activation maps revealed that the model was heavily influenced by clinically relevant spatial regions. Conclusions: Deep learning using color-coded Scheimpflug images achieved high diagnostic performance with regard to discriminating keratoconus, subclinical keratoconus, and normal corneal images at levels that may be useful in clinical practice when screening refractive surgery candidates. Translational Relevance: Deep learning can assist human graders in keratoconus detection in Scheimpflug camera color-coded corneal tomography maps.

Safety of Trypan Blue Capsule Staining to Corneal Endothelium in Patients with Diabetic Retinopathy.

PURPOSE: To study the potential corneal endothelial cell toxicity of trypan blue (TB) when used for phacoemulsification to stain the anterior capsule in patients with diabetic retinopathy. METHODS: This was a single-center, prospective, randomized, individual cohort study. One eye in each patient with diabetic retinopathy underwent phacoemulsification without trypan blue capsule staining (control eye), while the other eye underwent phacoemulsification with trypan blue capsule staining (study eye). Both eyes underwent intraocular lens implantation. Preoperative and four-week postoperative quantitative and qualitative morphometric endothelial cell analyses of the cornea were performed using noncontact specular microscopy. RESULTS: There were no significant differences in endothelial cell density (mean ± SD for the study group: 2506.74 ± 413.99 cells/mm2; mean ± SD for the control eyes: 2466.34 ± 369.12 cells/mm2; P=0.316) or endothelial cell density (CD) loss% (mean CD loss% was 7.23 ± 13.31 for the study eyes and 9.94 ± 9.36 for the control eyes; P=0.157) four weeks after the operation. Additionally, no significant differences were seen in the percentage of hexagonal cells, coefficient of variation, or corneal thickness between the two groups preoperatively and 4 weeks postoperatively. CONCLUSIONS: Direct administration of trypan blue into the anterior chamber for staining of the anterior capsule during cataract surgery did not result in any significant corneal endothelial changes on specular microscopy in patients with severe nonproliferative diabetic retinopathy or high-risk proliferative diabetic retinopathy at 4 weeks postoperatively. This trial is registered with NCT03755752.