Accurate prediction of glaucoma from colour fundus images with a convolutional neural network that relies on active and transfer learning.

PURPOSE: To assess the use of deep learning (DL) for computer-assisted glaucoma identification, and the impact of training using images selected by an active learning strategy, which minimizes labelling cost. Additionally, this study focuses on the explainability of the glaucoma classifier. METHODS: This original investigation pooled 8433 retrospectively collected and anonymized colour optic disc-centred fundus images, in order to develop a deep learning-based classifier for glaucoma diagnosis. The labels of the various deep learning models were compared with the clinical assessment by glaucoma experts. Data were analysed between March and October 2018. Sensitivity, specificity, area under the receiver operating characteristic curve (AUC), and amount of data used for discriminating between glaucomatous and non-glaucomatous fundus images, on both image and patient level. RESULTS: Trained using 2072 colour fundus images, representing 42% of the original training data, the trained DL model achieved an AUC of 0.995, sensitivity and specificity of, respectively, 98.0% (CI 95.5%-99.4%) and 91% (CI 84.0%-96.0%), for glaucoma versus non-glaucoma patient referral. CONCLUSIONS: These results demonstrate the benefits of deep learning for automated glaucoma detection based on optic disc-centred fundus images. The combined use of transfer and active learning in the medical community can optimize performance of DL models, while minimizing the labelling cost of domain-specific mavens. Glaucoma experts are able to make use of heat maps generated by the deep learning classifier to assess its decision, which seems to be related to inferior and superior neuroretinal rim (within ONH), and RNFL in superotemporal and inferotemporal zones (outside ONH).

The Role of LOX and LOXL2 in the Pathogenesis of an Experimental Model of Choroidal Neovascularization.

PURPOSE: We investigated whether lysyl oxidase (LOX) and lysyl oxidase-like 2 (LOXL2) play a role in an experimental model of choroidal neovascularization (CNV). The therapeutic potential of antibodies against LOX (M64) and LOXL2 (AB0023) was evaluated in a murine laser-induced CNV model. METHODS: Expression of LOX and LOXL2 in the posterior eye cups (including retina, retinal pigment epithelium, choroid, and sclera) was studied by qRT-PCR and immunohistochemistry. In the murine model of CNV, both antibodies were administered intraperitoneally every other day until the day killed. On different time points after laser, treatment outcome was studied by immunohistochemical analysis of inflammation, angiogenesis and fibrosis, and by transcript analysis of different cytokines. RESULTS: Levels of LOX and LOXL2 in the posterior eye cups were increased after CNV-induction at different time points after laser. At day 35, their protein expression patterns appeared to correlate with retinal glial cells and endothelial cells, respectively. Both antibodies significantly inhibited fibrosis, whereas AB0023 also significantly reduced angiogenesis and inflammation. Transcript levels of α-1 type I collagen (COL1A1) in the posterior eye cups were significantly decreased in lasered mice treated with either M64 or AB0023. Vascular endothelial growth factor expression was also reduced only after AB0023 treatment, whereas activated fibroblast marker α-smooth muscle actin (αSMA) levels were not significantly changed. CONCLUSIONS: This study suggests that LOX and LOXL2 may play an important role in the pathogenesis of AMD. Targeting LOXL2 could have a broader efficacy than targeting LOX, by reducing angiogenesis and inflammation, as well as fibrosis.

Trabeculectomy with or without anterior chamber maintainer and adjustable sutures.

PURPOSE: To assess the long-term outcomes of trabeculectomy using an anterior chamber maintainer and a combination of adjustable and releasable sutures compared with trabeculectomy using a viscoelastic and solely releasable sutures. METHODS: In a prospective, randomized, single-center study, 2 different trabeculectomy techniques were compared. One eye of each patient was randomly assigned to trabeculectomy with an anterior chamber maintainer and a combination of 2 adjustable and 2 releasable flap sutures (technique 1), whereas the other eye had trabeculectomy with viscoelastic and 2 releasable flap sutures (technique 2). The postoperative management was comparable in both groups except for the manipulation of the adjustable sutures. The main outcome measures were the postoperative intraocular pressure (IOP), the frequency of early postoperative complications and interventions, and the number of glaucoma medications. RESULTS: Thirty-two patients were included and had a follow-up of at least 18 months. The preoperative and postoperative IOP at 18 months was 22.1±8 and 13.3±2 mm Hg in the group using technique 1 and 22.7±7 and 13.3±3 mm Hg in the group using technique 2, respectively (mean±SD; P=0.18 for IOP reduction between both techniques). Early postoperative complications and interventions were infrequent and comparable in both techniques. The reduction in number of medications was not significantly different between the 2 groups (mean±SD from 2.4±1.0 to 0.4±0.8 in the group using technique 1 and from 2.3±1.0 to 0.2±0.5 in the group using technique 2; P=0.50). CONCLUSIONS: There is no significant difference in IOP reduction, in frequency of postoperative complications and interventions, or in reduction of medications between the 2 techniques.

The role of LOX and LOXL2 in scar formation after glaucoma surgery.

PURPOSE: The aim of this study was to elucidate the role of lysyl oxidase (LOX) and lysyl oxidase like (LOXL) 2 in pathologic wound healing after glaucoma surgery. We therefore investigated the expression of LOX and LOXL2 and evaluated the therapeutic potential of anti-LOX (GS-639556, formerly M64) and anti-LOXL2 (GS-607601, formerly AB0023) antibodies in a rabbit model of glaucoma trabeculectomy. METHODS: Ocular expression of LOX and LOXL2 was investigated by immunohistologic staining at different time points after trabeculectomy. Treatment with GS-639556 or GS-607601 was initiated in rabbits immediately after trabeculectomy by giving both intracameral and subconjunctival injections. Thereafter, the antibodies were given twice a week subconjunctivally until day 30 after surgery (day of euthanization). Treatment outcome was studied by clinical investigation of the bleb and by immunohistochemical analysis of angiogenesis, inflammation, and collagen deposition. RESULTS: LOX and LOXL2 were both upregulated in Tenon's capsule and the conjunctiva after glaucoma surgery. Repeated administration of LOX- or LOXL2-targeting monoclonal antibodies increased bleb area and bleb survival. Analyses of immunohistologic stainings showed that both antibodies significantly decreased fibrosis, whereas the anti-LOXL2 antibody also significantly reduced blood vessel density and inflammation. CONCLUSIONS: Targeting LOXL2 with an inhibitory monoclonal antibody (GS-607601) reduced pathologic angiogenesis, inflammation, and fibrosis. These results suggest that LOXL2 could be an appealing target for treatment of scar formation after glaucoma surgery, and point to the potential therapeutic benefits of simtuzumab, a humanized monoclonal antibody derived from GS-607601.

The role of different VEGF isoforms in scar formation after glaucoma filtration surgery.

Vascular endothelial growth factor (VEGF) plays an important role in both physiological and pathological angiogenesis. Our previous studies showed a differential role of VEGF isoforms in retinal physiological angiogenesis. We also demonstrated that non-selective inhibition of VEGF by bevacizumab had a beneficial effect on surgical outcome after glaucoma filtration surgery by reducing angiogenesis as well as fibrosis. However, the function of the VEGF isoforms in pathological angiogenesis and wound healing in the eye still remains unidentified. This study was designed to elucidate the differential roles of VEGF isoforms in scar formation after trabeculectomy. Furthermore, we also investigated whether pegaptanib (Macugen™, Pfizer), an aptamer which specifically blocks VEGF(165), could improve surgical outcome by reducing postoperative scarring. VEGF-R2 and neuropilin-1 (NRP-1) expression was analyzed in vitro by RT-PCR, and were found to be expressed at higher levels in human umbilical vein endothelial cells (HUVEC) as compared to Tenon fibroblasts (TF). The effect of the different VEGF isoforms (VEGF(121), VEGF(165) and VEGF(189)) and pegaptanib on cell proliferation was determined via WST-1 assay. Endothelial cell proliferation was stimulated after addition of VEGF(121) and VEGF(165), whereas VEGF(121) and VEGF(189) increased fibroblast growth. These effects on proliferation were associated with an activation of the ERK pathway, as revealed using the TransAM c-Myc assay. Inhibition of the ERK pathway, by PD98059 administration, significantly reduced VEGF isoform induced cell growth. A dose-dependent reduction of endothelial cell proliferation was observed after pegaptanib administration, while only the highest dose was able to inhibit fibroblast growth. Next, the in vivo effect of pegaptanib was investigated in a rabbit model of trabeculectomy. The surgical outcome was evaluated by performing clinical investigations (IOP, bleb area, height and survival), as well as histomorphometric analyses of angiogenesis (CD31), inflammation (CD45) and fibrosis (Sirius Red). A single postoperative application of pegaptanib had a beneficial impact on surgical outcome, mainly by reducing angiogenesis, but not inflammation or collagen deposition. Repeated injections slightly improved surgical outcome, but again solely by reducing angiogenesis. In summary, our results revealed that the VEGF isoforms play a differential role in ocular wound healing: VEGF(165) and VEGF(121) predominantly affect blood vessel growth, whereas VEGF(189) is rather involved in fibrosis, an important process in wound healing.

Reduced retinal neovascularization, vascular permeability, and apoptosis in ischemic retinopathy in the absence of prolyl hydroxylase-1 due to the prevention of hyperoxia-induced vascular obliteration.

PURPOSE: Prolyl hydroxylases (PHDs) are oxygen sensors that stabilize hypoxia-inducible factors (HIFs) to induce proinflammatory, vasopermeability, and proapoptotic factors. These may be potential targets to reduce the complications of ischemic retinopathies. METHODS: Oxygen-induced ischemic retinopathy (OIR) was generated as a model for retinopathy of prematurity (ROP) by placing 7-day-old mice in 75% oxygen for 5 days and returning them to the relative hypoxia of room air for 5 days. Neovascularization (NV) and avascular areas were assessed on retinal flat-mounts by image analysis. Blood-retinal barrier breakdown was assessed using ³H-mannitol as a tracer. Apoptosis was detected with TUNEL staining. HIF-1α and VEGF were quantified using Western blot analysis and ELISA. RESULTS: PHD1-deficient mice demonstrated reduced hyperoxia-associated vascular obliteration during oxygen-induced ischemic retinopathy. This was associated with subsequent reduced avascularity, vascular leakage, and pathologic NV during the hypoxic phase, which could be accounted for by a reduced expression of HIF-1α and VEGF. Apoptosis in the retina was also reduced in PHD1-depleted mice after 2 days in hyperoxia. CONCLUSIONS: PHD1 deficiency is associated with a reduction of ischemia-induced retinal NV. The regulatory mechanism in this model appears to be: PHD1 depletion prevents HIF-1α degradation in hyperoxia, which induces VEGF, thus preventing hyperoxia-related vessel loss. Without a vessel deficiency, there would not be relative hypoxia when the mice are returned to room air and there would be no need to initiate angiogenesis signaling. Blocking PHD1 may be beneficial for ischemic retinopathies and inflammatory and neurodegenerative disorders.

Preoperative nonsteroidal anti-inflammatory drug or steroid and outcomes after trabeculectomy: a randomized controlled trial.

PURPOSE: To investigate the benefit of preoperative treatment with either topical nonsteroidal anti-inflammatory drug (NSAID) or steroid in terms of clinical outcomes following trabeculectomy. DESIGN: Prospective, randomized placebo-controlled trial. PARTICIPANTS: Sixty-one patients. METHODS: Between July 2005 and October 2007, 61 consecutive medically uncontrolled glaucoma patients scheduled for first-time trabeculectomy were randomized to 1 of 3 study topical medication groups: nonsteroidal anti-inflammatory drugs (ketorolac), steroids (fluorometholone), or placebo (artificial tears). Patients instilled 1 drop 4 times daily for 1 month before the procedure and were examined on days 1 and 2, at weeks 1, 2, and 4, and at months 3, 6, 12, 18, and 24 after trabeculectomy. MAIN OUTCOME MEASURES: Incidence of postoperative surgical or medical interventions (needling, laser suture lysis, needling revision, and intraocular pressure [IOP]-lowering medication). RESULTS: Fifty-four patients (54 eyes) were entered for analysis. The mean number of preoperative medications was 2.3+/-0.9. The mean baseline IOP was 21.0+/-6.0 mmHg. The mean postoperative target IOP was 16.5+/-1.8 mmHg. The mean follow-up was 23.6+/-4.0 months. The percentage of patients requiring needling within the first year was 41% in the placebo group, 6% in the NSAID, and 5% in the steroid group (P = 0.006). The percentage of patients requiring IOP-lowering medication to reach the target IOP at 1 year was 24% in the placebo group, 18% in the NSAID group, and 0% in the steroid group (P = 0.054 overall; P = 0.038 for steroids vs. others). The log-rank test showed a significant (P = 0.019) difference in medication-free survival curves between the different groups. More specifically, patients in the steroid group needed significantly less medication over the total follow-up (P = 0.007). CONCLUSIONS: Topical ketorolac or fluorometholone for 1 month before surgery was associated with improved trabeculectomy outcomes in terms of likelihood of postoperative needling. In the steroid group, there was a significantly reduced need for additional postoperative IOP-lowering medication compared with the other groups. FINANCIAL DISCLOSURE(S): The author(s) have no proprietary or commercial interest in any materials discussed in this article.

Further pharmacological and genetic evidence for the efficacy of PlGF inhibition in cancer and eye disease.

Our findings that PlGF is a cancer target and anti-PlGF is useful for anticancer treatment have been challenged by Bais et al. Here we take advantage of carcinogen-induced and transgenic tumor models as well as ocular neovascularization to report further evidence in support of our original findings of PlGF as a promising target for anticancer therapies. We present evidence for the efficacy of additional anti-PlGF antibodies and their ability to phenocopy genetic deficiency or silencing of PlGF in cancer and ocular disease but also show that not all anti-PlGF antibodies are effective. We also provide additional evidence for the specificity of our anti-PlGF antibody and experiments to suggest that anti-PlGF treatment will not be effective for all tumors and why. Further, we show that PlGF blockage inhibits vessel abnormalization rather than density in certain tumors while enhancing VEGF-targeted inhibition in ocular disease. Our findings warrant further testing of anti-PlGF therapies.

Inhibition of vascular endothelial growth factor reduces scar formation after glaucoma filtration surgery.

PURPOSE: Filtration failure due to excessive postoperative scarring remains a major problem after glaucoma surgery. The authors have investigated whether glaucoma and filtration surgery are associated with increased levels of vascular endothelial growth factor (VEGF), and whether a humanized monoclonal antibody against VEGF, bevacizumab, can reduce postoperative scar formation and improve surgical outcome. METHODS: The levels of VEGF in samples of aqueous humor were measured by ELISA. The expression of the VEGF receptors Flt-1 and KDR was analyzed in cultured Tenon fibroblasts by real-time RT-PCR and Western blotting. The effect of VEGF and bevacizumab on Tenon fibroblasts in vitro was determined using a proliferation assay. The in vivo effect of the antibody was investigated in a rabbit model of trabeculectomy by measuring the intraocular pressure (IOP) and bleb area, and by immunohistological analysis of angiogenesis, inflammation, and fibrosis. RESULTS: VEGF levels were increased significantly in the aqueous humor of glaucoma patients and rabbits that had undergone surgery. Both VEGF receptors were expressed on Tenon fibroblasts. Fibroblast proliferation in vitro was stimulated by delivery of VEGF, and was inhibited by administration of bevacizumab. The antibody also reduced angiogenesis and collagen deposition significantly, and improved the outcome of glaucoma surgery in rabbits. CONCLUSIONS: VEGF was upregulated in the aqueous humor of glaucoma patients and in the rabbit model, and it stimulated fibroblast proliferation in vitro. This suggests that it is involved in the scarring process after filtration surgery. Bevacizumab reduced the proliferation of fibroblasts in vitro and improved surgical outcome.

Influences of atmospheric pressure and temperature on intraocular pressure.

PURPOSE: To determine whether the atmospheric pressure (ATM) change experienced during diving can induce changes in the intraocular pressure (IOP) of eyes in a normal population. METHODS: The IOP of 27 healthy volunteers (ages, 23.8 +/- 4.9 years; range, 18-44) was measured with a Perkins applanation tonometer by two independent investigators who were masked to the previous measurements. Measurements were taken at baseline (normal ATM, 1 Bar and 24 degrees C), at 28 degrees C and 24 degrees C after the ATM was increased to 2 Bar in a hyperbaric chamber, at baseline again, and finally at the normal ATM of 1 Bar but a temperature of 28 degrees C. Multivariate regression analysis was used to evaluate the RESULTS: results. The mean IOP decreased significantly from 11.8 mm Hg in the right eye (RE) and 11.7 mm Hg in the left eye (LE) at 1 Bar to 10.7 mm Hg (RE) and 10.3 mm Hg (LE) at 2 Bar (P = 0.024, RE; P = 0.0006, LE). The IOP decrease remained constant during the ATM increase period (40 minutes) and was independent of the temperature change. The temperature increase alone did not significantly influence the IOP. CONCLUSIONS: An increase of the ATM to 2 Bar (equal to conditions experienced during underwater diving at 10 meters) modestly but significantly decreased the IOP independently of the temperature change. During the period of increased ATM (60 minutes), the IOP decrease remained stable and was independent of blood pressure change or corneal thickness.