Deep Learning-Based Retinal Layer Segmentation in Optical Coherence Tomography Scans of Patients with Inherited Retinal Diseases.

BACKGROUND: In optical coherence tomography (OCT) scans of patients with inherited retinal diseases (IRDs), the measurement of the thickness of the outer nuclear layer (ONL) has been well established as a surrogate marker for photoreceptor preservation. Current automatic segmentation tools fail in OCT segmentation in IRDs, and manual segmentation is time-consuming. METHODS AND MATERIAL: Patients with IRD and an available OCT scan were screened for the present study. Additionally, OCT scans of patients without retinal disease were included to provide training data for artificial intelligence (AI). We trained a U-net-based model on healthy patients and applied a domain adaption technique to the IRD patients' scans. RESULTS: We established an AI-based image segmentation algorithm that reliably segments the ONL in OCT scans of IRD patients. In a test dataset, the dice score of the algorithm was 98.7%. Furthermore, we generated thickness maps of the full retinal thickness and the ONL layer for each patient. CONCLUSION: Accurate segmentation of anatomical layers on OCT scans plays a crucial role for predictive models linking retinal structure to visual function. Our algorithm for segmentation of OCT images could provide the basis for further studies on IRDs.

Anti-angiogenic properties of rapamycin on human retinal pericytes in an in vitro model of neovascular AMD via inhibition of the mTOR pathway.

PURPOSE: Choroidal neovascularizations (CNV) are partially stabilized through a coverage of pericytes leading to a partial anti-VEGF resistence. Drugs licensed for neovascular AMD (nAMD) do not take this mechanical and growth factor-driven CNV stability into account. The purpose of this work was to see if inhibiting the mammalian target of rapamycin (mTOR) may successfully block angiogenic cellular pathways in primary human retinal pericytes in an in vitro model of nAMD. METHODS: The mTOR inhibitor rapamycin was used to treat human retinal pericytes (HRP) at doses ranging from 0.005 to 15 g/ml. A modified metabolism-based XTT-Assay was used to assess toxicity and anti-proliferative effects. A scratch wound experiment showed the effects on migration. On Cultrex basement membrane gels, the influence of rapamycin on the development of endothelial cell capillary-like structures by human umbilical vein vascular endothelial cells (HUVEC) in the absence and presence of pericytes was investigated. RESULTS: Rapamycin showed no signs of toxicity within its range of solubility. The drug showed dose dependent anti-proliferative activity and inhibited migration into the scratch wound. Endothelial cell tube formation in a HUVEC monoculture was effectively inhibited at 45%. A co-culture of HUVEC with pericytes on Cultrex induced endothelial tube stabilization but was disrupted by the addition of rapamycin leading to degradation of 94% of the tubes. CONCLUSIONS: Rapamycin allows for an efficient modulation of aspects of angiogenesis in pericytes via mTOR-modulation in vitro. Further studies are needed to elucidate whether rapamycin may have an impact on CNV in nAMD in vivo.

Combined VEGF/PDGF inhibition using axitinib induces αSMA expression and a pro-fibrotic phenotype in human pericytes.

PURPOSE: Large trials on anti-VEGF/PDGF (vascular endothelial/platelet-derived growth factor) combination therapy have been established to improve management of neovascular activity in age-related macular degeneration. Targeting pericytes, PDGF is thought to induce vessel regression and reduce fibrovascular scarring. The fate of pericytes exposed to anti-VEGF/PDGF combination therapy is not clear. Therefore, this study was designed to study the influence of anti-VEGF/PDGF on pericyte phenotype and cellular behavior. METHODS: Human pericytes from placenta (hPC-PL) were treated with axitinib, a tyrosine kinase inhibitor targeting VEGFR1-3 and PDGFR. Toxic effects were excluded using live/dead staining. Phenotypic changes were evaluated using phalloidin staining for actin cytoskeleton and the expression of stress fibers. MRNA and protein expression levels of α-smooth muscle actin (αSMA) as a marker of proto-myofibroblastic transition were evaluated with real-time PCR and Western blotting. Influences of fibrotic cellular mechanisms were evaluated with a scratch wound migration and a collagen gel contraction assay. RESULTS: Treatment with 0.5, 1, and 2.5 µg/ml axitinib strongly induced a proto-myofibroblast-like actin cytoskeleton with a marked increase in stress fibers. Quantitative real-time PCR and Western blotting revealed these changes to be linked to dose-dependent increases in αSMA mRNA and protein expression. However, fibrotic cellular mechanisms were significantly reduced in the presence of axitinib (scratch wound closure: up to - 78.4%, collagen gel contraction: up to - 37.4%). CONCLUSIONS: Combined anti-VEGF/PDGF inhibition seems to induce a proto-myofibroblast-like phenotype in human pericytes in vitro, but reduce profibrotic cellular mechanisms due to prolonged anti-PDGF inhibition.

Optical coherence tomography-based misdiagnosis and morphological distinction in pachychoroid neovasculopathy vs. polypoidal choroidal vasculopathy.

PURPOSE: To evaluate the rate of misdiagnosis of aneurysmatic pachychoroid type 1 choroidal neovascularization/polypoidal choroidal vasculopathy (PAT1/PCV) among cases diagnosed as non-aneurysmatic pachychoroid neovasculopathy (PNV) and to define optical coherence tomography (OCT) features facilitating their distinction. METHODS: The database of the Department of Ophthalmology, Ludwig-Maximilians University Munich, was screened for patients diagnosed with PNV. Multimodal imaging was screened for the presence of choroidal neovascularization (CNV) and aneurysms/polyps. Imaging features facilitating the diagnosis of PAT1/PCV were analysed. RESULTS: In total, 49 eyes of 44 patients with a clinical PNV diagnosis were included, of which 42 (85.7%) had PNV and 7 (14.3%) represented misdiagnosed PAT1/PCV. SFCT was comparable (PNV: 377 ± 92 vs. PAT1/PCV: 400 ± 83 µm; p = 0.39). Whereas no difference was detected in total pigment epithelium detachment (PED) diameter (p = 0.46), maximum PED height was significantly higher in the PAT1/PCV group (199 ± 31 vs. 82 ± 46, p < 0.00001). In a receiver operating characteristic (ROC) analysis, the optimum cutoff for defining "peaking PED" was 158 µm with an area under the curve of 0.969, a sensitivity of 1.0 (95% CI: 0.59-1.0), and a specificity of 0.95 (95% CI: 0.84-0.99). Sub-retinal hyperreflective material (SHRM; p = 0.04), sub-retinal ring-like structures (SRRLS; p < 0.00001), and sub-RPE fluid (p = 0.04) were significantly more frequent in eyes with PAT1/PCV. CONCLUSION: A relevant percentage of eyes diagnosed with PNV might instead suffer from PAT1/PCV. The detection of a maximum PED height ("peaking PED") exceeding approximately 150 µm, SHRM, SRRLS, and sub-RPE fluid might greatly aid in the production of a more accurate diagnosis.

Reconstructing disease dynamics for mechanistic insights and clinical benefit.

Diseases change over time, both phenotypically and in their underlying molecular processes. Though understanding disease progression dynamics is critical for diagnostics and treatment, capturing these dynamics is difficult due to their complexity and the high heterogeneity in disease development between individuals. We present TimeAx, an algorithm which builds a comparative framework for capturing disease dynamics using high-dimensional, short time-series data. We demonstrate the utility of TimeAx by studying disease progression dynamics for multiple diseases and data types. Notably, for urothelial bladder cancer tumorigenesis, we identify a stromal pro-invasion point on the disease progression axis, characterized by massive immune cell infiltration to the tumor microenvironment and increased mortality. Moreover, the continuous TimeAx model differentiates between early and late tumors within the same tumor subtype, uncovering molecular transitions and potential targetable pathways. Overall, we present a powerful approach for studying disease progression dynamics-providing improved molecular interpretability and clinical benefits for patient stratification and outcome prediction.

Progression of Pachychoroid Neovasculopathy into Aneurysmal Type 1 Choroidal Neovascularization or Polypoidal Choroidal Vasculopathy.

PURPOSE: To describe the progression of pachychoroid neovasculopathy (PNV) into pachychoroid aneurysmal type 1 choroidal neovascularization (PAT1)/polypoidal choroidal vasculopathy (PCV). DESIGN: Retrospective longitudinal cohort study. SUBJECTS: Patients diagnosed with PNV with a follow-up of ≥2 years. METHODS: Multimodal imaging, including OCT and fluorescein and indocyanine green angiography, was reviewed for the presence of choroidal neovascularization (CNV), aneurysms within/at the margins of the CNV, and subfoveal choroidal thickness (SFCT). MAIN OUTCOME MEASURES: Rate of PNV to PAT1/PCV conversion and risk factors thereof. RESULTS: In total, 37 eyes of 32 patients with PNV with a mean follow-up of 3.3 ± 1.1 years (range, 2.0-5.2) were included in the study. At PNV diagnosis, the mean age was 59.7 ± 8.7 years (range, 38.5-78.0 years) and mean SFCT was 357 ± 92 µm (185-589). During the follow-up, 5 (13.5%) eyes developed aneurysms after a mean 3.4 ± 0.8 years (2.3-4.2), defining PAT1/PCV. The risk of PAT1/PCV conversion was 7.4% at year 3, 13.6% at year 4, and 30.7% at year 5. A mean of 5.2 ± 4.0 to 7.9 ± 3.6 intravitreal anti-VEGF injections were given per year, resulting in a significant reduction of SFCT to 317 ± 104 µm (122-589) (P = 0.0007). The age at diagnosis of PNV was significantly lower in eyes that later went on to develop PAT1/PCV (54.0 ± 5.6 [45.9-60.5] vs. 61.2 ± 8.4 [38.5-78.0] years; P = 0.025). At the end of the follow-up, SFCT had on average decreased by -14.0% ± 17.6% (-55.9% to 23.1%) in the PNV group, whereas it had increased by mean 6.9% ± 4.4% (0.00%-10.8%) in the PAT1/PCV conversion group (P = 0.0025). CONCLUSIONS: PNV can develop aneurysms within its type 1 CNV, defining the conversion to PAT1/PCV. In this study, the conversion to PAT1/PCV was seen in 13.5% of eyes, resulting in Kaplan-Meier estimates of risk for conversion of 7.4% at year 3, 13.6% at year 4, and 30.7% at year 5. Younger age at diagnosis of PNV and sustained choroidal thickening despite anti-VEGF therapy might be risk factors for PNV to progress into PAT1/PCV.

3D Heads-Up Display vs. Standard Operating Microscope Vitrectomy for Rhegmatogenous Retinal Detachment.

Purpose: To assess the efficacy and outcomes of 23-gauge vitreoretinal surgery for rhegmatogenous retinal detachment using a three-dimensional heads-up display (3D HUD) surgical platform as compared to a standard operating microscope (SOM) setting. Design: Retrospective cohort study. Participants: One hundred and forty consecutive eyes of 140 patients with primary retinal detachment. Methods: All eyes underwent 23-gauge pars plana vitrectomy for primary retinal detachment using either a 3D HUD (NGENUITY; Alcon Inc., Fort Worth, Texas, USA; n = 70 eyes) or a SOM setting (n = 70 eyes); in cases of significant cataract, additional phacoemulsification with intraocular lens (IOL) implantation was performed. Minimum follow-up was 2 months. Main Outcome Measures: Primary retinal reattachment rate, rate of proliferative vitreoretinopathy (PVR), best-corrected visual acuity (BCVA), and duration of surgery. Results: There were 70 eyes each in the 3D HUD and the SOM group. Both groups did not differ concerning age (p = 0.70), extent of retinal detachment (p = 0.07), number of retinal tears (p = 0.40), macular involvement (p = 0.99), and preoperative BCVA (p = 0.99). Postoperatively, 3D HUD and SOM were comparable concerning the primary retinal reattachment rate (88.6 vs. 94.3%; p = 0.37), the development of postoperative PVR (12.9% vs. 7.1%; p = 0.40) and final BCVA (0.26 ± 0.40 vs. 0.21 ± 0.38 logMAR; p = 0.99). Duration of surgery was significantly longer in the 3D HUD group (66.2 ± 16.5 vs. 61.2 ± 17.1 min; p = 0.04), an effect which however vanished after a "learning curve" of the first 35 eyes (p = 0.49). Conclusions: On par results to a conventional operating microscope can be achieved with a 3D HUD setting when performing 23-gauge vitreoretinal surgery for rhegmatogenous retinal detachment, including the primary retinal reattachment rate, the incidence of postoperative PVR and final BCVA. However, duration of surgery might initially be slightly longer with 3D HUD, suggesting the effect of a learning curve.