Deep Learning-Based OCTA Quantification and Modeling for Predicting Long-Term Anti-VEGF Efficacy in mCNV

Deep Learning-Based OCTA Quantification and Modeling for Predicting Long-Term Anti-VEGF Efficacy in mCNV

Myopic choroidal neovascularization (mCNV) is one of the sight-threatening complications secondary to pathological myopia. Intravitreal injection of anti-vascular endothelial growth factor (VEGF) is its first-line therapy. However, mCNV is prone to recurrence and long-term visual decline, and there is currently no definitive method for predicting long-term prognosis. This project aims to conduct a long-term follow-up and multi-dimensional quantitative analysis of mCNV using optical coherence tomography angiography (OCTA) images. It seeks to evaluate long-term prognostic indicators, such as recurrence and long-term visual acuity, following anti-VEGF therapy for mCNV, and to construct a deep learning (DL) prediction model for anti-VEGF efficacy based on multi-modal clinical data, ultimately enabling treatment personalization.

First, deep learning technology will be utilized to segment and quantitatively analyze mCNV in OCTA images, obtaining multi-dimensional quantitative parameters. These include OCTA-based The mCNV area and vessel junction(VJ). Patients will be followed up regularly for two years post-treatment, monitoring the number of injections, mCNV recurrence, OCT and OCTA quantitative parameters, fundus chorioretinal atrophy lesions, and visual acuity status. A multi-modal DL prediction model will be constructed, primarily based on the multi-dimensional quantitative characteristics of mCNV from OCTA images. This model will aim to identify sensitive indicators for predicting best-corrected visual acuity and recurrence after anti-VEGF therapy for mCNV and to clarify the relationship between therapeutic efficacy, baseline lesion status, and treatment regimen selection.

This research will open new avenues for clinically assessing the long-term efficacy of anti-VEGF therapy for mCNV, significantly improve the efficiency and accuracy of efficacy evaluation, and provide a critical reference for personalizing anti-VEGF treatment plans, holding substantial clinical significance.

Inclusion Criteria:

• Myopic spherical equivalent greater than -6.0 diopters and/or axial length > 26.5 mm.
• Patients presenting with myopic fundus changes, as defined by the International Photographic Classification System for Myopic Maculopathy.
• Completion of a comprehensive ophthalmic examination at initial diagnosis, including Best Corrected Visual Acuity , slit-lamp examination, dilated fundus examination, intraocular pressure measurement, refractometry, axial length measurement, color fundus photography, Optical Coherence Tomography, OCT Angiography, and Fundus Fluorescein Angiography.
• Patients with active mCNV at initial diagnosis who are treatment-naïve.
• Final inclusion of all images required confirmation by two retinal specialists. Images with discrepant assessments were adjudicated by a senior researcher to reach a final decision.

Exclusion Criteria:

• Patients with choroidal neovascularization (CNV) due to causes other than myopia, such as age-related macular degeneration, adult-onset foveomacular vitelliform dystrophy, multifocal choroiditis, punctate inner choroidopathy, or retinoschisis.
• Poor-quality OCTA images, for example, those containing projection artifacts from overlying vessels at the mCNV lesion site, or excessively dark images dominated by thick outer choroidal vessels.
• Patients with a history of intraocular surgery, such as prior pars plana vitrectomy.
• Subjects who developed any of the above conditions during the follow-up period or who requested to withdraw from the study were removed from the experimental cohort.