Deep Learning for Diagnosing and Segmenting Choroidal Neovascularization in OCT Angiography in a Large Real-World...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Choroidal neovascularization (CNV) is a leading cause of vision loss.
• Accurate diagnosis and segmentation of CNV are crucial for effective treatment.
• Current diagnostic methods can be labor-intensive and subjective.

Purpose of the Study:

• To develop and validate a deep learning algorithm for diagnosing and segmenting CNV.
• To assess the model's performance on a large, real-world, multicenter OCT angiography dataset.
• To evaluate the potential for automated CNV screening and monitoring in clinical practice.

Main Methods:

• A hybrid multitask convolutional neural network was trained on 105,66 OCT angiography scans from 3135 eyes.
• The dataset included scans with and without CNV from various diseases and healthy controls.
• The model was designed to output both CNV diagnosis and membrane segmentation.

Main Results:

• The model achieved high accuracy in CNV diagnosis (AUC = 0.97) with 95% sensitivity and 95% specificity.
• CNV lesions were segmented with a high F1 score of 0.78 ± 0.19.
• Performance remained consistent across different scan qualities and various disease types, enabling multiyear monitoring of subclinical lesions.

Conclusions:

• The proposed deep learning method accurately diagnoses and segments CNV in a large, real-world clinical dataset.
• The algorithm demonstrates potential for automated CNV screening and quantification.
• This technology can significantly aid in improving CNV diagnosis and treatment evaluation.