Accuracy of a Machine-Learning Algorithm for Detecting and Classifying Choroidal Neovascularization on...

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Age-related macular degeneration (AMD) is a leading cause of vision loss.
• Choroidal neovascularization (CNV) is a key complication of AMD.
• Accurate detection and classification of CNV are crucial for effective treatment.

Purpose of the Study:

• To assess the performance of a machine-learning (ML) algorithm for detecting and classifying CNV.
• To evaluate the algorithm's accuracy on spectral-domain optical coherence tomography (SD-OCT) images.
• To validate the ML model using data from clinical trials.

Main Methods:

• Developed and trained an ML pipeline using deep learning for SD-OCT B-scan segmentation and CNV classification.
• Utilized data from the HARBOR trial (NCT00891735) for model development and cross-validation.
• Externally validated the ML model using SD-OCT scans from the AVENUE trial (NCT02484690).

Main Results:

• The ML algorithm achieved high accuracy (AUROC = 0.99) in discriminating CNV presence versus absence.
• It accurately classified occult versus predominantly classic CNV types (AUROC = 0.91) on HARBOR images.
• External validation on AVENUE data showed good performance (AUROC = 0.88) for occult and predominantly classic CNV types.

Conclusions:

• The developed ML model demonstrates high accuracy in detecting CNV presence and subtypes on SD-OCT images.
• This AI tool shows promise for aiding in the diagnosis and management of neovascular AMD.
• The findings support the utility of ML in analyzing retinal imaging for AMD-related conditions.