A column-based deep learning method for the detection and quantification of atrophy associated with AMD in OCT scans

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Objective quantification of retinal atrophy in age-related macular degeneration (AMD) is crucial for clinical management and research.
• Spectral Domain Optical Coherence Tomography (OCT) is a key imaging modality for macular evaluation.

Purpose of the Study:

• To develop and validate a novel automated method for identifying and quantifying retinal atrophy in OCT scans.
• To visualize and measure atrophy lesions in corresponding infrared (IR) images.

Main Methods:

• A custom column-based convolutional neural network (CNN) classifies light scattering patterns in OCT A-scans.
• 3D column patches from volumetric OCT scans are used for classification.
• Identified atrophy columns are projected onto IR images for lesion segmentation and measurement.

Main Results:

• The automated method achieved a mean F1 score of 0.78 and an AUC of 0.937 on 106 clinical OCT scans.
• Performance metrics are comparable to human observer variability.
• Successfully identified and quantified 2,952 atrophy segments and 1,046 atrophy lesions.

Conclusions:

• Automated detection and quantification of AMD-related atrophy using a column-based CNN in OCT scans is feasible at an expert level.
• This tool can serve as valuable clinical decision support for diagnosing and managing retinal degenerations.
• Potential applications include patient follow-up and treatment efficacy evaluation in clinical research.