Deep Neural Networks for Automated Outer Plexiform Layer Subsidence Detection on Retinal OCT of Patients With...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Outer plexiform layer (OPL) subsidence on optical coherence tomography (OCT) is a key indicator of early outer retinal atrophy.
• This subsidence is a risk factor for geographic atrophy progression in intermediate age-related macular degeneration (AMD).
• Automated detection of OPL subsidence can aid in early diagnosis and monitoring.

Purpose of the Study:

• To develop and validate a deep neural network (DNN) system for automated detection and localization of OPL subsidence in retinal OCT images.
• To assess the performance of the DNN system in identifying early signs of AMD progression.

Main Methods:

• A DNN system comprising a detection module (DM) and a classification module (CM) was developed.
• The system predicted OPL subsidence locations on retinal OCT scans, combining and scoring overlapping detections.
• The system was trained and validated on a development dataset (140 AMD patients) and an independent external dataset (26 AMD patients).

Main Results:

• The DNN system achieved over 85% detection of OPL subsidences with less than one false-positive per scan.
• The average area under the curve for volume-level detection was 0.94 ± 0.03.
• Comparable or superior performance was observed on the independent external dataset.

Conclusions:

• Deep neural networks (DNNs) offer an efficient method for automated OPL subsidence detection in OCT images.
• The proposed DNN system demonstrates high sensitivity and minimal false positives for OPL subsidence detection.
• DNNs facilitate objective identification of early AMD signs, supporting screening and intervention assessment.