Whole-Volume Automated Segmentation of Retinal Optical Coherence Tomography (OCT) Images Using Deep Learning

Treatment decisions for retinal conditions like wet age-related macular degeneration (AMD) and diabetic macular edema (DME) rely on subjective assessment of retinal fluid as a marker of disease severity.

OCT segmentation offers the potential to objectively quantify disease burden and standardize treatment decisions.

Researchers developed a deep learning model to quantify volumes of clinically relevant pathology in OCT scans in individuals with AMD and DME.

This video illustrates example model segmentations, model success cases, model failures, and examples of model-specialist disagreements when retinal experts graded automated vs manual OCT segmentations for intraretinal fluid (IRF) and subretinal fluid (SRF) (for DME scans) or for IRF, SRF, subreintal hyperreflective material, and pigment epithelial detachment (PED) (for AMD scans).

Click the Related Article link for complete study details, and the Related Article Supplement PDF for full video image findings (eFigures 2 to 12 in the Supplement).

(Note: Set 1: 15 OCT scans from patients with new severe AMD imaged using the 3D OCT-2000 device from Topcon Corporation.

Set 2: 164 OCT scans acquired using the Topcon device or a Spectralis OCT device from Heidelberg Engineering GmbH, resulting in 4 subsets: (1) Topcon-AMD, (2) Heidelberg-AMD, (3) Topcon-DME, and (4) Heidelberg-DME.)