Segmentation of Macular Edema Datasets with Small Residual 3D U-Net Architectures

Abstract

This paper investigates the application of deep convolutional neural networks with prohibitively small datasets to the problem of macular edema segmentation. In particular, we investigate several different heavily regularized architectures. We find that, contrary to popular belief, neural architectures within this application setting are able to achieve close to humanlevel performance on unseen test images without requiring large numbers of training examples. Annotating these 3D datasets is difficult, with multiple criteria required. It takes an experienced clinician two days to annotate a single 3D image, whereas our trained model achieves similar performance in less than a second. We found that an approach which uses targeted dataset augmentation, alongside architectural simplification with an emphasis on residual design, has acceptable generalization performance -despite relying on fewer than 15 training examples.

Self-supervised contrastive learning improves machine learning discrimination of full thickness macular holes from epiretinal membranes in retinal OCT scans

“…The last several years have seen an explosion of deep learning models applied to ophthalmic clinical technologies including OCT and fundus imaging. These applications may be divided into the broad areas of classification/diagnosis [8][9][10][11][12][13][14][15][16][17][18], segmentation [19][20][21][22][23][24][25][26], image quality [27], and demographics prediction [28]. The current ophthalmic deep learning models focus primarily on diabetic retinopathy, age-related macular degeneration, retinopathy of prematurity, and glaucoma [29][30][31].…”

“…Deep learning based OCT image analysis for FTMH has also received attention lately, with models for classification [16,17], segmentation [23][24][25][26], and prognosis of success for FTMH corrective surgery [18,32,33]. A review is also available [34].…”

Robust 3D U-Net Segmentation of Macular Holes

“…The U-Net CNN architecture (Ronneberger et al, 2015) is a highly-utilized CNN architecture for biomedical image segmentation. It has had success in segmenting other eye diseases such as macular edema, even when dataset sizes are limited (Frawley et al, 2020). The no free lunch theorem states that there is no single model that performs optimally across all task distributions, whereas Occam's razor indicates we should choose the simplest or smoothest architecture where available.…”