Automated Segmentation of Cystic Macular Edema in OCT B-Scan Images

Mr Luke Greenwood, Mr Maged Habib, Professor Andrew Hunter, Dr Bashir Al-Diri

Abstract:

The analysis of retinal Spectral Domain Optical Coherence Tomography (SD-
OCT) images by trained medical professionals can be used to provide useful in-
sights into many various diseases. It is the most popular method of retinal imaging
due to it’s non invasive nature and the useful information it provides for making an
accurate diagnosis. In this paper, we present a deep learning approach for the au-
tomating the segmentation of cystic macular edema (fluid) in retinal OCT B-Scan
images. Our network makes use of atrous convolutions, skip connections, weight
decay and significant image augmentation to ensure the most accurate segmenta-
tion result possible without the need for any features to be manually constructed.
The network is evaluated against a publicly available dataset and achieved a max-
imal Dice coefficient of 95.2%, thus making it the current best performer on that
dataset.

Keywords: Deep Learning, Machine Learning, Semantic Segmentation, Convolutional Neural Networks, Medical Imaging, Computer Vision, TensorFlow, Python, SciPy, NumPy

Author Affiliation: University of Lincoln, Sunderland Eye Infirmary