Simultaneous detection and quantification of retinal fluid with deep learning
Proceedings of the MICCAI Retinal OCT Fluid Challenge (RETOUCH)
Abstract
We propose a new deep learning approach for automatic detection and segmentation of fluid within retinal OCT images. The proposed framework utilizes both ResNet and Encoder-Decoder neural network architectures. When training the network, we apply a novel data augmentation method called myopic warping together with standard rotation-based augmentation to increase the training set size to 45 times the original amount. Finally, the network output is post-processed with an energy minimization algorithm (graph cut) along with a few other knowledge guided morphological operations to finalize the segmentation process. Based on OCT imaging data and its ground truth from the RETOUCH challenge, the proposed system achieves dice indices of 0.522, 0.682, and 0.612, and average absolute volume differences of 0.285, 0.115, and 0.156 mm$^3$ for intaretinal fluid, subretinal fluid, and pigment epithelial detachment respectively.
Abstract sourceThis study describes a deep-learning approach to segmenting and quantifying retinal fluid in optical coherence tomography images. The paper is listed in the 2017 MICCAI Retinal OCT Fluid Challenge proceedings; a separate program confirms a presentation of the work at the Retina Society’s 51st Annual Scientific Meeting in 2018.
Summary
The paper presents a convolutional neural network for identifying three retinal fluid classes in optical coherence tomography (OCT) images: intraretinal fluid, subretinal fluid, and pigment epithelial detachment. The approach processes individual image slices, then uses graph-cut optimization and additional post-processing to produce segmentations and estimate fluid volumes. The authors evaluated the method with two-fold cross-validation on OCT data from the RETOUCH challenge.
The authors report Dice indices of 0.522, 0.682, and 0.612 for intraretinal fluid, subretinal fluid, and pigment epithelial detachment, respectively. They also note limitations in the challenge’s reference annotations, particularly for intraretinal fluid, and report differences in performance across imaging devices. These results describe an evaluation on challenge data; they do not establish clinical readiness.
Cite this work
Dustin Morley, Hassan Foroosh, Saad Shaikh, Ulas Bagci (2017). Simultaneous detection and quantification of retinal fluid with deep learning. Proceedings of the MICCAI Retinal OCT Fluid Challenge (RETOUCH) https://arxiv.org/abs/1708.05464
BibTeX