Deep Learning-based Diagnosis of Glaucoma Using Wide-field Optical Coherence Tomography Images

Purpose: (1) To evaluate the performance of deep learning (DL) classifier in detecting glaucoma, based on wide-field swept-source optical coherence tomography (SS-OCT) images. (2) To assess the performance of DL-based fusion methods in diagnosing glaucoma using a variety of wide-field SS-OCT images and compare their diagnostic abilities with that of conventional parameter-based methods. Methods: Overall, 675 eyes, including 258 healthy eyes and 417 eyes with glaucoma were enrolled in this retrospective observational study. Each single-page wide-field report (12×9 mm) of wide-field SS-OCT imaging provides different types of images that reflect the state of the eyes. A DL-based automated diagnosis system was proposed to detect glaucoma and identify its stage based on such images. We applied the convolutional neural network to each type of image to detect glaucoma. In addition, 2 fusion strategies, fusion by convolution network (FCN) and fusion by fully connected network (FFC) were developed; they differ in terms of the level of fusion of features derived from convolutional neural networks. The diagnostic models were trained using 382 and 293 images in the training and test data sets, respectively. The diagnostic ability of this method was compared with conventional parameters of the thickness of the retinal nerve fiber layer and ganglion cell complex. Results: FCN achieved an area under the receiver operating characteristic curve (AUC) of 0.987 (95% confide...
Source: Journal of Glaucoma - Category: Opthalmology Tags: Novel Insights: Original Studies Source Type: research