Glaucoma Detection From Raw Circumpapillary OCT Images Using Fully Convolutional Neural Networks

Abstract

Nowadays, glaucoma is the leading cause of blindness worldwide. We propose in this paper two different deep-learningbased approaches to address glaucoma detection just from raw circumpapillary OCT images. The first one is based on the development of convolutional neural networks (CNNs) trained from scratch. The second one lies in fine-tuning some of the most common state-of-the-art CNNs architectures. The experiments were performed on a private database composed of 93 glaucomatous and 156 normal B-scans around the optic nerve head of the retina, which were diagnosed by expert ophthalmologists. The validation results evidence that finetuned CNNs outperform the networks trained from scratch when small databases are addressed. Additionally, the VGG family of networks reports the most promising results, with an area under the ROC curve of 0.96 and an accuracy of 0.92, during the prediction of the independent test set.

Circumpapillary OCT-focused hybrid learning for glaucoma grading using tailored prototypical neural networks

“…A recent review of the literature found that most of the deep learning-focused studies using B-scans are conducted in a combination of fundus images to detect glaucoma via RNFL probability maps [42][43][44]. To fill this gap in the literature, we previously proposed different hand-driven and data-driven learning strategies for glaucoma diagnosis in [20,41,45], whose outcomes support the basis of this paper.…”

Strategies to Improve Convolutional Neural Network Generalizability and Reference Standards for Glaucoma Detection From OCT Scans

“…A number of studies have developed approaches based upon convolutional neural networks (CNNs) to detect glaucoma from optical coherence tomography (OCT) images. 6 – 10 These studies, in general, show excellent performance in terms of reasonably high sensitivity and specificity. However, to demonstrate clinical usefulness, it is essential to test a deep learning system's ability to be effective with a new dataset from a different clinic.…”

A Self-Training Framework for Glaucoma Grading In OCT B-Scans

“…In addition to the presented learning strategy, we propose several architectural changes that improve over existing ones both quantitatively and from an interpretability perspective. In our recent works focused on glaucoma detection from raw OCT samples [7]- [9], we conducted an in-depth experimental analysis of different deep-learning architectures, both trained from scratch and fine-tuning the most popular models in the literature. Specifically, the VGG family of networks reported the best results in [7].…”

“…In our recent works focused on glaucoma detection from raw OCT samples [7]- [9], we conducted an in-depth experimental analysis of different deep-learning architectures, both trained from scratch and fine-tuning the most popular models in the literature. Specifically, the VGG family of networks reported the best results in [7]. So, in [9], we employed them as a strong backbone to develop a new feature extractor able to learn useful representations from the slides of a spectraldomain (SD)-OCT volume.…”

“…Nevertheless, despite the rise of these models in many computer vision and medical problems, their application on OCT images for glaucoma assessment still presents several limitations. First, most of the literature focuses on discerning between healthy and glaucoma classes via OCT B-scans [7], [8], SD-OCT volumes [9]- [11] or probability RNFL maps by combining fundus images and OCT samples [12], [13]. Indeed, to the best of our knowledge, only [2] proposes a glaucomabased scenario beyond the healthy-glaucoma classification by including the suspect label.…”