The SUSTech-SYSU dataset for automatically segmenting and classifying corneal ulcers

Abstract

Corneal ulcer is a common ophthalmic symptom. Segmentation algorithms are needed to identify and quantify corneal ulcers from ocular staining images. Developments of such algorithms have been obstructed by a lack of high quality datasets (the ocular staining images and the corresponding goldstandard ulcer segmentation labels), especially for supervised learning based segmentation algorithms. In such context, we prepare a dataset containing 712 ocular staining images and the associated segmentation labels of flaky corneal ulcers. In addition to segmentation labels for flaky corneal ulcers, we also provide each image with three-fold class labels: firstly, each image has a label in terms of its general ulcer pattern; secondly, each image has a label in terms of its specific ulcer pattern; thirdly, each image has a label indicating its ulcer severity degree. This dataset not only provides an excellent opportunity for investigating the accuracy and reliability of different segmentation and classification algorithms for corneal ulcers, but also advances the development of new supervised learning based algorithms especially those in the deep learning framework.

High-Performance Detection of Corneal Ulceration Using Image Classification with Convolutional Neural Networks

“…Schwartz et al [8] revealed that regarding the present levels of blindness in a Central African Republic's population that they investigated, 95.5 percent of all existing cases are preventable and treatable. While it is clear that prevention contributes to a cost-effective treatment of corneal blindness [1], manual analysis of the ocular surface is a widely used method for assessing patterns of CU [9].…”

“…Although several approaches have aimed to automate the identification of CUs using machine learning techniques, they are either based on a small dataset, do not distinguish between different types of the disease, or do not manifest a sufficient level of accuracy to be implemented in the healthcare domain [9,11,[21][22][23][24][25][26][27][28][29].…”

The SUSTech-SYSU dataset for automated exudate detection and diabetic retinopathy grading

“…DRIVE and STARE are two classic fundus datasets for retinal vessel segmentation, and STARE also provides diagnostic information for a larger set of fundus images [17][18][19][20] . In one of our previous works, we also developed a dataset containing 712 ocular staining images for corneal ulcer segmentation and classification 21,22 . However, to the best of our knowledge, existing large-scale and well-annotated fundus image datasets with lesion annotations are relatively limited.…”

Development and validation of deep learning algorithms for automated eye laterality detection with anterior segment photography

“…Anterior segment imaging has largely improved documentation and understanding of anterior segment structures and pathological processes. With the rapid advancement of deep learning (DL) in healthcare, it is now possible to perform automated detection of several anterior segment eye diseases, such as pterygium 1 , corneal ulcer 2 , and cataracts using anterior segment photographs 3 . There is an increasing need for evaluation of corneal, glaucoma, and lens disease on DL algorithms with anterior segment imagery in telemedicine approach or clinical practice 4 – 6 .…”

“…A minibatch gradient descent of size 32 was employed for training, with an Adam optimizer learning rate of 0.0001 for better convergence. All the DL models and strategies were implemented in the TensorFlow framework (Google; v. 2 Validation of DL model. To better understand and demonstrate the DL model, we applied the gradientweighted Class Activation Mapping (CAM) to highlight the area in which the DL model may focus on eye laterality detection.…”