Abstract
Purpose :
Diabetic Retinopathy (DR) is one of the leading causes of blindness in the United States and other developed countries. Early detection is the key to prevention, which can be achieved effectively with high accuracy and fully automated screening tool suitable in primary care settings. We have built an artificial intelligence-based diabetic retinopathy (DR) screening tool which utilizes a cloud-based platform for large scale screening through primary care settings for early diagnosis and prevention of DR. Here, we aim to validate this tool.
Methods :
The cloud-based screening model was developed using deep learning techniques and tested using 88702 images from the Kaggle dataset and externally validated using 1748 high-resolution fundus images from the Messidor-2 dataset. For validation of primary care settings, we have used 264 images taken in two primary care clinics in Queens, New York City. The images were uploaded in the cloud-based iPredict software for testing the automated DR screening platform. An automated referable and non-referable DR is compared against the expert ophthalmologists' evaluation. The results are demonstrated with constructing the receiver operating characteristic curve (ROC), area under the curve (AUC), sensitivity and specificity of screening model with reference to professional graders.
Results :
The screening system achieved a high sensitivity of 99.21% and a specificity of 97.59% on the Kaggle dataset with an AUC of 0.9992. The system was also externally validated in Messidor-2, where it achieved a sensitivity of 97.63% and a specificity of 99.49% with an AUC of 0.9985. In the primary care settings, the system achieved a sensitivity of 92.3% overall (12 out of 13 referable images are correctly identified) and an overall specificity of 94.8% (233 out of 251 non-referable images).
Conclusions :
The proposed DR screening tool achieves state-of-the-art performance among the publicly available datasets - Kaggle and Messidor-2 – to the best of our knowledge. The excellent performance on various clinically relevant measures demonstrates that the tool is suitable for screening and early diagnosis of DR in the primary care settings effectively, and helps its prevention.
This is a 2020 ARVO Annual Meeting abstract.