Purchase this article with an account.
Sandeep Bhat, Malavika Bhaskaranand, Chaithanya Ramachandra, Todd P Margolis, Daniel A. Fletcher, Kaushal Solanki; Fully-automated Diabetic Retinopathy Screening Using Cellphone-based Cameras. Invest. Ophthalmol. Vis. Sci. 2015;56(7 ):1428.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Inexpensive retinal imaging and computerized screening algorithms are critical for screening the large, growing diabetic population for preventable vision loss due to diabetic retinopathy (DR). To achieve this, we present an end-to-end point-of-care diabetic retinopathy diagnostic device comprising an Ocular CellScope, a cellphone-based retinal imaging camera and an image analysis software enabling fully-automated screening for DR.
The Ocular CellScope is a retinal imaging device (shown in Figure 1) that easily attaches to an iPhone without needing any phone modification. The image analysis system utilizes novel techniques customized for DR screening. The core steps include: (i) image normalization, (ii) interest region detection, (iii) image gradability assessment, (iv) multi-scale image description, and (v) advanced machine learning techniques for multi-level classification. DR lesions including microaneurysms, hemorrhages, exudates, cotton wool spots, and neo-vascularization are detected and a Refer/No Refer DR screening recommendation is generated.<br /> We have obtained a dataset of 30 images (with 10 images having known signs of DR) obtained from the first prototype of the Ocular CellScope, which has since been improved. We evaluate our automated DR screening software on this dataset by conducting a cross-dataset test where the training is performed on a completely independent dataset.
On the cross-dataset test, our DR screening software achieves AUROC of 0.94 (100% sensitivity at 80% specificity) for identifying referable DR (moderate NPDR or higher on the ICDR scale). Figure 2 shows examples of DR lesion detection using our software on an Ocular Cellscope retinal image
The excellent performance of our DR screening software on retinal images captured using the Ocular Cellscope, demonstrates the robustness of our image analysis algorithms to varying lighting conditions and image qualities. It also proves the feasibility of retinal imaging using cellphone retinal cameras for DR screening.
This PDF is available to Subscribers Only