Abstract
Purpose :
Smartphone based fundus cameras are a relatively new, useful clincial tool to image the fundus. This technology combined with artificial intelligence provides the promise of an incredibly powerful way to screen for diabetic retinopathy. The purpose of this study is to determine the current state of studies examining artificial intelligence in smartphones for diabetic retinopathy screening, and to determine whether this technology is a feasible alternative to more traditional methods of diabetic retinopathy screening.
Methods :
The aim of this systematic review was to determine studies that used smartphone-based fundus cameras and deep learning algorithms to screen for diabetic retinopathy. A search strategy using keywords were developed to search Embase. Embase’s “Similar Records” function was used on relevant articles from the initial search. Each study underwent a title review, abstract review, and full-text review. Study methods, population data, and outcomes were gathered for each study.
Results :
Out of the initial 433 articles screened, 10 were ultimately included in this review. Reasons for exclusion included the type of study (reviews, meta-analysis), retrospective studies, and non-smartphone handheld cameras. An average of 300 (range 89-900) patients were included in each study. The most common devices included the Remidio Fundus on Phone (n=5) and Ocular Cell Scope (n=2). The majority of studies were conducted in India (n=6). The method of artificial intelligence implemented included Medios AI (n=4), the AUTO-DR algorithm (n=1), and other unspecified machine learning programs (n=5). The average sensitivity and specificity were 92.34 (range 73.3 – 100) and 85.28 (range 45-97), respectively across all studies.
Conclusions :
Artificial intelligence applied to diabetic retinopathy in the form of smartphone-based fundus images is a promising technology that will assist ophthalmologists screen for diabetic retinopathy, especially in rural or low-income areas. The high sensitivities and specificities presented in the current literature reaffirm the possibility of these technologies being used in the clinical setting. However, more studies with greater sample sizes, different artificial intelligence methods, and more diverse populations need to be conducted before this technology can be incorporated into clinical settings for diabetic retinopathy screening.
This is a 2020 Imaging in the Eye Conference abstract.