Artificial intelligence in diabetic retinopathy screening: clinical assessment using handheld fundus camera in a real-life setting

Abstract

Aim Diabetic retinopathy (DR) represents the main cause of vision loss among working age people. A prompt screening of this condition may prevent its worst complications. This study aims to validate the in-built artificial intelligence (AI) algorithm Selena+ of a handheld fundus camera (Optomed Aurora, Optomed, Oulu, Finland) in a first line screening of a real-world clinical setting. Methods It was an observational cross-sectional study including 256 eyes of 256 consecutive patients. The sample included both diabetic and non-diabetic patients. Each patient received a 50°, macula centered, non-mydriatic fundus photography and, after pupil dilation, a complete fundus examination by an experienced retina specialist. All images were after analyzed by a skilled operator and by the AI algorithm. The results of the three procedures were then compared. Results The agreement between the operator-based fundus analysis in bio-microscopy and the fundus photographs was of 100%. Among the DR patients the AI algorithm revealed signs of DR in 121 out of 125 subjects (96.8%) and no signs of DR 122 of the 126 non-diabetic patients (96.8%). The sensitivity of the AI algorithm was 96.8% and the specificity 96.8%. The overall concordance coefficient k (95% CI) between AI-based assessment and fundus biomicroscopy was 0.935 (0.891–0.979). Conclusions The Aurora fundus camera is effective in a first line screening of DR. Its in-built AI software can be considered a reliable tool to automatically identify the presence of signs of DR and therefore employed as a promising resource in large screening campaigns.

Automated Identification of Different Severity Levels of Diabetic Retinopathy Using a Handheld Fundus Camera and Single-Image Protocol

“…Other authors have recently assessed the performance of AI systems for DR evaluation using portable retinal cameras and yielding variable outcomes, including the detection of any DR, referable DR and sight-threatening DR; the main results of those studies are displayed in Table 3 . 5 , 23 , 24 , 25 The performance of our systems was comparable and often superior to the ones evaluated in those studies. Besides individual cameras’ and AI systems’ characteristics, such heterogeneity of performances may also be explained by different study designs; uneven sample sizes and variable data sets composition, especially the proportion of patients with and without the condition of interest, may also have influenced such variable results.…”

“… 10 The achieved values also compare well with those reported in the literature, both for strategies with traditional, tabletop retinal cameras and with portable devices. 5 , 21 , 22 , 23 , 24 , 25 …”

“… 26 In order to overcome challenges brought by the increasing numbers of diabetes cases worldwide, along with social and economic barriers to screening, telemedicine and AI have been established as cost-effective solutions. 5 , 17 The employment of a smartphone-based device with an offline, embedded AI system could allow incremental coverage because of the portability and the potential of point-of-care results, even in areas without internet connection, further improving adherence rates and aiding in the prevention of visual loss. 9 , 27 …”

“…Recent modalities, such as telemedicine, artificial intelligence (AI) systems, and portable retinal cameras, have been proposed to increase the uptake rates of DR screening. 5 Among these, handheld retinal cameras are interesting tools for such tasks, because of their affordability and portability, enabling a wider coverage of screening programs. These devices can be utilized by primary care physicians or nonmedical personnel after proper training.…”

Clinical use cases in artificial intelligence: current trends and future opportunities

“…In Singapore, the most successful example is Selena+, a fundal photography AI solution that democratises ophthalmology expertise to optometrists at the primary care level, which gained overseas adoption and was further validated using hand-held fundal photography units. [ 2 ]…”