In this study, we will prospectively evaluate the accuracy of a deep-learning based software algorithm in the detection of diabetic retinopathy from 60° wide single-field retinal fundus images.
We will obtain 60° wide single-field retinal fundus images from subjects who are diabetic patients in primary care environments (i.e. non-eye care settings, such as internal medicine, family medicine, and endocrinology). The fundus images will be uploaded to the RetinaVue Network software using the AutoDx-DR with Over-read modality, where images are transmitted to both AutoDx-DR and a remote ophthalmologist. AutoDx-DR will generate an initial Refer or Not Refer recommendation and the remote ophthalmologist will provide a detailed diagnostic image interpretation which will be used for patient management and referral during the course of the study. AutoDx-DR can also generate a third output which is "Inadequate for Interpretation". In this case, the images will still be sent to the remote ophthalmologist for interpretation. If the ophthalmologist also cannot interpret the images, the patient will be recommended to have an in-person comprehensive ophthalmologic exam. Subjects participating in this study will undergo further retinal fundus imaging: four mydriatic, stereoscopic 45° field of view (4W) retinal images and spectral domain optical coherence tomography (SD-OCT) captured with the Reference Standard Camera (i.e. Zeiss Cirrus 600 photo, or other appropriate FDA-cleared imaging device with stereo fundus photography and SD-OCT capability that meets the requirements of the certified fundus photography reading center such as Topcon, Optovue, and Heidelberg). The 4W and SD-OCT imaging will be performed by a certified technician located at or near the enrolling study site. The 4W and SD-OCT images will be interpreted by a certified fundus photography reading center (FPRC) for more than mild DR or any diabetic macular edema. The FPRC's Not Refer/Refer determination will be used as the "gold standard" interpretation in this study. The AutoDx-DR interpretation of the 60° wide single-field images will be compared to the FPRC "gold standard" interpretation for the Not Refer/Refer recommendation. Accuracy metrics of sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio (PLR), and negative likelihood ratio (NLR), with 95% confidence intervals will be calculated. All diabetic patients who are attending routine primary care or endocrinology appointments and meet the enrollment criteria will be invited to participate in the study. All study procedures will be performed during a single study visit.
Study Type
INTERVENTIONAL
Allocation
NA
Purpose
DIAGNOSTIC
Masking
NONE
Enrollment
1,539
AutoDx-DR provides basic image interpretation for the detection of DR and diabetic macular edema (DME) and a Refer or Not Refer recommendation.
Sensitivity
The ability of AutoDX-DR to correctly identify those with the disease (true positive rate)
Time frame: Each subject will only have one day study visit and will provide the AutoDX-DR and gold standard reading during their one visit.
Specificity
The ability of AutoDX-DR to correctly identify those without the disease (true negative rate)
Time frame: Each subject will only have one day study visit and will provide the AutoDX-DR and gold standard reading during their one visit.
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.