Automated Screening Device to Detect MTMROP

Overview

The purpose of the study is to evaluate the i-ROP DL system for autonomous ROP screening. To demonstrate the sensitivity and specificity of the i-ROP DL system on images obtained by healthcare providers in the neonatal intensive care unit (NICU) setting to detect more than mild ROP (MTMROP), defined as type 2 or worse (as defined by the Early Treatment for ROP \[ETROP\] study), or pre-plus, compared to an image-based reference standard diagnosis (RSD). This study will utilize a large multicenter dataset of images obtained as part of the Multicenter Telemedicine Approaches to Evaluating Acute-phase ROP (e-ROP) Study, a National Institute of Health multicenter study.

Retinopathy of prematurity is a disease that develops when babies are born prematurely, before the retina has fully developed. It is a leading cause of childhood blindness worldwide, but blindness is usually avoidable with accurate diagnosis, and timely treatment of babies with treatment-requiring disease. The purpose of diagnostic eye exams therefore (sometimes referred to as "ROP screening" examinations by clinicians) is to identify severe disease that requires treatment, and administer treatment, to reduce the risk of blindness. Each country has accepted criteria for which babies require eye examinations for ROP. In the United States, all babies born less than 31 weeks post-menstrual age (PMA) or under 1500 grams require diagnostic eye examinations starting at 31 weeks PMA or 4 weeks after birth, whichever is later. Exams are repeated at 1-2 week intervals depending on the level of disease identified on examination. Diagnosis of ROP is made using either ophthalmoscopy (in person examination at the bedside) or using telemedicine with remote digital fundus imaging. Telemedicine has become an acceptable standard of care for ROP diagnostic examinations, and in most telemedicine workflows, exams are repeated weekly to minimize the risk of missing ROP. Whether exams occur in person or via telemedicine, the ophthalmologist is responsible for documenting the presence or absence of ROP using the International Classification of ROP (ICROP) guidelines, including documenting the presence and degree of zone, stage, and plus disease. The zone, stage and plus disease classifications lead to overall disease severity categories, as defined by the Early Treatment for ROP (ETROP) study (see Table). The ETROP study determined that urgent treatment (within 48 hours) should occur when babies develop "type 1" ROP, but was not indicated for babies with type 2 ROP or less. Follow-up guidelines are determined by the responsible physician based on the level of disease, although in telemedicine these typically occur weekly. ROP "screening" criteria are designed to be highly sensitive so that babies with any risk of developing type 1 ROP are captured in the screening population and receive diagnostic eye examinations. As a result, most babies who receive examinations do not have severe ROP on examination. ROP examinations (especially in person ophthalmoscopic examinations where scleral depression is utilized) are stressful to the neonates, and time consuming to the ophthalmologist, and thus the current system is inefficient and exposes neonates to added stress, which can have significant short and long-term consequences . The ideal solution would be to reduce the number of in person examinations a baby receives without missing any cases of type 1 ROP, and limit the number of diagnostic eye examinations an ophthalmologist needs to perform to those babies with more than mild ROP (type 2 or worse). This scenario is directly analogous to diabetic retinopathy, where the recommendation is for all diabetics to have an eye examination every year, even though only a small minority will have treatment-requiring disease. In diabetic retinopathy, autonomous AI has provided an alternative care delivery model where screening examinations occur in the primary care office, and only positive screens are sent for full ophthalmoscopic examination by an eye care provider. In this clinical study, we are proposing to use the i-ROP DL system as an autonomous screening software as medical device for the detection of "more than mild ROP" (MTMROP) defined as type 2 or type 1 ROP using ETROP criteria. Digital images can be acquired by a technician in a standardized fashion. The system can analyze the images and output a positive or negative exam result. * A positive result triggers a referral for telemedical review by an ophthalmologist using the same set of images (does not require additional stress to babies). The responsible clinician can document the level of ROP and recommend either in person examination for possible treatment, or repeat screening examination or telemedical review at the appropriate interval. * A negative result automatically triggers a scheduled follow-up appointment for repeated screening examination in 1 week. The i-ROP DL system was originally trained to diagnose plus disease at 3 levels (no plus, pre-plus or plus ). Subsequent work using the outputs for plus disease classification, we developed the concept of a "vascular severity score" (VSS) as an output of the system. The VSS is applied at the eye level from 1-9 and reflects the spectrum of vascular changes in pre-plus and plus disease. Because pre-plus and plus disease develop in more severe ROP, the VSS can be used as an objective assessment of ROP severity, increasing with increasing stage of disease, and overall category of ROP. Thus, with the appropriate cut-point, the VSS can be used for screening purposes ensuring nearly 100% sensitivity for detection of treatment-requiring (type 1) ROP, and high sensitivity for detection of type 2 (or worse) ROP. The cut-point for this study will be \>=2.9. This study will utilize a large multicenter dataset of images obtained as part of the Multicenter Telemedicine Approaches to Evaluating Acute-phase ROP (e-ROP) Study, a National Institute of Health multicenter study. Importantly, the e-ROP study was designed and executed prospectively to evaluate telemedical eye examinations compared to in person ophthalmoscopic examinations. Digital fundus images were acquired using the Retcam Shuttle by trained technicians using a standardized protocol for image acquisition. Images were then subsequently read by telemedical graders. All babies received in person ophthalmological examinations by ophthalmologists who were masked to the images and were not part of the image acquisition process. In other words, the study was perfectly designed to evaluate an autonomous screening model for ROP where images are acquired by non-physicians using a defined protocol at the point of care and only positive exams are subsequently reviewed by the ophthalmologist.