Variability in Plus Disease Identified Using a Deep Learning-Based Retinopathy of Prematurity Severity Scale

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Retinopathy of prematurity (ROP) is a leading cause of childhood blindness.
• Clinical diagnosis of ROP, particularly 'plus disease,' is subjective, leading to inconsistent treatment.
• Objective diagnostic criteria are needed to improve ROP management.

Purpose of the Study:

• To determine objective differences in the diagnosis of plus disease among clinicians.
• To evaluate the utility of an automated retinopathy of prematurity (ROP) vascular severity score for diagnosing plus disease.

Main Methods:

• Retrospective cohort study using fundus photographs from 871 infants across 8 North American centers.
• A deep learning (DL) algorithm generated an ROP vascular severity score (1-9) for each image.
• Analyzed 5255 examinations, focusing on 168 eyes diagnosed with plus disease by 11 examiners.

Main Results:

• The mean ROP vascular severity score for plus disease was 7.4 ± 1.9 (median 8.5).
• Significant variability was observed in the diagnosis of plus disease both between and within individual examiners.
• One examiner consistently diagnosed plus disease at lower severity scores compared to others (P < 0.01).

Conclusions:

• Deep learning-based scoring revealed inter- and intra-examiner variability in diagnosing ROP plus disease.
• Objective vascular severity measurements could help define diagnostic thresholds for plus disease.
• Further research integrating clinical features with objective scores is needed for standardized ROP treatment decisions.