# Does field of view matter? Assessing its impact on Plus disease diagnosis in retinopathy of prematurity

## TL;DR

This study shows that the field of view in retinal images affects how experts diagnose Plus disease in retinopathy of prematurity.

## Contribution

The study reveals that restricting the field of view improves expert agreement but causes significant diagnostic shifts.

## Key findings

- Inter-expert reliability was slightly higher for 3DD images compared to entire-view images.
- 25% of Plus1 diagnoses shifted to pre-Plus when using 3DD images.
- AI could help reduce diagnostic variability in retinopathy of prematurity.

## Abstract

This study investigates the impact of field of view (FOV) on the diagnosis of Plus disease in retinopathy of prematurity (ROP), focusing on how different FOVs influence expert assessments.

Fundus images from 91 ROP infants were captured using a RetCam, each including the optic disk (OD) and a ridge/demarcation line. Cropped versions with a radius of three disc diameters (3DD) centered on the OD were automatically extracted using a computer-based algorithm. Images were categorized into five (normal, pre-Plus, Plus1, Plus2, Plus3), three (non-Plus, pre-Plus, Plus), or two groups (non-Plus, Plus). Five experts graded both entire-view and 3DD images. Inter-expert reliability was assessed using Cronbach’s alpha.

Seventy-four 3DD and 77 entire-view images were graded. Cronbach’s alpha values were 0.933 (entire-view) and 0.942 (3DD) in the 5-level grading system, indicating slightly higher consistency among experts for 3DD images. Notable diagnostic shifts were observed when comparing entire-view and 3DD images, with 25.0% of Plus1 diagnoses shifting to pre-Plus and 16.9% of pre-Plus diagnoses shifting to normal in the 5-level grading system. Two experts showed substantial changes in their grades, with one expert upscaling 33.7% and downscaling 16.9% of diagnoses, and the other upscaling 21.7% and downscaling 15.7% when assessing entire-view compared to 3DD images.

FOV significantly affects the diagnosis of Plus disease, with peripheral retinal features influencing expert judgments. Restricting the FOV to a 3DD area improved inter-expert reliability but led to notable diagnostic shifts. These findings highlight the need for standardized imaging protocols and the potential role of AI in reducing diagnostic variability in ROP.