Validating DevCheck: a semi-automated computer algorithm for measuring foveal immaturity in premature infants.

Area of Science:

• Ophthalmology
• Medical Imaging
• Neonatology

Background:

• Foveal immaturity is a key biomarker for retinopathy of prematurity (ROP).
• Existing semi-automated MATLAB programs for foveal development analysis require significant user input and are not optimized for spectral domain optical coherence tomography (SD-OCT).

Purpose of the Study:

• To validate the "DevCheck" Python algorithm for analyzing foveal development using handheld SD-OCT in premature infants.
• To assess the feasibility of DevCheck for routine ROP examinations.

Main Methods:

• Handheld SD-OCT imaging was performed on awake premature infants during ROP screening.
• The "DevCheck" Python algorithm was used by two independent graders.
• Intergrader agreement and agreement with a validated MATLAB program were assessed using intraclass correlation coefficients (ICCs).

Main Results:

• Strong intergrader agreement was observed for foveal and parafoveal retinal thicknesses and foveal angle (ICCs ranging from 0.75 to 0.91).
• DevCheck showed comparable results to the MATLAB program, with high ICCs for most parameters (ranging from 0.83 to 0.98), except for outer retinal thickness at the parafovea.
• Older gestational age was associated with a decreasing inner retinal fovea/parafovea ratio (P=0.004).

Conclusions:

• The "DevCheck" Python algorithm is a user-friendly tool for measuring foveal development from handheld SD-OCT images.
• DevCheck demonstrates good to excellent intergrader agreement and comparable performance to existing validated programs.
• These findings support the potential of DevCheck for future research and clinical applications in ROP screening.