Validating a computer algorithm measuring foveal maturity from spectral domain optical coherence tomography images in premature infants

Abstract

Purpose : Semi-automated foveal development analysis was validated on handheld swept source optical coherence tomography (SS-OCT) images. However, this MATLAB program requires heavy user input and has not been optimized for spectral domain (SD)-OCT. This prospective observational study validates a more user-friendly Python computer program (Python Software Foundation, Wilmington, DE, USA) for this purpose.

Methods : A total of 71 images from 47 awake premature infants (46.8% male; mean gestational age 27.99 ± 1.10 weeks; mean birthweight 974.30 ± 279.54 grams) were obtained at ROP screening sessions between 2015 and 2018 using the Envisu C2300 handheld SD-OCT (Leica, Deerfield, IL, USA). Two independent trained graders executed the Python program at 36 weeks (34-38 weeks) postmenstrual age. One grader performed the MATLAB program on a subset of 10 images. The intraclass correlation coefficient (ICC) was calculated between graders and between programs.

Results : For intergrader agreement (n=71), the ICC for inner and outer retinal thicknesses at the fovea were 0.91 (95% CI: 0.86-0.95) and 0.75 (95% CI: 0.60-0.84), respectively. At the parafovea they were 0.83 (95% CI: 0.72-0.89) and 0.81 (95% CI: 0.67-0.88), respectively. Foveal angle ICC was 0.82 (95% CI: 0.71-0.89). For agreement between computer programs (n=10), the ICC for inner and outer retinal thicknesses at the fovea were 0.98 (95% CI: 0.90-0.99) and 0.83 (95% CI: 0.29-0.96), respectively. At the parafovea they were 0.93 (95% CI: 0.73-0.98) and 0.48 (95% CI: -0.43-0.86), respectively. The foveal angle ICC was 0.95 (95% CI: 0.80-0.99).

Conclusions : We developed a user-friendly Python algorithm measuring foveal development from handheld SD-OCT with good to excellent agreement between graders and comparable results to a prior validated MATLAB program, except for outer retinal thickness at the parafovea. This program is feasible for future research and clinical applications.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.

 

(A) Original handheld spectral domain optical coherence tomography image. (B) The Python program measures foveal maturity by automatically segmenting layers. After user input (3 dots near fovea), it provides the following: outer and inner retinal thickness at the fovea and parafovea, foveal angle. ILM = internal limiting membrane; OPL = outer plexiform layer; RPE = retinal pigment epithelium; C/S Junction = conjunctival/scleral junction.