Estimation of Axial Length Using Metadata from Spectral-Domain and Swept-Source Optical Coherence Tomography

Abstract

Purpose : Axial length (AL) affects image magnification and consequently, the accuracy of retinal thickness and vascular measurements on optical coherence tomography (OCT). Conventional methods to measure AL, although non-invasive and easily obtained, require costly devices and may not be readily available, especially in the research setting. We aim to evaluate whether the use of OCT metadata can be used to effectively estimate AL.

Methods : Prospective, observational study. Eyes with best-corrected visual acuity ≤ 20/200, pseudophakia, recent intraocular surgery, eccentric fixation or history of macular edema, choroidal neovascularization, retinal detachment, amblyopia, strabismus and corneal or lenticular pathology were excluded. Patients underwent biometry measurements using the IOLMaster (Zeiss 500; Carl Zeiss Meditec), SD-OCT (SPECTRALIS®; Heidelberg Engineering, Inc.) and SS-OCTA imaging (PLEX®Elite 9000, Carl Zeiss Meditec). Focus factor (FF) was extracted from the Spectralis using a custom software developed at Massachusetts Eye and Ear, in addition to chinrest (CR) and lens (L) position from the Plex Elite. Univariate analyses were performed for potential confounder, lens status, and to evaluate the linear and quadratic relationships between FF, CR, L and corresponding AL. All variables with p ≤ 0.25 were included in the initial multivariate models. Final multivariate models were achieved with a backwards stepwise elimination procedure based on p < .10.

Results : Using Spectralis metadata, 70 eyes of 55 patients were included. Median AL was 24.5 (23.6–26.2) mm. Lens status and FF² were eliminated from the multivariate regression (p = .85, p = .27), while FF was included (p < .001) to create a final regression equation with R² = 0.43:

AL_Estimated = 24.359537 – 0.400459(FF)

Using Plex Elite metadata, 56 eyes of 46 patients were included. Median AL was 24.8 (23.6 – 25.9) mm. The final regression equation included L² and CR² (p<.001, p=.06) with R² = 0.54:

AL_estimated = 24.892 + 2.233*10^-5*L² + 8.684*10^-8*CR²

Conclusions : The use of metadata within the OCT in multivariate regression models shows potential in the estimation of axial length without the need to use conventional biometry methods. Further investigation may improve the accuracy of these models.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.