Axial length estimation using automated OCT scans

Abstract

Purpose : To construct an axial length (AL) estimation model using calibration and scan-time data from Spectral Domain (SD)-OCT devices. And to evaluate its use for aiding clinical evaluation.

Methods : 12 x 9 mm 3D OCT wide scans using Maestro2 (Topcon Healthcare, Tokyo, Japan) were collected. Parameters obtained during device calibration include the step size for reference mirror (m_step), baseline reference mirror position (m₀), OCT depth digital resolution (z_step), baseline z position (z₀), and baseline AL at m₀ (AL₀). Each scan’s reference mirror position (m) was recorded at capture time and the average retinal tissue position (z) was calculated post-process. Using these parameters, an equation to estimate AL was derived and validated using two datasets of eyes with corresponding OCT scans and measured AL. Root-mean-squared error (RMSE) was calculated between estimated and measured AL.

To evaluate the use of estimated AL, a reference database (RDB) dataset, previously collected through a prospective clinical study, and a test dataset, collected from four optometry practices, were obtained. All test scans were reviewed and found to be healthy. The average, temporal, superior, and inferior circumpapillary retinal nerve fiber layer thickness (RNFLT) were computed with and without magnification correction using estimated AL. Non-AL-corrected RDB was applied on non-AL-corrected test data, and AL-corrected RDB was applied on AL-corrected test data. The percentages of scans with RNFLT below 5% were computed.

Results : The derived equation is AL = m_step × (m − m₀) + z_step × (z − z₀) + AL₀. Figure 1 shows the correlation between estimated and measured AL. Dataset A had 232 scans with measured AL from 21.31 to 27.51 mm and a RMSE of 0.799 mm. Dataset B had 11360 scans with measured AL from 19.79 to 30.48 mm and a RMSE of 0.587 mm.

The RDB and test data consisted of 398 and 2119 scans, respectively. AL correction lowered the percentage of scans with RNFLT below 5% in all four sectors (Fig. 2A). With AL correction, there is a reduced dependency between estimated AL and flagged results (Figs. 2B−C).

Conclusions : An equation for AL estimation using SD-OCT calibration and scan-time parameters was derived and validated. In healthy eyes, flagging and its correlation with AL were reduced when OCT measurements were corrected by estimated AL.

This abstract was presented at the 2024 ARVO Imaging in the Eye Conference, held in Seattle, WA, May 4, 2024.

 

Estimated versus measured AL.

 

(A) Percentage of scans < 5% in each sector. (B) Boxplots of estimated AL in three flagged groups.