Customized Epi-On Crosslinking Algorithms for Automated Keratoconus Treatment Planning

Abstract

Purpose : Customized crosslinking treatments for keratoconus enhance visual and keratometric outcomes compared to non-customized treatments by tailoring the ultraviolet (UV) light pattern to the corneal geometry of the individual patient. Currently, physicians manually customize UV patterns using heuristics. Here, we assess 12 automated epi-on treatment planning algorithms in an in silico study to identify which have the greatest potential for aberration reduction and corneal flattening while eliminating user subjectivity.

Methods : 20 digital corneas were created using 3D tomography from keratoconic subjects. 12 planning algorithms were applied to each cornea representing 4 input maps (anterior tangential curvature, pachymetry, anterior and posterior elevation) and 3 scaling parameters (small, medium, and large UV zones). Each algorithm automatically detected and segmented the keratoconic defect and created stacked UV dose patterns of 11, 13, and 15 J/cm² fit to the individual corneas.

A photochemical-biomechanical model simulated epi-on treatment effects and predicted post-op corneal shapes. Outputs included change in max corneal curvature (Kmax), cone location and magnitude index, keratoconus severity index, vertical coma (VC), astigmatism, and spherical and higher-order aberrations. Comparisons were made to non-customized 9mm UV spots and manually-customized treatments.

Results : In silico analysis showed a spectrum of corneal flattening and aberration reduction tradeoffs. Pachymetry-based algorithms generated UV patterns that were central, circular, and gave the largest predicted change in Kmax (-2.4D). Elevation algorithms were peripheral, elliptical, and gave the largest predicted change in VC (-0.6um). Tangential curvature algorithms gave a balance between Kmax (-2.3D) and VC (-0.5um). All algorithms outperformed a 9mm spot (-0.2D Kmax, -0.1um VC) and manual customization (-1.7D Kmax, -0.3um VC).

Conclusions : Automated UV treatment plans may outperform non-customized and manually-customized crosslinking treatments for keratoconus while eliminating subjectivity.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.

 

Example UV treatment plans (black circles) generated by 4 of the 12 algorithms for one digital cornea.

 

Modeled change in vertical coma after application of 4 UV algorithm classes at 3 spatial scales (S1-S3) compared to 9mm spots and manually-customized treatments. Automation may improve aberrations and flattening while removing subjectivity.