Performance of raytracing, artificial intelligence and paraxial optics formulae for IOL calculation in very short eyes

Session Details

Session Title: Presented Poster Session: FLACS & Surgical Devices

Venue: Poster Village: Pod 1

First Author: : P.Hoffmann GERMANY

Co Author(s): :                           

Abstract Details

Purpose:

Very short eyes are still very challenging when it comes to precise IOL calculation. Main reason if the high influence of axial IOL position on the result, but lens shape factors, centration, pupil size and spherical aberration are also more important than in normal eyes. We examined two Raytracing packages (Okulix, PhacoOptics), one AI software (Hill RBF), four classical formulae (Haigis, Hoffer Q, Holladay, SRK/T) and two multi-variate formulae (Holladay II, Barrett Universal 2).

Setting:

Private eye clinic in Germany. Ongoing retrospective cohort study.

Methods:

80 short eyes (mean_axial_length 21.10mm) with full biometry and postop followup data sets were identified from our Haag-Streit Lenstar database. All eyes received either a Alcon_SA60AT oder J&J_Tecnis_1_piece. IOL. With the software or formulae mentioned above, the predicted refraction with the implanted lens was calculated based on preop biometry data. When constants were required, we used values that yielded a mean prediction error of zero in normal eyes. These constants were derived inhouse over a long period of time using either Zeiss IOLMaster or Haag Streit Lenstar and did not differ much from ULIB or IOLcon databases.

Results:

Median IOL power was 30.0 D. Mean axial length was 21.10±0.63 mm, mean r 7.66±0.26 mm. Pupil width was assumed as 2.5 mm. Software or formulae delivered the following prediction error [D], mean ± standard devation: Okulix –0.07±0.51, PhacoOptics –0.07±0.54, Hill RBF –0.16±0.58 D (54% of eyes “out of bounds”), Haigis –0.16±0.59, Hoffer Q –0.40±0.64, Holladay –0.11±0.58, SRK/T±0.29±0.60, Barrett Universal 2 –0.45±0.79, Holladay 2 –0.27±0.65. Variances of Raytracing results are significantly smaller than other methods (p<0.05).

Conclusions:

The Raytracing packages Okulix and PhacoOptics yield significantly better results in very short eyes than classical formulae, multivariate formulae or the Artifical Intelligence approach propagated by Warren Hill. Surprisingly, the multivariate formulae both perform worse than the old Holladay or Haigis formulae. Taking away relevant biometric information like WTW and LT from the multivariate formulae does not make a relevant difference for the refractive result. In contrast, Okulix and PhacoOptics rely heavily on ACD and LT measurements to predict the IOL position. In the near future, anterior segment OCT biometry may further improve the results.

Financial Disclosure:

None