A new sizing tool for the implantable collamer lens (ICL) based on neural network modelling

Session Details

Session Title: Phakic IOLs

Session Date/Time: Monday 09/10/2017 | 08:30-10:20

Paper Time: 09:31

Venue: Meeting Center Room I

First Author: : K.Vandekerckhove SWITZERLAND

Co Author(s): :                        

Abstract Details

Purpose:

Correct ICL sizing is crucial to avoid postoperative complications. White-to-white (WtW) measurement is the key element for the US FDA-approved method for ICL sizing. This method is associated with a 17% rate of suboptimal sizing, with either over- or under-vaulting. Alternatives approaches are based on sulcus-to-sulcus (StS) or on angle-to-angle (AtA) measurements. StS measures are time consuming and reportedly highly variable (2-3%) in comparison to automated WtW measures (<0.5%) or AtA measures (<0.5%).

Setting:

We implanted the most recent ICL version (V4c) in 66 consecutive myopic eyes (36 patients) in the Vista Alpina eye clinc (one surgeon). Postoperative vault (1 day, 1 week, 6 weeks, 6 months) was measured with OCT (Heidelberg Spectralis).

Methods:

First, the predictive value for vault of different pre-operative measurements was evaluated using classical multiple stepwise regression. Second, the predictive value for vault of the same pre-operative measurements was evaluated using artificial neural network modeling (MathWorks, Model Based Calibration Toolbox). WtW (Galilei 6, Scheimpflug), StS (Sonomed VuMax, UBM), anterior chamber depth (ACD; Galilei 6, optical biometry), AtA and lens rise (Heidelberg Spectralis, OCT), average keratometry, axial length and lens thickness (Galilei 6, Scheimpflug), ICL power, as well as ICL size were entered into both models.

Results:

Postoperative vault correlated well with AtA (R=0.45, p=0.002) and StS (R=0.54, p=0.007), but poorly with WtW (R=0.22, ns). The final multiple stepwise regression formula included ICL size, AtA, WtW, ACD, lens thickness and lens rise. Postoperative vault was within 100 µm, 200 µm and 300 µm of the calculated vault in 55%, 83% and 97% of eyes respectively. Using the same preoperative measurements, predictability with the neural network model was significantly better: postoperative vault was within 100 µm, 200 µm and 300 µm of the calculated vault in 70%, 97% and 98% of eyes respectively (chi-square = 8.0, p<0.05).

Conclusions:

OCT based measurement of AtA appears to correlate more strongly with post-operative vault than the traditional WtW measure. Furthermore, the predictability of vault was better with a neural network model than with classical multiple stepwise regression. We continue to feed the neural network model with clinical data on a weekly basis to further optimize its predictability and to validate it as an alternative sizing tool to the standard WtW based ICL sizing algorithm.

Financial Disclosure:

NONE