Abstract
Purpose :
Using Artificial Intelligence techniques to develop a new nomogram for the Wavelight® Refractive Suite, taking into account the spherical equivalent that is to be corrected, preoperative clinical parameters, and type of planned surgery (Laser-assisted in situ keratomileusis (LASIK) or photorefractive keratectomy (PRK)).
Methods :
5000 LASIK cases and 2000 PRK cases were analyzed. Actually delivered treatments were calculated by subtracting preoperative refractions from postoperative refractive outcomes. The Wavelight® Laser treatment delta (Δ) was calculated by subtracting actually delivered treatments from planned treatments, and corresponded to the "target" to enter in the laser software during treatment planning in order for the patient to achieve emmetropia. The data were divided in a train set and a test set and used to train a machine learning algorithm, in conjunction with preoperative clinical parameters, to predict the treatment delta on data never seen previously by the algorithm.
Results :
Predictions from the trained model were 20% more accurate than the existing nomogram for the spherical part of the treatment, and 25% more accurate than the existing nomogram for the cylindrical part of the treatment.
Conclusions :
Our nomogram allows a more accurate adjustment of the refractive target making it superior to the nomogram currently available for the Wavelight® Laser Suite. Artificial intelligence techniques have the potential to improve the predictability of treatment results we aim to deliver in laser refractive surgery.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.