Validation Of An Artificial Intelligence (Ai) Algorithm To Predict Implantable Collamer Lens (Icl) Size And Predict Post-Operative Vault Using Swept Source Oct Images.

Validate an automated algorithm of prediction of adequate ICL size and post-operative vault.

External Validation Group of 376 eyes implanted with ICL.

We developed a deep learning model which uses the OCT images as inputs and outputs the estimated vault and a probability of being inside the 250-750 microns range (P250-750) . The model selects the optimal size as the one exhibiting the higher P250-750 probability.

A group of 376 eyes implanted in an external hospital with an ICL using the STAAR nomogram were included in this study. The pre-operative anterior segment swept-source optical coherence tomography (AS SS-OCT) images were analyzed by our model blinded from the post operative results. The suggested ICL size and post-operative vault using this algorithm was compared to the ICL size and post-operative vault obtained using the regular nomogram.

A total of 376 eyes were included in this study. The average predicted  P250-750 was 59% for the implanted size using our model. 67% of the eyes actually achieved a vault in the range 250-750 microns. The absolute mean error between the predicted vault and the achieved vault was 142+/-115 microns, with 85% of cases having an absolute error <250 microns. In cases where the measured vault was inadequate (<250 or >750 microns), the algorithm suggested a more adequate ICL size in 81% of cases. In cases of ICL exchange, the algorithm predicted the right ICL size in 93.75% of cases (15 out of 16 exchanged ICL).

This developed algorithm of ICL size and post-operative vault prediction uses unique image data to analyze the anterior segment and can help clinicians guide their ICL size choice in a precise way.