3 minute read
AI may lead to higher levels
Improving calculations
Artificial intelligence-based approach seen as the best solution. Cheryl Guttman Krader reports
Artificial intelligence (AI) is here, and it will provide the pathway forward for achieving higher levels of refractive accuracy in cataract surgery, predicted John Ladas MD, PhD.
Speaking at the 37th Congress of the ESCRS in Paris, France, Dr Ladas described the work being done to develop a universal “super algorithm” that can improve any existing IOL formula at an accelerated rate.
“Artificial intelligence will transform IOL calculations just like it has transformed countless industries,” said Dr Ladas, Director, Maryland Eye Consultants & Surgeons, Silver Spring, MD, USA. “The shared vision of all cataract surgeons is to keep improving the accuracy of our IOL calculations. Just 10 years ago, the benchmark for success was to achieve an outcome within 0.5D of the target refraction in 55% of eyes. We have shown that we can raise that rate to 86% when we introduce AI, and we think the success rate can be improved even further, exceeding 90% and even reaching 95%.”
He added that major companies are exploring partnerships for adoption of the AI-based method.
“We have already started to develop the infrastructure to develop the methodology and hope to bring it to you soon,” Dr Ladas said.
A BIG DATA APPROACH Dr Ladas has named the AI method “PLUS” (Precision Ladas Universal Superalgorithm, Advanced Euclidean Solutions). It is a dynamic system that will automatically adjust an IOL formula using data submitted in real time from surgeons all over the world, and it can evolve over time to incorporate new variables that may be determined useful in the future.
“Axial length, corneal power and anterior chamber depth are the standard measurements used in IOL formulas, but there is a list of other biometric parameters that might be considered, such as lens thickness, preoperative refraction, white-to-white distance, posterior corneal astigmatism and equatorial lens position, to name a few. There is no human intelligence that can assimilate all of these factors,” said Dr Ladas.
“Vergence formulas and ray-tracing formulas are great formulas but they are static in the sense that they can never evolve. The Hill-RBF formula is AI-based, but it is dataset limited, meaning that if you want to add another variable, the formula has to be reinvented,” Dr Ladas said.
“Our approach is a combination of AI plus the framework that is applied to a framework or baseline formula. This can be any existing formula, such as the Barrett Universal II, Haigis, Holladay, SRK/T, Hoffer Q, etc.”
Rather than predicting an outcome, the PLUS AI-based model starts with the input variables for a case and predicts an adjustment from the baseline formula. As other variables emerge, they are weighted appropriately and the algorithm evaluates what the error is from what was predicted.
Findings from a study that included data from 1,471 eyes operated on in a university setting with IOL calculations done using the Barrett formula showed how application of the AI methodology could improve refractive outcomes after cataract surgery, Dr Ladas said.
“In every AI model we used, we found statistically significant improvement in mean absolute error compared with the result achieved using the Barrett formula,” he said.
NEXT STEPS The pathway forward relies on accumulating massive amounts of data. To meet this need, Dr Ladas envisions that IOL power calculation will be performed using the optical biometry measurements and the AI-based formula. Then, measurements obtained three weeks postoperatively will lead to algorithm optimisation. Ultimately, he foresees use of a “self-calibrating” biometer that uses objective data from millions of eyes and performs an AI-based lens calculation.
“It can be tailored to individual surgeons or groups and will evolve in perpetuity. No human can write that evolution,” Dr Ladas said.
Now, Dr Ladas and nine other high-volume surgeons are participating in a pilot project in which they will collect massive amounts of data and evaluate how the AI-based system can lead to better refractive outcomes.