3 minute read
Using Artificial Intelligence to Improve IOL Formulas
With more than 23 million procedures performed each year, cataract surgery is the world’s most common surgical procedure. To achieve the best possible visual outcome for the patient, it’s crucial to implant an intraocular lens (IOL) with the optimal power. To select the IOL’s power, the surgeon relies on a formula to calculate which power will achieve a postoperative refraction that best matches the patient’s goals.
A number of such formulas are available, using different methodologies to weigh variables like the eye’s length, corneal shape and lens thickness. While the accuracy of IOL power calculation has risen over the years, so have patient expectations for their refractive outcomes, creating demand for still more powerful formulas.
Cornea and cataract surgeon Nambi Nallasamy, M.D., saw this as an ideal application for an artificial intelligence (AI) technology called machine learning (ML).
“In ML, computers ‘learn’ from processing data sets far larger and more multifaceted than a human mind can absorb,” he explains. “In this case, we taught the computer to make IOL power decisions by processing data from nearly 10,000 patients who underwent cataract surgery at Kellogg.” That data set included key patient demographics plus eight different eye measurements.
The resulting Nallasamy Formula was then tested against seven of the most commonly used formulas, using retrospective patient data. It outperformed them all. Dr. Nallasamy and fellow cataract surgeons at Kellogg now employ it side-by-side with other formulas.
Having demonstrated that ML can improve IOL calculations for his patients, Dr. Nallasamy is now applying it to challenges specific to other populations.
First, he hopes ML can help make cataract surgery outcomes around the world more equitable by overcoming data biases inherent in IOL formulas. “Eye size and shape vary throughout the world,” he explains. “But our data sets reflect our local population. We need an ML tool that can easily adjust to different populations and IOL types with minimal additional information—a challenge called domain adaptation.”
Dr. Nallasamy is also putting his ML approach to work to help surgeons navigate more complex conditions, including patients requiring cataract surgery along with corneal transplantation, and those with corneal thinning disorders like keratoconus.
Selecting the right lens to meet these patients’ needs is especially challenging, because the cornea continues to change as they recover from surgery.
“Again, ML can generate a more customized lens calculation formula when we fine tune the dataset we use,” he explains, “emphasizing measurements from optical biometry and tomography to better model the postoperative evolution of the cornea.”
Header image caption: Nambi Nallasamy, M.D.