Using artificial intelligence to predict the success of laser eye surgery

Algorithms modelled on biological neural networks can predict whether patients will need further corrective treatment, study shows

What if you knew that a medical procedure you intended to have was likely to be unsuccessful? What if your doctor could change things before the procedure to maximize the chance of success, specifically for you?

Now, researchers have developed an algorithm that can predict whether one of the most common surgical procedures – laser eye surgery – will be successful in a given patient. The technique, described in the study “Using Neural Networks to Predict the Outcome of Refractive Surgery for Myopia” in open-access journal 4Open, could allow surgeons to tailor the procedure for each patient to maximize the chances of success.

Predicting success

For most patients, laser eye surgery for nearsightedness is effective and safe. However, a small percentage of patients do not experience a successful outcome and require subsequent corrective treatment.

Before many medical procedures, a doctor can give a patient a general idea of the chances of success. However, these success rates are often based on the general population and typically don’t reflect an individual patient’s chances.

In fact, given the number of unique factors affecting each patient’s outcome, it’s often too complex for a doctor to predict the success of a procedure in any one patient. Thankfully, with the advent of artificial intelligence, computers are ideally suited to this task.

Neural networks for individualized treatment

To better predict laser eye surgery patient outcomes, the authors of this latest study turned to neural networks – computing systems inspired by biological neural systems. These computer systems can learn to perform specific tasks by studying large numbers of relevant examples and identifying patterns that would be difficult or impossible for a human to spot.

“Neural networks are increasingly becoming part of our everyday life, whether they play chess or bridge, expertly translate between languages or predict the movement of the stock market,” says lead research Professor George Anogeianakis. “Ophthalmology has also benefitted from the power of neural networks, and so far, many of the predictions made by such computer systems are as good or better than those from experts.”

Developing a predictive tool

The researchers used data from over 2,000 patients who had undergone laser eye surgery to establish the link between surgery failure and 13 factors that have been reported to influence treatment outcomes. These included five patient-specific factors, such as age and thickness of a layer of the cornea, and eight non-patient factors, such as surgical technique and the temperature and humidity of the surgical suite.

By feeding these data into their neural network, the team found that they could predict the success or failure of a procedure with a high degree of accuracy. “The system is a simple clinical decision tool that identifies patients who are more likely to require a second corrective procedure,” explains Anogeianakis. “The surgeon can simply input the 13 required patient and non-patient parameters into the system before a procedure and the computer can then inform them if the patient is at high risk or not.”

Excitingly, as eight of the parameters – such as the planned surgical procedure – can potentially be modified by the surgeon, it may be possible to adapt surgical plans in response to the computer predictions and thus increase the patient’s chances of a successful procedure.

“It’s a question of medical ethics for doctors to use the most advanced technology to improve their patients’ outcomes and give their patients the most accurate information about their chances of success,” says Anogeianakis. “This system contributes to that goal.”

This "Key Summaries" article is brought to you in collaboration with SciencePOD.