Exploring the optometric application of expert computer systems: refractive error correction

Using the commercially available shell programs NeuroShell and EXSYS, example-based and rule-based expert computer systems were developed to explore how such systems could be applied in an optometric environment. To demonstrate such an application, the example-based NeuroShell system was taught to prescribe sphere and cylinder powers for low refractive error myopes. When the system learned with only the chief complaint, habitual prescription and acuity through the habitual, and best refraction with resulting acuity, it predicted the sphere and cylinder powers prescribed by human optometrists within 0.25 diopter 97 and 98 percent of the time, respectively, for 60 test eyes. With a set of only three rules, the rule-based EXSYS system predicted sphere powers within 0.25 diopter of the prescribed powers for 100 percent of the 60 test eyes. Results with these very simple refractive problems suggest that teaching expert computer systems to prescribe ophthalmic lens corrections is possible, and further suggest that the concept could be extended to more complex and challenging patient populations.