GPT-5–assisted versus expert surgeon refractive planning in smooth incision lenticule keratomileusis (SILK): comparative analysis and visual outcomes
摘要
To assess the performance of GPT-5 in refractive surgery planning by comparing its recommendations with expert surgeons and reporting visual outcomes of surgeon-planned Smooth Incision Lenticule Keratomileusis (SILK) procedures.
MethodsThis retrospective, observational study included 134 eyes from 67 patients who underwent SILK procedure using the ELITA femtosecond laser platform from January 2024 to September 2025. GPT-5 generated refractive plans based on manifest and cycloplegic refraction, autorefractometry, visual acuity, keratometry, and pachymetry. These were compared with surgeon-derived values using paired t-tests and correlation analysis. Preoperative and postoperative (1- and 3-month) uncorrected and corrected distance visual acuity (UDVA, CDVA), mean keratometry (Km), and thinnest corneal point were analyzed. Axis concordance was quantified as the absolute angular deviation (Δ Axis) between AI- and surgeon-planned astigmatism axes.
ResultsAI-generated spherical corrections were more conservative than surgeons’ (− 4.29 ± 2.42 D vs. − 4.70 ± 1.98 D; p = 0.0025), whereas cylindrical power did not differ significantly (p = 0.9241). The mean Δ Axis was 61.9° ± 50.1°, indicating substantial misalignment. UDVA improved from 0.74 ± 0.36 logMAR preoperatively to − 0.06 ± 0.07 logMAR at 3 months (p < 0.0001), and postoperative UDVA did not differ significantly from preoperative CDVA, confirming excellent visual recovery.
ConclusionsGPT-5 generated refractive plans that were partially consistent with expert surgeon decisions but lacked the clinical precision required for clinical implementation. In contrast, surgeon-planned SILK procedures achieved excellent visual acuity, corneal stability, and refractive predictability, reinforcing the safety and efficacy of Kerato Lenticule Extraction (KLEx) in real-world clinical practice.