AI-guided surgical decision support system for corneal laceration repair: a prospective, non-randomized controlled feasibility study

To evaluate the feasibility and application value of a transfer learning-based artificial intelligence (AI) system for wound recognition and suture position planning in patients with corneal lacerations. In this prospective, non-randomized controlled feasibility study, 25 consecutive patients were managed with either conventional suturing (n = 14) or AI-guided planning (n = 11) based on real-time system availability. The AI system automatically identified corneal laceration images and generated suture points. Best-corrected visual acuity (BCVA, logMAR) and cylinder (Cyl, diopters (D)) were recorded. Anterior corneal surface parameters, including corneal surface variation index (ISV), vertical asymmetry index (IVA), high asymmetry index (IHA), high eccentricity index (IHD), and minimum curvature radius (Rmin), were measured using Pentacam. All patients were followed up for at least 6 months. The transfer learning model achieved over 85% accuracy and sensitivity in corneal laceration recognition. AI guidance significantly improved visual outcomes, with a higher BCVA (0.677 vs. 0.331 logMAR, P < 0.05) and reduced astigmatism (Cyl: 2.023 D vs. 3.542 D, P < 0.05) compared to conventional suturing, and improved ISV, IVA, and IHA (all P < 0.05) compared to the conventional group. This feasibility study provides preliminary evidence that transfer learning-based wound recognition and suture position planning can be applied to corneal laceration treatment, suggesting its potential to provide more precise and effective suturing strategies for corneal trauma patients. Further validation with larger cohorts is warranted.