Scholarlink

Purpose <p>To assess the accuracy of artificial intelligence (AI)-based intraocular lens (IOL) power calculation formulas compared with traditional methods in highly myopic eyes, and to evaluate their performance across varying axial lengths and corneal curvatures.</p> Methods <p>This retrospective case series included 115 highly myopic eyes that underwent phacoemulsification with IOL implantation. IOL power was calculated using four conventional formulas (SRK/T, Haigis, Holladay 2, Barrett Universal II) and seven AI-based formulas (Hill-RBF 3.0, Karmona, Hoffer QST, PEARL-DGS, Ladas Super Formula, Kane, HM-ZL). The outcomes were evaluated using standard deviation (SD), assessed with Heteroscedastic test; root-mean-square absolute error (RMSAE), assessed with bootstrap-t method; mean absolute error (MAE), assessed with Friedman test; and the percentage of eyes within ± 0.25 D to ± 1.00 D of prediction error, assessed with Cochran’s Q test. Subgroup analyses were performed based on axial length (AL) and corneal curvature (Kmean).</p> Results <p>Most AI-based formulas—especially Hill-RBF 3.0, PEARL-DGS and Kane—demonstrated higher accuracy than traditional formulas. Results: Overall, the MAEs of Hill-RBF 3.0, PEARL-DGS, and Kane were significantly lower than that of Holladay 2 (<i>P</i> &lt; 0.05). The SD of PEARL-DGS also differed significantly from Holladay 2 (<i>P</i> &lt; 0.05). In the long axial length group, Hill-RBF 3.0, PEARL-DGS, and Kane showed significantly lower MAEs than Holladay 2 (<i>P</i> &lt; 0.05). In the moderate corneal curvature group, BUⅡ, Hill-RBF 3.0, Hoffer QST, PEARL-DGS, and Kane had significantly lower MAEs than Holladay 2, and the SDs of Hill-RBF 3.0 and PEARL-DGS differed significantly from both Holladay 2 and SRK/T (<i>P</i> &lt; 0.05). Trend lines showed that AI-based formulas exhibited more consistent and stable performance across different AL and K<sub>mean</sub>.</p> Conclusion <p>AI-based formulas provide superior refractive prediction in highly myopic eyes compared with traditional methods. Tailored formula selection based on biometric profiles may enhance refractive outcomes in cataract surgery.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparative evaluation of traditional and AI-based intraocular lens power calculation formulas in highly myopic eyes

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月23日

查看全文
  • Xiaopeng Jiang,
  • Jiangjie Wang,
  • Qingmin Jiang,
  • Xiangyu Zhou,
  • Fei Xia,
  • Meng Gao

摘要

Purpose

To assess the accuracy of artificial intelligence (AI)-based intraocular lens (IOL) power calculation formulas compared with traditional methods in highly myopic eyes, and to evaluate their performance across varying axial lengths and corneal curvatures.

Methods

This retrospective case series included 115 highly myopic eyes that underwent phacoemulsification with IOL implantation. IOL power was calculated using four conventional formulas (SRK/T, Haigis, Holladay 2, Barrett Universal II) and seven AI-based formulas (Hill-RBF 3.0, Karmona, Hoffer QST, PEARL-DGS, Ladas Super Formula, Kane, HM-ZL). The outcomes were evaluated using standard deviation (SD), assessed with Heteroscedastic test; root-mean-square absolute error (RMSAE), assessed with bootstrap-t method; mean absolute error (MAE), assessed with Friedman test; and the percentage of eyes within ± 0.25 D to ± 1.00 D of prediction error, assessed with Cochran’s Q test. Subgroup analyses were performed based on axial length (AL) and corneal curvature (Kmean).

Results

Most AI-based formulas—especially Hill-RBF 3.0, PEARL-DGS and Kane—demonstrated higher accuracy than traditional formulas. Results: Overall, the MAEs of Hill-RBF 3.0, PEARL-DGS, and Kane were significantly lower than that of Holladay 2 (P < 0.05). The SD of PEARL-DGS also differed significantly from Holladay 2 (P < 0.05). In the long axial length group, Hill-RBF 3.0, PEARL-DGS, and Kane showed significantly lower MAEs than Holladay 2 (P < 0.05). In the moderate corneal curvature group, BUⅡ, Hill-RBF 3.0, Hoffer QST, PEARL-DGS, and Kane had significantly lower MAEs than Holladay 2, and the SDs of Hill-RBF 3.0 and PEARL-DGS differed significantly from both Holladay 2 and SRK/T (P < 0.05). Trend lines showed that AI-based formulas exhibited more consistent and stable performance across different AL and Kmean.

Conclusion

AI-based formulas provide superior refractive prediction in highly myopic eyes compared with traditional methods. Tailored formula selection based on biometric profiles may enhance refractive outcomes in cataract surgery.