長眼における9つの人工知能ベースの眼内レンズ度数計算式予測

To compare accuracy of nine artificial intelligence (AI)- based intraocular lens (IOL) power calculation formulas in eyes with axial length > 26 mm.

Retrospective accuracy and validity analysis.

Myopic patients with cataract who underwent uneventful phacoemulsification with in-the-bag implantation of a PARTIAL-RoF narrow IOL in MW-med Eye Center, Krakow, Poland; or in Kyiv Clinical Ophthalmology Hospital Eye Microsurgery Center, Ukraine METHODS: Prior to cataract surgery IOL power was calculated. The power of the implanted IOL was selected from the predictions of SRK/T, Holladay 2 or Barrett Universal II. Three months after phacoemulsification, refraction was measured. Post-surgery IOL power calculations were performed utilizing the following formulas: 3C 2.0, Hill-RBF 3.0, Hoffer QST, Kane, Karmona, Ladas Super Formula AI (LSF AI), Nallasamy, PEARL-DGS, and Zhu-Lu.

Root mean square absolute error (RMSAE), median absolute error (MedAE) and percentage of eyes with prediction error (PE) within ±0.25 D, ±0.50 D, ±0.75 D and ±1.00.

Three hundred twenty one eyes with axial length ≥ 26.00 mm were studied. Considering RMSAE, all tested formulas except Karmona demonstrated statistical superiority over Zhu-Lu (0.532, p<0.001) and 3C 2.0 (0.510, p<0.001). In terms of MedAE, Hill-RBF 3.0, Pearl-DGS and Kane achieved statistically better outcomes (p<0.05) than 3C 2.0 and Zhu-Lu (0.260, 0.270, 0.280, 0.342, 0.371; respectively). Based on percentage of eyes with PE within ±0.50 D, the Pearl-DGS (80.06%), Hill-RBF 3.0 (77.57%), Kane (77.26%), Hoffer QST (76.95%), LSF AI (76.95%), and Nallasamy (76.64%) formulas exhibited statistically superior accuracy (p<0.05) compared with the Zhu-Lu (66.67%) and 3C 2.0 (68.85%) formulas.

Among the nine AI-based formulas evaluated, Pearl-DGS, Hill-RBF 3.0, Kane, and Hoffer QST demonstrated comparably the highest accuracy in IOL power calculation in eyes with an AL > 26 mm, in terms of RMSAE and percentage of eyes within 0.5 D of target.