Efficacy of segmented axial length and artificial intelligence approaches to intraocular lens power calculation in short eyes.

Abstract

In short eyes, to compare the predictive accuracy of newer intraocular lens (IOL) power calculation formulas using traditional and segmented axial length (AL) measurements. Cullen Eye Institute, Baylor College of Medicine, Houston, Texas and East Valley Ophthalmology, Mesa, Arizona. Multi-center retrospective case series. Measurements from an optical biometer were collected in eyes with AL <22 mm. IOL power calculations were performed with 15 formulas using 2 AL values: (1) machine-reported traditional AL (Td-AL) and (2) segmented AL calculated with the Cooke-modified AL nomogram (CMAL). 1 AL method and 7 formulas were selected for pairwise analysis of mean absolute error (MAE) and root mean square absolute error (RMSAE). The study comprised 278 eyes. Compared with the Td-AL, the CMAL produced hyperopic shifts without differences in RMSAE. The ZEISS AI IOL Calculator (ZEISS AI), K6, Kane, Hill-RBF, Pearl-DGS, EVO, and Barrett Universal II (Barrett) formulas with Td-AL were compared pairwise. The ZEISS AI demonstrated smaller MAE and RMSAE than the Barrett, Pearl-DGS, and Kane. K6 had a smaller RMSAE than the Barrett formula. In 73 eyes with shallow anterior chamber depth, the ZEISS AI and Kane had a smaller RMSAE than the Barrett. ZEISS AI outperformed Barrett, Pearl-DGS, and Kane. The K6 formula outperformed some formulas in selected parameters. Across all formulas, use of a segmented AL did not improve refractive predictions.