Anterior Chambers Shallower Than 2.5 mm in Eyes With Short to Normal Axial Length: Can Artificial Intelligence-Based Lens Formulas Achieve Reliable Accuracy?

PURPOSE: To evaluate the performance of artificial intelligence (AI)-based intraocular lens (IOL) power calculation formulas in eyes with extremely shallow anterior chambers (anterior chamber depth [ACD] < 2.5 mm).

METHODS

This retrospective study enrolled 120 eyes of 120 patients with ACD < 2.5 mm who underwent phacoemulsification and IOL implantation. Seven AI formulas were evaluated: Hill-RBF 3.0, Kane, PEARL-DGS, LSF Al, Hoffer QST, Karmona, and Nallasamy. Subgroup analysis was performed based on lens thickness (LT).

RESULTS

The Kane, PEARL-DGS, and Hoffer QST formulas exhibited no significant systematic bias (all P > .05) and demonstrated comparable refractive predictability in the overall cohort. In the thin lens subgroup (LT < 4 mm), the PEARL-DGS, LSF Al, and Hoffer QST yielded superior precision. In the intermediate subgroup (4 mm ⩽ LT < 5 mm), the Hill-RBF 3.0, Kane, and PEARL-DGS achieved higher accuracy. Notably, in eyes with pronounced lens-related crowding (LT ⩾ 5 mm), the Hill-RBF 3.0, Kane, and Hoffer QST remained the most reliable.

CONCLUSIONS

Lens thickness is a critical determinant of refractive outcomes in eyes with extremely shallow ACD. Given that no individual AI-based formula maintains optimal predictive performance across the entire spectrum of lens morphology, a stratified formula selection strategy is recommended. Specifically, the PEARL-DGS is prioritized for thin lenses, Kane for intermediate profiles, and Hoffer QST for eyes with pronounced lens disproportion.