Best-Corrected Visual Acuity Quantitative Prediction for Cataract Patients: AI-Assisted Clinical Diagnostics Facilitation via the Inverse Problem Algorithm.

Objective

This study provided a quantitative prediction of best-corrected visual acuity (BCVA) for cataract patients using the inverse problem algorithm (IPA) technique earlier proposed by the authors.

Methods

To this end, seven risk factors (age, BMI, MAP, IOP, HbA1c, LDL-C, and gender) were linked by a semi-empirical formula by normalizing each factor into a dimensionless range of -1.0 to +1.0. The adopted inverse problem algorithm (IPA) technique was run via a self-developed program in STATISTICA 7.0, featuring a 29-term nonlinear equation considering seven risk factors, cross-interaction between various pairs of factors, and one constant term [7 + (7 × 6)/2 + 1 = 29]. The IPA neglected quadratic, triple, or quadruple factors' cross-interactions. This study used a dataset of 632 cataract patients to attain a reliable BCVA prediction with a variance of 0.929. A verification dataset of 160 patients with similar symptoms was used to verify this approach's feasibility, reaching a good correlation with R2 = 0.909.

Results

The verification group's derived average AT (agreement) (9.12 ± 27.00%) indicated a slight deviation between the theoretical prediction and practical BCVA. The significant factors were age, body mass index (BMI), and intraocular pressure (IOP), whereas mean arterial pressure (MAP), hemoglobin A1c (HbA1c), low-density-lipoprotein cholesterol (LDL-C), and gender insignificantly contributed to BCVA.

Conclusions

The proposed approach is instrumental in AI-assisted clinical diagnosis, yielding robust BCVA predictions for individual cataract patients based on their biological indices before the ophthalmological examination procedure.