ESCRS - PO458 - A Novel Intraocular Lens Power Calculation Method Based On Artificial Intelligence For Chinese Cataract Patient With Long Axial Length.

A Novel Intraocular Lens Power Calculation Method Based On Artificial Intelligence For Chinese Cataract Patient With Long Axial Length.

Published 2024 - 42nd Congress of the ESCRS

Reference: PO458 | Type: Free paper | DOI: 10.82333/64y8-nh65

Authors: Xue Ding 1 , Xin Xin Li 2 , Shao Wei Li* 1 , Yan Ling Wei 3 , Jia Wang 3 , Qiong Yan Tang 4 , Yong Wang 5 , Jing Wang 6 , Yan Wen Zhou 7 , Juan Sun 8 , Zhe Ming Wu 9 , Yan Hong Jia 10 , Zhen Lin Lin 11

1Department of Ophthalmology,Beijing Aier-Intech Eye Hospital,Beijng,China, 2Central South University,Changsha,China, 3Beijing Nine stars intelligent Medical Equipment CO.,Ltd,Beijng,China, 4Changsha Aier Eye Hospital,Changsha,China, 5Wuhan Aier Eye Hospital,Wuhan,China, 6Shenyang Aier Excellence Eye Hospital,Shenyang,China, 7Liaoning Aier Eye Hospital,Liaoning,China, 8Hubin Aier Eye Hospital,Binzhou,China, 9Guangzhou Aier Eye Hospital,Gangzhou,China, 10Nanning Aier Eye Hospital,Nanning,China, 11Fuzhou Aier Eye Hospital,Fuzhou,China

Purpose

To develop and evaluate an accurate new machine learning-based intraocular lens (IOL) power estimation method (LSWAI) for Chinese cataracts with axial length (AL) >26.0mm.

Setting

Aier Eye Hospital Group Co., Ltd, China.

Methods

The dataset was collected from 10 deferent centers, and we divided the data set into a training set containing 1123 cases and a test set containing 281 cases. The formula adopts a total of 5-dimensional features including eye AL, coreal curvature (K1, K2), lens thickness (LT) and anterior chamber depth (ACD). In the process of formula construction, we first perform polynomial transformation on the features and then analyze them using Principal Component Analysis (PCA). The AI model is trained with selected features and undergoes 5-fold cross-validation on the training set to prevent overfitting. The accuracy of the AI models is compared with that of the Barrett Universal II (BUII), Kane, and EVO 2.0 power formulas.

Results

The MAEs±SD were: 0.34±0.44D for LSWAI, 0.36±0.50D for BUII, 0.36±0.53D for EVO, 0.47±0.84D for Kane. The LSWAI formula has the highest proportion of prediction errors (PEs) within ±0.25D, ±0.50D, ± 0.75D, ±1.0D(42.7%, 78.29%, 91.10%, 96.80%). The proportion for BUII was 40.93%, 71.17%, 87.19%, 93.24%, respectively, for EVO 2.0 was 37.72%, 71.89%, 87.19%, 92.53%, and for Kane was 39.15%, 70.11%, 86.48%, 91.81%. 

Conclusions

The LSWAI formula predicts IOL power with high accuracy and stability, For Chinese cataracts with AL >26.0mm.