Acta ophthalmologica · 2021
Prediction of total corneal power from measured anterior corneal power on the IOLMaster 700 using a feedforward shallow neural network.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Langenbucher Achim, Cayless Alan, Szentmáry Nóra, Weisensee Johannes, Wendelstein Jascha, Hoffmann Peter
Аннотация
BACKGROUND: The corneal back surface is known to add some astigmatism against-the-rule, which has to be considered in cataract surgery with toric lens implantation. The purpose of this study was to set up a deep learning algorithm which predicts the total corneal power from keratometry and biometric measures.
METHODS: Based on a large data set of measurements with the IOLMaster 700 from two clinical centres, data from N = 21 108 eyes were included, each record containing valid data for keratometry K, total keratometry TK, axial length AL, central corneal thickness CCT, anterior chamber depth ACD, lens thickness LT and horizontal corneal diameter W2W from an individual eye. After a vector decomposition of K and TK into equivalent power (.EQ) and projections of astigmatism to the 0°/90° (.AST0° ) and 45°/135° (.AST45° ) axis, a multi-output feedforward shallow neural network was derived to predict TK from K, AL, CCT, ACD, LT, W2W and patient age.
RESULTS: After some trial and error, the neural network having a Levenberg-Marquardt training function and three hidden layers (10/8/5 neurons) performed best and showed a fast convergence. The data set was split into training data (70%), validation data (15%) and test data (15%). The prediction error (predicted corneal power CPpred minus TK) of the network trained with the training and cross-validated with test data showed systematically narrower distributions for CPEQ-TKEQ, CPAST0° -TKAST0° and CPAST45° -TKAST45° compared with KEQ-TKEQ, KAST0° -TKAST0° and KAST45° -TKAST45° . There was no systematic offset in the components between CPpred and TK.
CONCLUSION: Unlike any fixed correction term, which can compensate only for a static intercept of the astigmatic components TKEQ, TKAST0° and TKAST45° compared with KEQ, KAST0° and KAST45° , our trained neural network was able to reduce the variance in the prediction error significantly. This neural network could be used to account for the corneal back surface astigmatism for biometers where the corneal back surface measurement or total keratometry is not available.
METHODS: Based on a large data set of measurements with the IOLMaster 700 from two clinical centres, data from N = 21 108 eyes were included, each record containing valid data for keratometry K, total keratometry TK, axial length AL, central corneal thickness CCT, anterior chamber depth ACD, lens thickness LT and horizontal corneal diameter W2W from an individual eye. After a vector decomposition of K and TK into equivalent power (.EQ) and projections of astigmatism to the 0°/90° (.AST0° ) and 45°/135° (.AST45° ) axis, a multi-output feedforward shallow neural network was derived to predict TK from K, AL, CCT, ACD, LT, W2W and patient age.
RESULTS: After some trial and error, the neural network having a Levenberg-Marquardt training function and three hidden layers (10/8/5 neurons) performed best and showed a fast convergence. The data set was split into training data (70%), validation data (15%) and test data (15%). The prediction error (predicted corneal power CPpred minus TK) of the network trained with the training and cross-validated with test data showed systematically narrower distributions for CPEQ-TKEQ, CPAST0° -TKAST0° and CPAST45° -TKAST45° compared with KEQ-TKEQ, KAST0° -TKAST0° and KAST45° -TKAST45° . There was no systematic offset in the components between CPpred and TK.
CONCLUSION: Unlike any fixed correction term, which can compensate only for a static intercept of the astigmatic components TKEQ, TKAST0° and TKAST45° compared with KEQ, KAST0° and KAST45° , our trained neural network was able to reduce the variance in the prediction error significantly. This neural network could be used to account for the corneal back surface astigmatism for biometers where the corneal back surface measurement or total keratometry is not available.