Posters
Forecasting the refractive error post cataract surgery using Tensorflow and Tensorboard, two open source machine learning tools developed by Google brains
Poster Details
First Author: O.Richoz SWITZERLAND
Co Author(s): D. Tabibian P. Aleksandra A. Konstantinos M. El Wardani K. Hashemi G. Kymionis
Abstract Details
Purpose:
To improve the refractive accuracy post cataract surgery using the SRK-T calculation and a Deep Neural Network Classifier using the Tensorflow library.
Setting:
Jules Gonin Eye hospital
University of Lausanne
Methods:
122 eyes had an uncomplicated cataract surgery at the Jules Gonin Eye hospital all received the same IOL brand (AMO Technis ZCB00), all the capsulorhexis size were between 4.5-6.0 mm of diameter. A deep neuronal network was trained using 2/3 of the eyes as a training set and 1/3 of the eyes as a testing set. 11 different variables feed the neuronal network (age ,AL ,K1 ,K2…) and 3 different results were expected representing the mismatch between the 6 weeks post refractive values and SRK-T calculated results (between -0.25 +0.25, more than -0.25 and more than +0.25.
Results:
By using 50’000 learning steps with a three hidden layer unit of respectively 100, 50 and 25 nodes, the deep neuronal network shows an accuracy of up to 77.8%.
Conclusions:
Deep Neural Network technology freely accessible with Tensorflow offers a great opportunity for improving the refractive outcome of cataract surgery. More data are needed for improving the accuracy of the algorithm.
Financial Disclosure:
None