An Approach of Combining Convolution Neural Network and Graph Convolution Network to Predict the Progression of Myopia

To develop an approach of combining convolution neural network and graph convolution network to predict the progression of myopia. The working distance (WD) and light intensity (LI) of three hundred and seventeen children were recorded by Clouclip. The spherical equivalent refraction (SER) of the children were recorded by ophthalmologists. The data of WD and LI were filtered and mapped into a two-dimensional WD-LI space. The percentage of time (PoT) falling into each pixel in the space was calculated for each subject. The space of each subject can be thought of as an image and it is the input of our neural network model that combining several convolution layers and graph convolution layers. The output of the model is the SER. With tenfold cross validation, the validation error is 0.79 D when the L1 loss function is used. This study provides an innovative way to predict the development of myopia by WD and LI. The convolution neural network and graph convolution network are used to predict the myopia with WD and LI simultaneously, which has not been done before.