This is python code for implementation in the paper:
Nicole Hallett, Kai Yi, Josef Dick, Christopher Hodge, Gerard Sutton, Yu Guang Wang, Jingjing You. Deep Learning Based Unsupervised and Semi-supervised Classification for Keratoconus. arXiv preprint arXiv:2001.11653, 2020.
The transparent cornea is the window of the eye, facilitating the entry of light rays and controlling focusing the movement of the light within the eye. The cornea is critical, contributing to 75% of the refractive power of the eye. Keratoconus is a progressive and multifactorial corneal degenerative disease affecting 1 in 2000 individuals worldwide. Currently, there is no cure for keratoconus other than corneal transplantation for advanced stage keratoconus or corneal cross-linking, which can only halt KC progression. The ability to accurately identify subtle KC or KC progression is of vital clinical significance. To date, there has been little consensus on a useful model to classify KC patients, which therefore inhibits the ability to predict disease progression accurately.
In this paper, we utilised machine learning to analyse data from 124 KC patients, including topographical and clinical variables. Both supervised multilayer perceptron and unsupervised variational autoencoder models were used to classify KC patients with reference to the existing Amsler-Krumeich (A-K) classification system. Both methods result in high accuracy, with the unsupervised method showing better performance. The result showed that the unsupervised method with a selection of 29 variables could be a powerful tool to provide an automatic classification tool for clinicians. These outcomes provide a platform for additional analysis for the progression and treatment of keratoconus.
For training, one can run
python train.py
There are some parameters for the train program.
TRAIN_DIR - the path training data
model_checkpoint_dir - model weights save directory
epoch - how many epochs in training
For cluster, one can run
python cluster.py
There are some parameters for the cluster program.
normalize_constant - the normalize_constant, default = 50
weight - the pretrained weight file name, if no pretrained weight, then ignore
epoch - how many epochs in training
For example, you can run
python cluster.py --weight '2weights.09-2.05.h5' \
--normalize_constant 40 \
--epoch 100
Please cite our paper if you use this code in your own work:
@article{hallett2020deep,
title={Deep Learning Based Unsupervised and Semi-supervised Classification for Keratoconus},
author={Hallett, Nicole and Yi, Kai and Dick, Josef and Hodge, Christopher and Sutton, Gerard and Wang, Yu Guang and You, Jingjing},
journal={arXiv preprint arXiv:2001.11653},
year={2020}
}