Keratoconus Screening Based on Deep Learning Approach of Corneal Topography

Bo-I Kuo; Wen-Yi Chang; Tai-Shan Liao; Fang-Yu Liu; Hsin-Yu Liu; Hsiao-Sang Chu; Wei-Li Chen; Fung-Rong Hu; Jia-Yush Yen; I-Jong Wang

doi:10.1167/tvst.9.2.53

Keratoconus Screening Based on Deep Learning Approach of Corneal Topography

Transl Vis Sci Technol. 2020 Sep 25;9(2):53. doi: 10.1167/tvst.9.2.53. eCollection 2020 Sep.

Authors

Bo-I Kuo^{1

2}, Wen-Yi Chang³, Tai-Shan Liao⁴, Fang-Yu Liu^{1

5}, Hsin-Yu Liu¹, Hsiao-Sang Chu¹, Wei-Li Chen^{1

6}, Fung-Rong Hu¹, Jia-Yush Yen⁷, I-Jong Wang^{1

6}

Affiliations

¹ Department of Ophthalmology, National Taiwan University Hospital, Taipei, Taiwan.
² Department of Ophthalmology, Taipei City Hospital, Renai branch, Taipei, Taiwan.
³ National Center for High-Performance Computing, National Applied Research Laboratories, Hsinchu, Taiwan.
⁴ Taiwan Instrument Research Institute, National Applied Research Laboratories, Hsinchu, Taiwan.
⁵ Department of Ophthalmology, Tri-Service General Hospital, National Defense Medical Center, Taipei, Taiwan.
⁶ Graduate Institute of Clinical Medicine, College of Medicine, National Taiwan University, Taipei, Taiwan.
⁷ Department of Mechanical Engineering, National Taiwan University, Taipei, Taiwan.

Abstract

Purpose: To develop and compare deep learning (DL) algorithms to detect keratoconus on the basis of corneal topography and validate with visualization methods.

Methods: We retrospectively collected corneal topographies of the study group with clinically manifested keratoconus and the control group with regular astigmatism. All images were divided into training and test datasets. We adopted three convolutional neural network (CNN) models for learning. The test dataset was applied to analyze the performance of the three models. In addition, for better discrimination and understanding, we displayed the pixel-wise discriminative features and class-discriminative heat map of diopter images for visualization.

Results: Overall, 170 keratoconus, 28 subclinical keratoconus and 156 normal topographic pictures were collected. The convergence of accuracy and loss for the training and test datasets after training revealed no overfitting in all three CNN models. The sensitivity and specificity of all CNN models were over 0.90, and the area under the receiver operating characteristic curve reached 0.995 in the ResNet152 model. The pixel-wise discriminative features and the heat map of the prediction layer in the VGG16 model both revealed it focused on the largest gradient difference of topographic maps, which was corresponding to the diagnostic clues of ophthalmologists. The subclinical keratoconus was positively predicted with our model and also correlated with topographic indexes.

Conclusions: The DL models had fair accuracy for keratoconus screening based on corneal topographic images. The visualization mentioned in the current study revealed that the model focused on the appropriate region for diagnosis and rendered clinical explainability of deep learning more acceptable.

Translational relevance: These high accuracy CNN models can aid ophthalmologists in keratoconus screening with color-coded corneal topography maps.

Keywords: convolutional neuronal network; corneal topography; deep learning; keratoconus.

Publication types

Research Support, Non-U.S. Gov't

MeSH terms

Cornea / diagnostic imaging
Corneal Topography
Deep Learning*
Humans
Keratoconus* / diagnosis
Retrospective Studies