Free Papers
Screening refractive surgery candidates by corneal tomography-based deep learning
Free Paper Details
First Author: Y.Xie CHINA
Co Author(s): L. Zhao Q. Liu H. Lin
Abstract Details
Purpose:
To improve the accuracy and efficiency of screening for refractive surgery candidates using an artificial intelligence (AI) model.
Setting:
The corneal tomography data were collected from a Pentacam HR (OCULUS, Wetzlar, Germany) at the Zhongshan Ophthalmic Center, China, with examination dates extending from July 2016 to March 2019. The investigation was undertaken from July 2018 to June 2019.
Methods:
The images were classified as showing normal corneas (1887 images), suspected irregular corneas (799 images), early-stage keratoconus (731 images), keratoconus (1978 images), or myopic postoperative corneas (1070 images). We used a deep learning algorithm, InceptionResNetV2, to build our AI model to identify at-risk corneas. An independent test dataset of 100 images from 94 patients was used to compare the accuracy of our model with that of human specialists as well as other classifiers in the Pentacam HR system.
Results:
The model achieved an overall detection accuracy of 94.7% (95% CI: 93.3%-95.8%) on the validation dataset. Moreover, on the independent test dataset, our model achieved an overall detection accuracy of 95% (95%CI: 88.8%-97.8%) comparable to that of senior ophthalmologists in our clinic who are refractive surgeons (92.8% (95%CI: 91.2%-94.4%); P=0.72). In distinguishing corneas with contraindications for refractive surgery, our AI model performed better than the classifiers (95% VS 81%, P<0.001) in the Pentacam HR system on an Asian patient database.
Conclusions:
Our corneal tomography-based deep learning system effectively classified images that provided corneal information and identified at-risk corneas. This system could provide guidance to refractive surgeons in screening candidates for refractive surgery and has potential for generalized clinical application.
Financial Disclosure:
-
