Eur. Phys. J. Plus (2020) 135: 266
https://doi.org/10.1140/epjp/s13360-020-00127-y

Regular Article

Automatic corneal nerve fiber segmentation and geometric biomarker quantification

¹ Department of Mathematics and Computer Science, Eindhoven University of Technology, Eindhoven, The Netherlands
² Department of Biomedical Engineering, Eindhoven University of Technology, Eindhoven, The Netherlands
³ Laboratory of Neuro Imaging (LONI), USC Stevens Neuroimaging and Informatics Institute, Keck School of Medicine of University of Southern California, Los Angeles, USA
⁴ University Eye Clinic Maastricht, Maastricht, The Netherlands
⁵ Key Laboratory of Biomedical Information Engineering of Ministry of Education, School of Life Science and Technology, Xi’an Jiaotong University, Xi’an, China
⁶ Department of Surgery, Netherlands Cancer Institute, Amsterdam, The Netherlands
⁷ Department of Electrical Engineering, Eindhoven University of Technology, Eindhoven, The Netherlands

^* e-mail: t.tan1@tue.nl

Received: 11 June 2019
Accepted: 2 December 2019
Published online: 20 February 2020

Abstract

Geometric and topological features of corneal nerve fibers in confocal microscopy images are important indicators for the diagnosis of common diseases such as diabetic neuropathy. Quantitative analysis of these important biomarkers requires an accurate segmentation of the nerve fiber network. Currently, most of the analysis are performed based on manual annotations of the nerve fiber segments, while a fully automatic corneal nerve fiber extraction and analysis framework is still needed. In this paper, we establish a fully convolutional network method to precisely enhance and segment corneal nerve fibers in microscopy images. Based on the segmentation results, automatic tortuosity measurement and branching detection modules are established to extract valuable geometric and topological biomarkers. The proposed segmentation method is validated on a dataset with 142 images. The experimental results show that our deep learning-based framework outperforms state-of-the-art segmentation approaches. The biomarker extraction methods are validated on two different datasets, demonstrating high effectiveness and reliability of the proposed methods.