Classification of Corneal Nerve Images Using Machine Learning Techniques
Keywords:
Machine learning, image classification, corneal confocal microscopy, medical image analysis, neuropathyAbstract
Recent research shows that small nerve fiber damage is an early detector of neuropathy. These small nerve fibers are present in the human cornea and can be visualized through the use of a corneal confocal microscope. A series of images can be acquired from the subbasal nerve plexus of the cornea. Before the images can be quantified for nerve loss, a human expert manually traces the nerves in the image and then classifies the image as having neuropathy or not. Some nerve tracing algorithms are available in the literature, but none of them are reported as being used in clinical practice. An alternate practice is to visually classify the image for neuropathy without quantification. In this paper, we evaluate the potential of various machine learning techniques for automating corneal nerve image classification. First, the images are down-sampled using discrete wavelet transform, filtering and a number of morphological operations. The resulting binary image is used for extracting characteristic features of the image. This is followed by training the classifier on the extracted features. The trained classifier is then used for predicting the state of the nerves in the images. Our experiments yield a classification accuracy of 0.91 reflecting the effectiveness of the proposed method.
Downloads
Downloads
Published
Issue
Section
License
Open access licenses
Open Access is by licensing the content with a Creative Commons (CC) license.
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.