Introduction

Our task was to design a program that can automatically analyze images of corneal nerve layer taken from a corneal confocal microscope. Parameters such as length, width, branching frequency and branching angles can be extracted through our program. And such parameters will be important to analyze the status of the corneal nerve system and its relationship with diabetic neuropathy.

Diabetic neuropathy is one of the common complications associate with diabetes mellitus which mostly affect peripheral nerves. Though it plays an important role in diabetic medicine, it is difficult to diagnose neuropathy or quantify its development. One common way to study neuropathy is using esthesiometer, but it is rather subjective and quantitative. Skin biopsy is a definitive method in neuropathic diagnosis, but it is invasive, inconvenient, and uncomfortable. Therefore people have attempted to find a better way of diagnosis that is convenient and non-invasive to evaluate and quantify nerves in order to assess and quantify systemic peripheral nerve health.

One hypothesis to achieve this is to analyze the image of nerves of the corneal nerve layer. The images are easy to acquire using the corneal confocal microscope. However, the correlation between the corneal neuromorphology and the development of diabetic neuropathy remains unknown. Furthermore, diagnosing neuropathy based on neuromophology requires deep knowledge and experience in this area, and it’s still subjective even for experts. Therefore, medical researchers were driven to develop an objective approach to diagnose diabetic neuropathy based on corneal neuromophology.

In order to do so, a large amount of corneal nerve images need to be examined, so that the correlation between morphology and neuropathy can be revealed. This process would be extremely tedious and time consuming if the images were to examine manually one after another. As a result, our team were invited to develop an algorithm to rescue researches from this heavy labor.