Analysis of corneal images for the recognition of nerve structures

The recognition of nerve structures in the cornea appears to be an important clinical issue, e.g. to investigate about damages from surgical interventions (LASIK/PRK) or severity of diabetic neuropathy. We addressed the problem of recognizing and tracing corneal nerves in confocal microscopy images. After image luminosity and contrast are normalized, the nerve tracking procedure is run. After nerve segments are recognized, a post-processing procedure removes false recognitions and links sparse segments. A prototype of the algorithm was implemented in the Matlab language and run on a personal computer. An evaluation was performed on a data set containing 12 images. The percent of tracked nerves (with respect to manual tracing) was on average 87%; the rate of false positive recognitions (with respect to total tracing) was on average 9%. Running times of the prototype were 4-5 minutes per image.