Nerve fiber tracing in bright-field images of human skin using deep learning

Abstract

The goal of this thesis is to find an automated method that can trace all nerve fibers in bright-field images of skin tissue. This is an important step towards the automated quantification of intra-epidermal nerve fiber density, an important biomarker in the diagnosis of small-fiber neuropathy.
Deep learning is a popular new field of research in computer vision. In recent years it has been successfully applied to a lot of computer vision problems. Here we try to use deep learning for nerve fiber segmentation.

It will be shown how a convolutional neural network can be implemented and trained to produce nerve fiber segmentation maps. This involves the optimization of many layers of computations (summing up to tens of millions of parameters), which we did using a modern machine learning toolkit. Statistical analysis of the obtained results show that the neural network has a performance that is comparable to a human control, and out-competes an earlier method that was developed using conventional image analysis tools, by a big margin. A number of improvements are proposed to further increase the neural network performance.