In order to facilitate medical research on diabetic neuropathy, we successfully implemented an algorithm such that can automatically trace out nerves from corneal images.
Substituting our algorithm into the program developed by our partner team, which is also invited by our sponsor and one of our teammates belongs to, we integrated a program that can batch a whole folder of images in a same time, refine traced trajectories, label different nerves, calculate and document the length of each nerves. However, since these functions were not derived from our efforts, we decided not to elaborate them here.
For further study, instead of merely constructing a traced image, we will develop a structure for nerve information, storing every traced points in one nerve in sequence and serving as another type of outputs. In this way, the strength of our exploratory algorithm can be maintained and be fully exploit in further studies. For example, with this structure, we will study the “curvature” base on the direction change between adjacent segments. Similarly, branching will be studied base on this structure. Since the width of the nerves in these corneal images was around 3 pixels, it was too thin to reveal significant change through the development of diabetes. As a result, width was cast out of the desired parameters.
After the functions in order to extract neural parameters from images are fully established, we will compare these automatically extracted parameters with those manually obtained, either by manual tracing or observation. This comparison will give us a final validation on the program and reveal its applicability to medical research substituting manual labor.