Exponential filters to extract small structures in noisy images

Segmentation of small structures in noisy images is inherently challenging because the edge information is contained in the same high frequency component as the noise. We have overcome this obstacle in the analysis of the sural nerve in the ankle by processing images to reduce noise and detecting edges with a detector which has reduced sensitivity to noise. Effective segmentation delineates the nerve boundary without breaking the nerve structure into sub-regions and was evaluated in two ways. Firstly, by visual comparison with specific anatomy in each of 40 subjects, comprising a partial population from a nerve hydration study. Secondly, by quantitative comparison with nerve hydration measurements obtained by previously described manual methods. The measurements obtained from this semi-automated approach show close correlation with those obtained manually