Hierarchical models for epidermal nerve fiber data

• Published in: Statistics in medicine Vol. 37; no. 3; pp. 357 - 374
• Main Authors: Andersson, Claes, Rajala, Tuomas, Särkkä, Aila
• Format: Journal Article
• Language: English
• Published: England Wiley Subscription Services, Inc 10.02.2018
• Subjects: adjusted Rand index; algorithm; blister; classification; Diabetes; Diabetic neuropathy; disease severity; EM-algorithm; epidermal nerve fiber; hierarchical model; human; Medical statistics; nerve fiber; nerve tree; Neurology; Probability Theory and Statistics; Rand index; Sannolikhetsteori och statistik; skin biopsy; skin nerve; statistical model; unsupervised classification; nerve tree; adjusted Rand index; diabetic neuropathy; unsupervised classification; EM-algorithm
• ISSN: 0277-6715; 1097-0258; 1097-0258
• DOI: 10.1002/sim.7516

While epidermal nerve fiber (ENF) data have been used to study the effects of small fiber neuropathies through the density and the spatial patterns of the ENFs, little research has been focused on the effects on the individual nerve fibers. Studying the individual nerve fibers might give a better understanding of the effects of the neuropathy on the growth process of the individual ENFs. In this study, data from 32 healthy volunteers and 20 diabetic subjects, obtained from suction induced skin blister biopsies, are analyzed by comparing statistics for the nerve fibers as a whole and for the segments that a nerve fiber is composed of. Moreover, it is evaluated whether this type of data can be used to detect diabetic neuropathy, by using hierarchical models to perform unsupervised classification of the subjects. It is found that using the information about the individual nerve fibers in combination with the ENF counts yields a considerable improvement as compared to using the ENF counts only.