Spatial analysis and modeling of nerve fiber patterns
Doctoral thesis, 2018
The ENF samples are mainly analyzed as point patterns, where the points are the locations where nerve fibers enter the epidermis or terminate. The analysis is partly based on existing summary statistics for point patterns, but we also propose a new summary statistic to quantify the proportion of the skin covered by the nerve fibers. Two cluster processes are introduced as models for the patterns consisting only of the locations where the nerve fibers enter the epidermis. For one of the models, a Bayesian hierarchical method for parameter estimation is proposed. A model for the end points is also presented, and non-spatial models for individual nerve fibers, which are used to perform unsupervised classification of the subjects.
From the results we find that while all patterns are aggregated, the level of aggregation tends to increase with increased severity of the neuropathy. The results from the modeling indicate that the increased aggregation is caused by a decrease in the number of clusters, while the structure within clusters appears to be similar in all disease groups. The results from the non-spatial analysis indicate that the nerve fibers from healthy subjects tend to extend further than those from subjects with diabetic neuropathy.
The use of methods and models developed in this thesis is not limited to ENF data, but can be applied to point pattern data in general. In particular, the models for the base point patterns and the methods for estimating the parameters of these models are contributions to the point process literature.
hierarchical models
Bayesian estimation
epidermal nerve fibers
cluster processes
spatial point processes
diabetic neuropathy
Author
Claes Andersson
Chalmers, Mathematical Sciences, Applied Mathematics and Statistics
Discovering early diabetic neuropathy from epidermal nerve fiber patterns
Statistics in Medicine,;Vol. 35(2016)p. 4427-4442
Journal article
A Bayesian hierarchical point process model for epidermal nerve fiber patterns
Mathematical Biosciences,;Vol. 313(2019)p. 48-60
Journal article
Andersson, C., Mrkvicka, T. Inference for cluster point processes with over- or under-dispersed cluster sizes
Hierarchical models for epidermal nerve fiber data
Statistics in Medicine,;Vol. 37(2018)p. 357-374
Journal article
I större delen av arbetet analyserar vi datan som punktmönster, där punkterna är positionerna där en nervtråd börjar eller slutar. Det finns en stor uppsättning statistiska verktyg för att beskriva olika aspekter av punktmönster, och med hjälp av dessa kan vi beskriva vissa skillnader mellan nervmönster från friska och sjuka patienter. Vi introducerar också vissa nya verktyg som beskriver
hur stor del av huden som täcks av nerverna. Dessutom utvecklar vi modeller för punktmönstren, vilka kan ge en bättre förståelse för hur nervåterväxten förändras av sjukdomen. De metoder och modeller vi utvecklar är inte specifika för nervfiberdata, utan kan användas till punktmönster generellt. De nya modellerna och metoderna för att skatta parametrarna till dessa är viktiga bidrag till litteraturen på området.
Subject Categories
Other Computer and Information Science
Bioinformatics (Computational Biology)
Probability Theory and Statistics
ISBN
978-91-7597-735-5
Doktorsavhandlingar vid Chalmers tekniska högskola. Ny serie: 4416
Publisher
Chalmers
Pascal, Chalmers Tvärgata 3
Opponent: Chief research scientist Thordis L. Thorarinsdottir, Norwegian Computing Center, Norway.