Multidimensional Characterisation of Time-dependent Image Data: A Case Study for the Peripheral Nervous System in Ageing Mice

Abstract

Segmentation of µm-resolution image data of irregularly shaped objects poses challenges to existing segmentation algorithms. This is especially true, when imperfections like noise, uneven lightning or traces of sample preparation are present in the image data. In this paper, considering electron micrographs of femoral quadriceps nerve sections of mice, a segmentation method to extract single axons surrounded by myelin sheaths is developed which is able to cope with various imperfections and artefacts. This approach successfully combines established methods like local thresholding and marker-based watershed transform to achieve a reliable segmentation of the given data. Indeed, the resulting segmentation map can be used to quantitatively determine geometrical characteristics of the axons and myelin sheaths. This is exemplified by modelling the joint probability distribution of axon area and myelin sphericity using a parametric copula approach, and by analysing the evolution of the model parameters for image data obtained from mice of different ages.

Keywords

• automated segmentation
• copula
• electron microscopy
• parametric modelling
• quadriceps nerve
• statistical image analysis