An image-based model of the interstitial cells of cajal in the gastrointestinal tract

Abstract

Gastrointestinal (GI) motility refers to the movements of the walls of the GI tract, resulting in the movement of digesta. GI motility is mediated by a complex integration of neural, hormonal and myogenic mechanisms. Interstitial cells of Cajal (ICC) play a key role in generating and mediating a rhythmic bioelectrical event, known as slow waves to govern gastric motility. Regional variations in the structure and distributions of ICC networks in the stomach affect slow wave activity in the proximal and distal regions, for example the corpusantrum transition region. Thus, it is important to study the variations in ICC network characteristics at a large spatial scale by combining experimental knowledge with imaging and modelling techniques to relate the effects of ICC networks on slow wave activity and gastric functions; particularly in the corpus-antrum transition region. Current imaging techniques, such as immunofluorescence microscopy, are limited to small field of view (~200x200 μm2 to 400x400 μm2) and are not large enough to study microstructural changes in cell networks over a large spatial scale, for example 4x4 mm2. In this research, an image-based approach was employed to study ICC variations from a series of immunofluorescence images taken from the murine stomach, coupled with a number of mathematical models in an attempt to reproduce the ICC network characteristics over an extended spatial scale. The first aim of the research was to apply image processing and feature extraction techniques on tissue biopsies of size 200x200 μm2 to obtain key characteristics of ICC networks. The images were of proliferating ICC from a murine stomach obtained via immunofluorescence dual-staining. Four images were processed in total – two images were from ICC-IM layer consisting of two muscle networks propagating in adjacent directions, one image was from ICC-IM single muscle layer and the last image was from ICC-MY muscle layer. ICC nuclei and cell-bodies were segmented from each tissue image by separating the colour channel associated with each structure. Centroids of red clusters were collocated with corresponding cell-bodies to isolate ICC nuclei in plane. ICC nuclei counts were obtained for ICC-IM dual-muscle layers (20, 13), ICC-IM single muscle layer (10) and ICC-MY (13). The Hough transform was applied to the cell-bodies sub-images to obtain the orientation of the ICC networks. In the ICC-IM dual-muscle images, the Hough angle orientations (reported with respect to x-axis and origin at top left of the image) confirmed that the dual-muscle layers were aligned at 90±10° to each other (Image 1 = -75±5° and 20±5°, Image 2 = -30±5° and 70±5°) corresponding to longitudinal and circular muscle layers. ICC-IM single layer had major Hough lines aligned in one direction (-15±5°). ICC-MY image had a web-like structure with Hough lines at three major angles (-80±5°, -30±5° and 30±5°). The Hough angle orientations obtained for each tissue image confirmed that ICC networks lie at varying orientations in different muscle layers. Other characteristics relating to lengths, widths, centroid locations and neighbourhood connectivity of cell-bodies were also obtained confirming spatial and density variations. The second aim of this research was to create a one-dimensional (1D) reconstructed model of the ICC network which incorporated the key characteristics obtained in the image processing step. A longitudinal network model (consisting of ICC-IM single muscle layer) and a grid network (consisting of overlapping ICC-IM networks) were created. Each model consisted of the same resolution and spatial density of nuclei and cell-bodies as determined from the original tissue images. The network characteristics of the image-based models were validated through the same image processing and features extraction methods applied to the original tissue images. In the longitudinal models (0.4x0.4 mm2), the cell-bodies lengths were in a range of 23 to 113 μm, nuclei count was 80 and cell-bodies orientations were -15° (±5°). In the grid network, the cell-bodies were in range of 23 to 113 μm, nuclei count was 140 and cell-bodies orientations were between -75° and 20° (±5°). All the characteristics from the validated network model fall within the range of the characteristics obtained from the tissue images. Finally, the image-based models were extended to encapsulate ICC variations across a larger spatial scale than the immunofluorescence image field. The longitudinal and grid models were extended to 4x4 mm2, which was 400 times larger than the ICC network in the original tissue images (200x200 μm2). The extended models were created by calculating a nuclei distribution factor, based on the extended image size and field, to determine the total nuclei in the network. The extended longitudinal model consisted of 8,000 nuclei representing a single-muscle layer whereas the grid model consisted of 8,000 nuclei in each of the dualmuscle layers. Each model introduced a variation where the network was dense in one region of the image and became sparse towards the opposite region to represent regional variations in ICC networks across a larger area in the stomach. In conclusion, the work presented in this thesis has provided a framework for efficiently extracting key characteristics relating to ICC count, and network orientations from histological images. In addition, a series of network models have been developed to reproduce the ICC networks characteristics obtained from the images to an extended spatial scale. The work presented in this thesis can be applied to study the regional variations in healthy versus depleted ICC networks which have been associated with changes in slow wave propagation and gastric functions, particularly in the corpus-antrum transition region.