| dc.description.abstract | Abstract
The purpose of this thesis is the investigation of the intratumoural network through image analysis of histological sections. Tumour vasculature is characterized by complexity, irregularities and poorly regulated growth. Fractal analysis has been used to establish that tumour vasculature has a different network architecture from that of the normal arterio-venous system or the capillary network. The
vasculature is responsible for the transportation of oxygen to tumour cells, however its many pathological features results in, among others, the presence of hypoxic regions. Hypoxia is a challenge
to the treatment of cancer, both through its indirect biological effects, such as a reduced progression through the cell cycle, but also through direct chemical effects. In particular, the oxygen effect reduces the effectiveness of radiation therapy. Furthermore, the network morphology relates to many other parameters as well, such as the angiogenic and the metastatic capability of the cancer. This raises the possibility of using image analysis, and fractal analysis in particular, to quantify different aspects of the network morphology.
The study limits itself to parameters which may be obtained from digitized images of histological sections with endothelial-specific staining. The investigated parameters are primarily obtained through fractal analysis and syntactic structure analysis. A few more parameters, such as the number of vessels, the size of the vessels, the total vascular area, and cumulative histograms of distances to
the nearest vessel, were obtained directly from the images. The investigated parameters depend on both the number of vessels in the image, and the distribution of the vessels. Two particular areas
have been emphasized, the first is the identification of how strongly the parameters relate to the vessel distribution, and the second is the implementation of fractal analysis on vascular cross sections.
Four different CD34-stained immunohistological sections have been analysed. They were obtained from malignant carcinomas of the breast and exhibited qualitatively different vascular patterns. A routine has been developed to segment out the vessels from the background staining before the image analysis.
The investigated fractal dimensions include the Box Counting dimension, the Sandbox dimension, the Correlation dimension, the Mass dimension and the Fourier dimension. These have been applied to images processed in three different ways. The first contained the entire vessel lumens, the second only the outer vessel wall perimeter and the last only the vessels’ geometric centre of mass. In addition fractal analysis has been performed on Gabriels’s Graph and the Euclidean Minimum Spanning Tree, both of which belong to the Syntactic Structure Analysis graphs. The different methods and images provided both different dimensions and different curve shapes. Some of the curves did not have any meaningful power-law scaling regions at all, however, most of them did. The Sandbox dimension in general and the mass centre images in particular, have been considered the most promising of these methods. Although it may be argued that the term dimension does not, in any meaningful way, relate to most of the parameters obtained through these methods, they do most certainly appear capable of differentiating various vessel distributions from each other. In addition to the fractal analysis methods, all other investigated methods have been applied to the four cases as well.
In order to identify the relationship between the parameters and the number of vessels in a particular image, two simulations have been performed. The first simulation generated the images through a uniform random distribution probability (10.000 images), while the second used a three-dimensional invasion percolation cluster to generate the vessel positions (15.560 images). The mean and standard deviations of the results at each vessel count have been investigated. Large standard deviations have been interpreted as a strong dependency on the vessel distribution. The slope of the mean, on the other hand, shows the parameters dependency on the number of vessels. The sizes of the standard deviations are considered relative to the slopes. Most of the analysis parameters showed large variations for low vascular densities. A subset of the parameters had large variations even at very high vascular densities.
In conclusion, most of the investigated parameters appear to be promising candidates for further studies. Fractal analysis may be applied to vascular cross-sections. It is, however, important to rigorously specify how the analysis is performed, as a large number of possible results may be acquired through these methods. In particular the Sandbox dimensions of the mass centre images, Gabriel’s Graph, and the Euclidean Minimum Spanning Tree, at large sandbox diameters, are recommended for further study, with the possible additon of the EMST dimension at small diameters, as this require no extra computation time. At this point in time it is not recommended to exclude any of the SSA-parameters from further studies. The next adviceable step would be to perform a correlation study, comparing these parameters to other data of clinical value, related to treatment, diagnosis or prognosis. | nor |