Statistical Indicators for the Analysis of Digitalized Brain Tumor Images

• Main Authors: Matthias Templ, Semaguel Aklan, Peter Filzmoser, Matthias Preusser, Johannes A. Hainfellner
• Format: Article
• Language: English
• Published: Austrian Statistical Society 2016-02-01
• Series: Austrian Journal of Statistics
• Online Access: http://www.ajs.or.at/index.php/ajs/article/view/156

In this contribution, indicators for computer-based analysis and assessment of tumor cell proliferation in human brain tumors are developed. The methods are applied on (two) samples of digitized human brain tumor tissue sections immunostained with an antibody against the Ki67 epitope. The Ki67 immunostaining highlights cells undergoing cell division and is thus a surrogate marker for tumor growth. The challenges are related to the enormous size of the images (“big data”) analyzed, some of them are larger than 100 GB. Thus, efficient methods to extract relevant information have to be applied. Before starting with the statistical analysis, the digitized images are preprocessed to extractthe highlighted cells. Then the distribution of Ki67 immunostaining patterns is analyzed. Starting with a bivariate kernel density estimation, the proposed indicators are used to evaluate and compare the resulting density estimates. Moreover, the spatial distribution of clusters of Ki67-labeled tumor cells is of particular interest. The results allow to evaluate and compare images or sectors of the images. This evaluation and comparisons of samples or sectors could turn out to be useful in practice since it allows for a pre-selection of interesting sectors and samples. Thus, the time-consuming part of manual inspection of the huge images could be reduced.