Assessment of tumour angiogenesis in tissue section images based on a self-organising map (SOM)
Constantinos Loukas · Computer Methods in Biomechanics and Biomedical Engineering Imaging & Visualization · 2013
The majority of computational approaches encountered in histological image analysis focus on classification of biological structures visualised under high magnification. In addition to histological grading, image analysis has contributed in the assessment of angiogenesis, a crucial factor in the radiation response of solid tumours. Evaluation of neovascularisation is usually performed qualitatively via manual examination of the tissue sample with a standard transillumination microscope. In this paper, we present an alternative methodology for the unsupervised segmentation and analysis of microvessels in images acquired under low magnification. We have employed an unsupervised learning technique based on a self-organising map (SOM) that considers only the pixels' colour content as input. Various colour transformations and feature vectors have been examined. The best configuration model yielded a minimum recognition accuracy of 86% in pixel classification experiments, and 88% in microvessel counts after comparison with an expert histopathologist. The proposed technique provides several benefits such as speed of analysis and extraction of important shape characteristics for the detected microvessels.