Segmentation of tiny objects in very poor-quality angiogenesis images

Abstract

This paper deals with a straightforward and effective solution that isolates tiny objects from very poor-quality angiogenesis images. The objects of interest consist of the cross-section of blood vessels present in histological cuts of malign tumors that grow in soft parts of the human body through a natural process known as angiogenesis. The proposed strategy applies a conditional morphological closing operator using a structuring element based on criteria resulting from local statistical properties. This approach gives in all cases a lower percent of false target count (FTC) and false non-target count (FNTC) errors, with respect to the error equally calculated for two other strategies discussed briefly in this paper, when the results are compared with images segmented manually by pathologists.