BACR5

A segmentation algorithm for quantitative morphological analysis of microvessels in immunostained histological tumour sections

Constantino Reyes-Aldasoro, Leigh Willams, Chryso Kanthou, Gillian Tozer

University of Sheffield, UK

Background
Immunostaining of fixed tumour sections for CD31 (PECAM-1) is a common technique for identifying endothelial cells (EC) in pre-clinical and clinical studies of the effects of novel vascular-targeted treatments. However, image analysis is often restricted to simple manual vascular counts due to limitations of available segmentation techniques for more detailed semi-automated analysis.

Aim
The aims of this work were two-fold: ﬁrst, to segment ECs cells from background in CD-31-immunostained tumour sections (peroxidase-activated 3,3’-diaminobenzidine (DAB)) for measurements of stained area, blood vessel eccentricity and ‘wiggliness’ and vessel wall thickness. Second, to investigate whether the vascular disrupting agent combretastatin-A-4 3-O-phosphate (CA-4-P, 100 mg/kg), induced early morphological changes in tumour microvessels that could contribute to vascular shut-down.

Method and Results

Eight SW1222 human colon carcinoma xenografts were grown sub-cutaneously in immuno-deprived (SCID) mice; 4 tumours were treated with saline and 4 with CA-4-P and excised 30 minutes later. Tumours were zinc-fixed and sections immunostained for CD31 using DAB (brown), with haematoxylin counterstaining (blue). High power images from across whole tumour sections were acquired. The algorithm exploited the distinctive hues of stained vascular endothelial cells, cell nuclei and background. Pre-processing corrected the shading and thresholded in the 3D Hue, Saturation, Value colour model, which provided the seeds for a region-growing algorithm that segmented the endothelial cells from the background and nuclei. Three morphological tasks were performed on the segmented objects: joining objects that should have been a single object, closing objects that had a narrow gap around their periphery, and splitting objects that had two or more inner holes. The algorithm was tested on 149 images, from which 20,003 vessels were segmented: 10,623 from tumours treated with CA-4-P (79 images) and 9,380 from saline-treated (70 images). Statistical difference was observed in 3 measurements: vessel eccentricity, wall thickness and wiggliness. CA-4P treatment resulted in rounder vessels with thicker walls in the tumour periphery and less wiggly contours in large vessels. Large treated vessels had higher eccentricity than controls.

Conclusion
These results suggest that the effects of CA-4-P vary in different tumour regions. The algorithm has general application to other studies involving investigations of tumour vascular-targeted agents.

Acknowledgments

This work was funded by Cancer Research UK