A scale-space tracing algorithm for analysis of tumour blood vessel morphology from transmitted light optical images

Background Limited contrast in optical images is problematic for analysing tumour vascular morphology. We describe an algorithm for segmenting tumour vasculature in transmitted light optical images, without the need for contrast enhancement. Effects of angiopoietin-1 (Ang-1) and -2 (Ang-2) ± treatment with the vascular disrupting agent combretastatin A4P (CA4P) were investigated. Method SW1222 human colorectal carcinoma cells were transfected with Ang-1 or Ang-2 cDNA, or with empty vector and implanted into ‘window’ chamber-bearing mice. Transmitted light images of tumours (x10 objective) were acquired before and up to 24h after treatment with 30 mg/kg CA4P or saline. Vessel tracing used a scale-space approach, employing differences in intensity of transmitted light between the vessels and surrounding tissues, as well as differences in the chromatic components, hue and saturation. The centreline of vessels was detected as a “ridge” in a topographical analogy and successive levels of filtering provided different scales to detect sharp to diffuse ridges. Morphological parameters were measured from the traced images - average vessel length (AL) width (AW), and relative vascular area (RA). Results The algorithm successfully identified the majority of tumour microvessels. CA4P-treated tumours showed a decrease in RA and increase in saturation balance up to 1-3h, with recovery by 24h. Saline had no effect. Ang-2 over-expressing tumours had lower values of AL, AW and RA than Ang-1 and wild-type (WT) tumours. Ang-1 tumours were similar to the WT except that AL was longer.