Whole slide quantification of stromal lymphatic vessel distribution and peritumoral lymphatic vessel density in early invasive cervical cancer: a method description

Abstract

Peritumoral Lymphatic Vessel Density (LVD) is considered to be a predictive marker for the presence of lymph node metastases in cervical cancer. However, when LVD quantification relies on conventional optical microscopy and the hot spot technique, interobserver variability is significant and yields inconsistent conclusions. In this work, we describe an original method that applies computed image analysis to whole slide scanned tissue sections following immunohistochemical lymphatic vessel staining. This procedure allows to determine an objective LVD quantification as well as the lymphatic vessel distribution and its heterogeneity within the stroma surrounding the invasive tumor bundles. The proposed technique can be useful to better characterize lymphatic vessel interactions with tumor cells and could potentially impact on prognosis and therapeutic decisions.

Figure 1

Immunohistochemical detection of lymphatic vessels. Lymphatic vessels were detected on a cervical cancer section by immunostaining with antipodoplanin antibody (D2/40) (red) (a). They are detected by automated segmentation (blue lines) (b). Blood vessels are not stained (arrows).

Figure 2

(a) Virtual image of whole cervical cancer section with lymphatic vessel detection (blue lines) achieved at high resolution (insert). (b) Binarized image of lymphatic vessels (D2/40 positive) (red) and tumor tissue (green). (c) Original image overlaid by binarized structures.

Figure 3

Immunohistochemical detection of lymphatic vessels. Vessel sections detected by using an antipodoplanin antibody appear in red (a) and are automatically segmented in blue (b) by image processing (blue lines). Some tumor cells positive for podoplanin are manually eliminated (green circles in b).

Figure 4

(a) Euclidean distance function applied to the binary image generated in Figure 2(b). The intensity of each pixel of the blue background indicates the distance of this pixel to the nearest point to the tumor. The more a pixel is far from the tumor, the more its intensity is bright. (b) Normalized histogram representing the number of vessels in function of their distance to the tumor.

Figure 5

Euclidean distance function applied to the region of interest around the tumor. Vessels (in red) oriented at 0° (a), 90° (b), 180° (c) and 270° (d) are shown on the left (green = tumor cells). The histograms on the right correspond to the vessel distribution (normalized frequency) as a function of the distance to the closest tumor nodules (mm).

Figure 6

Polar representation of the density of vessels by unit of length measured from the tumor mass centre.