Optimal Resolution for Automatic Quantification of Blood Vessels on Digitized Images of the Whole Cancer Section

Abstract

Discrepancies concerning the prognostic significance of cancer vascularization can be partly explained by biases due to quantification protocols. We recently recommended a swift, inexpensive and automatic analysis of 2,700 dpi slide scanner images of the whole immunostained sections. Another team, proposed, quite at the same time, to work at 4,000 dpi. The aim of the present paper is to check if information contained in images scanned at 2,700 and 4,000 dpi are relevant and equivalent, when compared to the low magnification of the microscope, in order to propose the best compromise between precision and time expense. To evaluate precisely the amount of information gained or lost according to the resolution used, we compared the number and size of blood vessel profiles, manually detected, on twenty one Hodgkin lymphoma acquired with a scanner (2,700 and 4,000 dpi) and with a microscope (16,000 dpi). Results obtained at 4,000 dpi were equivalent to the estimation performed at microscopical level either by a biologist or a pathologist, while tiny vessels were lost at 2,700 dpi. Scanning whole histological sections at 4,000 dpi provides a relevant method for evaluating tumour vascularization, which can be easily automated and standardized.

Improved computer-assisted analysis of the global lymphatic network in human cervical tissues

“…Owing to high image size that hampers calculation, all binary images of the detected structures were decimated according to the procedure described elsewhere. 31 Figure 2 illustrates the different steps of image processing.…”

Automatic quantification of the microvascular density on whole slide images, applied to paediatric brain tumours

“…Finally, let us remark that, even with these possible limitations in mind, our method should be much more reproducible than manual measurements performed on a few chosen excerpts in each slide even by a trained pathologist [15], if the latter is at all doable on such a large set of slides. We prefer to save the pathologist's time for quality control and make our overall study cheaper by the extensive use of the computer.…”

“…Luckily, virtual microscopy and the digitisation of pathology slides have become quite common over the last few years, allowing the use of the computer to perform various quantitative and reproducible measurements on histology sections [13]. In the beginning of this digital era, for cost and material reasons, it was not possible to measure the parameters of microvessels at full resolution (20x or 40x), and it was suggested to use images at resolution around 1x [14,15]. However, slide scanners produce now high resolution microscopy images of whole slides in a short time (at most a few minutes) and for a reasonable price [16], and using them for quantification should improve the precision of the results.…”

Whole Slide Quantification of Stromal Lymphatic Vessel Distribution and Peritumoral Lymphatic Vessel Density in Early Invasive Cervical Cancer: A Method Description

“…Once the binary image was obtained, high image size hampers calculations. To overcome this limitation, before quantification, binary images were decimated according to the procedure previously described [ 18 ]. Image analysis was performed using image analysis library Pandore (GREYC, Caen France) and tool box of MATLAB software (9.2).…”

“…These software allows to process the image in order to extract vessels from the background and to perform measures such as vessel number and/or vessel size. However, in most cases, standard software is unable to accurately detect vascular structures and different teams have developed home-made image processing methods for specific applications [ 14 , 16 – 18 ]. Their success depends critically on the immunohistochemical staining quality [ 19 ] which allows the discrimination of objects of interest by a coloration and/or a modification of grey-level intensities.…”