Using computer vision algorithms for implementation of hematological analysis based on Price-Jones curve

Author(s):  V.G. Sinyuk, Belgorod State Technological University named after V.G. Shukhov, Belgorod, Russia, vgsinuk@mail.ru

D.S. Batishchev, Belgorod State University, Belgorod, Russia, batishchev@bsu.edu.ru

E.S. Soynikova, LLC «Technologies Reliability», Belgorod, Russia, 831468@bsu.edu.ru

V.M. Mikhelev, candidate of Sciences, associate Professor, Belgorod National Research University, Belgorod, Russia

Issue:  Volume 45, № 3

Rubric:  Computer simulation history

Annotation:  This article is about implementing a hematological analysis through computer vision algorithms. This type of analysis is one of the basic analyses providing huge amount information about patient and his state. We propose a pipeline with a few steps for image preprocessing thus image become more contrast and noiseless. At first image color space converting – so we separate a luminance channel and ignore other channels (due to source image features). Then we blur image with Gaussian filter and apple CLAHE filter for contrast improvement, so background pixels form more homogenous areas and become less bright in comparison to cell’s pixels. The next step is background removal and image binarization based on Otsu algorithm for border pixel luminance level detection. Afterwards we extract an array of contours fr om binary image and use this array as an input source for Watershed algorithm. As a result, we have a color image wh ere every single class of object has its own color and an array of object. This array then used as a source for cells diameters distribution histogram – a Price-Jones curve. All described steps implemented in Python 2.7 with OpenCV and Seaborn libraries.

Keywords:  hematological analysis, blood cells, image segmentation, computer vision, Price-Jones curve

Full text (PDF):  Download

Downloads count:  433