Article,
Segmentation of White Blood Cells with Colour Space Transformation and use of Transfer Learning for Optimization
CENTRAL ASIAN JOURNAL OF MEDICAL AND NATURAL SCIENCES, (2022)
Abstract
White blood cells (WBC), or leukocytes, are an essential part of the human immune system, constantly protecting the body against viruses, bacteria and other foreign invaders. Determining the WBC is crucial for curing various diseases, especially leukemia. The abnormality of WBC count causes leukemia, failing the autoimmune system. Image segmentation, an application of pattern recognition techniques, is employed in this paper to find the WBC count. WBC are identified by the colour difference between their nucleus and cytoplasm. Using CNN’s U-net architecture, the cell borders of WBC are marked, enabling us to find the area of abnormality in the given sample. This method is trained on the images in the known training data. The model can produce an accuracy of around 87% for segmenting WBC from the blood smear. And for the final output with the concept of swarm intelligence in AI. The WBC count in the image is found with the OpenCV method considering optimization purposes. Secondly, the system transfers and modifies the model with transfer learning models VGG/ResNet and counts cells with the deep neural model. The counting model can be used for other modelling and application purposes.
