White Blood Cells Classification Model using Automatic Soft Computing

Accurate classification of blood cells, especially WBCs, is crucial for disease diagnosis and management. Traditional optical microscopic imaging provides limited information, but hyperspectral imaging (combining spatial and spectral data) offers richer insights. There are challenges and opportunities in white blood cell (WBC) classification using both Convolutional Neural Networks (CNN) including the following: (1) Important data, such as representative characteristics at various frequencies and orientations, may be lost due to the CNN model's usage of local receptive fields, shared weights, and subsamples. (2) To avoid over fitting, deep neural networks need a large pool of labeled training models. and (3) the higher computational overhead is yet another challenge. Therefore, these challenges motivate us the need of the proposed study. The scope of the proposed study is limited to the classification of WBC cells. Framework to classify and count the leukocytes in microscopic images and developed framework supports the pathologist and boosts the CAD system accuracy. The process of analyzing leukocytes in microscopic blood images is the major concern of this research. In this thesis, first we work on WBCs segmentation from microscopic blood images.