An hybrid soft attention based XGBoost model for classification of poikilocytosis blood cells

Poikilocytosis affects the human body because it causes changes in the shape of Red Blood Cells (RBCs). This aberrant shape and structure of RBCs lead to a number of health issues in our bodies. These issues include lack of oxygen, protein, nutrients, and other substances. In addition, it leads to a number of disorders such as Anemia, Thalassemia, etc. Detection and classification of such cells in Peripheral Blood Smear (PBS) is time-consuming, especially in geographically underserved regions where there is a paucity of hematologists. Hence, the deployment of deep learning for the classification of such cells reduces the workload of hematologists. In this work, a novel hybrid approach using K-mean color quantization segmented-based attention-deep Convolution Neural Network (CNN) with an Extreme Gradient Boosting (XGBoost) algorithm for classifying poikilocytosis has been proposed for detection and classification of abnormal cells.The proposed hybrid model has outperformed the lightweight CNN model and benchmark models on two data sets: one publicly accessible Chula-PIC-Lab data set and one privately collected dataset. On the Chula-PIC-Lab and CCHRC datasets, the proposed model obtains 95.38%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$95.38\%$$\end{document} and 93.43%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$93.43\%$$\end{document} of F1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_1$$\end{document}-score on segmentation, 98.44%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$98.44\%$$\end{document} and 98.24%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$98.24\%$$\end{document} of accuracy on classification; 98.55%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$98.55\%$$\end{document} and 98.28%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$98.28\%$$\end{document}of F1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_1$$\end{document}-score on poikilocytosis classification respectively.