RBC Classification using Deep Learning
Vinay Bidari · 2021
The therapeutic investigation of microscopic blood smear starts with recognizing blood cells of various groups as well as estimating cell toll in blood sample. The idiosyncratic blood cell grading and counting generates priceless knowledge to the pathologist about assorted infections. This exercise can be easily concluded if the shapes of blood cells are identified first and then using the shapes, we classify the blood cells. In this research work we build and test an automatic microscopic blood smear red blood cell (RBC) classification by using Convolutional Neural Network (CNN) based deep learning approach. We train and test the statistical CNN data models based on probabilistic pattern recognition to classify the RBC of a blood smear into Normal Cells, Echinocytes, Elliptocytes and Sickle cells. The H - minimum Transform (HmT) and Watershed Transform (WT) are used in pre-processing of images to increase the accuracy if segmentation shape extraction of the blood cells. Then the Image Data Store created considering every strongest feature of each type of blood cell convolutional layers of various sizes. Training takes place through 11 layered CNN whose performance measured by using Mean Average Precision (mAP) justifies that the deep learning-based classifiers provide satisfactory results.