Automated leukocyte classification based on transfer learning for heterogeneous dataset
Diana Baby, Sujitha Juliet, M M Anishin Raj · AIP conference proceedings · 2022
Leukocytes are the most important element of blood which protects us by retaining immunity, thereby providing the defense mechanism. There are several challenges such as structure, size change, time complexity, poor image quality, and varying angle of rotation in the classification of WBC which affect the accuracy. The purpose of this study was to use transfer learning based on a pre-trained model to detect and classify the leukocytes into four separate classes such as lymphocyte, neutrophil, eosinophil, and monocyte. We employed a blend of deep learning-based models and machine learning for the classification, and have compared the results using the features extracted using MobileNetv1 and MobileNetv2 learning model, which is trained and estimated using Logistic Regression. The results show an accuracy of 88.46% for a heterogeneous dataset of leukocytes while using MobileNetv2 and Logistic regression without data augmentation.