Leukocyte classification based on feature selection using extra trees classifier: a transfer learning approach

The criticality of investigating the white blood cell (WBC) count cannot be underestimated, as white blood cells are an important component of the body's defence system. From helping to diagnose hidden infections to insinuating the presence of comorbidities like immunodeficiency, an accurate white blood cell count can contribute significantly to shape a physician's assessment. The manual process performed by the pathologists for the classification of WBCs is a time consuming and tedious task, which is further disadvantaged by a lack of accuracy. This study concentrates on the automatic detection and classification of WBC without data augmentation into four subtypes such as eosinophil, monocyte, lymphocyte and neutrophil based on images from three different datasets. The methodology adopted in this paper is transfer learning approach in which the features are extracted using ResNet50, DenseNet121, MobileNetv2, Inceptionv3 and Xception deep learning models.The extra trees classifier is used as an intermediate stage for selecting most predominant features, which reduce the execution time. When evaluating the performance on the basis of recall, precision, F-measure and accuracy parameters, the classification of ResNet50 features selected by extra trees classifier using multi-class support vector machine (SVM) provides the highest accuracy of 90.76% .

Automated summary

This study focuses on automatic detection and classification of white blood cells using deep learning models to improve accuracy and reduce time. By selecting ResNet50 features with extra trees classifier, the highest accuracy of 90.76% is achieved in multi-class SVM evaluation.