Comparative Study of Data Augmentation Strategies for White Blood Cells Classification

Abstract

The paper is devoted to the study of strategies that can be applied to unbalanced data in solving the task of classifying white blood cells. The main goal of the proposed paper is to determine the best approach to combat data imbalance when working with images of blood cells. Description of classical and state-of-the-art methods to deal with imbalance is given. In the course of the research, biomedical data is collected, annotated, and preprocessed, as well as selected strategies are applied to form datasets. Base model of artificial neural network for image classification is selected and built. Also, the dependence of trained models accuracy on the applied strategies is studied. Thus, the best approach to data augmentation for white blood cells classification problem is determined.

Keywords

• deep learning
• medical image processing
• data augmentation
• imbalanced data problem
• digital cytological analysis