An efficient deep learning approach to classify white blood cells

Abstract

The human immune system’s white blood cells (WBCs) fight against infection and prevent the body from potentially harmful substances. They consist of neutrophils, eosinophils, basophils, monocytes, and lymphocytes, each of which comprises a various amount and has a specific task to do. Identifying white blood cells has been one of the most critical parts in medical science because it helps in diagnosing and monitoring various diseases and disorders. The manual microscopic analysis is difficult and subjective, and its time-consuming nature reduces the statistical dependability of the results. Problem with existing deep learning methods is that they can be heavy on computation. In this paper, we have proposed a very lightweight and efficient methodology which is called L100K-NetV2 with only 97,704 trainable parameters to classify white blood cells. The experiment, done with the Raabin- WBC (R-WBC) dataset, managed to achieve an accuracy of 98.11% in the TestA set. The proposed deep-learning methodology outperformed many other pre-trained deep learning models in terms of accuracy and parameter counts which helps to decrease the computational cost and training time.