A CNN Based Deep Learning Approach for Leukocytes Classification in Peripheral Blood from Microscopic Smear Blood Images

Abstract

As a cardinal sign of the pathological condition of human health, WBC also known as leukocytes has prominently regulated a functional role in the body’s immune system. So, the classification of white blood cells has been kept on great precedence in the diagnosis of hematological diseases such as leukemia or any kind of blood cancers even AIDS so. But the evaluation process is quite complicated and painstaking as leukocytes include a series of variety within it. Following the drawbacks, deep learning can offer much refinement in the classification manner of leukocytes, especially, the convolutional neural network, i.e., CNN, an advanced and sophisticated technology that has been conceded widespread as a compatible approach regarding the visual image analysis. Thus considering the pledges, a new approach has been introduced in this paper based on a precisely established CNN model with an adjuration of outrun the traditional approaches proposed before that includes multiple complicated steps such as preprocessing, segmentation, feature extraction, and so on in classifying four types of white blood cells. The model is established on a feasible study and a series of experiments conducted on large-scale training sets of 9957 annotated blood smear images of WBCs (neutrophils, eosinophils, monocytes, and lymphocytes) to perceive a hierarchical manner of learning and understanding automatically. And then, the detection model is evaluated on test sets of 2487 annotated blood smear images of WBCs. This diligent research results in salient recognition performance in classifying four types of WBCs with an accuracy of 98% on average. So, our proposed customized model can be used in the real-life healthcare sector for identifying diseases that also reduce the disease detection cost.