Automatic White Blood Cell Detection and Identification Using Convolutional Neural Network

Differential blood count is a very common medical test which determines relative percentage of each type of white blood cell (WBC) in a blood sample. This test is usually performed by visual inspection of a blood sample which is time consuming and tedious task for a medical specialist. This test can be performed automatically with appropriate equipment as well. However, such equipment is quite expensive and available only at larger medical centers. In this paper alternative approach is proposed which is based on low cost microscope and digital camera coupled with appropriate algorithm for WBCs detection and identification in a blood image. The proposed algorithm consists of two steps. An image of blood sample is segmented in order to detect possible WBCs which are then further classified with Convolutional Neural Network (CNN) into 5 classes. The proposed approach shows promising results obtaining accuracy of 81.11% on created dataset.