Multi-Network Blood Cell Classification System Based on Decision Fusion

The classification and identification of blood cells from microscopic images is very important for the detection of serious diseases. To increase the efficiency of the identification and counting of cells from 8 classes such as Basophils, Eosinophils, Erythroblasts, Immature Granulocytes, Lymphocytes, Monocytes, Neutrophils, and Thrombocytes, a special advance was taken using artificial intelligence for image analysis. This article proposes a collective intelligence decision system based on the fusion of individual decisions of four efficient convolutional neural networks, VGG 16, Xception, ResNet-50, and NasNetLarge. The fusion is based on the efficiency of each network's performance in recognizing (correctly classifying) the membership class of the analyzed cell. In this sense, a weight is established for each network associated with a certain class. The global system has better performances than each individual network and is better or comparable to those reported in recent works.