Analyze Microscopic Images to Detect Blood Diseases

Abstract

This chapter uses the geometric and shape features to provide unique information of the white blood cells and leukemia cells for classification. Compares between PSO and GSA to train the Feed Forward Neural Network (FNN). To classify the Agranulocytes that contains lymphocytes and monocytes cells and Granulocytes that contains neutrophils, eosinophils and basophils cells. The FNNs evaluated different iteration number and different number of particle and agents for applying PSO and GSA. The experiment shows that the number of particles, number of agents, and the iteration numbers affect the classification results. The average accuracy for classifying Granulocytes type is 95.16% and 94.5%. The average accuracy for classifying Agranulocytes type is 99.7% and 99.5%. Optimized MLP with gray wolf optimized algorithm helps in avoiding traps in the local minimum and slow convergence rate for white blood cell and leukemia classification. Using the gray wolf optimizer to adopt the weights and biases of the multilayer perceptron. To enhance the accuracy of classification between normal and leukemia cells, classify the normal cells to their five categories and classify the leukemia to their four categories. The proposed system applies to different data sets (ALL-IDB2, LISC and ASH-Image bank) and overcomes the segmentation and classification problems of microscopic images. Shows an outstanding results, the average classification accuracy between normal and leukemia cells is 98.58%, between white blood cell categories is 98.9% and between leukemia types is 98.93%.