An approach to classify white blood cells using convolutional neural network optimized by particle swarm optimization algorithm

Blood cell count is an important parameter in analysing a person’s health condition. White blood corpuscles (leukocytes) are responsible for deciding the immunity system of a person. White blood cells (WBC) are classified into: lymphocytes, monocytes and granulocytes. Granulocytes are sub-classified into: neutrophils, eosinophils, and basophils. These five types perform different functions in acting as a defence mechanism of the body. Manual investigation performed at laboratories for WBC count is prone to errors due to certain factors like human fatigue, inter-operability errors, etc. Further, there is serious issue in using trained data which cater to the changes in morphology of the white blood corpuscles, in order that trained classifiers could capitalize well. In a way to reduce misclassification rate, a methodology wherein a deep learning approach integrated with an evolutionary algorithm is proposed. Convolutional neural network (CNN) hyper-parameters were optimized using particle swarm optimization algorithm (PSO) to improve the network performance in classifying white blood cells into five types. The method is tested on merged LISC and BCCD datasets which achieved classification accuracy of 99.2% with 94.56% sensitivity, 98.78% specificity and 0.982 AUC. The results are compared with similar proven algorithms like genetic algorithms (GA), differential evolution (DE) and grey wolf optimization (GWO) algorithms. The experimental outcomes demonstrated PSO’s potential in optimizing the CNN hyper-parameters for white blood cell classification enhancing the sensitivity rate and serve a best second opinion in assessing blood cell count.