Cell Group Recognition Method Based on Adaptive Mutation PSO-SVM

Abstract

The increased volume and complexity of flow cytometry (FCM) data resulting from the increased throughput greatly boosts the demand for reliable statistical methods for the analysis of multidimensional data. The Support Vector Machines (SVM) model can be used for classification recognition. However, the selection of penalty factor c and kernel parameter g in the model has a great influence on the correctness of clustering. To solve the problem of parameter optimization of the SVM model, a support vector machine algorithm of particle swarm optimization (PSO-SVM) based on adaptive mutation is proposed. Firstly, a large number of FCM data were used to carry out the experiment, and the kernel function adapted to the sample data was selected. Then the PSO algorithm of adaptive mutation was used to optimize the parameters of the SVM classifier. Finally, the cell clustering results were obtained. The method greatly improves the clustering correctness of traditional SVM. That also overcomes the shortcomings of PSO algorithm, which is easy to fall into local optimum in the iterative optimization process and has poor convergence effect in dealing with a large number of data. Compared with the traditional SVM algorithm, the experimental results show that, the correctness of the method is improved by 19.38%. Compared with the cross-validation algorithm and the PSO algorithm, the adaptive mutation PSO algorithm can also improve the correctness of FCM data clustering. The correctness of the algorithm can reach 99.79% and the time complexity is relatively lower. At the same time, the method does not need manual intervention, which promotes the research of cell group identification in biomedical detection technology.

Keywords: adaptive mutation PSO-SVM; biomedicine; cell clustering; flow cytometry; fluorescent reagent; supervised clustering.

Figure 1

Generation principle of flow pulse signal.

Figure 2

Support vector machine classification schematic diagram.

Figure 3

Flow chart of PSO algorithm for adaptive mutation.

Figure 4

Optimal individual fitness (a) PSO algorithm; (b) Adaptive mutation PSO algorithm.

Figure 5

Experimental control group (a) Cell staining strategy; (b) Artificial clustering results.

Figure 6

The correctness of four kernel functions.

Figure 7

Number of support vectors for four kernel functions.

Figure 8

Clustering results of human peripheral blood cells (a) Fitness curve; (b) Clustering results.

Figure 9

Clustering results of Lymphocyte (a) Fitness curve (b) Clustering results.