Classification of White Blood Cells Using L-Moments Invariant Features of Nuclei Shape

Automated classification of white blood cells from microscope images is still challenging, particularly in terms of feature representations choice considering its complexity, compactness and efficiency. Particularly, in this scenario, the feature representations have to be invariant to non-uniform illumination, shape of the nuclei, stage of maturity, change in topology due to rotation, scale and shifting. This paper proposes a new white blood cell feature representation which aims at increasing robustness to those challenges. The proposed feature representation is designed based on L-moments (L-skewness, L-mean, L-scale and L-kurtosis) of the Radon projection of segmented nuclei shape. Coupled with Linear Discriminant Analysis, the proposed feature representation has been shown to be highly effective at encoding the discriminative properties of the white blood cells, and invariant to intra-class cell variations. Support Vector Machine (SVM) based (ones-vs-all) schema and a classification tree are applied to separate the multiple classes of cells. The proposed approach is evaluated for a 10-class problem. It achieves an average classification accuracy of 97.23% and outperforms all other feature representations, including bispectral invariant, local binary pattern, and histogram of oriented gradients using the same classifier on the same dataset. The proposed method is also compared and benchmarked with the other 12 existing techniques for classification of white blood cells into 10 classes over the same datasets and the results show that the proposed method achieves high accuracy in comparison with other approaches. The proposed method is also highly competitive in terms of computation and efficiency in comparison with other approaches.