Statistical classification of erythrocytes in hereditary spherocytosis based on the spectral features of cell surfaces’ AFM images

Abstract

The method of classification of erythrocytes (red blood cells) based on spectral features of the cell surface images (of physical-mechanical properties maps) obtained with an atomic-force microscope (AFM) is proposed. Each scan line of the original AFM image is considered as a random sequence realization and the discrete Fourier transform is applied to compute its spectral features. The spectral estimates are smoothed on the map and the informative characteristics are computed as the medians of the spectrogram values for each frequency. The classification or two classes of erythrocytes (spherocytes and discocytes) taken from patients with hereditary spherocytosis was carried out by the obtained informative characteristics using the decision trees and boosted decision trees methods. The frequency interval was found with the best classification accuracy – over 82 % for the boosted decision trees method.

Keywords

• afm images
• erythrocytes
• maps of physical-mechanical properties
• method of classification
• discrete fourier transform
• decision trees
• boosted decision trees