Rapid classification of micron-sized particles of sphere, cylinders and ellipsoids by diffraction image parameters combined with scattered light intensity. (February 2019)

• Record Type: Journal Article
• Authors: Wang, Wenjin; Wen, Yuhua; Lu, Jun Q.; Zhao, Lin; Al-Qaysi, Safaa A.; Hu, Xin-Hua
• Is Part Of: Journal of quantitative spectroscopy & radiative transfer. Volume 224(2019)
• Page Start: 453
• Page End: 459
• Publication Date: 2019-02
• Subjects: Single light scattering -- Diffraction imaging -- Image analysis -- Nonspherical particle analysis; Spectrum analysis -- Periodicals; Radiation -- Periodicals; Analyse spectrale -- Périodiques; Rayonnement -- Périodiques; Radiation; Spectrum analysis; Periodicals; 543.0858
• DOI: 10.1016/j.jqsrt.2018.12.010
• Languages: English
• ISSNs: 0022-4073

Abstract

Highlights: An accurate method for diffraction imaging simulation has been validated and applied to generate unpolarized diffraction images of 1965 single particles, composed of single and double spheres, cylinders and ellipsoids, under coherent illumination. A Gaussian mixture model (GMM) based clustering algorithm has been developed to classify these particles into 3 types of spheres, cylinders and ellipsoids with the texture parameters of the calculated diffraction image data. GMM classifiers with selected diffraction image texture parameters and integrated forward scattered light intensity have been found to classify particles of same refractive index with accuracies ranging from 82.6% to 97.2%. The effective texture parameters of diffraction images extracted with a gray-level co-occurrence matrix were identified as correlation, sum entropy, entropy, dissimilarity and inverse difference moment. Abstract: Spatial distributions of light scattered by single particles correlate closely with their morphologies in terms of refractive index (RI) distribution. Diffraction imaging of scattered light under coherent excitation presents a unique approach to acquire and extract feature parameters for particle classification. A validated method has been applied in this study to accurately simulate diffraction imaging of light scattered by homogeneous particles and obtain calculated diffraction image (DI) data. The feature parameters of DI data have been extracted by the gray-level co-occurrence matrix (GLCM) algorithm. We have developed an unsupervised machine learning algorithm based on Gaussian mixture model (GMM) to classify 1965 particles made of single and double spheres, cylinders and ellipsoids with varied RI values in parameter space. It has been shown that selected GLCM parameters combined with integrated forward scatter intensity can provide effective markers for accurate and morphology based classification. For 1791 particles of the same RI, the mean accuracy values of classifying particles into 3 particle types range from 82.6% to 97.2%. These results demonstrate the strong potential of diffraction imaging method for rapid analysis and classification of nonspherical and homogeneous particles by the GMM classifiers that is very challenging in comparison to distinguishing biological cell types.