A modeling strategy for cell dynamic morphology classification based on local deformation patterns. (September 2019)

• Record Type: Journal Article
• Authors: Li, Heng; Pang, Fengqian; Liu, Zhiwen
• Is Part Of: Biomedical signal processing and control. Volume 54(2019)
• Publication Date: 2019-09
• Subjects: Cell dynamic morphology -- Local temporal feature -- Local deformation pattern -- Unsupervised learning; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2019.101587
• Languages: English
• ISSNs: 1746-8094

Abstract

Graphical abstract: In this paper, we report on a classification strategy of cell dynamic morphology. A schematic overview of the proposed strategy. In Step A, the local temporal features of cell deformation are extracted from 100 time-lapse images and compose 180 feature vectors of 6 dimensions for each cell. The feature vectors are reduced to 2 dimensions in Step B and clustered in Step C to model local deformation pattern. In Step D the distribution of the three patterns is calculated as the input of SVM. Classification is performed by SVM in Step E. Data dimension is indicated below each step. Highlights: A strategy based on local deformation patterns is introduced to classify cell dynamic morphology. Cell dynamic morphology is decomposed into local temporal features, and then local deformation patterns is captured from these features. Unsupervised learning algorithms are employed to discover the underlying regularities of the dynamic morphology and construct the local deformation patterns. We have studied the comments of the reviewer carefully and have made corrections, which we hope meet with approval. Abstract: Cell morphology is often used as an indicator of cell status to understand cell physiology. Therefore, the interpretation of cell dynamic morphology is a meaningful study in biomedical research. In this paper, a strategy based on local deformation patterns is introduced to classify cell dynamic morphology. The strategy decomposes dynamic morphology into local temporal features, and then captures local deformation patterns from these features through unsupervised learning. As the patterns contain underlying regularities of the dynamic morphology, they are employed to classify cell dynamic morphology. In our study, mouse lymphocytes were collected to observe the dynamic morphology, and two datasets were thus set up to investigate the performances of the proposed strategy. Experimental results validated the capacity of the proposed strategy. By considering the spatial heterogeneity and the temporal regularity of cell dynamic morphology, the strategy was competent to classify the dynamic morphology and provided remarkable advances in the accuracy and robustness of the classification on both datasets.