Quantatitave Analysis Of Cell Morphological Changes In Microimages

Cell image processing has a wide range application in the fields of biomedicalengineering.It is based on the pattern recognition and digital image processing technique torealize the quantitative analysis of cells in microscopic images. Especially, the descriptionof cellular morphology has been the research focus for decades,which could assist theclinical diagnosis and treatments.Regarding to the living cell, the cellular morphology willchange along the physiological environment and could reflect the dynamic processes. Sothe description of cell morphological change is critical for understanding the mechanism ofphysiological and pathological processes.This paper focuses on the feature extraction of cellular movements, aiming to achievethe quantitative analysis of morphological deformation. To address the problem, a novelframework is proposed to measure morphological changes from both boundary motion andregional dynamic behaviors, which are also the most intuitive features of visualization.Concretely, motion history image (MHI) is introduced to record the history of cellularmovements to a static image, and then combined with shape descriptors to extractgeometric features from the cell boundary, representing the energy and direction ofboundary movements. MHI could covert the complex motion description to ordinary shapedescription. Dynamic texture is utilized to describe the intracellular motion, achieving acomplete representation of morphological changes. Above all, the proposed features arecontaining motion information and more suitable to operate on the image sequences.A model of multiple temporal scales of bags of words (BOW) is employed to reducethe influence of different temporal dimensions for motion estimation, since theanalysis ofmotion in2D+t images is based on the temporal dimension.The BOW model is introducedto deal with image sequences and combined with multiple temporal scales to handle theheterogeneous cellular motion. Experiments were provided to assort the lymphocytes intofour groups according to their morphological changes. The lymphocytes were obtainedfrom the blood samples of mice undergoing back skin transplantation. The classificationaccuracy achieves76.7%, which is superior to the methods of circumference radicaldifferences and dynamic time warping. Experiment results indicate that the proposedalgorithm issensitive to the dynamic movements and discriminative for variantmorphological changes, which can be used for detection of abnormal cell morphological changes and assist in early diagnosis.