Yonggang Shi

Modeling dynamic cellular morphology in images

This paper presents a geometric method for modeling dynamic features of cells in image sequences. The morphological changes in cellular membrane boundaries are represented as sequences of parameterized contours. These sequences are analyzed as paths on a shape space equipped with an invariant metric, and matched using dynamic time warping. Experimental results show high sensitivity of the proposed dynamic features to the morphological changes observed in lymphocytes of healthy mice after undergoing skin transplantation when compared with standard representation methods and shape features.

Quaternion generic Fourier descriptor for color object recognition

By introducing quaternions, Fourier transforms, the fundamental tool in signal and image processing, were extended to treat a color image in a holistic manner. On the other hand, color image description has long been a major topic of pattern recognition. In this work, inspired by the quaternion Fourier transforms, we propose a quaternion generic Fourier descriptor to describe color images holistically and effectively without losing any color information. The theoretical framework to construct a set of invariants with respect to geometric transformation is also provided. Moreover, several experiments are conducted on color image databases to illustrate the effectiveness of the proposed invariants. The results provide supports for that the quaternion generic Fourier descriptor invariants have striking robustness to geometric transformation and noise. Remarkable progress in accuracy, robustness and efficiency of color object recognition has been made by the proposed descriptor. In addition, it presents superior performances on real recognition scenarios. HighlightsWe propose an image descriptor with the quaternion algebra to describe color images.A framework to develop robust invariants to geometric transformation is provided.Experiments are conducted to validate the performance of the proposed descriptor.The proposed descriptor performs robustly to geometric transformation and noise.A progress in accuracy of recognition has been made by the proposed descriptor.On real recognition scenarios, the proposed descriptor also performs superiorly.

Quantitative Analysis of Lymphocytes Morphology and Motion in Intravital Microscopic Images

Studying the morphology and interior movement of lymphocytes in intravital microscopic images is essential to understanding and treating various biological processes and pathological situations. A method combing features of shape, deformation, and intracellular motion for quantitatively characterizing the dynamic behavior of a single lymphocyte is proposed in this paper. The method is tested on a set of image sequences of lymphocytes obtained from the peripheral blood of mice undergoing skin transplantation using a phase contrast microscope. Experimental results coincide with the clinical observation and pathological analysis, demonstrating that the extracted cell morphology and motion features can provide new insights into the relationship between the dynamic behavior of lymphocytes and the occurrence of graft rejection.

Analyzing dynamic cellular morphology in time-lapsed images enabled by cellular deformation pattern recognition.

Computational analysis of cellular morphology aims to provide quantitative information of the global organizational and physiological state of cells, and has long been a major topic of biomedical research. Instead of analyzing morphology of static cells, we concentrate on live-cell deformation in a period of time. According to our observation of dynamic cell behavior, we have assumed that the pattern of cellular deformation is relevant to the cellular state. Moreover, based on our assumption an innovative approach for characterizing the deformation pattern is described and applied into cell classification. After normalizing and aligning cell image sequences, we extract the continuity of deformation at each angle through time-lapse. Then the deformation pattern is given by the histogram of the continuity of deformation. Experimental results demonstrate that the cellular deformation pattern provided by our approach can be applied to discriminate cellular activation. In addition, the deformation pattern recognition makes remarkable progress in the classification of cells.

Quantitative analysis of live lymphocytes morphology and intracellular motion in microscopic images

Abstract Cellular morphology and motility analysis is a key issue for abnormality identification and classification in the research of relevant biological processes. Quantitative measures are beneficial to clinicians in making their final diagnosis. This article presents a new method for measurement of live lymphocyte morphology and intracellular motion (motility) in microscopic images acquired from peripheral blood of mice post skin transplantation. Our new method explores shape, deformation and intracellular motion features of live lymphocytes. Especially, a novel way of exploiting intracellular motion information based on optical flow method is proposed. On the basis of statistical tests, optimal morphological and motility features are chosen to form a feature vector that characterizes the dynamic behavior of the lymphocytes (including shape, deformation and intercellular motion). In order to evaluate the proposed scheme, the above feature vector is used as input to a probabilistic neural network (PNN) which then classifies the dynamic behavior of lymphocytes in a set of cell image sequences into normal and abnormal categories. Comparative experiments are conducted to validate the proposed scheme, and the results revealed that joint features of shape, deformation and intracellular motion achieve the best performance in expressing the dynamic behavior of lymphocytes, compared with Fourier descriptor and Zernike moment methods.

Multi-classification of cell deformation based on object alignment and run length statistic.

Cellular morphology is widely applied in digital pathology and is essential for improving our understanding of the basic physiological processes of organisms. One of the main issues of application is to develop efficient methods for cell deformation measurement. We propose an innovative indirect approach to analyze dynamic cell morphology in image sequences. The proposed approach considers both the cellular shape change and cytoplasm variation, and takes each frame in the image sequence into account. The cell deformation is measured by the minimum energy function of object alignment, which is invariant to object pose. Then an indirect analysis strategy is employed to overcome the limitation of gradual deformation by run length statistic. We demonstrate the power of the proposed approach with one application: multi-classification of cell deformation. Experimental results show that the proposed method is sensitive to the morphology variation and performs better than standard shape representation methods.

Multi-modal diffeomorphic demons registration based on mutual information

Diffeomorphic demons algorithm is an efficient and robust method in nonrigid image registration. It uses mean squared error as the similarity measure and set the gradient and gray level difference as the interior and exterior force respectively to make the points move. However it cannot deal with multiple modality images. In this paper, mutual information is used as the similarity measure instead of mean squared error and the gradient of mutual information respect to parameters is set as the force to make the points move. Then the diffeomorphic demons algorithm is extended to match multi-modal images. The experiment with magnetic resonance T1 image and magnetic resonance T2 image shows that this method is effective and performs better and more quickly compared with B-spline free-form deformation method.

Interactive region-based MRF image segmentation

An interactive region-based Markov random field (MRF) image segmentation method is proposed for solving inaccurate parameter estimation and mis-segmentation of MRF method. Because color and texture features in natural image are very complex, unsupervised method cannot accurately achieve segmentation. The proposed method also introduces human-computer interaction to improve segmentation. The segmentation is achieved by classifying pixels into different classes. All these classes can be represented by multivariate Gaussian distributions. In the proposed method, image is firstly separate into homogeneous regions, and interactive information is carried out as manual marks on over segmentation regions to roughly indicate object and background. Feature parameters of object and background can be accurately calculated from marked regions. To solve partial mis-segmentation might appear in MRF model, we use adjacent potential energy as region merging metric to automatically correct mis-segmentation. Empirical results show that the proposed algorithm can accurately segment object from background. Compared with traditional MRF algorithm and unsupervised Graph Cut algorithm, the proposed algorithm achieve better results. Based on more accurate initial parameters and automatic correction of mis-segmentation, the proposed method can well extract object from background.

A Method of Quantitative Analysis for Dynamic Cellular Morphological Change

In recent years, with the advancement of micro- scopic imaging technique, cellular morphological analysis has become one of the most important branches of biomedical image processing field. Cell image analysis, an assisting method of clinical information acquisition, has played an important role in facilitating the appropriate diagnosis and treatment against some serious diseases such as cancer, leukemia and etc. Early studies of this field focuses on still image information and mainly concerns either cell counting or the morphological characteristics of cell image. Furthermore, much effort has been devoted to analyzing the morphological change of cell image qualitatively, while less attention has been paid to analyzing it quantitatively which in fact is more effective, objective and accurate. Under this very circumstances, our work concentrates on analyzing the cellular morphological change by means of quantitative measurement, and a new method—circumference radial difference (Rdc )i s pro- posed. The basic idea is to measure the morphological change by estimating the radial differences from 360 different angles. Two sets of lymphocyte video database provided by the cooperation hospital are used to verify the proposed approach. Experimental results show that the proposed (Rdc) measurement can produce reasonably good agreement with subjective judgment by people.

Cell dynamic morphology classification using deep convolutional neural networks: Cell Dynamic Morphology

Cell morphology is often used as a proxy measurement of cell status to understand cell physiology. Hence, interpretation of cell dynamic morphology is a meaningful task in biomedical research. Inspired by the recent success of deep learning, we here explore the application of convolutional neural networks (CNNs) to cell dynamic morphology classification. An innovative strategy for the implementation of CNNs is introduced in this study. Mouse lymphocytes were collected to observe the dynamic morphology, and two datasets were thus set up to investigate the performances of CNNs. Considering the installation of deep learning, the classification problem was simplified from video data to image data, and was then solved by CNNs in a self‐taught manner with the generated image data. CNNs were separately performed in three installation scenarios and compared with existing methods. Experimental results demonstrated the potential of CNNs in cell dynamic morphology classification, and validated the effectiveness of the proposed strategy. CNNs were successfully applied to the classification problem, and outperformed the existing methods in the classification accuracy. For the installation of CNNs, transfer learning was proved to be a promising scheme.