Modelling and identification of neutrophil cell dynamic behaviour

• Main Author: Zhang, Xiliang
• Other Authors: Kadirkamanathan, Visakan ; Billings, Stephen A.
• Published: University of Sheffield 2016
• Subjects: 612.1
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.694457

This thesis is focussed on the shape analysis and tracking of neutrophil cells to facilitate the understanding of their behaviour. Neutrophil, one of the important type of white blood cell, protects humans and animals from infections and inflammations. The underlying mechanism is believed to be that when inflammation happens, a changing chemotaxis field causes neutrophils to move to inflammatory sites. During this migration process, neutrophils also change shapes. For example, pseudopod protrusions are formed on the boundary in response to the local gradient of the chemotactic field. After inflammation resolution, some neutrophils go to "sleep" and some move back but what drives these mechanisms are still not fully understood. If the mechanisms were known, it would be helpful to accelerate or slow down the process of treating some diseases. The thesis attempts to provide a quantitative analysis based on time lapse microscopic images of in vivo zebrafish animal models. The underlying premise governing the analysis is to identify if cell states from their motion and shape. The thesis begins with cell centroid tracking and uses three common kinematic models commonly used in the target tracking literature. The interacting multiple model framework, which is underpinned by multiple Kalman filters, is used to determine probabilities of most likely model to explain the cell motility pattern. These different models are then compared to identify if the motion pattern (motility) can be attributed to different cell behaviours. This is then followed by cell shape tracking to characterise not only the cell shape but also to identify regions of protrusions. It addresses the problem of estimating the chemotactic field that acts to recruit neutrophil cells to the inflammation sites based only on observed cell tracks, without any direct measurement associated with the external cell environment. By assuming that the cell velocity is proportional to the local chemotactic gradient, a least squares method in combination with the Kalman filter, was used to estimate this field. Results on a set of real data show the estimated field. Cell shapes were modelled as B-spline parametric active contours. By casting the parametric active contour model in state space form, Kalman filter was employed to track the shapes. Shape tracking required solving the problem of cell boundary association between two time frames which required identification of correspondence points at cell boundaries. This required improvement to a nearest neighbour filter method to give continuity of the cell shapes of the same neutrophils across the different frames. Characterisation of cell shape was carried out by employing Fourier descriptors from which two features, magnitude of the highest descriptor and the magnitude of the lowest frequency, were chosen as proxy for associating cell shapes with cell states. These features for instance are associated with cell roundedness. Both tracking methods, cell centroid and shape tracking, are employed on the real data and results shown to demonstrate the effectiveness of the methods. This thesis makes the following novel contributions. Firstly, a new framework was established to solve the cell centroid and shape tracking towards characterising cell shape behaviour. Secondly, the nearest neighbourhood filter employed to solve the association problem was improved to solve the problem of neutrophil cells disappearing and reappearing in image frames. Thirdly, the low frequency Fourier descriptor, combined with other methodologies, was successfully implemented in detecting the modes of neutrophil behaviour. In addition, the chemotactic field was estimated by using the centroid velocities. Furthermore, the multiple model filter was used for behavioural mode detection of cells.