Research On Randomization Method Of Multi-cell Motion Analysis

Biomedical image plays a significant role in medical diagnosis and treatment.The research of cell image is an important branch of medical image.To obtain feature and motion trajectoryof cells from cell image sequence is an important basic work in medical analysis.In recent years,researchers in related fields have achieved a lot of beneficial results.But due to the large amount of information,the complexityof cell structure and shape,therandomness of cell movement and the influence by cell acquiring technology,image quality and other factors,it is difficult to estimate the varying number and trajectories of cells.According to these challenges,exploits the spatio-temporal information from the image sequence,several multi-cell motion analysis methods are developed for potential engineering applications.The main work contributionsand innovation points are as listed follows:1.According to the problem of multi-cell adhesion in image sequences,a hybrid cell detection algorithm based on threshold is proposed.The simulation results show that the"Recall" and "Precision" of this image sequence can reach 98.32%and 97.03%.According tomulti cell overlapping problem in dense cases,an improved watershed hybrid detection algorithm isproposed.Overlapping cells can be very well segmented to a single cell,and hardly appear incorrect or incomplete segmentation phenomenon and the accuracy of the segmentation can be up to 96%or so.2.According to the problem of state coupling caused by the cells in the presence of splitting and collision,an extended interacting multiple model particle filter cell motion analysis algorithm is proposed.The method of cell hybrid detection method is designed,which is based on threshold processing and holes filling.Three modes are modled based on three kinds of interaction behavior between cells(independent,collision,division),and parameters of the angular velocity and area feature are augmented in cell state.Strategy of cell data correction is presenteded based on difference measurement of cell area and distance.Simulation results demonstrate the effectiveness of the proposed design method.3.According to the difference of dynamic characteristicsandthe varing number of cells,an ant colony multi-cell motion analysis method based on background subtraction is proposed.Priori colony distribution is generated by the nonparametric kernel density estimation.The multimodality of pheromone is established by Multi-colony reconstruction.Cell labeling and state extraction is accomplished by fast clustering algorithm.Simulation results demonstrate the effectiveness of the proposed design method.4.According to spatially adjacent cell state estimate in the low signal to noise ratio image sequences,a novel ant system with multiple tasks is modeled.The initial ant colony distribution is obtained by the approximate average method and divided into several groups by K-means clustering.Collaboration and competition working mode is modled and the multimodality of pheromone is established.The estimates of multi-cell state and the number of cells at each frame are extracted through merging multiple similar tasks.The simulation results show that this method is more accurate than other algorithm.5.According to different cell population density distribution in low signal to noise ratio image sequences,an efficient and effective ant-based algorithm for tracking cells in varying density is presented.The initial ant colony distribution is obtained according to the average likelihood degree on various regions of a real cell image.Interactive cooperation mode and competition mode are modled based on the sparse and dense cell events.Modes update based on ant pheromone is designed.Simulation results demonstrate the effectiveness of the proposed design method.