A generalized data association approach for cell tracking in high-density population

The nearest neighbor association method is the most common approach for multi-cell tracking, but it is easily prone to errors in the case of high-density population with cell division, occlusion and uneven motion. In this paper, we propose a generalized data association approach with a combination of information including the distance of cell position, dynamics and morphology. To address the incompatibility of cell number and positions among two or more consecutive frames, a novel contour similarity measurement based on optimal subpattern assignment with particle swarm optimization is developed. Afterwards, five events during cell motion are further analyzed and corresponding data association approach is proposed respectively. Experiment results show that our algorithm could give a more accurate association results and outperforms other methods in high-density cell population.