CNN-based two-stage cell segmentation improves plant cell tracking

Abstract Automated tracking of densely packed plant cells in time-lapse live imaging datasets is a challenging task. A local graph matching algorithm was proposed to track such cells by exploiting the tight spatial topology of neighboring cells. However, it is not easy to segment the plant cells in noisy region by the existing cell segmentation approaches, and false cell segmentation will cause serious errors in subsequent tracking procedure. In this letter, we present to segment plant cells via a CNN-based two-stage segmentation approach. In the first stage, the cells are preliminarily segmented by the existing plant cell segmentation methods such as the watershed algorithm. In the second stage, a fine CNN-based discrimination model retains the segmented cell candidates with clear boundaries in the original noisy cell images. With the CNN-based two-stage cell segmentation method, an accurate cell segmentation result is obtained as the input of the next tracking procedure. The experimental results demonstrate that our proposed segmentation approach achieves great segmentation performance for various plant cell datasets. Moreover, the proposed segmentation method is combined with three existing plant cell tracking algorithms, to demonstrate that the CNN-Based two-stage segmentation approach is very generic and greatly improves the plant cell tracking accuracy.