Date of Award

January 2015

Degree Type


Degree Name

Master of Science (MS)


Biomedical and Chemical Engineering


James J. Henderson


Automated Cell Tracking, MATLAB, Orthogonal Design, Weighting Factor Optimization

Subject Categories



Localization and tracking of cells generates raw digital information from microscopy images, including images of stained nuclei and highly precise determination of central positions of cells, which can be analyzed for investigation of cell motility. In a previous study by this group, an algorithm termed automated contour-based tracking for in vitro environments (ACTIVE) was established for tracking large cell populations for long periods of time. For the two-cell interaction events on which ACTIVE was initially focused, error rate was reduced as much as 43% compared to a traditional positional analysis algorithm by Kilfoil and colleagues. In the present thesis, we investigated whether the ACTIVE algorithm could be improved when applied to a more complicated condition: three-cell interactions. To determine whether modification of the ACTIVE algorithms could allow ACTIVE to outperform the Kilfoil benchmark method when applied not only to two-cell interaction cases but also to three-cell interaction cases, two approaches were developed and studied: 1) optimization of the existing ACTIVE cost-function weighting factors by orthogonal design with addition of two new factors, velocity and directionality, and detection of ranges and effects for all factors, and 2) modification of the circumstances under which the Kilfoil approach and the cost function approach were executed. We found the position factor to be the most important and accurate among all the factors, and optimized all factors. What is more, the directionality was determined to be the second most significant factor of the cost function for correctly tracking cells. However, modification of neither the position nor directionality factor could achieve higher accuracy than the Kilfoil method. Having evaluated the new strategy that combines both the cost function and the Kilfoil method, we found that the new strategy did not result in higher accuracy for three-cell interactions, as compared to the pure Kilfoil benchmark method. The accuracy of the new strategy was 6% lower on average than the Kilfoil method. Although the results of the present work do not yet achieve a method for analysis of three-cell interactions that outperforms purely positional analysis, the work provides a method for optimization of the cost function and new understanding of characteristics of three-cell interactions that lead to reduced accuracy in the cost function and/or positional (Kilfoil) approaches.


Open Access

Included in

Engineering Commons