Rat Mesenchymal Stem Cell Segmentation on Time-lapse Fluorescence Microscopy Video with Two-Dimensional Bandpass Filters

This study focuses on enhancing rat mesenchymal stem cell (MSC) segmentation in fluorescence microscopy images. It optimizes parameters of the two-dimensional bandpass filter, crucial for addressing challenges with thin, elongated cells. The bandpass segmentation, ranking second in the Cell Tracking Challenge (CTC) 2020 for rat MSC segmentation, indicates potential performance improvement through parameter optimization. The optimization involves two Gaussian filters and a tophat filter, utilizing a grid search and perceptron model on the Fluo-C2DL-MSC dataset in the CTC. Optimized parameters include bandpass filters with a Mexican Hat shape and a standard deviation ratio between two Gaussian filters around (-0.28) to (-0.99). Tophat filter sizes ranging from 300 to 1000 pixels produce segmentation results with Jaccard Index scores of 0.708±0.002, 0.728±0.002, and 0.739±0.004 across various images. Post-optimization, segmentation success remains at 66.67 %, struggling to accurately segment approximately 33.34 % of thin and long cells. Further improvements require noise reduction methods and exploration of filters with different shapes to accommodate elongated cells.