An Automatic Cell Counting Algorithm for Fluorescent Images with Non-ideal Background
Date Issued
2011
Date
2011
Author(s)
Lee, Ting-Chung
Abstract
Cell count is among the fundamental information in cytopathology and cell biology and hematology; most of the researchers count cells and observe morphology manually through microscopes to acquire statistical information. In general, cell counting is labor-intensive and time-consuming and often very operator-dependent, especially when counting cells in images with non-ideal background. And commercial products may not be able to facilitate counting for such images, either.
In this thesis, sample was prepared from mixing two kind of cell lines (HUVEC labeled with fluorescence and Jurkat not), and cells in the micrograph of samples were analyzed and recognized by image processing techniques and multivariate statistic. First, the histogram of image was classified by Gaussian mixture model method for foreground and background extraction and processed with morphological filter for noise removal. Nearest Neighbor method was used to identify targets according to their features extracted from images. The robustness of classifier was verified by k-fold cross validation. This algorithm can analyze and count cells for two fluorescence-stained cells out of non-ideal background.
Results show the image segmentation by Gaussian mixture model is virtually independent to the environmental condition of images (exposure time, contrast, and etc.) and the accuracy of recognition is around to 97% for extracting cell according to the built feature database. The algorithm serves the needs of cell counting of medical research very well.
Subjects
Cell count
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-100-R98543002-1.pdf
Size
23.54 KB
Format
Adobe PDF
Checksum
(MD5):1464fd3f4607f1a0d3203c83446f3fdd