Flow cytometry mainly used for detecting the characteristics of a number of
biochemical substances based on the expression of specific markers in cells. It
is particularly useful for detecting membrane surface receptors, antigens,
ions, or during DNA/RNA expression. Not only can it be employed as a biomedical
research tool for recognising distinctive types of cells in mixed populations,
but it can also be used as a diagnostic tool for classifying abnormal cell
populations connected with disease. Modern flow cytometers can rapidly analyse
tens of thousands of cells at the same time while also measuring multiple
parameters from a single cell. However, the rapid development of flow
cytometers makes it challenging for conventional analysis methods to interpret
flow cytometry data. Researchers need to be able to distinguish
interesting-looking cell populations manually in multi-dimensional data
collected from millions of cells. Thus, it is essential to find a robust
approach for analysing flow cytometry data automatically, specifically in
identifying cell populations automatically. This thesis mainly concerns
discover the potential shortcoming of current automated-gating algorithms in
both real datasets and synthetic datasets. Three representative automated
clustering algorithms are selected to be applied, compared and evaluated by
completely and partially automated gating. A subspace clustering ProClus also
implemented in this thesis. The performance of ProClus in flow cytometry is not
well, but it is still a useful algorithm to detect noise.

研究了自动化门控算法在流式细胞术中的局限性，并比较了三种典型自动聚类算法和子空间聚类算法 ProClus 在识别细胞群中的性能。