Cell painting uses fluorescent agents to label the compositions or organelles of cells to evoke morphological profiling. Imaging flow cytometry (IFC) is a multi-channel imaging technique to acquire individual cell images, including the brightfield and multiple single fluorescence channels. Thus, it is necessary to assess whether cell painting combined with IFC can provide sufficient phenotypic information to distinguish morphological changes in cells. This thesis investigated changes in morphological characteristics and classification of images under different drug perturbations by employing this novel combination. The analysis procedure of IFC images of U-2 OS cells and leukemia blood cells was the focus of this thesis, which could be broadly divided into two stages. The first stage was the preprocessing of images, exploring a preprocessing framework that uses montage processing of cell images to reshape the specification of individual cell images and involve cell segmentation. The following stage was image analysis, which could be further branched into two approaches. The first approach consisted of quantifying brightfield features using CellProfiler (CP) and performing feature classification using CellProfiler Analyst (CPA). Three machine learning classifiers in CPA were utilized: Random Forest, AdaBoost, and Gradient Boosting. The investigation found that the brightfield intensity, size of the cells, and texture complexity were the most distinguishing features. The second approach employed a convolutional neural network model to conduct image classification from two image resources: the brightfield images and the merged brightfield and fluorescence channel images. This study found that brightfield images alone for phenotypic classification were not sufficient, but the accuracy of classification can be further improved by superimposing fluorescence information into the brightfield images. Nevertheless, the availability of IFC for differentiation of cell phenotypic changes under different drug effects is still proven to be viable. Furthermore, this thesis also discussed some measures to improve the image analysis procedure both regarding the image preprocessing and image analysis stages.
Identifer | oai:union.ndltd.org:UPSALLA1/oai:DiVA.org:uu-480151 |
Date | January 2022 |
Creators | Dai, Xinyi |
Publisher | Uppsala universitet, Institutionen för farmaceutisk biovetenskap |
Source Sets | DiVA Archive at Upsalla University |
Language | English |
Detected Language | English |
Type | Student thesis, info:eu-repo/semantics/bachelorThesis, text |
Format | application/pdf |
Rights | info:eu-repo/semantics/openAccess |
Page generated in 0.0014 seconds