LB9
Advanced microscope-based cytometry using brightfield image analysis
Rehan Ali, Mark Gooding, Ella Bentin, Martin Christlieb, Michael Brady
University of Oxford, UK
Aim
Flow cytometers are invaluable instruments in cancer biology research as they enable the high-throughput quantification of cell fluorescence, which can be used to characterise differences in cancer cell populations. However, they currently do not provide detailed single cell-level information. Automated microscopy with adherent cells potentially enables the high-throughput acquisition of high resolution cell images, which yield additional information, particularly when used to generate detailed cell-level time-series which can better describe differences in cellular dynamics. A key capability of such a system is automated cell segmentation. Ideally, this would be performed on images from an anatomic modality such as brightfield microscopy in order to minimise photobleaching, locate the whole cell body (as opposed to the region described by the fluorophore), and to distinguish touching cells. Until recently, this has been difficult, largely due to the poor contrast observed for adherent cells in brightfield images. We have now developed a system that automates image acquisition, brightfield cell segmentation and fluorescence quantification for adherent cells.
Method
Multi-modality cell images were automatically acquired using an Nikon Eclipse 90i epi-fluorescence microscope controlled by an interface developed by the Gray Cancer Labs. Cells were segmented from the brightfield images using an algorithm we have developed which uses a combination of advanced image analysis techniques from the computer vision and medical imaging fields, namely local phase-based feature extraction, refractive index recovery using a physical model, texture analysis and active contours. Segmentations were validated against expert manual segmentations. Using images of fixed HeLa cells exposed to H2O2 and stained with annexin-FITC and propidium iodide, cell population fluorescence was quantified using our microscope approach (measuring total intracellular fluorescence) and traditional FACS, and the results were compared. Our software is written in Matlab and is freely available online at www.sephace.com .
Discussion
Our method was able to quantify 81% (±19%, 134 cells) of the cell body pixels from brightfield images of adherent HeLa cells, which is a significant improvement on the current state-of-the-art. Log-plots of FITC vs PI compared favourably with those from FACS. We aim to use our method to obtain accurate timeseries data in order to validate mathematical models of the pharmacokinetics of ZnATSM, a fluorescent analogue of the PET hypoxic marker CuATSM.
