Automated CTC Classification, Enumeration and Pheno Typing: Where Math meets Biology

Circulating tumor cells (CTCs) found in the blood of cancer patients are a promising biomarker in precision medicine. Yet, CTCs are very rare and it is very challenging to manually detect, characterize and count them in images of fluorescently labeled cells. Therefore, this thesis aims to develop image analysis and machine learning models to automatically and accurately detect and classify CTCs in fluorescent images and prove the added benefit of automated methods for cancer patients. CTCs and their fluorescent signals strongly vary in size, shape and signal intensity. We addressed the presence of multiple size and intensity scales by the development of a multiscale segmentation model. We combined nonlinear segmentation with scale spaces and spectral decompositions. This results in an accurate segmentation of the cells combined with a cell clustering based on their size and intensity. We further adapted the model to a purely size-based clustering of the cells which excels in segmenting also very dim fluorescent signals. Based on the segmentation of all cells, we extracted quantifiable features of the immunofluorescent expression of each cell. These features were used for an accurate and reproducible assessment of expression levels of CTC treatment targets. Moreover, they allowed the identification of new cell populations which were also overexpressed by cancer patients compared to healthy subjects. We further analyzed the consensus of multiple reviewers in manually scoring cells as a “CTC” or “no CTC”. Their significant disagreement motivated the development of an automated learning algorithm to reproducibly classify CTCs and other cell populations of interest. Therefore, we used a convolutional neural network with an architecture based on autoencoders and combined it with an advanced visualization of the low-dimensional latent space. The chosen network allows to not only classify cells but to reveal new cell populations and subclasses of known populations. The work presented in this thesis resulted in the development of an open-source software for CTC analysis called ACCEPT (Automated CTC Classification, Enumeration and PhenoTyping). The toolbox facilitates and automates the process of detecting and classifying CTCs and other cell populations in fluorescent images.