Search

Additional file 3: Supplementary Table S1. Linear ACCEPT gates used for the automated enumeration of CTCs and tdEVs. Supplementary Table S2. Univariable Cox regression analyses of CK+ and CK- CTCs and tdEVs after log transformation. Supplementary Table S3. Sensitivity, specificity and accuracy of ≥ 23% CTCs and ≥ 7% tdEVs, double positive for CK and HER2, as tests to predict the HER2 status of the tissue. The accuracy increases with the total CTCs and tdEVs detected (≥ 1, 5, 10, 20, 50, 100) at the cost of number of eligible patients to be assessed.

Circulating Tumor Cell (CTC) counts, as determined by the CellSearch® system, associate with poor overall survival (OS) in castration-resistant prostate cancer (CRPC), breast, colorectal and non-small cell lung (NSCLC) cancer patients. The cut-offs used to discriminate between patients with favorable and unfavorable prognoses are 5 CTCs / 7.5 mL of blood for breast and CRPC, 3 for colorectal and 1 for NSCLC. The low numbers of CTCs frequently limit the accurate discrimination of patients into favorable or unfavorable prognosis groups. We previously reported the presence of CK+, DNA-, CD45- tumor-derived extracellular vesicles (tdEVs) in ~20x higher frequencies in CRPC patients and showed their equivalence to CTC counts in predicting clinical outcome. In this study, we explored the presence of tdEVs in the blood of metastatic CRPC, breast, colorectal, and NSCLC patients, determined the association of tdEVs with OS and whether they aid in stratifying patients with favorable CTC counts. Digitally stored CellSearch® system images obtained from previously reported studies from 190 CRPC, 450 colorectal, 179 breast and 117 NSCLC patients before the initiation of a new treatment were used to enumerate tdEVs automatically using the open-source ACCEPT software. tdEV counts highly correlated with CTC counts in all cancer types (Spearman’s Rho tests p

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.

In the field of Circulating Tumor Cell (CTC) research many new technologies are emerging to isolate CTCs. Some of them provide accompanying automated image analysis tools that present possible CTCs to the user. Others need fully manual image analysis. For all CTC isolation technologies the definition of a CTC based on the immuno-morphologic criteria is either customized to the specific platform or subjective to the user causing high interreader differences – a problem which may condemn many CTC-based clinical studies to failure. Thus, an important issue that the field is confronted with is the lack of a unified and standardized definition to classify a cellular object as a CTC. This problem is addressed within the European FP7 consortium CTCTrap and the Innovative Medicines Initiative (IMI) consortium CANCER-ID by the development of an open-source image analysis toolbox for CTC identification and enumeration. This toolbox is baptized ACCEPT (Automated CTC Classification, Enumeration and Phenotyping) and can process images generated by various CTC isolation technologies. The main software components are the Marker Characterization, the Full Detection and the Automatic Classification. The Marker Characterization tool aims at quantifying the antigens expressed by previously selected CTCs. The Full Detection tool is based on advanced mathematical methods to reliably detect all objects in the images, visualize the objects in scatter plots and enable the user to classify the cell types by the use of gates or selection of specific objects in the scatter plots or on the actual images. The Automatic Classification tool first detects all objects in the images followed by an automated classification approach that – as a result – presents found CTCs to the user. We demonstrate the effectiveness of these tools on two different datasets. The Marker Characterization tool was tested for Her2 expression on archived CTC images isolated and classified by the CellSearch system from patients with metastatic breast cancer. Investigators from three different institutes were asked to score these cells for Her2 positivity first on the images generated by the CellTracks Analyzer and afterwards using ACCEPT. We show that the improved CTC visualization provided in ACCEPT, combined with several measurements which we extract for each cell, can reduce the inter-user variability. The Full Detection and Automatic Classification tools of ACCEPT were tested on archived samples of patients with castration resistant prostate cancer processed with the CellSearch system as well as on microsieves obtained after filtration of the blood discarded by the CellSearch system. Results were compared with manually scored CTCs and showed the improvement of CTC classification by the availability of quantitative image analysis tools.