There has been increased interest in utilizing non-invasive “liquid biopsies” to identify biomarkers for cancer prognosis and monitoring, and to isolate genetic material that can predict response to targeted therapies. Circulating tumor cells (CTCs) have emerged as such a biomarker providing both genetic and phenotypic information about tumor evolution, potentially from both primary and metastatic sites. Currently, available CTC isolation approaches, including immunoaffinity and size-based filtration, have focused on high capture efficiency but with lower purity and often long and manual sample preparation, which limits the use of captured CTCs for downstream analyses. Here, we describe the use of the microfluidic Vortex Chip for size-based isolation of CTCs from 22 patients with advanced prostate cancer and, from an enumeration study on 18 of these patients, find that we can capture CTCs with high purity (from 1.74 to 37.59%) and efficiency (from 1.88 to 93.75 CTCs/7.5 mL) in less than 1 h. Interestingly, more atypical large circulating cells were identified in five age-matched healthy donors (46–77 years old; 1.25–2.50 CTCs/7.5 mL) than in five healthy donors <30 years old (21–27 years old; 0.00 CTC/7.5 mL). Using a threshold calculated from the five age-matched healthy donors (3.37 CTCs/mL), we identified CTCs in 80% of the prostate cancer patients. We also found that a fraction of the cells collected (11.5%) did not express epithelial prostate markers (cytokeratin and/or prostate-specific antigen) and that some instead expressed markers of epithelial–mesenchymal transition, i.e., vimentin and N-cadherin. We also show that the purity and DNA yield of isolated cells is amenable to targeted amplification and next-generation sequencing, without whole genome amplification, identifying unique mutations in 10 of 15 samples and 0 of 4 healthy samples.

Deep learning is transforming the analysis of biological images but applying these models to large datasets remains challenging. Here we describe the DeepCell Kiosk, cloud-native software that dynamically scales deep learning workflows to accommodate large imaging datasets. To demonstrate the scalability and affordability of this software, we identified cell nuclei in 10 6 1-megapixel images in ~5.5 h for ~$250, with a sub-$100 cost achievable depending on cluster configuration. The DeepCell Kiosk can be downloaded at https://github.com/vanvalenlab/kiosk-console; a persistent deployment is available at https://deepcell.org. Main Text While deep learning is an increasingly popular approach to extracting quantitative information from biological images, its limitations significantly hinder its widespread adoption. Chief among these limitations are the requirements for expansive sets of training data and computational resources. Here, we sought to overcome the latter limitation. While deep learning methods have remarkable accuracy for a range of image-analysis tasks including classification 1 , segmentation 2-4 , and object tracking 5,6 , they have limited throughput even with GPU acceleration. For example, even when running segmentation models on a GPU, typical inference speeds on megapixel-scale images are in the range of 5-10 frames per second, limiting the scope of analyses that can be performed on images in a timely fashion. The necessary domain knowledge and associated costs of GPUs pose further barriers to entry, although recent software packages 7-11 have attempted to solve these two issues. While cloud computing has proven effective for other data types 12-15 , scaling analyses to large imaging datasets in the cloud while constraining costs is a considerable challenge. To meet this need, here we have developed the DeepCell Kiosk (Fig. 1a). This software package takes in configuration details (user authentication, GPU type, etc.) and creates a cluster in the cloud that runs predefined deep learningenabled image-analysis pipelines. This cluster is managed by Kubernetes, an open-source framework for running software containers (software that is bundled with its dependencies so it can be run as an isolated process) across a group of servers. An alternative way to view Kubernetes is as an operating system for cloud computing. Data is submitted to the cluster through either a web-based front-end, a command line tool, or an ImageJ plugin. Once submitted, it is placed in a database where the specified image-analysis pipeline can pick up the dataset, perform the desired analysis, and make the results available for download. Results can be visualized by a variety of visualization software tools 16,17. To ensure that image-analysis pipelines can be run efficiently on this cluster, we made two software design choices. First, image-analysis pipelines access trained deep learning models through a centralized model server in the cluster. This strategy enables the cluster to efficiently allocate resources, as the various co...

Live-cell imaging experiments have opened an exciting window into the behavior of living systems. While these experiments can produce rich data, the computational analysis of these datasets is challenging. Single-cell analysis requires that cells be accurately identified in each image and subsequently tracked over time. Increasingly, deep learning is being used to interpret microscopy image with single cell resolution. In this work, we apply deep learning to the problem of tracking single cells in live-cell imaging data. Using crowdsourcing and a human-in-the-loop approach to data annotation, we constructed a dataset of over 11,000 trajectories of cell nuclei that includes lineage information. Using this dataset, we successfully trained a deep learning model to perform cell tracking within a linear programming framework. Benchmarking tests demonstrate that our method achieves state-of-the-art performance on the task of cell tracking with respect to multiple accuracy metrics. Further, we show that our deep learning-based method generalizes to perform cell tracking for both fluorescent and brightfield images of the cell cytoplasm, despite having never been trained on those data types. This enables analysis of live-cell imaging data collected across imaging modalities. A persistent cloud deployment of our cell tracker is available at http://www.deepcell.org.

Circulating tumor cells (CTCs) have a great potential as indicators of metastatic disease that may help physicians improve cancer prognostication, treatment and patient outcomes. Heterogeneous marker expression as well as the complexity of current antibody-based isolation and analysis systems highlights the need for alternative methods. In this work, we use a microfluidic Vortex device that can selectively isolate potential tumor cells from blood independent of cell surface expression. This system was adapted to interface with three protein-marker-free analysis techniques: (i) an in-flow automated image processing system to enumerate cells released, (ii) cytological analysis using Papanicolaou (Pap) staining and (iii) fluorescence in situ hybridization (FISH) targeting the ALK rearrangement. In-flow counting enables a rapid assessment of the cancer-associated large circulating cells in a sample within minutes to determine whether standard downstream assays such as cytological and cytogenetic analyses that are more time consuming and costly are warranted. Using our platform integrated with these workflows, we analyzed 32 non-small cell lung cancer (NSCLC) and 22 breast cancer patient samples, yielding 60 to 100% of the cancer patients with a cell count over the healthy threshold, depending on the detection method used: respectively 77.8% for automated, 60–100% for cytology, and 80% for immunostaining based enumeration.

Extraction of rare target cells from biosamples is enabling for life science research. Traditional rare cell separation techniques, such as magnetic activated cell sorting (MACS), are robust but perform coarse, qualitative separations based on surface antigen expression. We report a quantitative magnetic separation technology using high-force magnetic ratcheting over arrays of magnetically soft micro-pillars with gradient spacing, and use the system to separate and concentrate magnetic beads based on iron oxide content (IOC) and cells based on surface expression. The system consists of a microchip of permalloy micro-pillar arrays with increasing lateral pitch and a mechatronic device to generate a cycling magnetic-field. Particles with higher IOC separate and equilibrate along the miro-pillar array at larger pitches. We develop a semi-analytical model that predicts behavior for particles and cells. Using the system, LNCaP cells were separated based on the bound quantity of 1μm anti-EpCAM particles as a metric for expression. The ratcheting cytometry system was able to resolve a ±13 bound particle differential, successfully distinguishing LNCaP from PC3 populations based on EpCAM expression, correlating with flow cytometry analysis. As a proof of concept, EpCAM-labeled cells from patient blood were isolated with 74% purity, demonstrating potential towards a quantitative magnetic separation instrument.