FogBank: a single cell segmentation across multiple cell lines and image modalities

Joe Chalfoun 0 Michael Majurski 0 Alden Dima 0 Christina Stuelten 1 Adele Peskin 0 Mary Brady 0 0 Information Technology Laboratory, National Institute of Standards and Technology , Gaithersburg, MD , USA 1 Laboratory of Cellular and Molecular Biology, National Cancer Institute, National Institutes of Health , Bethesda, MD , USA Background: Many cell lines currently used in medical research, such as cancer cells or stem cells, grow in confluent sheets or colonies. The biology of individual cells provide valuable information, thus the separation of touching cells in these microscopy images is critical for counting, identification and measurement of individual cells. Over-segmentation of single cells continues to be a major problem for methods based on morphological watershed due to the high level of noise in microscopy cell images. There is a need for a new segmentation method that is robust over a wide variety of biological images and can accurately separate individual cells even in challenging datasets such as confluent sheets or colonies. Results: We present a new automated segmentation method called FogBank that accurately separates cells when confluent and touching each other. This technique is successfully applied to phase contrast, bright field, fluorescence microscopy and binary images. The method is based on morphological watershed principles with two new features to improve accuracy and minimize over-segmentation. First, FogBank uses histogram binning to quantize pixel intensities which minimizes the image noise that causes over-segmentation. Second, FogBank uses a geodesic distance mask derived from raw images to detect the shapes of individual cells, in contrast to the more linear cell edges that other watershed-like algorithms produce. We evaluated the segmentation accuracy against manually segmented datasets using two metrics. FogBank achieved segmentation accuracy on the order of 0.75 (1 being a perfect match). We compared our method with other available segmentation techniques in term of achieved performance over the reference data sets. FogBank outperformed all related algorithms. The accuracy has also been visually verified on data sets with 14 cell lines across 3 imaging modalities leading to 876 segmentation evaluation images. Conclusions: FogBank produces single cell segmentation from confluent cell sheets with high accuracy. It can be applied to microscopy images of multiple cell lines and a variety of imaging modalities. The code for the segmentation method is available as open-source and includes a Graphical User Interface for user friendly execution. - Background Many cell lines that are currently being studied for medical purposes, such as cancer cell lines, grow in confluent sheets. These cell sheets typically exhibit cell line specific biological properties such as the morphology of the sheet, protein expression, proliferation rate, and invasive/metastatic potential. However, cell sheets are comprised of cells of different phenotypes. For example, individual cells in a sheet can have diverse migration patterns, cell shapes, can express different proteins, or differentiate differently. Identifying phenotypes of individual cells is highly desirable, as it will contribute to our understanding of biological phenomena of tumor metastasis, stem cell differentiation, or cell plasticity. Time-lapse microscopy now enables the observation of cell cultures over extended time periods and at high spatiotemporal resolution. Furthermore, it is now possible not only to label cells with fluorescent markers, but also to express fluorescently labeled protein, enabling spatiotemporal analysis of protein distribution in a cell sheet at a cellular level. To assess properties of individual cells within the observed sheet, however, it is necessary to accurately track these cells in a fully automated fashion. Thus, one of the requirements of an automated image analysis method is high accuracy single cell segmentation for individual time steps and its applicability to a wide range of cell types. Additionally, it is preferred that the developed method can analyze a multitude of image types, for example, phase contrast, differential interference contrast, and fluorescence images, as they are typically obtained in biomedical science. Segmentation methods based on morphological watersheds are used for object separation and appear throughout the image processing and analysis literature and patents, since the method was first applied to image segmentation [1]. Most watershed methods work by dividing the image surface into regions based on pixel intensity gradient contours. However, the high level of noise in biological images leads to over-segmentation - a major problem when morphological watersheds are used [2-5]. This noise creates small minima across the regions of interest in an image, and gives rise to numerous small segmented regions that do not have biological significance. Therefore, a new segmentation method that accurately separates confluent cells into single cells for a wide range of applications is needed. In general, watershed regions are formed either by a flooding process, expanding out from gradient minima, or by a watershed transform which computes a direct solution. Either of these methods can include the entire image, or begin from user-defined seed points. For flooding techniques, typically the regions are flooded according to intensity levels, through an immersion simulation [6] creating a topographic surface. Automatic minima detection can occur, for example, from low frequency components in the morphological gradient of an image [7]. Distance transforms can also be used for watershed segmentation, flooded from localized distance maxima [8]. Traditional watershed flooding by gradient level has been improved by adding local neighborhood comparisons and geodesic distance checking as the flooding occurs [9]. Gradient vector flow (GVF) [10], a diffusion of the classical gradient, has been used to give more weight to important feature edges. The viscous watershed technique [11] simulates flooding on a filtered relief of the image. More user-dependent methods extract regions through selected localized watershed flooding [12]. A variety of different watershed transforms are available, dating back from Meyer's watershed transform, which uses topographic distance to solve a shortest path function [11]. The Image Foresting Transform (IFT) [13] transforms an image into a weighted graph, in which each pixel is represented by a node in the graph. Cost functions are calculated for all possible paths within the graph to find the optimal region separation. The Tie-Zone Watershed (TZWS) transform [14] is derived from the IFT transform, and defines tie-zones, where regions overlap and the forests could produce multiple solutions, and defines unique optimal partitions between regions. Defining an energy minimization function to partition regions [15] more effi (...truncated)