CellPose vs. CellProfiler segmentation comparison

Apr 18, 2023 | Knowledge

Image segmentation is an essential process and one of the first challenges in analyzing images of complicated cellular phenotypes at the single cell level. Accurate and efficient segmentation is crucial for extracting meaningful information from images and analyzing biological processes. In recent years, several software tools have been developed to automate the segmentation process, including CellPose and CellProfiler.

CellPose is a deep learning-based tool that can accurately segment individual cells from complex images, including those with multiple channels, fluorescence, and high-density images. Its human-in-the-loop system allows you to manually correct and refine segmentation results, making it particularly useful for complex images with overlapping or irregularly shaped cells. This system allows you to review and adjust segmentation results manually, improving the accuracy of the output.

Figure 1. Manual correction of the segmentation in CellPose.

CellProfiler, on the other hand, is a modular, open-source software platform that can be used to analyze a variety of biological images, including microscopy, histology, and flow cytometry data. CellProfiler allows you to create custom pipelines for image segmentation, analysis, and visualization.

CellPose is best suited for fluorescence microscopy images, including those with overlapping and densely packed cells. Additionally, CellProfiler can handle a variety of image types, including brightfield, phase contrast, and fluorescent images, as well as multi-channel and time-lapse images.

Both CellPose and CellProfiler have user-friendly interfaces, but they have different approaches to segmentation. CellPose has a simple, one-click segmentation process that allows you to easily segment cells in complex images. CellProfiler however, requires more advanced knowledge of image analysis and offers a more customizable approach to segmentation, allowing you to create custom pipelines that can handle specific image types and segmentation tasks.

We evaluated the performance of two cell segmentation programs by comparing the nuclei segmentation results of MCF-7 and U2OS cells. Both CellPose and CellProfiler performed well in segmenting nuclei, with similar numbers of correctly segmented nuclei. However, we observed differences in the number of regions of interest (ROIs) identified by the two programs (Figure 2 and Figure 3). This difference can be attributed to CellPose discarding overexposed cells that may be too bright for downstream analysis, while CellProfiler includes them. Despite these differences, both programs achieved great nuclei segmentation results overall.

Figure 2. Nuclei segmentation comparison of MCF-7 cells.

Figure 3. Nuclei segmentation comparison of U2OS cells.

Our comparison of the whole cell segmentation found that both CellPose and CellProfiler performed well in segmenting the cells. However, we observed that in areas where cells are densely packed, CellPose excelled at accurately segmenting individual cells, while CellProfiler tended to fill in the gaps between cells. This difference in performance may be attributed to the deep learning-based approach used by CellPose, which can better handle complex images with overlapping cells, while CellProfiler’s modular approach may require more fine-tuning to handle these types of images.

Figure 4. An example of segmentation comparison between CellPose and CellProfiler.

We also evaluated a CellPose plugin for CellProfiler, called RunCellPose, which performed similarly to the standalone CellPose software in cell segmentation. However, we noted that the plugin does not automatically connect to different objects in the CellProfiler pipeline, which can affect downstream processing and analysis. Custom Python functions may be required to overcome this limitation and integrate the plugin effectively into a CellProfiler workflow. Despite this limitation, RunCellPose provides a convenient way to use CellPose within the CellProfiler framework, allowing you to combine the strengths of both programs in their image analysis workflows.

Overall, both CellPose and CellProfiler offer powerful tools for cell segmentation, and you should choose the program that best fits your specific needs and the complexity of your images. The unique features of each program, such as CellPose’s deep learning-based approach and human-in-the-loop system, and CellProfiler’s modular design and compatibility with other image analysis tools, provide valuable options for cell segmentation in scientific research.