Nucleus segmentation across imaging experiments: the 2018 Data Science Bowl

Abstract

Segmenting the nuclei of cells in microscopy images is often the first step in the quantitative analysis of imaging data for biological and biomedical applications. Many bioimage analysis tools can segment nuclei in images but need to be selected and configured for every experiment. The 2018 Data Science Bowl attracted 3,891 teams worldwide to make the first attempt to build a segmentation method that could be applied to any two-dimensional light microscopy image of stained nuclei across experiments, with no human interaction. Top participants in the challenge succeeded in this task, developing deep-learning-based models that identified cell nuclei across many image types and experimental conditions without the need to manually adjust segmentation parameters. This represents an important step toward configuration-free bioimage analysis software tools.

Cellpose: a generalist algorithm for cellular segmentation

“…Note that the advantage of Cellpose compared to other models grew on these visually-distinct images, suggesting that Cellpose has better generalization per-formance. Finally, we assembled a large dataset of images of nuclei, by combining pre-segmented datasets from several previous studies, including the Data Science Bowl kaggle competition [14]. Because nuclear shapes are simpler, this dataset did not have as much variability as the dataset of cells, as illustrated by the t-SNE embedding of the segmentation styles (Figure S6a).…”

“…Recognizing this problem in the context of nuclear segmentation, a recent Data Science Bowl challenge amassed a dataset of varied images of nuclei from many different laboratories [14]. Methods trained on this dataset can generalize much more widely than those trained on data from a single lab.…”

BIAFLOWS: A collaborative framework to reproducibly deploy and benchmark bioimage analysis workflows

“…The image datasets have been selected to recapitulate common BIA analysis tasks: spot detection (2D/3D), nuclei segmentation (2D/3D), nuclei tracking (2D), landmark detection (2D), filament tracing (3D), and tissue detection (2D) in whole-slide histology images. All image datasets are imported from previously organized challenges (DIADEM [21], ISBI Cell Tracking Challenge [22], ISBI Particle Tracking Challenge [23], Kaggle Data Science Bowl 2018 [24]), created from synthetic data generators (CytoPacq [25], TREES toolbox [26], Vascusynth [27], SIMCEP [28]), or contributed by NEUBIAS members [37]. To showcase the versatility of the platform, the image analysis workflows available to process these images are running on different BIA platforms: ImageJ macros [29], Icy protocols [30], CellProfiler pipelines [31], Vaa3D plugins [32], ilastik pipelines [33], Python scripts leveraging Scikit-learn [34] for supervised learning algorithms, and Keras/PyTorch [35] [36] for deep learning.…”

Deep learning-enabled phenotyping reveals distinct patterns of neurodegeneration induced by aging and cold-shock

“…In this work, we sought to integrate cutting-edge deep learning approaches to segment beads in PVD fluorescence images from live animals (Figure 1b). Convolutional Neural Networks (CNNs) have recently shown state-of-the-art performance in image segmentation tasks across a wide range of biological and biomedical images datasets [41][42][43][44][45][46][47] . Here, we utilize Mask R-CNN 48 , a CNN model that is designed to predict binary instance masks (one mask per predicted bead object) from an image to detect PVD beads.…”