FIGURE SUMMARY
Title

A novel deep learning-based 3D cell segmentation framework for future image-based disease detection

Authors
Wang, A., Zhang, Q., Han, Y., Megason, S., Hormoz, S., Mosaliganti, K.R., Lam, J.C.K., Li, V.O.K.
Source
Full text @ Sci. Rep.
Figure 1

3DCellSeg: A two-stage light-weight, fast, and robust pipeline for 3D cell segmentation. [Note: There are two stages in the pipeline. The first stage is a semantic segmentation, where the input is a 3D cell membrane image and the output consists of three masks, which indicate whether a voxel is the cell foreground, membrane, or background. The second stage is an instance segmentation performed on the basis of these three masks. The cellular images and segmentation results were generated by Python Matplotlib (https://matplotlib.org) using the HMS dataset].
Figure 2

Model comparison and representative slices. [Note (a) shows the accuracies of different cell segmentation models for the HMS dataset. 3DCellSeg achieves the second best accuracy in ARE, VOIsplit, and VOImerge, and achieves the best accuracy in Avg JI, JI > 70%, and JI > 50% (the plots for DSC-related metrics are of high similarity to JI-related metrics). (b) and (c) show representative slices of different model segments. ACME tends to under-segment (see the dark green region which mis-classifies different cells as one cell) while U-Net + SWS tends to over-segment (see the over-segmented small cells in the central region). PanopticFCN, Mask R-CNN FPN, and Mask R-CNN C4 are accurate on the HMS dataset but they are severely under-segment on the ATAS dataset. The cellular images in (b) and (c) were generated by Python Matplotlib (https://matplotlib.org) using the HMS and ATAS49 datasets].
Figure 3

3DCellSeg performance on the ATAS, LRP, and Ovules datasets. [Note: Different cell instances were randomly assigned different colors. The LRP dataset images are annotated: the yellow circle shows where 3DCellSeg has made a mistake and the green circle shows that 3DCellSeg can segment cells that were not labelled in the ground truth. The cellular images were generated by Python Matplotlib (https://matplotlib.org) using the ATAS49, LRP50, and Ovules51 datasets].
Figure 4

The structure of 3DCellSegNet. [Note: The extra voxels on the edge of the feature maps are removed after each deconvolution operation, in order to ensure the size of the up-sampled feature map is identical with that of the corresponding down-sampled feature map].
Figure 6

TASCAN algorithm for cell clustering.
Acknowledgments
This image is the copyrighted work of the attributed author or publisher, and ZFIN has permission only to display this image to its users. Additional permissions should be obtained from the applicable author or publisher of the image. Full text @ Sci. Rep.
Your Input Welcome
We welcome your input and comments. Please use this form to recommend updates to the information in ZFIN. We appreciate as much detail as possible and references as appropriate. We will review your comments promptly.
Please check the highlighted fields and try again.
Thank you for submitting comments. Your input has been emailed to ZFIN curators who may contact you if additional information is required.
Oops. Something went wrong. Please try again later.