Skip to main content
. 2021 Feb 17;7(8):eabe2480. doi: 10.1126/sciadv.abe2480

Fig. 2. CLS segmentation with Mask R-CNN algorithm.

Fig. 2

(A) Workflow of Mask R-CNN for CLS detection and segmentation. ROI, region of interest. (B) Free-response receiver operating characteristic (FROC) curve for testing results with Mask R-CNN with an average number of 100 ground truth CLSs. (C and D) Typical 3D macrophage-labeled light sheet images and respective segmentation images for adipose tissue from (C) obese and (D) lean mice. All images share the same scale bar of 200 μm. (E and F) Distributions and four-component Gaussian fits for CLSs in perirenal adipose tissues in (E) obese and (F) lean mice. Components are given in the legends together with fractional contribution to the total counts in parentheses. (G) CLS and (H) cell number densities across adipose tissue in both obese and lean mice (n = 4). Asterisks indicate the following: *P = 0.0143 and **P = 0.0034, Student’s t test.