Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

< Back to Article

Figure 1.

Acquisition and image processing of confocal images.

Organotypic culture of E13 SMGs (a) control or (b) treated with ROCK inhibitor (140 µM Y27632), showing reduced branching with ROCK inhibitor treatment. Explants were immunostained with anti-E-cadherin antibody as an epithelial marker (red) and SYBR green as a total nuclei marker (green). Multiple overlapping confocal images through the mid-section of (c) control- and (d) ROCK inhibitor-treated explants were captured to cover the whole explant. Images were stitched using the inverse Fourier transform of the phase correlation matrix and blended to provide composite images of (e) control (f) and ROCK inhibitor treated explants. Scale bars: 200 µm (a, b), 100 µm (c), (d), and (e), and (f). In our study, the sublingual tissues were discarded and only the submandibilar gland was used, (Figure S2).

More »

Figure 1 Expand

Figure 2.

Generation of Cell Graphs.

Stitched images were segmented using the active contour method to define epithelial (white) vs mesenchymal tissue (black) in control (a) and ROCK inhibitor-treated explants (d). These masks were used to identify the epithelial nuclei (b, e) and mesenchymal nuclei (c, f). Using each nucleus as a vertex, cell-graphs were constructed for control and ROCK inhibitor-treated tissues, respectively (g, h), where zoomed regions of cell graphs corresponding to regions of the original images (shown as red boxes in a and d) are shown in detail. Epithelial tissue is respresented by the blue graph and the mesenchymal tissue is represented by the red graph. We discarded the sublingual tissues and only used the submandibilar gland, (Figure S2).

More »

Figure 2 Expand

Table 1.

Global structural features.

More »

Table 1 Expand

Table 2.

Spectral features.

More »

Table 2 Expand

Table 3.

Local structural features.

More »

Table 3 Expand

Table 4.

Morphological (shape based) features.

More »

Table 4 Expand

Figure 3.

Direct validations of cell-graph features using standard image analysis methods.

Plots of (a) area, (b) perimeter and (c) circularity from images using conventional image analysis methods and plots of cell-graph-derived raw data pertaining to (d) area, (e) perimeter and (f) circularity are shown. Control refers to untreated epithelium and Y27632 refers to the ROCK inhibitor treatment. The same trends for control vs ROCK inhibitor treatment were observed for the features obtained using image analysis and cell-graphs. The percent differences between the conventional image analysis and our image segmentation technique are found to be 1.16% and 0.73% for the area; 5.66% and 5.94% for the perimeter; 11.0532 and 16.1463 for the circularity of the control and ROCK inhibitor-treated samples, respectively.

More »

Figure 3 Expand

Figure 4.

Indirect validations of cell-graph features using standard image analysis methods.

Control refers to untreated epithelium and Y27632 treated epithelium. (a) Diameter of explants was measured using MetaMorph image analysis tools from single confocal images (b) Total nuclei were measured from single confocal images. (c) Thickness was measured from confocal Z-stacks of images. With Y27632 treatment, diameter of the explants increases and thickness and number of cells decreases thus reducing the overall compactness of the tissue structure. Cell-graph-derived features, such as clustering coefficient (d), average path length (e) and number of connected components (f) show that Y27632 treatment increases the distance between two cells, thereby lowering the number of linked cells and decreasing the overall compactness in the epithelial and mesenchymal regions.

More »

Figure 4 Expand

Figure 5.

Bipartite graph analysis.

The changes in the correlation clusters of the four tissue samples are studied through bi-partite graph analysis for the untreated vs. treated epithelial tissue comparison in (a) and for the untreated vs. treated mesenchymal tissue comparison in (b).

More »

Figure 5 Expand

Figure 6.

Geometric interpretation of changes in cell-graph features.

A geometrical understanding of example cell-graph features is provided together with corresponding representative tissue samples. Geometrical interpretations of the changes for the example features are studied.

More »

Figure 6 Expand

Figure 7.

Illustration of a Tucker3 model for tensor analysis.

and indicate the number of components extracted from the first, second and third mode (), respectively, and and are the component matrices. is the core tensor and represents the error term.

More »

Figure 7 Expand

Figure 8.

Multiway modeling by tensor analysis.

Our dataset is modeled as a higher order array to capture the multilinear structures. (a) Tissue type analysis reveals that the untreated epithelial, untreated mesenchymal and treated mesenchymal tissues are grouped together. (b) Hotelling's T2 versus sum squared residuals to reveals features that the tensor analysis cannot fit with the model.

More »

Figure 8 Expand

Table 5.

Epithelial vs Mesenchymal comparison in control tissue samples.

More »

Table 5 Expand

Table 6.

Epithelial vs Mesenchymal comparison in ROCK-inhibitor-treated tissues.

More »

Table 6 Expand

Table 7.

Control vs ROCK-inhibitor-treated comparison of epithelial tissues.

More »

Table 7 Expand

Table 8.

Control vs ROCK-inhibitor-treated comparison of mesenchymal tissues.

More »

Table 8 Expand

Table 9.

Comparison of the learning accuracies using all the multi-scale features or only global graph features, spectral features, local graph features or morphological features.

More »

Table 9 Expand