Journal: Frontiers in Bioengineering and Biotechnology

Article Title: Fibrillar Collagen Quantification With Curvelet Transform Based Computational Methods

doi: 10.3389/fbioe.2020.00198

Figure Lengend Snippet: Collagen orientation calculation on SHG images of human pancreatic tissue with CurveAlign (CA) and its comparison with manual measurement and two other open-source tools including structure tenor based OrientationJ (OJ) and Fourier transform-based CytoSpectre (CS). (A–D) show the original SHG images and the location of eight ROIs. (E–H) show the manually labeled fibers (in magenta) overlaid on the original images of ROIs 1, 3, 5, and 7, respectively. (I–L) show curvelets (in green color) represented orientations on the original images of ROIs 1, 3, 5 and 7, respectively. The bottom row shows the comparison in orientation (left) and alignment (right). For the ROIs with relative larger alignment coefficient including ROIs 3–8, all the methods yield similar orientation. For ROIs 1–2 with relative small alignment, the discrepancy becomes bigger, with the CurveAlign Curvelets analysis mode (CA-CT) being closest to the manual measurement. All the methods share some similar trends in alignment measurement, with the CurveAlign fiber segments analysis mode (CA-CTF) being closest to the manual measurement. The differences are mainly due to the different fiber orientation detection algorithm and fiber alignment definition. Images (A–D) are shown at the same scale while images (E–L) are shown at the same scale. Scale bar equals 50 microns.

Article Snippet: Moreover, we have developed a prototype of KNIME ( ) nodes for CurveAlign and CT-FIRE for a convenient feature statistical analysis, classification, and visualization.

Techniques: Comparison, Labeling

Journal: Frontiers in Bioengineering and Biotechnology

Article Title: Fibrillar Collagen Quantification With Curvelet Transform Based Computational Methods

doi: 10.3389/fbioe.2020.00198

Figure Lengend Snippet: An example of using CurveAlign to create tumor boundary from a bright-field image and quantify the relative angle on a breast cancer TMA core. (A) The SHG image, (B) Original bright-field image, (C) SHG image (in yellow) overlaid on top of the registered bright-field image. (D) Segmentation generated boundary mask. (E) Heatmap of the relative angle with red color shows the region where one or more fibers have an angle larger than 60°. (F) CurveAlign overlay image with the lines indicating the fiber center locations and orientations overlaid on the top of the SHG image; the outline of the tumor boundary is highlighted in yellow, and the two rectangular ROIs are in magenta; a blue line is used to associate the center of fiber with the corresponding boundary locations, and the red lines indicate the fibers located either beyond the distance range or inside the boundary, and green lines are the fibers of interest. (G) and (H) Zoomed-in ROI results. The arrows in these images show that the fibers either more parallelly aligned in (G) with respect to the boundary or are more perpendicularly aligned in (H) with respect to the tumor boundary. In boundary creation, pixel per micron ratio was set to 1.65; fiber extraction in CT-FIRE used default settings; the distance parameter was set to 150 pixels; the SHG image size is 2048 by 2048 pixels, and the sizes of ROI are identically set to 256 by 256 pixels. Images (A,C–F) are shown at the same scale while images (G) and (H) are shown at the same scale. Scale bar equals 100 microns.

Article Snippet: Moreover, we have developed a prototype of KNIME ( ) nodes for CurveAlign and CT-FIRE for a convenient feature statistical analysis, classification, and visualization.

Techniques: Generated, Extraction

Journal: Frontiers in Bioengineering and Biotechnology

Article Title: Fibrillar Collagen Quantification With Curvelet Transform Based Computational Methods

doi: 10.3389/fbioe.2020.00198

Figure Lengend Snippet: Validation of individual fiber tracking and orientation detection. (A) shows the 30 straight fibers with given fiber properties. (B) CT-FIRE individual fiber overlay image of (A) . (C) CurveAlign CT-curvelets overlay image of (A) . (E) shows the 30 curvy fibers with given fiber properties. (F) CT-FIRE individual fiber overlay image of (E) ; CurveAlign CT-curvelets overlay image of (A) . Both CT-FIRE fibers [color lines in (B) and (F) ] and curvelets [green lines with red center point in (C) and (G) ] are mostly faithfully overlaid on the actual fiber directions. The box plots of average orientation and alignment of the 100 synthetic images in each dataset are shown in (D) and (H) , respectively. The box plots (D) and (H) of the 100 synthetic images in each dataset show that the calculated orientations and alignment are close to the setting parameters. In the boxplot, the red line represents the median; the blue lines represent the 25th and 75th percentiles, respectively; the dashed lines and black lines indicate the lower and upper limits of the data points that are not considered as outliers; and the red crosses represent outliers. The t test shows that all the analysis modes yield the same mean values as the control values for both orientation and alignment at the 5% significance level except for the alignment calculated under the CT mode. CTF-S, CT-FIRE individual fiber mode for straight fiber images; CT-S, CT-curvelets mode for straight fiber images; CTF-C, CT-FIRE individual fiber mode for curvy fiber images; CT-C, CT-curvelets mode for curvy fiber images.

Article Snippet: Moreover, we have developed a prototype of KNIME ( ) nodes for CurveAlign and CT-FIRE for a convenient feature statistical analysis, classification, and visualization.

Techniques: Biomarker Discovery, Control