CurveAlign (Download Instructions) is a quantitative tool for interpreting the regional interaction between collagen and tumors by assessment of up to thirty-four stromal fiber features, including angle, alignment, and density (Windows64 APP package download, Mac APP package download).
CurveAlign can incorporate fiber extraction from CT-FIRE, another open source tool best suited for individual fiber analysis and can use a machine learning Support Vector Machine (SVM) approach to rank disease-associated features. CurveAlign retains features of v2.3, including statistics for relative angles and heat maps to show the fiber angle alignment and fiber-boundary association.
When do I use CurveAlign and when do I use CT-FIRE?
CurveAlign quantifies all fiber angles within a region of interest relative to a user-defined boundary. CT-FIRE analyzes individual fiber metrics such as length, width, angle, and curvature. Besides the relative angle quantification, the newest version of CurveAlign can be used to extract other collagen fiber features, such as localized fiber density, fiber alignment, and the spatial relationship between fiber and the defined boundary. In addition, the extracted individual fibers extracted by CT-FIRE can be imported into the CurveAlign for the feature extraction mentioned above. For now, CurveAlign should be used for bulk assessment of collagen features including angles/density and CT-FIRE for individual fiber quantification.
CurveAlign was developed to search for stromal changes correlated with disease in images of collagen and epithelial cells. Collagen images may be obtained by a number of imaging approaches, however we have focused here on Second Harmonic Generation (SHG) images of collagen. Epithelial cell information is input into CurveAlign as an 8-bit mask file, that must be pre-registered with the collagen image, where white pixels correspond to epithelial cell regions and black pixels correspond to anything else in the image, i.e. the background, collagen fibers etc. These mask files can be generated by any appropriate means, such as manual ROI annotation in FIJI or using segmentation tools in MATLAB or FIJI. The output features can then be used to potentially classify images or fibers using machine learning techniques. A SVM approach is incorporated into CurveAlign for ranking extracted features, and these features can also be used or ranked through many open source machine learning tools such as Weka and R.
The primary change in CurveAlign version 3.0 is feature extraction from CT-FIRE output. Compared to the v2.3, the major updates are: (1) output up to thirty four fiber features, including angle, alignment, and density; features are saved as both .csv files and .mat files, for potential image or fiber classification using machine learning techniques; (2) tiff boundary can be loaded to investigate the interaction between fiber and cell; (3) CT-FIRE fiber extraction results can be read in; (4) a feature ranking using SVM approach is included; and (5) keep the most features of the older version (2.3), such as the statistics for the relative angles, heat map to show the fiber(angle) alignment, and the fiber-boundary association, etc.
|Jeremy Bredfeldt||Carolyn Pehlke|