Introduction:: Collagen fiber organization is known to change drastically under stresses induced by the cancer tumor microenvironment.1 These changes could have prognostic potential and help unlock a deeper understanding of cancer development and metastasis. Second Harmonic Generation (SHG) is a two-photon microscopy methodology that allows for label-free imaging of collagen.2 Multiple studies have observed anisotropic collagen fiber signatures in the tumor microenvironment from SHG images.2 One computational analysis technique recently adapted for SHG collagen images, the 2D Wavelet Transform Modulus Maxima (WTMM) Anisotropy Method, has demonstrated success in both melanoma mouse models and H&E biopsy slides from human pancreatic ductal adenocarcinoma patients.3,4 The continuous wavelet transform allows for the mathematical differentiation of directional preferences of collagen signatures at different micron scales. While 2D imaging and analysis has found anisotropic signatures in the tumor microenvironment, a deeper study is warranted to characterize the 3D environment with anisotropy existing along multiple dimensional planes. SHG microscopy can generate 3D images consisting of multiple slices due to its localized two-photon excitation which are more biologically relevant for understanding changes in collagen organization. Analysis techniques capable of characterizing collagen anisotropy from multiple angles and anisotropy in general will serve to benefit researchers and clinicians as more 3D imaging technologies are developed. This work adapts the 2D WTMM Anisotropy Method into a 3D technique using orthogonal projections of 3D images. The method is applied to the projections of simulated ellipsoidal fibers, calculating the anisotropy factor, and reporting preferential rotation angles if sufficiently anisotropic.
Materials and Methods:: Fibers were modelled using extremely eccentric ellipsoids with a long axis of 128 pixels and two short axes of four pixels. Fiber centroids were randomly positioned within the central 128x128x128 subregion of an 8-bit, 256x256x256-pixel image cube. Each fiber was oriented at an angle between 0 and 180 degrees from the X-axis towards the Y-axis, known as theta, and an angle between 0 and 180 degrees, known as phi, from the Z-axis towards the X-axis. Voxel values were set to 255 where a fiber exists and zero elsewhere. Three experimental sets of faux fiber cubes, each with 20 fibers, were generated for development and testing of the 3D method (Table 1). The first set had four categories, each with n=25 cubes: anisotropic (both angles phi and theta range from 85-95o), slightly anisotropic (60-120o), slightly isotropic (30-150o), and isotropic (0-180o). The second set of cubes followed the same methodology, but phi was always fixed at zero. The third set of cubes followed the same methodology, but theta was fixed at zero. Orthogonal max intensity projections were obtained along each set of axis pairs (XY, XZ, YZ) for each cube and then the 2D Wavelet Transform Modulus Maxima (WTMM) Anisotropy Method was applied. Theta and phi, as well as their respective anisotropy factors, were then calculated for each cube from the results of the 2D WTMM Anisotropy Method.
Results, Conclusions, and Discussions:: The 2D WTMM Anisotropy Method is a multiscale technique that calculates an anisotropy factor (higher values correspond to higher directional preference) at different wavelet size scales. An example output for a cube from set one's slight anisotropy category (i.e. with phi and theta being in the range 60-120o) is shown in Figure 1. An example for each of the four categories from set two is shown in Figure 2, where the calculated angles from cube axis projections fall within the ranges of the allowed fiber angles. To categorize the level of anisotropy, the mean anisotropy factor of all three views is plotted on the Y-axis and the Z-projections anisotropy subtracted from the X-projection, and Y-projection’s combined anisotropy is plotted on the X-axis. This produced Figure 3 where clear clustering of the different sources of anisotropy can be found. The slight anisotropy and isotropy data from sets two and three are added in Figure 4 with set one’s slight datasets removed for sake of graphical clarity. The clustering is less clear in Figure 4 as set two and set three have overlapping values between anisotropy with slight anisotropy categories. Therefore, determining conclusive thresholds that incorporate slight changes in angle range will need to be further explored. The isotropic datasets are still clearly separated from the anisotropic datasets which means theta and phi can be reported confidently from the anisotropy factor. The 2D WTMM Anisotropy Method applied on orthogonal projections of 3D faux fibers can determine if an image cube is anisotropic or not while reporting accurate angles. When one of the two rotational angles is fixed (anisotropic) and the other angle is allowed to vary, the nuances in anisotropy become harder to differentiate and both angles tend to be reported, but with high variability for one. Future work includes exploring other metrics to use from the 2D WTMM to create a more robust differentiation, running this method on datasets beyond just fibers, and using this tool on real SHG collagen images.
Acknowledgements (Optional): :
References (Optional): : 1. Xu, S., Xu, H., Wang, W. et al. The role of collagen in cancer: from bench to bedside. J Transl Med 17, 309 (2019). doi: 10.1186/s12967-019-2058-1
2. Natal R de A, Adur J, Cesar CL, Vassallo J. Tumor extracellular matrix: lessons from the second-harmonic generation microscopy. Surg Exp Pathol (2021) 4:1–8. doi: 10.1186/s42047-021-00089-0
3. Hamilton J, Breggia A, Fitzgerald TL, Jones MA, Brooks PC, Tilbury K and Khalil A. Multiscale anisotropy analysis of second-harmonic generation collagen imaging of human pancreatic cancer. Front. Oncol. 12:991850. (2022) doi: 10.3389/fonc.2022.991850
4. Tilbury K, Han X, Brooks PC, Khalil A. Multiscale anisotropy analysis of second-harmonic generation collagen imaging of mouse skin. J BioMed Opt (2021) 26:1–19. doi: 10.1117/1.jbo.26.6.065002