Multiscale and multi-modality visualization of angiogenesis in a human breast cancer model

Abstract

Angiogenesis in breast cancer helps fulfill the metabolic demands of the progressing tumor and plays a critical role in tumor metastasis. Therefore, various imaging modalities have been used to characterize tumor angiogenesis. While micro-CT (μCT) is a powerful tool for analyzing the tumor microvascular architecture at micron-scale resolution, magnetic resonance imaging (MRI) with its sub-millimeter resolution is useful for obtaining in vivo vascular data (e.g. tumor blood volume and vessel size index). However, integration of these microscopic and macroscopic angiogenesis data across spatial resolutions remains challenging. Here we demonstrate the feasibility of 'multiscale' angiogenesis imaging in a human breast cancer model, wherein we bridge the resolution gap between ex vivo μCT and in vivo MRI using intermediate resolution ex vivo MR microscopy (μMRI). To achieve this integration, we developed suitable vessel segmentation techniques for the ex vivo imaging data and co-registered the vascular data from all three imaging modalities. We showcase two applications of this multiscale, multi-modality imaging approach: (1) creation of co-registered maps of vascular volume from three independent imaging modalities, and (2) visualization of differences in tumor vasculature between viable and necrotic tumor regions by integrating μCT vascular data with tumor cellularity data obtained using diffusion-weighted MRI. Collectively, these results demonstrate the utility of 'mesoscopic' resolution μMRI for integrating macroscopic in vivo MRI data and microscopic μCT data. Although focused on the breast tumor xenograft vasculature, our imaging platform could be extended to include additional data types for a detailed characterization of the tumor microenvironment and computational systems biology applications.

Fig. 1

Multiscale and multi-modality imaging of tumor angiogenesis. a In vivo MRI with its range of intrinsic and extrinsic contrast mechanisms enables the assessment of in vivo angiogenic parameters, but has a spatial resolution limited to the macroscopic scale (~100 μm). c Micro-CT (μCT) allows us to visualize blood vessels directly and quantify angiogenesis down to the microscopic spatial scale (~8 μm) of a capillary. b Therefore, here we suggest that MR microscopy (μMRI) with its intermediate (~40 μm) spatial resolution is useful for integrating angiogenesis data from in vivo MRI and μCT.

Fig. 2

Illustration of vessel extraction steps for μMRI data. a Flow chart of segmentation procedure. Letters b–j indicate output images corresponding to each segmentation step. b Slice of raw T2*w μMRI data (TE = 4.9 ms). Blood vessels appear dark due to presence of waterless Microfil. c Result of the tubeness filter after combining the filter outputs using σ = 0.8, 1.0, 1.2, 1.4 with a maximum intensity arithmetic operation. d Result of C-means clustering with five classes. The blue arrow indicates class 1 and the red arrows denote classes 3 and 4, respectively. e Result of applying a conservative threshold to the output of the tubeness filter (green), overlaid on the raw μMRI data. f Class 1 from (d) shown in blue. g Result (red) of applying a lower threshold on the output of the tubeness filter and masking out regions corresponding to classes 3 and 4 in (d). h Overlay of (e), (f) and (g). Co-localization of green and red pixels shown in yellow, red and blue overlap shown in purple pixels and all three colors add up to white pixels. i Yellow structures are the combination of (e), (f) and (g) while red voxels were added by the 3D closing operation. j Final extracted blood vessels after 3D region removal of <6 connected voxels. (Color figure online)

Fig. 3

Illustration of vessel extraction steps for μCT data. a Flow chart of μCT vessel segmentation procedure. Letters b–e indicate output images corresponding to each segmentation step. b Slice of raw μCT data. Blood vessels appear bright due to presence of radio-opaque Microfil. c Result of the tubeness filter after combination of the filter outputs using σ = 1.0, 1.5 with a maximum intensity arithmetic operation. d Results (red) of the tubeness filter after undergoing a 3D mean filter and thresholding, and thresholded raw data shown in green. Yellow pixels indicate overlap of the red and green pixels. Holes in large vessels were filled by thresholding the raw data, as indicated by green pixels. e Final extracted blood vessels after 3D closing operation and isolated 3D region removal of <27 voxels in red. f Volume rendering of the segmented vasculature from a 0.5 mm tumor slice. (Color figure online)

Fig. 4

Assessment of vessel extraction and co-registration of μMRI and μCT. a Visualization of an isosurface view of the tumor vasculature extracted from the volumetric μCT data in Amira. b Fluorescence image of the breast tumor xenograft overlaid with a bright-field image of the microfilled vasculature (×4). The images were overlaid and processed in ImageJ. Arrows indicate vessels visible in both (a) and (b). c Overlay of the μMRI (green) and μCT (orange) derived blood vessels of a breast tumor xenograft after co-registration. On the right is a zoomed version wherein white arrows indicate vessels that can easily be seen in the μMRI and μCT data. Small vessels were not always visible in the μMRI image while they were resolved with μCT as indicated by red arrows, which is most apparent in the tumor rim. d Cropped view of a slice through the extracted μMRI-derived tumor blood vessels. Red arrows indicate structures that are also visible in the bright-field image in (e). e Bright-field image of a 1 mm tumor slice. Structures visible in (d) and (e) are indicated by red arrows. f Scatter plot of the μMRI FBV versus the μCT FBV for a representative breast tumor xenograft with the regression line shown in red. (Color figure online)

Fig. 5

Multi-modality blood volume measurements for tumors at different stages. Co-registered fractional blood volume (FBV) maps for a 1 mm thick tumor slice overlaid on the corresponding anatomical spin-echo images of a PIW3 tumor a–c and PIW5 tumor d–f computed using in vivo MRI, μMRI and μCT, respectively. Zero FBV values are not mapped. g, h Q–Q plots of pooled FBV values for five PIW3 (g) and three PIW5 (h) tumors illustrate the similarity of in vivo MRI (plus symbol), μMRI (open circle) FBV distributions to those of μCT. i Plots of tumor-wise median FBV versus tumor volume. j Table of mean FBV values and standard deviation for data pooled for PIW3 and PIW5 tumors

Fig. 6

Differences in vascularization between viable and necrotic tumor regions. a Smoothed in vivo ADC map of a 1 mm thick tumor slice of a PIW5 tumor wherein the tumor region is delineated in red; b Classification of tumor tissue by C-means clustering of the smoothed ADC map in (a). The class corresponding to the highest ADC values was defined as the necrotic area (visualized by yellow border); c color map of the Euclidean distance of every voxel from the nearest blood vessel, averaged over the 1 mm tumor slice. A volume rendering of the blood vessels is shown in gray/white with the necrotic ROI (outlined in black). The necrotic ROI co-localizes with a poorly vascularized tumor region with large distances to the nearest blood vessels; d, e enlarged regions from the necrotic ROI and rim. f In vivo ADC map and g–i nearest vessel distance map for an early stage PIW3 tumor. (Color figure online)