Figure 1.
A BLI photographic image of the mouse acquired to validate the proposed approach.
The arrows indicate the different tumor locations: humerus (red), femur (green), kidney (blue).
Figure 2.
Overview of the integrated, interactive local SRR of MRI mouse data applied to two case studies.
Case Study A: After rigidly registering CT to MRI, articulated atlas-based segmentation is performed (A1). Subsequently, articulated planar reformation is applied to the segmented MRI, and the data is visualized in the standardized atlas space (A2). The user can now interactively select any bone of interest guided by the BLI images for SRR reconstruction. A high-resolution SRR image of the humerus with a tumor is presented. Case Study B: BLI+MRI mouse data are first co-registered (B1) to define the VOIs (B2) using the BLI. A VOI is interactively selected for performing SRR. A high-resolution SRR image of the kidney with metastases is presented.
Figure 3.
(a, c, e, g) One transversal slice from the 3D GE image (gold standard), single low-resolution (1 LR), SRR (2), and SRR (4) reconstructions respectively. (b, d, f, h) The same slice with the boundary of the region used for quantification (white) and detected objects (green, red, and blue; all regions occupied by segmented objects have been dilated with one pixel for better visualization). Green color indicates true positive detections based on the high-quality GE scan, red—false positives, and blue—false negatives (note that several visually missing objects were detected on other slices, hence they are not highlighted as false negatives on the shown slice). (i) FLI image of the phantom. (j) One coronal slice from the 1 LR image. (k) Detection performance in terms of the F-score of different reconstruction methods as functions of the threshold on the bottom-hat image. (l) F-score as function of object size for optimal threshold on each of the reconstructed volumes.
Figure 4.
From left to right: a CT scan, a single low-resolution image (1 LR), and SRR reconstructions, each based on a different number of low-resolution images. Two orthogonal slices of the same VOI are shown to illustrate the effect of the SRR in a 3D volume. The orange dashed line approximately indicates where the yz-slice (3rd row) intersects the xy-slice (2nd row). The red arrows points to the (micro) tumor in the knee. The green arrow points to a location outside the tumor, at which recovery of the fine details is obvious: the margins of the tumor are clearly delineated. The CT and all the MR images are shown in the coordinate system associated with the principal axes of the bone, and the low-resolution volume is resampled to isotropic resolution beforehand. Image contrast on the MRI images was increased for visualization purposes. For all MR images the corresponding frequency spectra up to the Nyquist frequency of the reconstruction volume are shown, above and beneath the xy-slices and the yz-slices, respectively, demonstrating enhanced high-frequency content.
Figure 5.
From left to right: a CT scan, a single low-resolution image (1 LR), and SRR reconstructions, each based on a different number of low-resolution images. Two orthogonal slices of the same VOI are shown to illustrate the effect of the SRR in a 3D volume. The orange dashed line approximately indicates where the yz-slice (3rd row) intersects the xy-slice (2nd row). The red arrows point to the tumor. The green arrows point to some of the locations where recovery of the fine details is the most noticeable. The CT and all the MR images are shown in the coordinate system associated with the principal axes of the bone, and the low-resolution volume is resampled to isotropic resolution beforehand. Image contrast on the MRI images was increased for visualization purposes. For all MR images the corresponding frequency spectra up to the Nyquist frequency of the reconstruction volume are shown, above and beneath the xy-slices and the yz-slices, respectively, demonstrating enhanced high-frequency content.
Table 1.
SRR times in seconds for each reconstructed right bone and the whole-body of the mouse, using 2 and 4 low-resolution images.
Figure 6.
From left to right: a single low-resolution image (1 LR), and SRR reconstructions, each based on a different number of low-resolution images. The red and yellow arrows point to two different tumors. Two orthogonal slices of the same VOI are shown to illustrate the effect of the SRR in a 3D volume. The green arrows point to other locations where the improvement in image quality is particularly noticeable. The orange dashed line approximately indicates where the yz-slice (3rd row) intersects the xy-slice (2nd row). In all the MR images, the xy-view is the in-plane direction of the scans. Note that the metastatic lesion seen in the BLI image (Figure 1, blue arrow) actually consists of numerous lesions as shown on MRI scans. For the low-resolution image, the selected views are resampled to isotropic resolution and the image contrast on the MRI images was increased for visualization purposes. For all MR images the corresponding frequency spectra up to the Nyquist frequency of the reconstruction volume are shown, above and beneath the xy-slices and the yz-slices, respectively, demonstrating enhanced high-frequency content.