Fig 1.
Representative images showing manual qualitative scores.
A score of 0 contains no signs of tumor mass. The contrast in the lung space is derived from blood vessels and heart tissue. A score of 1 shows one or two small nodules within the lung field. A score of 2 shows more nodules that are beginning to coalesce, and a score of 3 shows a large coalescing mass that has infiltrated more than half the lung space.
Fig 2.
(A) Depiction of MLAST’s segmentation algorithm. The original microCT 2D image contains the densities of various tissues observed in a thoracic scan. The ribcage provides a high contrast for thresholding. The interior points of the rib regions are used to create a mask of the thoracic cavity. The densities of the voxels within the masked image are segmented with a k-means clustering algorithm and the intensities are classified into 3 tissue types: soft tissue (green), lung (blue), and intermediate (cyan). The final segmented result is shown in 2D and applied to all slices in the z-stack. (B) Depiction of how MLAST detects the cranial and caudal boundaries of the thoracic cavity. The cranial boundary is determined by the bifurcation of the trachea. Below the bifurcation, the two tracheal regions (colored blue and shown in inset) are separate. Tissues above the bifurcation, where the two regions are no longer separable, are excluded from the final segmentation. On the caudal end, the diaphragm is segmented based on density changes in the z-trace of each voxel. The resulting diaphragm (red) is removed from the thoracic counts of soft tissue, lung, and intermediate.
Fig 3.
Representative slices of lung from a tumor-bearing mouse showing the resulting segmentation from the MLAST algorithm at various levels (3, 6, 9 and 12 mm) above the base of the diaphragm.
Top panel is MLAST only, middle panel is microCT only and the bottom panel is the overlay of MLAST and microCT images. The tissue labels for MLAST segmentation are color-coded for visual purposes: soft tissue (green), intermediate (cyan), lung (blue), diaphragm (red), and the mutlislice bone mask (yellow).
Fig 4.
Comparison of MLAST with manual scoring system and manual segmentation.
(A) Results comparing MLAST values (% thoracic cavity) to manual scores on a 0–3 scale. MLAST values represent soft tissue (green bar), lung (blue bar) and intermediate density (cyan bar). The number of samples that corresponded to manual scores 0–3 is represented in Y-axis. Error bars are standard error of mean and * indicates statistical significance by Student’s t-test at a Bonferroni-corrected significance level of α = 0.0056. (B) Comparison of the three methods: manual scored samples, manually segmented volume and MLAST values. Linear regression analysis of automatically segmented thoracic tissue (intermediate + soft tissue) by MLAST vs manually segmented thoracic soft tissue (heart + tumor) volume on a log scale show good corellation.
Table 1.
Validation of MLAST with manual scoring (related to Fig 4A).
Fig 5.
Comparison of MLAST using contrast-enhanced microCT imaging and histology.
(A) Validation results showing linear regression of MLAST and manually segmented volumes vs contrast-enhanced manual segmentation on a log scale. The volumes of MLAST-segmented non-contrast scans, which are made up of soft tissue and intermediate, are shown in orange. The volumes of manually segmented non-contrast scans, which are made up of heart and tumor, are shown in black. The x-axis represents the volumes of manually segmented contrast-enhanced scans, which are also made up of heart and tumor. Regression lines (solid) are shown for both results, along with a 95% Confidence Interval (dotted). (B) Representative scans with low and high tumor burden by manual and MLAST segmentation. (C) MLAST validation with histology: Representative H&E stained images with low and high tumor burden at low and high magnification. The graph shows the correlation of tumor burden evaluation by MLAST to H&E methods.
Fig 6.
Utility of MLAST in an efficacy trial using Kras/Lkb1 GEM model.
(A) Manual scores assigned to scans from vehicle control (blue) and drug treatment (pink) groups before (baseline) and 3 weeks post-treatment regimen. (B) MLAST scores from vehicle (blue) and drug treatment (pink) groups before and post-treatment regimen. (C) Representative images from vehicle control and drug treated groups before and post-treatment. Images show profound progression of tumor in vehicle control group, but not in drug treated group. (n = 10/group; +/- SEM; ** = p-value < 0.0001).