A method for supervoxel-wise association studies of age and other non-imaging variables from coronary computed tomography angiograms

TL;DR

This work addresses how chronological age relates to cardiac morphology and tissue density in CCTA by introducing a supervoxel-wise Imiomics-inspired framework. It combines TotalSegmentator-based segmentation, a two-stage inter-subject deformable registration with joint semantic and intensity guidance, and robust supervoxel aggregation to map age-related changes onto the heart, revealing sex-specific patterns such as age-related LA expansion in females and LVV decline in both sexes. The method is validated through registration metrics (Dice, JD, ICE), explicit-voxel correlations, and proof-of-concept analyses, and applied to a large SCAPIS subset ($n=1388$), highlighting localized associations beyond traditional ROIs. The approach enables exploratory discovery of aging-related cardiac features and paves the way for linking imaging phenotypes with non-imaging biomarkers, potentially informing aging research and cardiovascular risk assessment.

Abstract

The study of associations between an individual's age and imaging and non-imaging data is an active research area that attempts to aid understanding of the effects and patterns of aging. In this work we have conducted a supervoxel-wise association study between both volumetric and tissue density features in coronary computed tomography angiograms and the chronological age of a subject, to understand the localized changes in morphology and tissue density with age. To enable a supervoxel-wise study of volume and tissue density, we developed a novel method based on image segmentation, inter-subject image registration, and robust supervoxel-based correlation analysis, to achieve a statistical association study between the images and age. We evaluate the registration methodology in terms of the Dice coefficient for the heart chambers and myocardium, and the inverse consistency of the transformations, showing that the method works well in most cases with high overlap and inverse consistency. In a sex-stratified study conducted on a subset of $n=1388$ images from the SCAPIS study, the supervoxel-wise analysis was able to find localized associations with age outside of the commonly segmented and analyzed sub-regions, and several substantial differences between the sexes in the association of age and volume.

Figures (17)

• Figure 1: Illustration of the proposed methodology. Step 1: The images are segmented using TotalSegmentator. Step 2: The images are registered to a common reference space using a novel registration method, combining intensity information and segmentation masks. Step 3: Supervoxel-derived aggregate features of local volume and density are extracted for the entire cohort. Step 4: Pearson correlation is performed independently between each supervoxel/feature and a non-imaging variable such as age. Step 5: The statistically significant correlation coefficients are visualized in color overlayed on the original CCTA image volume, providing a detailed map of the association between the variable and sub-regions of the heart.
• Figure 2: Illustration of the image channels and masks used in the deformable image registration process for two selected axial slices. (a,f) The pre-processed CT images for two sample axial slices. As can be seen in the black areas, non-cardiac structures such as lungs, stomach, liver, and esophagus have been cut out, and replaced by the minimal density value (after limiting the density range to between -300 and +200 HU). (b,c,g,h) The six anatomical masks used for the image registration (LV, RV, LA, RA, MYO, Aorta), merged into two channels, to reduce the runtime, the amount of memory required, and the number of competing objectives. (d,e,i,j) The low and high-density regions, excluding the non-cardiac structures, as for (a,f), are used to further focus the registration on the relevant cardiac tissue, and aid in the registration of the vessels and other high-density sub-structures.
• Figure 3: Analysis of the template selection for the female cohort. The sub-figures (a-f) display the histograms of the features used to select the template with age measured in years and volumes measured in mL. The value of each feature in the selected template is shown as a black bar overlaying the histogram.
• Figure 4: Analysis of the template selection for the male cohort. The sub-figures (a-f) display the histograms of the features used to select the template with age measured in years and volumes measured in mL. The value of each feature in the selected template is shown as a black bar overlaying the histogram.
• Figure 5: Illustration of the SLIC supervoxels (volumetric connected segments), as used in this work for a more robust and efficient Imiomics analysis, here shown on top of selected slices of both the female and male reference images. The large homogeneous regions of the ventricles, the vessels, and the aorta have more regularly shaped supervoxels given that there is less variation in density to influence the shape, while the other parts have much more high-frequency variations in the boundaries.