A Spatial Atlas for Mapping Vascular Microcalcification using 18F-NaF PET/CT: application in Hyperphosphatemic Familial Tumoral Calcinosis

Abstract

Background:

Vascular calcification causes significant morbidity and occurs frequently in diseases of calcium/phosphate imbalance. Radiolabeled sodium fluoride (18F-NaF) positron emission tomography/computed tomography (PET/CT) has emerged as a sensitive and specific method for detecting and quantifying active microcalcifications. We developed a novel technique to quantify and map total vasculature microcalcification to a common space, allowing simultaneous assessment of global disease burden and precise tracking of site-specific microcalcifications across time and individuals.

Methods:

To develop this technique, four patients with hyperphosphatemic familial tumoral calcinosis (HFTC), a monogenic disorder of FGF23 deficiency with a high prevalence of vascular calcification, underwent 18F-NaF PET/CT imaging. One patient received serial imaging one year after treatment with an IL-1 antagonist. A 18F-NaF-based microcalcification score (mCS), as well as calcification volume, was computed at all perpendicular slices, which were then mapped onto a standardized vascular atlas (SVA). Segment-wise mCS_mean and mCS_max were computed to compare mCS levels at pre-defined vascular segments within subjects.

Results:

Patient with HFTC had notable peaks in mCS near the aortic bifurcation and distal femoral arteries, compared to a control subject who had uniform distribution of vascular 18F-NaF uptake. This technique also identified microcalcification in a 17-year-old patient, who had no CT-defined calcification. This technique could not only detect a decrease in mCS throughout the patient treated with an IL-1 antagonist, but it also identified anatomic areas that had increased-responsiveness while there was no change in CT-defined macro-calcification after treatment.

Conclusions:

This technique affords the ability to visualize spatial patterns of the active microcalcification process in the peripheral vasculature. Further, this technique affords the ability to track microcalcifications at precise locations, not only across time but also across subjects. This technique is readily adaptable to other diseases of vascular calcification and may represent a significant advance in the field of vascular biology.

Graphical Abstract

Background

Vascular calcification is pathological mineral deposition in the blood vessel walls, often occurring in the intimal and medial vessel layers, and predisposing individuals to vascular insufficiency.¹ Vascular calcification is associated with many genetic and acquired conditions, including kidney disease and diabetes.^2-4 In the case of either endothelial damage or primary mineral imbalance, such as chronic kidney disease (CKD), the end result is the deposition of hydroxyapatite (HA) in the vessel wall. Vascular macro- and microcalcification are calcific nodules >50 μm and <50 μm in size, respectively.^5,6 Due to the clinical significance of the onset of microcalcification, advanced imaging techniques are now being applied to detect, quantify, and monitor vascular microcalcifications with the goal of improving diagnosis and treatment of vascular disease.^7,8 Radiolabeled sodium fluoride (18F-NaF) positron emission tomography (PET) has emerged as a sensitive and specific method for detecting active vascular calcification, specifically, microcalcification.^5,8 18F-NaF, used most often to image bone metastases, also has powerful implications for the study and treatment of metabolic bone imaging.^9-11 The pathogenesis of vascular calcification involves mechanisms similar to that of bone formation and remodeling, thus, 18F-NaF is well suited for sensitively detecting HA deposits in arterial walls.^4,12

When injected, 18F-NaF is distributed throughout the vasculature and target tissues where the dissociated fluoride ion can exchange with the hydroxyl group of hydroxyapatite to form fluorapatite. The newly formed mineral is then captured and quantified using PET.⁹ 18F-NaF PET, in combination with computed tomography (CT) for three-dimensional (3D) localization. This imaging modality is extremely sensitive and can detect deposits of HA before they have coalesced into CT-detectable macro-calcifications.^5,13 As such, 18F-NaF PET/CT can differentiate between active and inactive calcifications, which is not only useful for assessing for disease progression, but can also improve the sensitivity and specificity of assessing treatment responses as inactive disease, which will not respond to most treatments, can be exclude from analsis.^14-16 Grouping plaques based on 18F-NaF avidity has been shown to identify high-risk atherosclerotic plaques and outperform coronary calcium scoring in predicting fatal or nonfatal myocardial infarction.^14-16

Recently, there has been an increased interest in moving beyond the individual plaque and studying total disease burden.^17-20 In addition, different vascular zones, or segments, such as the distal arteries of individuals with peripheral artery disease, may be more severely affected.^21-23 And most disease states, including hyperphosphatemic familial tumoral calcinosis (HFTC), can show heterogeneity in terms of vascular segment involvement^24,25, thus emphasizing the need for a technique to capture and quantify heterogeneity across the vascular tree.

HFTC is a rare, autosomal recessive disorder of deficient fibroblast growth factor-23 (FGF23) signaling, resulting in an elevated calcium-phosphate product.^24,26 While features of this disease have variable penetrance, the vascular calcification in both large and small vessels has been shown to lead to vascular insufficiency including amputations, emphasizing the need for improved techniques to further study and characterize this disorder.²⁷ Currently, there are no available tools for studying the spatial distribution of microcalcifications in a cohort. Thus, we aimed to use 18F-NaF PET/CT to develop a graphical and quantitative tool to rapidly study unique spatial patterns of microcalcification, as well as quantify microcalcification at various vascular segments.

We present here a novel method to map microcalcifications throughout the vasculature to a spatial referencing system or “common space”, which allows for comparison of areas of microcalcification between patients with vessels of different lengths. We utilize the 3D coordinates of each individual to map a microcalcification score (mCS) onto a standardized vascular atlas (SVA). We show how this method allows for both pattern exploration and a spatially-informed quantification of vascular microcalcification in four patients with HFTC, and in one patient of whom received treatment with an interleukin-1 (IL-1) antagonist (anakinra), the potential for quantification of response to treatment.

Methods

To minimize the risk of unintentionally disclosing information that can be used to reidentify private information in this human study, detailed clinical information data are available from the authors upon reasonable request. MATLAB code, as well as compiled code, developed and used throughout this study is available upon request to the authors.

Patients:

Four patients (a 17-, 20-, 53, and 73-year-old, all women) with genetically confirmed HFTC were evaluated at the National Institutes of Health as part of a natural history study of HFTC between 2019-2022. All patients are referred throughout by their identifiers: FTC-1, FTC-2, FTC-3, and FTC-4 (17-, 20-, 53-, and 73-year-old, respectively). All data were handled in a manner compliant with the Health Insurance Portability and Accountability Act and approved by the Institutional Review Board at the NIH, Bethesda, MD. All gave written informed consent to participate in a NIH IRB-approved clinical study. FTC-3 underwent two 18F-NaF PET/CT scans: one at baseline and one following 15 months of treatment with an IL-1 antagonist (anakinra) for systemic inflammation, a common feature of HFTC.

One healthy control (46-year-old woman) from the CAMONA study (clinical trial# NCT01724749) was included as a reference.^19,28 The healthy control was a never-smoker, had BP< 120/80 mmHg, as well as had a hemoglobin A1c, low-density lipoprotein (LDL) levels, and triglyceride levels within normal range. This subject had no CT-defined macro-calcifications.

18F-NaF PET/CT Image Acquisition:

Images of patients with HFTC were acquired at the NIH Clinical Center, Department of Radiology and Imaging Sciences. 18F-NaF PET/CT scans were obtained using a dedicated PET/CT Siemens Biograph 128 mCT scanner (Siemens Medical Solutions USA, Inc. Malvern, PA, USA) with FlowMotion positioning and 512 x 512 matrix size. Subjects were injected with an average ± standard deviation (SD) of 5.05 ± 0.62 millicurie (mCi) of 18F-NaF (range, 4.5 to 5.9 mCi). PET images from the vertex of the skull to the plantar surface of the feet were acquired, 61.1 ± 2.2 (mean ± SD) min (range, 58.1 to 62.8 minutes) post-tracer injection. Standardized uptake values (SUVs) were measured on time-of-flight (TOF) PET images. A non-contrast, low-dose CT scan was performed for attenuation correction and co-registration. CT images were registered to the corresponding PET images using rigid point feature registration with mutual information using MIM Software (version 7.2.3, MIM Software Inc., Cleveland, OH, USA).

Manual Vascular Segmentation and Target-to-Background Calculation:

The 18F-NaF PET information (Standardized Uptake Values – SUVs) within the vasculature were segmented from the rest of the body using MIM software. Regions of interest (ROIs) were drawn to capture the vasculature using axial slices of the CT image. Given the proximity of the abdominal aorta to the vertebral column, care was taken to exclude extra-aortic 18F-NaF uptake from the vertebral column using the co-registered 18F-NaF PET image. Briefly, the 18F-NaF signal window was lowered to 0-3 SUV to more easily visualize where focal 18F-NaF uptake from the vertebrae may confound aortic uptake. This vertebral spillover is defined as 18F-NaF uptake within the aorta that is contiguous with vertebral uptake, where there is a gradient of 18F-NaF signal extending from the vertebra (illustrated and detailed in Figure S1). Any 18F-NaF uptake within the original ROI that is contiguous with vertebral uptake and exceeds the uptake within the vascular lumen was manually excluded.

The labeled vasculature was divided into six segments (ascending aorta, descending aorta, abdominal aorta, common/external iliac arteries, femoral arteries, and popliteal arteries) using common anatomic landmarks as follows: the beginning of the ascending aorta was defined as the axial slice at the sternal angle; the beginning of the descending aorta was defined as the peak of the aortic arch; the beginning of the abdominal aorta, iliac arteries, femoral arteries, and popliteal arteries were defined by the axial slice at the level at which the vasculature intersects the aortic hiatus, aortic bifurcation, inguinal ligament, and adductor hiatus, respectively. Each axial slice position was recorded for later analyses.

Given that the clinically acquired SUVs calculated from 18F-NaF PET scans can be influenced by radiotracer concentration within the blood pool, all SUVs were converted to a target-to-background ratio (TBR). Voxel-wise TBRs were calculated by dividing the SUVs by the background SUV. The background SUV was defined as the median SUV calculated from an ROI inside the superior vena cava lumen. All voxels within the vasculature were then exported into MATLAB (version 2022b, MathWorks^®) for further processing.

Curvilinear Approach and Slice-wise mCS Calculation:

The ascending and descending aorta (referred to as the supradiaphragmatic aorta), has a curved and sometimes tortuous course. Unadjusted axial slices through such a curved structure will introduce over- and underassessments of the counts at a given level. To control for this potential confounding inaccuracy, using MATLAB, a curvilinear approach was used to extract voxel-level data from the supradiaphragmatic aorta, thus rendering more precise perpendicular slices. Further, by using a curvilinear approach, the data extracted at each perpendicular slice can be ordered spatially on a continuum beginning at the start of the aortic arch.

To do this in MATLAB, all voxel-level data inferior to the aortic hiatus were set aside, and a 3D volume of the supradiaphragmatic aorta was created from the 3D coordinates and HU values. The 3D volume was then converted to a binary volume (Figure 1, panel A). MATLAB’s “bwskel” function was used to create a skeletonization of the binarized volume, which determines a 1-voxel width centerline running through the volume while conserving the topology and Euler characteristic of the object within the same 3D voxel space as the original volume.²⁹ The x, y, and z coordinates of the centerline were defined as the original voxels coordinates where the centerline pass through (Figure 1, panel B and C). Next, the centerline points were ordered, starting from the beginning of the ascending aortic arch, and ending at the beginning of the abdominal aorta (Figure 1, panel D). The 3D points were then smoothed using a moving mean filter with a smoothing factor of 0.1 (Figure 1, panel E). To extract voxel-level data from perpendicular slices through the superior aorta, normal vectors (relative to the perpendicular plane) were defined for each central voxel. These normal vectors were defined by a 3D directional vector connecting two adjacent center voxels along the smoothed centerline (Figure 1, panel F). Voxel-level PET data were extracted from the perpendicular planes defined by the center voxel location and normal vector (Figure 1, panel G).

Figure 1: Method for slice-wise extraction of 18F-NaF data from supradiaphragmatic aorta.

(A) The supradiaphragmatic aorta was binarized and then (B) skeletonized to find the centerline throughout the volume. The skeletonization can be viewed within the binarized volume for quality control (C). Since axial slices (Z axis) are organized in descending order when they are exported from MIM Software, MATLAB was used to plot the X, Y, and Z coordinates of the skeletonized centerline (D) and order the coordinates starting from the beginning of the ascending arch and ending at the start of the abdominal aorta. Once ordered, the X, Y, and Z coordinates were smoothed using a moving mean filter (E). A normal vector was calculated for each center voxel by using the directional vector connecting the center voxel with the voxel downstream to it. Panel (F) displays a scatter plot with center voxels (every third voxel is displayed) along with its respective normal vector. Perpendicular slices (G) were defined by planes (shown by gray planes/boxes) centered at the center voxel and perpendicular to its respective normal vector.

Importantly, the perpendicular slice that marked the beginning of the descending aorta was defined as the slice where the tangent of the smoothed centerline was equal to 0, marking the peak of the aortic arch. For the vasculature below the aortic hiatus, voxel-level information was extracted from axial slices of the volume by grouping voxels based on their unique axial slice location.

Microcalcification Score (mCS) Definition:

A microcalcification score (mCS) was given to each perpendicular slice of the 18F-NaF PET/CT volume. The mCS is defined as the mean TBR of all voxels above the 90^th percentile TBR value. Here, the mCS served as an image biomarker to reflect the most metabolically active voxels per perpendicular slice along the vasculature.

Calcific Volume (CV) Definition:

The volume of calcification, or calcific volume (CV), was also calculated at each perpendicular slice throughout the vasculature. The CV was defined as the number of voxels above 200 HUs multiplied by the voxel volume.

Defining the SVA and Mapping mCS and CV:

The standardized vascular atlas (SVA) was defined as a fixed range of values, where each value within the range represents a relative distance from the beginning of the arch of the aorta. To map each slice-wise position to the SVA, each perpendicular slice location of the patient’s vasculature received a relative segment distance (RSD), as shown in Figure 2, panel A. The RSD is defined as the percentage distance of the slice location from the start of the relevant vascular segment of that patient. Therefore, each slice received a value from 0 to 1, with values closer to 1 meaning the slice is closer to the end of that patient’s vascular segment. For example, if the abdominal aorta began on axial slice 50 and the aortic bifurcation began on slice 100, then slice 51 would receive a value of (1/50) and slice 75 would receive a value of (25/50) because they are 2% and 50% between the abdominal aorta and the beginning of the iliac arteries, respectively.

Figure 2: Equations for mapping each slice position to the standardized vascular atlas (SVA).

(A) A representative maximum intensity projection (MIP) of the registered 18F-NaF PET/CT scan, where all 18F-NaF uptake outside of the vasculature is masked. The vasculature is divided into 6 segments (Seg. IDs). A yellow dotted line represents a hypothetical axial slice through the iliac arteries. The relative segment distance (RSD) is shown (blue text), which is the percentage distance from the beginning of the relative vascular segment (Seg.ID = 4, or iliac arteries, in this example). (B) Equation for mapping each slice to a position on the SVA (${\mathrm{P}}_{\text{sva}}$), which is a function of slice position. Here, $\overrightarrow{\text{RP}}$ is the relative position vector and $\overrightarrow{\text{SF}}$ is the scaling factor vector (Ba) Supporting equations for the equation in B. Of importance, the input of “slice” has 2 components: the Seg. ID in which the slice resides and the RSD. The $\overrightarrow{\text{RP}}$ is dependent on the Seg. ID and RSD ascribed to that slice (Ba). $\overrightarrow{\text{SF}}$ is a vector that adds spatial relevance to the SVA and consists of the average vascular segment lengths among the cohort. (Bb) Key for terms in equations B and Ba.

Next, knowing the vascular segment in which the slice resides (“Seg. ID” – Figure 2, panel A) and the RSD, the SVA position (P_sva) for each perpendicular slice location is calculated by the equation in panel B of Figure 2. Here, $\overrightarrow{\text{RP}}$ is the relative position vector and $\overrightarrow{\text{SF}}$ is the scaling factor vector. The $\overrightarrow{\text{RP}}$ is dependent on the Segment ID (Figure 2, panel Ba). $\overrightarrow{\text{SF}}$ is a vector that adds spatial relevance to the SVA, where each numerical component represents the population average for the length of each vascular segment, starting from the ascending aorta and ending with the popliteal arteries. The average segment lengths were determined by averaging the length of each segment of all three patients within the cohort to serve as a population average. The average lengths of the ascending aorta, descending aorta, abdominal aorta, iliac arteries, femoral arteries, and popliteal arteries were 50, 60, 183, 170, 300, and 87 mm, respectively. Thus, the sum of all average segment lengths is 850 and represents the range of the SVA.

As shown by the equation in panel B of Figure 2, the position on the SVA for each slice is equal to the sum of the product of the matrix multiplication of $\overrightarrow{\text{RP}}$ and $\overrightarrow{SF}$. Given the $\overrightarrow{\text{RP}}$ consists of values between 0 and 1, each slice is mapped to a position between 0 and 850.

The mCS can then be plotted versus the SVA, where atlas positions 0-50, 50-110, 110-293, 293-463, 463-763, 763-850 correspond to the range for the ascending aorta, descending aorta, abdominal aorta, iliac arteries, femoral arteries, and popliteal arteries, respectively. Because each mCS receives a spatial identifier based on its relative position within each vascular segment, microcalcification can be explored and quantified across a cohort of subjects while retaining spatial information.

Visualizing Microcalcification Patterns in HFTC:

The mCSs and CV throughout the vasculature of FTC-1, FTC-2, FTC-3, and FTC-4 were calculated and presented on the SVA to visualize the pattern of vascular micro- and macro-calcification, respectively. In this way, sites of vascular microcalcification and macro-calcification can be simultaneously compared in a spatially-informed manner. The mCS and CV plots were smoothed with a 0^th order filter using 15 neighbors for visualization. The smoothed plot of mCSs for the healthy control (46F) was included on the SVA for each patient to serve as a reference for the physiologic pattern 18F-NaF uptake. The control subject did not have CT-defined vascular calcification, so the CV for this subject was not included.

18F-NaF signal within the lumen and vascular wall should follow of normal (Gaussian) distribution when there are no focal 18F-NaF uptake. To test this, a Kolmogorov-Smirnov test for normality was performed for the set of mCSs for each patient to statistically test the distribution of 18F-NaF using GraphPad Prism (version 9.5.1). Note that this is a spatially-independent test to determine if there is abnormal vascular uptake and does not inform the anatomic location of the abnormal uptake.

Comparing mCS by Segment

Each vascular segment received a mean microcalcification score (mCS_mean) by averaging the mCS of all slices within each vascular segment. Thus, mCS_mean is a segment-wise measure summarizing the most active microcalcifications within specific vascular segments. The maximum mCS (mCS_max) per vascular segment was also calculated to capture the most active microcalcification per segment. Given that the mCS_mean and mCS_max values of each segment are paired by patient, a repeated measures (RM) Analysis of Variance (ANOVA), assuming sphericity, was performed to test whether there is a difference in mCS_mean or mCS_max among the different vascular segments. When appropriate (F-ratio suggests rejection of null hypothesis), the mean mCS_mean and mean mCS_max of each segment were compared to the ascending aorta using a Dunnett’s test to correct for multiple comparisons, with a p-value of 0.05 as the threshold for significance. The ascending aorta was used as the control, or baseline mCS_mean and mCS_max, for comparisons because it is the segment with the lowest mCS and is not routinely affected in the HFTC disease process by clinical experience. Both the RM ANOVA and Dunnett’s test were performed using GraphPad Prism (version 9.5.1).

Slice-wise Aggregation of mCS for Cohort:

Because each patient’s data are transferred to the SVA, the mCSs for all patients can be aggregated at each position of the vasculature. The mCSs for each patient at each vascular segment were uniformly resampled such that each patient had an equal number of mCSs per vascular segment. MATLAB was used to resample each vascular segment by applying an antialiasing lowpass filter that accounted for the delay introduced by the filter.³⁰ After resampling, the mean and standard deviation (SD) of the mCSs (n=4) was calculated for each position on the SVA. The mean (± SD) at each SVA position was plotted versus the SVA to represent how mCS data can be aggregated to enable cohort-level analyses. Sub-segment analyses (calculation of mCS_mean at precise SVA ranges) were carried out using precise vascular atlas locations that were identified as having peaks in mCS. A RM ANOVA and Dunnett’s test was done to compare the mean mCS_mean at these sub-segments to the mean mCS_mean at the ascending aorta.

Assessing Treatment Response:

FTC-3’s pre- and post-treatment 18F-NaF PET/CT scans were analyzed and transferred to the SVA. To explore the clinical utility of this method, both pre-treatment and post-treatment CV and mCS data were presented on the SVA and spatial responses to treatment were explored by comparing the percent change of mCS_mean and mCS_max from baseline at each vascular segment.

Results

Figure 3, panel A displays a representative original 18F-NaF PET maximum intensity projection (MIP), CT MIP, and PET/CT fusion MIP for a patient (FTC-2), where the vasculature from the aortic arch to the popliteal arteries had been segmented (blue outline in fused image). Figure 3, Panel B illustrates how all 18F-NaF PET uptake outside of the vasculature segmentation was masked to only visualize the uptake within the vasculature. The original vascular-only 18F-NaF PET/CT images for FTC-1, FTC-2, FTC-3, and FTC-4 are displayed in Figure 3, panel C, organized from youngest (17-year-old) to oldest (73-year-old) from left to right. The vascular 18F-NaF PET signal viewing window is set to 0-3.5 for all images.

Figure 3: Visualization of Vascular-only 18F-NaF Uptake and Original Images for HFTC patients.

Panel (A) displays a representative 18F-NaF PET maximum intensity projection (MIP), computed tomography (CT) MIP, and PET/CT fusion MIP for a patient (FTC-2), where the vasculature from the aortic arch to the popliteal arteries had been segmented (blue outline in fusion MIP). (B) All 18F-NaF PET uptake outside of the vasculature segmentation was masked to only visualize the uptake within the vasculature. The 18F-NaF PET signal window was adjusted to 0-3.5 SUV (red-orange color bar) to provide best contrast. (C) The vascular-only 18F-NaF PET/CT fusion MIPs are provided for each patient, including a 46-year-old control from the CAMONA study as a reference. HFTC patient images are arranged from youngest to oldest (left to right), and each patient’s vascular 18F-NaF uptake is windowed from 0-3.5 SUV.

The vascular-only 18F-NaF PET/CT image for the 46-year-old control is provided by the left-most image in Figure 3, panel C. This control subject is included as a visual reference for physiologic 18F-NaF uptake throughout the vasculature. The 18F-NaF uptake in the control’s vasculature shows a near uniform distribution, where there are no obvious focal 18F-NaF avidities along the vascular path. This contrasts with the vascular 18F-NaF uptake seen in the HFTC patients, where each have visibly increased 18F-NaF avidity at various regions throughout the vasculature, notably near the aortic bifurcation and distal femoral arteries in each patient (Figure 3, panel C)

Patient-level Exploration of Microcalcification in HFTC:

The mCS throughout the vasculature was quantified for each patient and plotted on the SVA (Figure 4). The CV throughout the vasculature is also provided for each patient to allow comparison of sites with macro-calcification (CV) with sites of microcalcification (mCS). The mCS plot for the healthy control is provided for each patient as a reference of normal 18F-NaF uptake throughout the vasculature (as shown by the black line in each panel of Figure 4). The control subject did not have CT-defined calcification, thus was not included in each plot.

Figure 4: Panels (A), (B), (C), and (D) represent the mCSs and calcific volume (CV, “Ca²⁺ vol.”) plotted on the SVA for FTC-1, FTC-2, FTC-3, and FTC-4, respectively.

For each plot, the mCSs of a healthy control (46F) from the CAMONA cohort is provided for reference (grey line); the shaded area represents the relative normal range of mCS based on the control subject. Each patient had a spike in mCS notably around two anatomic locations: the aortic bifurcation and the distal femoral arteries. However, the degree of mCS at each of these locations differed among the patients. (A) FTC-1 had a spike in mCS at the aortic bifurcation but had predominantly distal femoral artery involvement. (B) FTC-2’s vasculature also had increased mCS near the aortic bifurcation, with overall greater mCS in the femoral arteries. (C) FTC-3’s vasculature, on the other hand, had predominantly aorta involvement, with the peak mCS near the aortic bifurcation. (D). FTC-4’s vasculature showed a mCS pattern similar to that of FTC-1 and FTC-2, with peaks of macrocalcification near both the aortic bifurcation and distal femoral arteries. Note: each subject’s original vascular uptake from the clinically acquired 18F-NaF PET/CT scan is presented under the plot as a visual reference. The shaded lines (light red) represent the actual mCS value at each SVA location (not error), while the bolded lines represent the smoothed mCS scores using a box-sliding filter (0th order box-sliding filter using 15 neighbors). Fluctuations in the mCS plot represent the degree of microcalcification heterogeneity observed throughout the vascular tree.

The quantified mCS plot confirms that the control subject had a relatively uniform distribution of 18F-NaF uptake, with no focal peaks in mCS (Figure 4, all panels). This is statistically demonstrated by the finding that the mCSs for control subject followed a normal, Gaussian distribution, where the distribution of mCSs passed the Kolmogorov-Smirnov (KS) test for normality (p=0.1). However, all patients with HFTC had areas of increased mCS (Figure 4), relative to the control, either predominantly proximal involvement (FTC-3, panel C) or predominantly distal involvement (FTC-1, FTC-2, and FTC-4 in panels A, B, and D, respectively). Statistically, the distribution of mCSs for each HFTC patient failed the KS test for normality (p=1x10⁻¹⁴, p=2.4x10⁻², 4.7x10⁻⁹, and p=1x10⁻¹⁴ for FTC-1, FTC-2, FTC-3, and FTC-4, respectively). Importantly, this statistic is a spatially-independent measure of mCS distribution; thus, the mCS plot and SVA (Figure 4) must then be used to identify which vascular region is contributing to the abnormal 18F-NaF vascular distribution.

The younger patients (FTC-1 and FTC-2) displayed marked mCS near the aortic bifurcation and femoral arteries with low CV throughout their vasculature (Figure 4, panel A and B). Notably, the 17-year-old (FTC-1) showed no macro-calcification (CV) at the femoral arteries, while the level of microcalcification (mCS) exceeded the level of the older HFTC patients (Figure 4, panel A). It is likely that this is demonstrating the ability of 18F-NaF PET to capture the vascular disease process earlier than structural imaging and likely predicts calcific disease in the future if no intervention is successful in halting the disease process. The mCS and CV plot of the oldest patients (FTC-3 and FTC-4) support the previous point given that the CV plot nicely mirrors, or correlates, with the mCS plot. This demonstrates the presence of calcific plaques (evident by CV plot) that are actively mineralizing. Further, while FTC-3 has increased micro- and macro-calcification predominantly in the distal aorta and near the aortic bifurcation, there is a spike in mCS in the distal femoral arteries without the presence of calcification (Figure 4, panel C). It is possible that this increase is predicting the future calcification of this vascular segment, which is the same site of calcification as seen in FTC-4.

Comparing mCS_mean by Vascular Segment

The mCS_mean and mCS_max was calculated for each vascular segment of each patient (Figure 5) as a summary measure to quantitatively compare intra-subject patterns of microcalcification. Both graphs of mCS_mean and mCS_max (Figure 5) grouped by vascular segment displays the various patterns of vascular calcification seen in patients with HFTC. This also allows for a simplified comparison of overall mCS at the various vascular segments.

Figure 5: The mCS_mean (A) and mCS_max (B) per vascular segment, grouped by patient.

(A) On average, the mCS_mean was the greatest at the femoral arteries (2.43 ± .39 (n=4)), followed by the abdominal aorta (2.37 ± .65 (n=4)). The mCS_mean was the lowest at the ascending aorta (1.85 ± .49 (n=4)). However, statistically, there was no difference in mCS_mean among the six vascular segments (F(5,15)=0.67, p=0.65). (B) The graph of mCS_max showed a similar trend to the graph of mCS_mean, with the greatest mCS_max occurring at the femoral arteries (4.68 ± .88 (n=4)), followed by the abdominal aorta (4.55 ± 1.60 (n=4)). The mean mCS_max among the six vascular segments was found to be significantly different (RM ANOVA, F(5,15)=3.2, p=0.034). A multiple comparison test (Dunnett’s test) was performed and found the mean mCS_max at the femoral arteries was significantly greater than the mean mCS_max at the ascending aorta (p=0.039). The mean mCS_max at the descending aorta, abdominal aorta, iliac arteries, and popliteal arteries were not significantly greater compared to the ascending aorta.

The mean mCS_mean (±SD) at the ascending aorta, descending aorta, abdominal aorta, iliac arteries, femoral arteries, and popliteal arteries was 1.85 ± .49 (n=4), 1.88 ± .37 (n=4), 2.37 ± .65 (n=4), 2.13 ± .63 (n=4), 2.43 ± .39 (n=4), and 2.28 ± .76 (n=4), respectively (Figure 5, panel A). A RM ANOVA was performed and found that the mean mCS_mean was not significantly different among the vascular segments (F(5,15)=0.67, p=0.652).

The mean mCS_max (±SD) at the ascending aorta, descending aorta, abdominal aorta, iliac arteries, femoral arteries, and popliteal arteries was 2.22 ± .69 (n=4), 2.38 ± .45 (n=4), 4.55 ± 1.60 (n=4), 3.99 ± 1.10 (n=4), 4.68 ± .88 (n=4), and 3.92 ± 1.73 (n=4), respectively (Figure 5, panel B). The mean mCS_max of the six vascular segments were statistically different (RM ANOVA, F(5,15)=3.2, p=0.034). A Dunnett’s test for multiple comparisons was used to investigate which segments contribute to the difference in mean mCS_max. The mean mCS_max at the femoral arteries was significantly greater (2.11-fold) than the mean mCS_max at the ascending aorta (p=0.039) (Figure 5, panel B). While the mean mCS_max at the abdominal aorta was 2.05-fold greater than the mean mCS_max at the ascending aorta, this difference did not reach statistical significance (p=0.053). Additionally, there was no statistical difference between the mean mCS_max at the descending aorta, iliac arteries, or popliteal arteries and the ascending aorta (p=0.99, p=0.18, and p=0.20, respectively).

Slice-wise Aggregation of mCS for more Accurate Spatial Analysis:

While ascribing a summary measure such as mCS_mean to each vascular segment has utility (Figure 5, panel A), it likely misses or dilutes cases of severe mCS seen at specific vascular zones. Or, in other words, it can underestimate the true microcalcification burden within specific segments. Conversely, the summary measure mCS_max is more sensitive as is reports the highest mCS within a segment. This summary measure has the potential to overestimate microcalcification burden within the specified vascular segment, as it is a summary measure over a relatively long anatomic distance (e.g., entire femoral artery). A likely more robust method for exploring mCS patterns is to aggregate mCS in a slice-wise manner for all patients within a cohort and to then investigate mCS within more specific vascular zones.

To explore this, first the mean mCS at each position on the vascular atlas were calculated for all patients (n=4) and plotted on the SVA (Figure 6, panel A). When mCS data are aggregated at each SVA position for all HFTC patients, there is a clear pattern of increased mCS at a region near the aortic bifurcation (SVA position 250 – 350) and a distal region of the femoral artery (SVA position 675 – 750). By viewing each patient’s mCS in a common space, a more precise spatial location can be followed and analyzed.

Figure 6: Exploring mCS among HFTC Cohort in a Common Space.

(A) The mCSs were aggregated at each SVA position for all HFTC patients. The mean mCSs (bolded red line) for FTC-1, FTC-2, FTC-3, and FTC-4 plotted versus the SVA. The light red lines represent the mean plus and minus one standard deviation (SD). The mCS is greatest particularly at two regions: near the aortic bifurcation and distal femoral arteries. The SVA positions 250-325 was used for the aortic bifurcation sub-segment, and SVA positions 675-750 was used for the distal femoral artery sub-segment. (B) The mCS_mean for the ascending aorta, aortic bifurcation, and distal femoral arteries. The mCS_mean at the aortic bifurcation sub-segment was significantly greater (p=0.0003) than the ascending aorta (2.94 ± 0.55 (n=4) vs. 1.85 ± .49 (n=4)). While the mCS_mean at the distal femoral artery sub-segment was greater compared to the ascending aorta (3.05 ± 0.56 (n=4) vs. 1.85 ± .49 (n=4)), this difference did not reach statistical significance. Note: The mean and mean ± 1SD mCS values were smoothed using a 0th order box-sliding filter using 15 neighbors.

The mCS_mean was calculated at these two sub-segments (SVA positions 250-325 and 675-750), which correspond to the same relative region near the aortic bifurcation and distal femoral arteries, respectively, in all patients. At the aortic bifurcation sub-segment, the mCS_mean was 2.94 ± 0.55 (n=4), which is significantly greater than the mCS_mean at the ascending aorta (p=0.0003) (Figure 6, panel B). The mCS_mean at the distal femoral arteries (SVA positions 675-750) was 3.05 ± 0.56 (n=4). While this mCS_mean was greater than at the ascending aorta (1.64-fold), this difference did not meet statistical significance (p=0.16). The mCS_mean at this distal femoral artery sub-segment likely did not reach statistical significance due to one patient (FTC-3) having a lower mCS_mean at this location compared to the ascending aorta.

Spatially Visualizing Therapeutic Response:

FTC-3 underwent 18F-NaF PET/CT at baseline and 1 year after treatment with anakinra. Both pre- and post-treatment scans were analyzed and the CV and mCS were calculated and mapped onto the SVA (Figure 7). Panel A displays the plot of CV along the SVA for both pre- and post-treatment scans. The CT MIP is provided for both pre- and post-treatment scans below the plot. Further, panel B displays the plot for mCS for both pre-and post-treatment. The MIP of vascular 18F-NaF PET uptake at each timepoint is provided for reference in at the bottom of Figure 7, panel B.

Figure 7: 18F-NaF PET/CT scans of FTC-3 at baseline and 1 year post treatment with an IL-1 antagonist.

(A) the calcific volume (CV) along the vasculature is plotted on the SVA for the pre-treatment (red line) and post-treatment (blue line). For reference, the CT MIP of the vascular only is shown for the pre- and post- treatment (bottom of CV plot). (B) The plot of mCS for both pre-treatment (red) and post-treatment (blue) is shown. The 18F-NaF PET uptake within the vasculature only is shown for both pre- and post-treatment images at the bottom of panel B. The 18F-NaF PET images in are windowed equally (0 – 4.0 SUV). Note: The shaded lines represent the raw values (raw CV and mCS), while the bold lines represent the smoothed CV and mCS plots. Both CV and mCS plots were smoothed using a box-sliding filter (0^th order using 15 nearest neighbors) for visualization.

Notably, the plot of CV had negligible change after treatment (panel A). Thus, as expected in a relatively short timeframe (1 year), there is quantitatively no change in the volume of macro-calcification. Conversely, there is a noticeable decrease in mCS at various locations throughout the SVA (panel B).

The global mean mCS across all vascular segments was 2.66 at pre-treatment and 2.16 at post-treatment (↓18.8%). Further, the mCS_mean in each vascular segment decreased, some exhibiting a larger treatment response than others. The mCS_mean decreased from 2.95 to 1.84 (↓37.6%) at the ascending aorta, from 2.75 to 2.21 (↓19.6%) at the descending aorta, from 3.62 to 3.30 (↓8.8%) at the abdominal aorta, from 3.11 to 2.44 (↓21.5%) at the iliac arteries, and from 2.05 to 1.68 (↓18.0%) at the femoral arteries. The mCS_mean at the popliteal arteries increased slightly from 1.47 to 1.48 (↑0.7%), which is likely a clinically insignificant change.

While the mCS_mean within the entire abdominal aorta decreased by only 8.8%, a more focused analysis of the aortic bifurcation sub-segment (SVA positions 250-325) found that this region decreased from 3.76 to 3.17 (↓15.6%). Thus, by taking advantage of the SVA position, a more focused analysis can be done to investigate change in various vascular zones, thus potentially identifying areas most likely to progress, and/or most likely to respond to treatment. The mCS_max is also useful to report. The mCS_max at the abdominal aorta decreased from 7.58 to 5.82 (↓23.2%), and the mCS_max at the femoral arteries decreased from 4.57 to 3.52 (↓22.3%). Each of these changes from baseline are greater than the change seen in the mCS_mean at these segments, suggesting the mCS_max is likely more sensitive to treatment response.

Conclusions

Vascular calcification occurs in several disease states and is associated with serious adverse events, including myocardial infarction and vascular insufficiency. It is well characterized that increased flow velocity and wall shear stress at arterial bifurcations can contribute to endothelial damage and calcification^31-33, but questions remain regarding other vascular calcification patterns, especially those in rare diseases and in younger patients. As shown by our group and others, patients with HFTC can present with vascular calcification within large and small arteries, including the femoral arteries.^24,25,34 This work details a novel method for systematically studying patterns of vascular microcalcification. Mapping early stage, active microcalcifications (mCS) onto the SVA permits the study of microcalcification patterns within a cohort and provides a framework for quantifying microcalcification within specific vascular zones, allowing for population-level analysis. This method will not only be useful in identifying disease heterogeneity but will also aid in providing a more sensitive imaging biomarker to assess changes in vascular microcalcification compared to structural imaging.

By bringing the vasculature of each subject into a common space, specific vascular beds can be visually and quantitatively compared within a cohort. As observed in HFTC, patients may present with region-specific patterns of vascular involvement (e.g. FTC-1, FTC-2, and FTC-4’s distal involvement vs. FTC-3’s proximal involvement). Further, this technique allowed for the detection of significant vascular microcalcification in patients as young as 17 and 20 years old (FTC-1 and FTC-2, Figure 4, panels A & B). Notably, the 17-year-old patient exhibited marked microcalcification in the distal vasculature while showing no macro-calcification (CT-defined calcification). The identification of such significant vascular microcalcification warrants close monitoring and interventions, as indicated, with the goal to prevent the development of clinically significant macro-calcific disease.

Patients with HFTC were found to have visibly increased mCS notably near the aortic bifurcation and femoral arteries (Figure 4). The mCS_mean and mCS_max were both explored as summary measures of microcalcification at the six defined vascular segments. The mCS_max was significantly greater at the femoral arteries compared to the ascending aorta (Figure 5, panel B). There was no significant difference in mCS_max at the other vascular segments compared to the ascending aorta, nor where there any differences in mCS_mean along the segments. There are clear disadvantages to using a summary measure such as mCS_mean and mCS_max to ascribe to a vascular segment. For instance, mCS_mean likely grossly underestimates true burden of microcalcification as it can dilute a high mCS in more focal areas. Conversely, mCS_max can overestimate burden for an entire segment due to only reporting the mCS at a single SVA position within a large segment. Due to these realizations, a likely more robust approach is to use more precise SVA positions, or ranges, at specific vascular zones of interest. In the current study, we explored this by using the summary measure mCS_mean to summarize the microcalcification at the aortic bifurcation (SVA positions 250-325, Figure 6) and the distal femoral arteries (SVA positions 675-750, Figure 6). With this method, we found both vascular zones to have increased mCS_mean compared to the ascending aorta, with only the aortic bifurcation reaching statistical significance.

It is important to understand that the slice-wise aggregated mCS plot for the cohort is purely a tool for visualizing patterns among a cohort of patients. Due to the resampling method, the shape of each patient’s mCS plot is conserved, while the magnitude at each position (mCS) is not, as the new (resampled) mCS is technically an estimated value. However, mathematically, the function of mCS over space is conserved, allowing relative mCS_mean values to be compared across the various vascular segments within a cohort.

Another powerful application of this method is the assessment of treatment response. There was no change in CV along FTC-3’s vasculature after treatment with an IL-1 antagonist (Figure 7, panel A). However, there was a decrease in mCS (Figure 7, panel B). Through a spatially-informed analysis using the SVA, it was demonstrated that some vascular beds responded better than others. Thus, by identifying these differentially affected regions, we are potentially identifying regions that are more sensitive to treatment and/or more likely to progress, which undoubtedly helps in identifying target areas for future trials. In the assessment of decreased 18F-NaF uptake (mCS), one must consider two scenarios: (1) did the calcific plaque progress from a more “active” calcification to a more “stable” calcification (where the calcium volume would concurrently increase), or (2) did the decrease in activity (mCS) translate to less, more quiescent calcification. Given that FTC-3’s macro-calcific disease did not progress, while the activity diminished (scenario #2), we believe this is a demonstration of disease prevention, where the decrease in mCS can be correlated to the amount of macro-calcification prevented.

It is important to interpret the mCS values of the healthy control from the CAMONA cohort with caution since this patient was imaged at an outside center and likely affected by differences in scanning protocol and factors mentioned above. Therefore, we used this control purely as a visual reference for spatial distribution of 18F-NaF uptake and did not compare the control mCS values to our patients. To compare mCS values within our cohort to a reference range, there is a great need for future studies to systematically quantify mCSs within a large cohort of healthy, age-matched individuals. Further, for there to be an established normal range for mCS along various vascular segments, there is a need for a large subject size which will necessitate a concerted effort to standardize 18F-NaF PET imaging protocols so that subjects imaged across centers/institutes can be aggregated to meet sample size requirements.

The mCS was chosen as the biomarker for vascular microcalcification because it captures the mean of the most metabolically active regions per slice of the original 18F-NaF PET image. There are several important considerations and limitations of the current method. The mCS is affected by factors including image acquisition time, post-processing (i.e., attenuation correction methods), individual body compositions differences, and more.³⁵ This variability is minimized in single studies where noise and scanner fingerprint are the same across subjects. Therefore, while 18F-NaF PET has proven to be a clinically useful modality in assessing and categorizing individual vascular calcifications^14-16, it is important that future studies test precision and repeatability to translate this technique into clinical use. The use of a cohort with a rare, monogenic disease allowed for the development of this technique in a relatively homogenous cohort, as opposed to a disease of multifactorial pathophysiology, such as chronic kidney disease. As such, this method will need to be further validated in different cohorts. User harmonization and education will be essential to developing a repeatable and reliable method of SVA creation.

There are additional limitations with the method that must be considered, from image registration to segmentation. In the current method, a rigid registration was used to align the 18F-NaF PET to CT. It has been well described that bulk body motion is non-rigid; thus, in the case that a patient moves significantly between PET and CT imaging, this registration method can result in misalignments. Therefore, the current method could be improved by implementing advanced non-rigid registration techniques. Also, in the case of bulk body motion that necessitates additional alignment techniques, it is possible that the attenuation corrected PET values could be erroneous given they depend on the original PET/CT alignment. Further, these mismatch-induced PET attenuation correction errors could especially be erroneous if they occur near the dense vertebrate spine.

The current method of manual segmentation is a time-consuming task, especially for users not trained in medical image processing. It is recognized that both accuracy and efficiency must meet appropriate standards before it makes its way into clinical practice. As in most clinical PET scans, a non-contrast low-dose CT obtained, which are lower quality than standard diagnostic CT. This decreased quality can make segmentations a challenge for non-trained staff, especially at soft tissue structures. One potential improvement could be the implementation of iodinated intravenous (IV) contrast to CT, which could drastically improve the vascular segmentations and greatly improve segmentation speed and accuracy via the use of adaptive threshold-based methods. The use of CT angiography would likely also greatly improve the accuracy and repeatability of the curvilinear approach used to define the vascular centerline. It is also important to recognize that iodinated contrast is radio-dense, thus could alter the attenuation corrected PET values³⁶; however, this would likely not be an issue as long as scanning protocols are consistent and standardized across patients and institutions.

Even with the use of IV contrast or artificial intelligence (AI) based segmentation methods to improve speed and accuracy of vascular segmentations, great care must be taken to exclude extra-aortic uptake that can confound the 18F-NaF uptake within the vasculature, specifically at the vertebral spine. Further, given the necessary exclusion of some regions of the posterior aspect of the aortic due to spillover, the calcification volume and 18F-NaF signal may be underestimations at some SVA positions. This is a clear limitation, and there is active research to eliminate this issue via background correction, local projection and hybrid kernelized methods.³⁷ Furthermore, the advancement of total-body PET (EXPLORER scanner) may also eliminate this issue via its far superior resolution and the ability to reduce injection dose by 40-fold.³⁸ Nonetheless, the current method lays the foundation for quantifying and visualizing both calcific and micro-calcific vascular disease in a common space for a variety of applications.

This work clearly demonstrates that data acquired from standard 18F-NaF PET/CT scans and mapped onto a SVA can be used to identify patterns of vascular microcalcification, quantify microcalcification burden, quantify differences in microcalcification between patients and specific segments, and sensitively detect changes in vascular microcalcification following therapeutic intervention. As in several diseases, such as Takayasu arteritis³⁹ and giant cell arteritis⁴⁰, patterns often emerge that serve as hallmarks. By implementing this technique, patterns of microcalcification may rapidly emerge in a myriad of disease states, such as CKD. Further, by applying this technique when assessing treatment response, it may be found that some therapeutics also have spatially heterogeneous effects, which could result in a drug being deemed ineffective – as is likely the case in many trials which use global disease measures as primary outcomes. With personalized medicine at the forefront, this method could also be utilized to discover that certain microcalcification patterns warrant unique treatment approaches, or even suggest entirely different pathological processes. Expansion of the application of this technique has the potential to be a quantum advance in the study and treatment of vascular disease.

Highlights.

• 18F-NaF PET offers the ability to detect and quantify the active process of mineral deposition within the vasculature.

• By mapping 18F-NaF PET-based measures of microcalcification to a standardized vascular atlas (SVA), or spatial referencing system, the quantity and pattern of vascular disease can be explored across individuals and cohorts.

• When implemented in Hyperphosphatemic Familial Tumoral Calcinosis (HFTC), the SVA allowed the discovery that microcalcification at the distal femoral arteries and aortic bifurcation may be hallmark features of HFTC and was seen in patients as young as 17 years old.

• The mapping of a microcalcification score (mCS) and calcific volume (CV) to the SVA also discovered that treatment with an IL-1 antagonist (anakinra) resulted in decreased mCS in the absence of change in macro-calcification (CV), demonstrating the superior sensitivity of 18F-NaF PET.

• The SVA offers a spatially informed framework for mapping 18F-NaF PET-based measures of microcalcification which can be applied to many disorders to sensitively study pattern and quantity of vascular microcalcification across time and space.

Non-standard Abbreviations and Acronyms

3D

Three-dimensional

18F-NaF

Radiolabeled sodium fluoride

CKD

Chronic kidney disease

CT

Computed tomography

FGF23

Fibroblast growth factor-23

HA

Hydroxyapatite

HFTC

Hyperphosphatemic Familial Tumoral Calcinosis

IL-1

Interleukin-1

MIP

Maximum intensity projection

mCi

Millicurie

mCS

Microcalcification score

PET

Positron emission tomography

P_sva

Position on standardized vascular atlas

ROI

Region of interest

$\overrightarrow{RP}$

Relative position vector

RSD

Relative segment distance

$\overrightarrow{SF}$

Scaling factor vector

SUV

Standardized uptake value

SVA

Standardized vascular atlas

TBR

Target-to-background ratio

TOF

Time-of-flight

Data Availability

MATLAB code, as well as compiled code, is available upon request to the corresponding author.