Assessment of Quantitative [¹⁸F]Sodium Fluoride PET Measures of Knee Subchondral Bone Perfusion and Mineralization in Osteoarthritic and Healthy Subjects

Abstract

Objective:

Molecular information derived from dynamic [¹⁸F]sodium fluoride ([¹⁸F]NaF) PET imaging holds promise as a quantitative marker of bone metabolism. The objective of this work was to evaluate physiological mechanisms of [¹⁸F]NaF uptake in subchondral bone of individuals with and without knee osteoarthritis (OA).

Methods:

Eleven healthy volunteers and twenty OA subjects were included. Both knees of all subjects were scanned simultaneously using a 3T hybrid PET/MRI system. MRI MOAKS assessment was performed to score the presence and size of osteophytes, bone marrow lesions, and cartilage lesions. Subchondral bone kinetic parameters of bone perfusion (K₁), tracer extraction fraction, and total tracer uptake into bone (K_i) were evaluated using the Hawkins 3-compartment model. Measures were compared between structurally normal-appearing bone regions and those with structural findings.

Results:

Mean and maximum SUV and kinetic parameters K_i, K₁, and extraction fraction were significantly different between Healthy subjects and subjects with OA. Between-group differences in metabolic parameters were observed both in regions where the OA group had degenerative changes as well as in regions that appeared structurally normal.

Conclusions:

Results suggest that bone metabolism is altered in OA subjects, including bone regions with and without structural findings, compared to healthy subjects. Kinetic parameters of [¹⁸F]NaF uptake in subchondral bone show potential to quantitatively evaluate the role of bone physiology in OA initiation and progression. Objective measures of bone metabolism from [¹⁸F]NaF PET imaging can complement assessments of structural abnormalities observed on MRI.

Introduction

Abnormal bone physiology is a potential mechanism for the onset and progression of knee osteoarthritis (OA), a disease marked by degradation and loss of soft tissues like cartilage and the development of bone marrow lesions (BML) and osteophytes. Structural changes on magnetic resonance imaging (MRI) have been observed to precede changes in joint space width or structural bone changes on standard radiographs (1,2). The MRI Osteoarthritis Knee Score (MOAKS) has been widely used to semi-quantitatively classify and study multiple bone and soft tissue structural changes that are believed to be relevant for functional integrity of the knee and the pathophysiology of OA (3). This method characterizes bone and soft tissue findings by size within a specified subregion. The size of osteophytes, BML, and cartilage lesions have been associated with the severity of joint pain and progression of OA (4). While MRI is a useful tool for evaluating structural changes in cartilage and bone, quantitative assessment and imaging of subchondral bone physiology remains a challenge.

Molecular information derived from [¹⁸F]sodium fluoride ([¹⁸F]NaF) positron emission tomography (PET) imaging has shown promise as a marker of bone metabolism in multiple bone and joint disorders (5). Subchondral bone is metabolically active, and changes in subchondral bone remodeling and vascularization have been implicated in early OA (4,6–8). Uptake of [¹⁸F]NaF reflects bone metabolism and remodeling, and is correlated with mineral apposition rate assessed using bone histomorphometry (9). Following injection into the vasculature, [¹⁸F]NaF uptake is preferentially concentrated at sites of newly mineralizing bone, where fluoride ions exchange with hydroxyl ions of newly synthesized hydroxyapatite crystals on the bone surface (10,11). A common metric for PET image evaluation is the standard uptake value (SUV), the uptake of the radiotracer in a region normalized to the amount of tracer injected and body weight. Mean and maximum SUV measures enable simple, semi-quantitative evaluation of uptake in a specific region with low precision error (12–14). Increases in SUV measures are associated with the presence and severity of patellofemoral pain (15), BML and osteophytes (16), and adjacent cartilage changes (17,18). However, as SUV can be impacted by biologic factors and reconstruction methods, caution is advised when making comparisons between subject cohorts or in longitudinal studies (19,20). Additionally, SUV reflects total tracer uptake but does not differentiate between the specific mechanisms driving uptake. Fitting a kinetic model to dynamic changes in tissue [¹⁸F]NaF activity allows for fully quantitative evaluation of the rates of tracer delivery to bone (K₁, or bone perfusion), uptake into the intracellular bone compartment (k₃), and clearance back to plasma (k₂), as well as an overall metric of tracer uptake from plasma into bone (K_i) (21–25). These metrics have shown good agreement with other measures of bone metabolism: K_i is highly correlated with histomorphometric, serum, and urinary markers of bone formation and resorption in both high and low turnover bone diseases, and K₁ has been demonstrated to adequately represent bone blood flow as measured by [¹⁵O]H₂O (9,21,23,26–29). Kinetic measures of [¹⁸F]NaF uptake enable quantitative analysis of bone physiological parameters at specific locations in the knee to help us better understand mechanisms of tracer uptake in situations of altered bone physiology.

In combination with MRI, hybrid PET/MR imaging enables simultaneous assessment of PET functional data along with MRI high-resolution structural images and multiple soft tissue contrasts. Prior work applying PET/MRI in the knees of healthy volunteers characterized normal patterns of [¹⁸F]NaF uptake across bones tissue types and subchondral bone regions (30). Differences in SUV and kinetic parameters were observed between trabecular and cortical bone, in addition to regional differences in bone perfusion and tracer extraction fraction in subchondral bone (30). Additionally, in subjects with early OA, observed increases in [¹⁸F]NaF uptake measures of SUV and K_i were associated with quantitative MRI parameters of overlying cartilage (18) and also with semi-quantitative scores for cartilage and BML (31). However, the mechanisms behind the elevated tracer uptake in OA bone compared to healthy bone remain unclear. Kinetic modeling of [¹⁸F]NaF uptake to estimate rate parameters of tracer uptake and delivery holds promise as a technique that may help us understand patterns and physiological mechanisms of changes in bone metabolism associated with OA.

This preliminary study aims to utilize hybrid [¹⁸F]NaF PET/MR imaging and kinetic modeling of [¹⁸F]NaF uptake to evaluate differences in quantitative [¹⁸F]NaF uptake measures of bone vascularization and mineralization between healthy and OA subjects. Further, we assess how structural changes in bone and cartilage that are observable on MRI relate to quantitative subchondral bone metabolic parameters.

Methods

Study Population

This study was approved by the University Institutional Review Board. All subjects gave written informed consent and were requested to abstain from weight-bearing exercise for 24 hours prior to participating in the study. Eleven healthy volunteers with no knee pain (32.8 ± 6.3 years, BMI 22.66 ± 3.43 kg/m², 6 female) were compared with 20 OA subjects who were recruited based on knee pain and clinically-established osteoarthritis in at least one knee (56.6 ± 10.4 years, BMI 26.84 ± 4.78 kg/m², 14 female) (Table 1). In the OA group, 18 subjects had been diagnosed with mild to moderate OA with radiographs, and 2 subjects were diagnosed through symptomatic assessment including chronic pain and synovitis at least 2 years prior to the start of the study.

Table 1:

Subject information for Healthy and Osteoarthritic cohorts

		Healthy	Osteoarthritic
Subjects (N)		11	20
Female (N)		6	14
Age (years, mean (std))		32.8 (6.3)	56.6 (10.4)
BMI (kg/m2, mean (std))		22.66 (3.43)	26.84 (4.78)
Total bone subregions (N)		198	360
No MOAKS findings (N (%))		196 (98.9%)	144 (40.0%)
Bone Marrow Lesions	Total	1 (0.5%)	60 (16.6%)
	1	1	39
	2	0	20
	3	0	1
Osteophytes	Total	1 (0.5%)	188 (52.2%)
	1	1	108
	2	0	51
	3	0	29
Cartilage Lesions	Total	0 (0%)	122 (33.8%)
	1	0	27
	2	0	66
	3	0	29

PET/MR Image Acquisition

Both knees of all subjects were scanned simultaneously using a 3T whole-body time-of-flight hybrid PET/MRI system (GE SIGNA, GE Healthcare, Milwaukee WI) with two 16-channel flexible phased-array wrap coils (NeoCoil, Pewaukee WI) (32). Both knees were scanned in one PET bed in list mode (field of view 26 cm, matrix 384 × 384) for 50 ± 3.7 minutes immediately following intravenous injection of 93 ±4.4 MBq of [¹⁸F]NaF.

MR imaging was acquired simultaneously with PET acquisition. MR angiography images were acquired using a 3D inversion recovery spoiled gradient echo (IR-SPGR) sequence with repetition time (TR)/echo time (TE) = 21/2.1 ms, slices = 18 slices, slice thickness = 1.2 mm, and flip angle = 15°. MR-based attenuation correction of PET data was performed using a 2-point Dixon fat-water T1-weighted spoiled gradient echo MR sequence with TR/TE1/TE2 = 4.1/1.1/2.2 ms, field of view = 50 × 37.5 cm, matrix = 256 × 128, slice thickness = 5.2, overlap = 2.6 mm, 120 images/slab, and scan time = 18 seconds.

For morphologic MRI, coronal T2-weighted images were acquired using a 2D fat-suppressed FSE sequence with TR = 2500 ms, field of view = 16 cm, matrix = 320 × 224, slice thickness = 3 mm, echo train length = 12, TE = 54 ms. Sagittal fat-water separated images were acquired using a 2D IDEAL FSE sequence with TR/TE = 3000/30 ms, field of view = 16 cm, matrix = 320 × 256, slice thickness = 3 mm, echo train length = 8, and phase acceleration = 2. A double-echo in steady-state (DESS) sequence was used for tissue morphology and segmentation, with a TR/TE1/TE2 = 24.6/5.8/43.4 ms, field of view = 16 cm, matrix = 512 × 512, and slice thickness = 1.5 mm (33,34).

All PET image frames were reconstructed from acquired list-mode data using a time-of-flight reconstruction with resolution recovery corrections and a regularized reconstruction iterative algorithm (QClear, beta value of 350). To generate the image-derived input function and time activity curves (30), dynamic PET frames were reconstructed: for the image-derived input function (IDIF), frame times were 20 × 1 s, 10 × 10 s, 10 × 30 s, 5 × 1 min, and 2 min frames for the remaining duration of the scan; for time activity curves (TAC), frame times were 6 × 10 s, 10 × 1 min, and 2 min frames for the remaining duration.

Segmentation

To generate bone subregions, the first echo of the DESS image was first re-sampled to PET resolution using Horos software (Nimble Co LLC d/b/a Purview, Annapolis, MD) and used to manually create masks of the entire femur, patella, and tibia using ITK-Snap (35). Bone masks were then segmented further to create subchondral bone masks of between 0.5 and 3 mm using k-means clustering of registered water-only and fat-only Dixon images as described previously (36). Subchondral bone was manually subdivided into 9 subregions per knee representing the patella as well as the medial and lateral tibia and femur, then each portion of the femur was further manually subdivided into trochlear, central, and posterior regions (Figure 1A).

Figure 1:

Representative PD-weighted water-only IDEAL [A,D] anatomical water MRI images, [¹⁸F]NaF PET SUV images [B,E], and fusion images [C,F] of a healthy knee (top row) and a knee with osteoarthritis (bottom row). Anatomical subdivisions used for modified MOAKS scoring of bone marrow lesions, cartilage damage and osteophytes, as well as kinetic parameters for [¹⁸F]NaF uptake in subchondral bone, are indicated in [A]: patella (P); medial and lateral tibia (T); and trochlear (TF), central (CF), and posterior (PF) regions of the medial and lateral femur. In panel [D], note osteophytes at the inferior patella and posterior lateral femoral condyle (blue arrows), bone marrow lesions in the posterior lateral femoral condyle and posterior lateral tibia (yellow arrows), and full-thickness cartilage loss over the posterior lateral femoral condyle (green arrow).

MOAKS Scoring

Assessment of structural knee OA was performed using a modified MOAKS system. First, standard MRI Osteoarthritis Knee Score (MOAKS) (3) assessment was performed by a board-certified musculoskeletal radiologist (JM) with 8 years of experience using sagittal PD-weighted IDEAL, DESS (with multiplanar reformats), and coronal T2-weighted fat-saturated images. BML size as a percentage of MOAKS subregion volume was graded using scores of 0 (none), 1 (less than 33%), 2 (33–66%), or 3 (more than 66%), which included the volume of any associated subchondral cysts. Osteophytes were graded as MOAKS 0 (none), 1 (small), 2 (medium), or 3 (large) in the medial and lateral tibia; medial, lateral, inferior, and superior patella; and medial and lateral portions of the anterior, central, and posterior femur. The extent of cartilage loss, including partial and full-thickness loss, was graded as a percentage of the MOAKS subregion cartilage surface area using a score of 0 (none), 1 (less than 10%), 2 (10–75%), or 3 (more than 75%). Next, multiple MOAKS subregions within the patella and tibia were merged by using the maximum scores to arrive at a single score for the entire patella, medial tibia, and lateral tibia. Scores for regions representing the entire patella; medial and lateral tibia; and anterior, central, and posterior regions of the medial and lateral portions of the femur were used for comparison with [¹⁸F]NaF uptake (Figure 1A).

SUV and Kinetic Uptake Parameters

All SUV and kinetic parameter estimation was performed for individual subregions (Figure 1A). The mean and maximum standardized uptake values (SUVmean and SUVmax) were calculated from images obtained by averaging the last two frames of the dynamic study.

An IDIF was determined from [¹⁸F]NaF activity (kBq/mL) in the popliteal artery of each knee independently as previously described (30). The IDIF and TAC data were fit to a Hawkins two-tissue tracer kinetic model using a nonlinear regression (NLR) method as described previously (22,30,36) to estimate the rate constants: K₁ = bone perfusion (mL min⁻¹ mL⁻¹), k₂ = tissue clearance (min⁻¹), and k₃ = mineralization (min⁻¹). The rate constant k₄ representing bone clearance was defined as 0. NLR fitting to estimate these three rate parameters, along with parameters to account for partial volume fraction, blood fraction, and input dispersion estimate was performed for each bone subregion using COMKAT software (37).

From these parameters, the extraction fraction was defined as:

$$\mathit{Extraction}\mathit{fraction}=\frac{k_3}{k_2+k_3}$$ (1)

which represents the fraction of [¹⁸F] entering the bone tissue that binds to the bone matrix (as opposed to being cleared back into the plasma pool) and ranges in value from 0 to 1. The parameter K_i^NLR (K_i), the rate of clearance of [¹⁸F] from the plasma to the bone mineral compartment, was calculated using the formula (22):

$$K_i^{NLR}=K_1*extractionfraction$$ (2)

and has units of mL min⁻¹ mL⁻¹.

Statistical Analysis

Statistical comparisons of measures between: 1) Healthy and OA groups; 2) Healthy group regions and OA group regions that appear structurally normal (i.e. regions in healthy subjects with all MOAKS scores = 0 compared to regions in OA subjects with all MOAKS scores = 0); 3) structurally normal-appearing regions (all MOAKS scores = 0) in both healthy and OA groups, and OA group regions with BML, osteophytes and adjacent cartilage lesions (i.e. regions in OA group knees with MOAKS scores > 0); and 4) OA group regions with BML, osteophytes, and cartilage lesions (i.e. regions in OA group knees with MOAKS scores 1, 2, and 3) were made using a mixed effects model adjusted for clustering within subjects and for differences with bone region. The residuals met model assumptions of linearity and variance and were normally distributed. The mean and 95% confidence intervals of the measures were calculated, and a p value (alpha = 0.05) was calculated for comparisons. No post-hoc test for multiple comparisons was used due to the exploratory nature of this study (38,39). Healthy group knee regions that had a single MOAKS score greater than 0 were excluded from the analysis as these were assumed not to be healthy. All statistical comparisons were made using Minitab 19 software (Minitab LLC, State College PA). Violin plots were used to visualize the distribution of measures in different groups, where distribution plot size is scaled by the number of observations in each score category.

Results

Structural MRI Findings in Healthy and OA Cohorts

In the 22 Healthy-group knees studied, there were 196 regions (98.9%) that appeared structurally normal and had no MOAKS findings (all MOAKS scores = 0) (Table 1). In the 40 OA-group knees studied, only one knee had no MOAKS findings in all regions. There were a total of 144 regions (40.0%) with no MOAKS findings. Within the 216 regions where there were one or more MOAKS findings, there were 60 regions with BML, 188 regions with osteophytes, and 122 regions with adjacent cartilage loss. There were 104 regions with a single feature, 70 regions with two features, and 19 regions with all three features. Representative PET/MRI images (Figure 1) demonstrate the presence of osteophytes, BML, and adjacent cartilage loss in healthy and OA subjects on anatomical MRI images and on PET SUV images.

NaF Uptake in Healthy and OA Cohorts

On the whole, the OA group had significantly greater SUVmean (p = 0.001), SUVmax (p < 0.001), K_i (p = 0.009), and K₁ (p = 0.003), and significantly lower extraction fraction (p = 0.001) compared to the Healthy group (Figure 2, Table 2, and Supplemental Table 1). Furthermore, structurally normal-appearing regions (all MOAKS scores = 0) in OA-group knees (“Normal OA” or “N OA” bone) had significantly greater SUVmean (p = 0.004), SUVmax (p = 0.002), and K₁ (p = 0.020), and significantly lower extraction fraction (p = 0.004) compared to structurally normal-appearing Healthy-group regions (MOAKS=0) (“Normal Healthy” or “N H”). K_i was not significantly different between these groups (p = 0.116).

Figure 2:

There was a significant difference in SUVmean (not shown), SUVmax, K_i, K₁, and extraction fraction between healthy and osteoarthritic subchondral bone [A-D]. There was also a significant difference in SUVmean (not shown), SUVmax, K₁, and extraction fraction between bone that appeared structurally normal in healthy knees (“Normal Healthy”) and in OA knees (“Normal OA”) [E-H]. * = p < 0.05, ** = p < 0.01, *** = p < 0.001.

Table 2:

Mean and 95% confidence intervals for uptake parameters

Group		Count	SUVmean mean (95% CI)	SUVmax mean (95% CI)	Ki mean (95% CI)	K1 mean (95% CI)	Extraction fraction mean (95% CI)
Healthy		196	0.54 (0.50, 0.58)	1.55 (1.39, 1.71)	0.0084 (0.0078, 0.0090)	0.0098 (0.0091, 0.0104)	0.875 (0.858, 0.893)
OA Normal-appearing regions		144	0.92 (0.84, 1.01)	3.21 (2.73, 3.68)	0.0098 (0.0090, 0.0106)	0.0177 (0.0154, 0.0199)	0.690 (0.652, 0.728)
OA All regions		360	1.48 (1.34, 1.62)	6.48 (5.85, 7.12)	0.0143 (0.0133, 0.0153)	0.0261 (0.0240, 0.0282)	0.650 (0.626, 0.674)
Bone Marrow Lesions	1	39	1.88 (1.43, 2.34)	11.50 (9.05, 13.94)	0.0193 (0.0152, 0.0234)	0.0294 (0.0234, 0.0353)	0.705 (0.629, 0.781)
	2	20	3.56 (2.18, 4.94)	17.91 (14.37, 21.45)	0.0290 (0.0214, 0.0366)	0.0463 (0.0336, 0.0589)	0.669 (0.574, 0.765)
	3	1	3.43	25.84	0.0332	0.0468	0.708
Osteophytes	1	108	1.67 (1.40. 1.94)	7.56 (6.41, 8.70)	0.0162 (0.0144, 0.0180)	0.0292 (0.0258, 0.0327)	0.624 (0.583, 0.665)
	2	51	1.71 (1.40, 2.02)	9.22 (7.22, 11.22)	0.0167 (0.0139, 0.0196)	0.0317 (0.0263, 0.0371)	0.608 (0.536, 0.679)
	3	29	2.96 (2.07, 3.85)	12.05 (9.25, 14.86)	0.0236 (0.0176, 0.0295)	0.0488 (0.0369, 0.0607)	0.528 (0.453, 0.602)
Cartilage Lesions	1	27	1.57 (1.27, 1.87)	8.61 (6.14, 11.09)	0.0153 (0.0126, 0.0180)	0.0284 (0.0215, 0.0353)	0.610 (0.533, 0.687)
	2	66	2.42 (1.91, 2.93)	11.57 (9.76, 13.37)	0.0221 (0.0188, 0.0254)	0.0396 (0.0332, 0.0495)	0.645 (0.588, 0.703)
	3	29	2.24 (1.65, 2.83)	10.46 (7.88, 13.04)	0.0183 (0.0136, 0.0230)	0.0382 (0.0310, 0.0453)	0.506 (0.412, 0.599)

CI: confidence interval

OA: Osteoarthritis group

Normal-appearing regions: all MOAKS scores = 0

MOAKS Findings and NaF Uptake

Regions with MOAKS size 1–3 BML, osteophytes, and adjacent cartilage lesions were associated with significantly elevated SUVmean, SUVmax, K_i, and K₁ compared to Normal Healthy and Normal OA bone (Figure 3, Table 2, and Supplemental Table 1). While increased [¹⁸F]NaF uptake was present in regions with both cartilage loss and adjacent BML or osteophytes, it was also present in regions with cartilage loss alone (Figure 4). The extraction fraction was significantly lower in all OA bone with MOAKS findings compared to Normal Healthy. Larger MOAKS size 2 or 3 findings were associated with significantly higher SUVmean, SUVmax, K_i, and K₁ compared with size 1 findings. Additionally, the extraction fraction in bone adjacent to MOAKS size 3 cartilage lesions was significantly lower than bone with adjacent MOAKS size 2 cartilage lesions.

Figure 3:

Standardized uptake value (SUV) and kinetic parameters of [¹⁸F]NaF uptake in subchondral bone regions was compared with MOAKS scores for bone marrow lesions (left), osteophytes (middle) and adjacent cartilage loss (right) in the corresponding region. Regions with bone marrow lesions and osteophytes had significantly different SUVmean (not shown) and SUVmax [A,E,I] and rates K_i [B,F,J] and K₁ [C,G,K] compared to structurally normal bone in healthy knees (“N H”) and compared to structurally normal bone regions in an OA knee with MOAKS findings in other regions (“N OA”). Extraction fraction [D,H,L], though significantly altered in OA bone compared to healthy bone, was not significantly different between structurally normal bone in OA knees and regions with MOAKS findings. SUV, K_i, K₁, and extraction fraction in subchondral bone were significantly altered in regions with adjacent cartilage loss compared to regions with normal-appearing cartilage and bone. There was only one bone subregion with a MOAKS size 3 bone marrow lesions and the value of measures in this region are marked with an asterisk.

Figure 4:

Representative PD-weighted water-only IDEAL anatomical MRI image (left), [¹⁸F]NaF PET SUV image (center), and fusion images (right) demonstrating findings of cartilage loss but no adjacent bone marrow lesions or osteophytes [A] and of cartilage loss with adjacent bone marrow lesions and osteophytes [B]. In [A], there is partial-thickness cartilage loss over the posterior medial tibia (green arrow) without adjacent bone marrow lesions or ipsicompartmental osteophytes. Note also increased [18F]NaF uptake at the medial head of gastrocnemius origin (white arrow). In [B] there are areas of full-thickness cartilage loss over the central lateral femoral condyle and lateral tibial plateau (green arrows) with subtle underlying bone marrow lesions and large ipsicompartmental osteophytes (blue arrows).

Discussion

All tracer uptake measures were significantly different in OA subchondral bone compared to Healthy bone. Previous work has shown regions of increased [¹⁸F]NaF uptake in an OA population but did not compare uptake to healthy human subjects (15,16,31). Kinetic analysis was able to show that increased tracer uptake was driven by increased K₁, or delivery of the tracer to subchondral bone regions. Given the good agreement between K₁ and bone blood flow, this suggests that bone vascularity or perfusion is increased in OA subjects compared to healthy controls (21,26). Others have also found that areas of [¹⁸F]NaF uptake do not always correspond to structural changes in bone or cartilage on MRI (15,16). In this work, we showed that there are significant differences in quantitative [¹⁸F]NaF uptake parameters between normal-appearing regions in Healthy-group knees and normal-appearing regions in OA-group knees. This adds further evidence to prior work which has hypothesized that increased bone metabolic activity may serve as an early marker of OA disease (40). However, additional work is necessary to study longitudinal relationships between increased [¹⁸F]NaF uptake and disease onset and progression.

Compared to bone that appears structurally normal on MRI in Healthy-group and OA-group subjects, the measures SUVmean, SUVmax, K_i, and K₁ were significantly elevated in regions with BML, periarticular osteophytes, and adjacent cartilage lesions. These findings are in agreement with previous literature on MRI findings and [¹⁸F]NaF PET SUV and ^99mTc-HDP SPECT/CT SUV in OA (15,16,31,41). Results suggest that the abnormal bone metabolism in these regions is primarily the result of significantly greater bone perfusion rates when compared to bone regions that appear normal on MRI. Our results also suggest strong relationships between subchondral bone metabolic abnormalities and changes in overlying cartilage, in agreement with prior studies utilizing PET/MRI and CT arthrography (31,42). Cartilage lesion size was associated with altered K_i, K₁, and extraction fraction in adjacent subchondral bone. Changes were present in regions with cartilage loss adjacent to osteophytes and BML, but also in regions with cartilage loss alone. Others have similarly found strong spatial relationships between SUV and changes in overlying cartilage in the absence of BML or osteophytes (17,18). In ACL injured subjects, SUVmax was correlated with differences in deep and superficial cartilage T2 relaxation times, indicating local breakdown of cartilage microstructure (17). In early OA models, such as after ACL injury, increased subchondral bone remodeling is regarded as a potential mechanism for the progression of OA (6,8). Loss of cartilage structural and functional integrity may be associated with vascular invasion and increased remodeling in adjacent subchondral bone (7). Furthermore, histological studies have identified that vascular invasion of subchondral bone is present in the development of both osteophytes and BML (43–46). Measures sensitive to changes in bone perfusion and turnover may add insight to the role of specific features in the progression of OA.

Measures of [¹⁸F]NaF delivery and uptake into the bone can be used to objectively and quantitatively assess bone metabolism. Both SUV and K_i show similar trends with MRI findings, which has previously been suggested in situations where the disease status does not substantially perturb the arterial input function compared to a healthy population (11). SUV may be a preferred measure for its simplicity and lower precision error compared to tracer uptake rate, however the Hawkins model method of kinetic parameter estimation enables a complete evaluation of bone tracer kinetics to provide objective measures of the mechanisms of tracer uptake that may be compared across patient cohorts and studies. Further, while subject scan times are similar or slightly longer with dynamic imaging, the overall time commitment is decreased due to elimination of 30–60 minutes of time for tracer distribution in conventional static PET imaging. Furthermore, dynamic PET imaging is 2–3 times faster than alternative methods for imaging bone metabolism such as SPECT/CT (41). While there are advantages to this technique, there are also a number of limitations. Errors in the estimation of dynamic [¹⁸F]NaF uptake parameters can arise from errors in the arterial input function, a measure of the activity delivered to the region of interest through the vasculature. Here, we used an IDIF using measures of activity in the popliteal artery and calibrating the input function with measures of venous blood at the end of the scan, approximately 60 minutes after injection, when venous and arterial blood [¹⁸F] concentration is equal (47). This is the simplest method for deriving an arterial input function, as it avoids repeated blood sampling, but can introduce errors from partial volume and noise. This method can overestimate arterial activity, which can consequently affect estimates of kinetic uptake parameters K_i and K₁ (47). Efforts were made to minimize errors from partial volume and noise by using gradients to identify voxels in the artery with the highest activity and avoid including peripheral voxels when generating the arterial segmentation. Our method also used the assumption that the tracer is bound irreversibly to bone and so the rate k₄ can be approximated to 0. This assumption is generally believed to be reasonable as it improves the precision for estimating K_i due to the small value and poor precision of k₄ (12,48). Finally, we limited our evaluation to larger bone regions in the bones of the knee rather than evaluating specific structural features. As PET images have limited resolution, estimating kinetic parameters in larger regions reduced the impact of partial volume and low resolution on measure variability.

It is necessary to point out several other limitations of the study. Due to the nature of the study, which exposes subjects to ionizing radiation, small patient cohorts were used to examine the potential of this method to study bone metabolic abnormalities related to OA progression. Healthy subject data from a prior study were used in this comparison to avoid subjecting additional healthy volunteers to radiation. As such, the Healthy group was not age- or BMI-matched to the OA cohort. These factors were not accounted for in the mixed effects model as improved model fitting was achieved with a limited model, which was deemed more appropriate due to the exploratory nature of the study and relatively small sample size. Additionally, the OA cohort was recruited based on clinical OA diagnosis and not radiographic severity, and had varied numbers and distributions of structural findings on MRI. For example, there was only 1 instance of a MOAKS size 3 BML which limited our comparisons for this group. Furthermore, these subjects were scanned at one time point. Longitudinal studies are necessary to evaluate spatio-temporal relationships between structural features, [¹⁸F]NaF uptake, and changes in bone metabolism in the progression of OA.

In conclusion, kinetic parameters of the delivery and uptake of [¹⁸F]NaF were observed to be significantly different between Healthy and OA groups as well as between structurally normal-appearing regions between the two groups. Metabolic subchondral bone differences between Healthy subjects and OA subjects were observed in OA-group regions with BML, osteophytes, and adjacent cartilage loss. Kinetic analysis further showed that differences in total [¹⁸F]NaF bone uptake, a marker of bone metabolism, was driven primarily by increased delivery of the tracer, suggesting increased vascularity in subchondral bone in OA. Kinetic parameters of [¹⁸F]NaF uptake present an opportunity to probe numerous markers of bone metabolism and their role in disease progression. Coupled with MRI, hybrid PET/MRI can further our understanding of the impact of specific bone and cartilage features in the progression of OA as well as spatial relationships between them.