Quantitative Dynamic Contrast-Enhanced MRI Identifies Radiation-Induced Vascular Damage in Patients With Advanced Osteoradionecrosis: Results of a Prospective Study

Abstract

Purpose

We aim to characterize the quantitative dynamic contrast-enhanced (DCE) magnetic resonance imaging (MRI) parameters associated with advanced mandibular osteoradionecrosis (ORN) compared with the contralateral normal mandible.

Methods and Materials

Patients with a diagnosis of advanced ORN after curative-intent radiation treatment of head and neck cancer were prospectively enrolled after institutional review board approval and study-specific informed consent were obtained. Quantitative maps generated with the Tofts and extended Tofts pharmacokinetic models were used for analysis. Manual segmentation of advanced ORN 3-dimensional volume was done using anatomic sequences to create ORN volumes of interest (VOIs). Subsequently, normal mandibular VOIs were segmented on the contralateral healthy mandible of similar volume and anatomic location to create control VOIs. Finally, anatomic sequences were coregistered to DCE sequences, and contours were propagated to the respective parameter maps.

Results

Thirty patients were included. The median time to ORN diagnosis after completion of IMRT was 38 months (range, 6–184 months), whereas median time to ORN progression to advanced grade after initial diagnosis was 5.6 months (range, 0–128 months). There were statistically significant higher K^trans and V_e in ORN-VOIs compared with controls (0.23 vs 0.07 min⁻¹, and 0.34 vs 0.15; P <.0001 for both). The average relative increase of K^trans in ORN-VOIs was 3.2-fold higher than healthy mandibular control VOIs. Moreover, the corresponding rise of V_e in ORN-VOIs was 2.7-fold higher than in the controls. Using combined K^trans and V_e parameters, 27 patients (90%) had at least a 200% increase of either of the studied parameters in the ORN-VOIs compared with their healthy mandible VOIs.

Conclusions

Our results confirm that there is a quantitatively significant higher degree of leakiness in the mandibular vasculature as measured using DCE-MRI parameters of areas with advanced ORN versus healthy mandible.

Introduction

Osteoradionecrosis (ORN) of the mandible is a debilitating complication of external beam radiation therapy (EBRT) for patients with head and neck cancer.^1–3 Head and neck squamous cell carcinoma (HNSCC) has an estimated annual incidence of approximately 62,000 cases in the United States.⁴ The incidence of human papillomavirus (HPV)-associated HNSCC continues to rise unabated, and it is expected to continue to rise for the next 2 decades until the effects of immunization begin to affect incidence.^5,6

Recent data show the rate of ORN development in patients with HNSCC after EBRT is approximately 7% despite aggressive dental care and close follow-up.⁷ Even with a relatively low incidence, the prevalence and burden of ORN is expected to rise because of the excellent prognosis for HPV-positive patients (ie, 5-year overall survival of 80%−90% for most patients). Therefore, it is expected that the United States will accumulate a large population of adults with a history of mandibular radiation, a nearly 10-fold increase compared with historical trends when prevalence was lower owing to higher mortality of non–HPV-associated cancers.

Despite the use of more conformal EBRT techniques, such as intensity modulated radiation therapy (IMRT), the mandible remains exposed to significant radiation doses because of its close proximity to target volumes; this can eventually lead to the development of ORN, especially when coupled with infection or dental manipulation.^7,8 Early-stage ORN can be controlled with conservative measures such as antibiotics, surgical debridement, hyperbaric oxygen therapy, pentoxifylline, or tocopherol.^9,10 However, progression to advanced ORN typically requires extensive surgical resection and complex reconstruction and leads to a substantial reduction in the quality of life of HNSCC survivors.^11,12

Anatomic imaging using computed tomography (CT) or conventional magnetic resonance imaging (MRI) does not identify ORN-related bony changes until relatively late in the process, when the patient is generally already experiencing symptoms.^13,14 Dynamic contrast-enhanced (DCE) MRI is a clinically available imaging method that is shown to detect early-stage idiopathic osteonecrosis of the femur otherwise not visible on conventional MRI.¹⁵ DCE-MRI parameters can be used to monitor bone healing secondary to trauma or fracture and any chronic changes in bone health associated with age-related osteoporotic changes.^16–18

The most commonly accepted biologic mechanism of ORN development remains that summarized by Marx’s 3 H’s of hypoxic, hypovascular, and hypocellular tissue. Therefore, we expect that shifts in vascularity portend the development of ORN.¹⁹ Hence, we focused on DCE-MRI as opposed to other imaging modalities. Our group has recently demonstrated that DCE-MRI can be used to detect alterations in bone vascularity after definitive radiation therapy to patients with head and neck cancer.²⁰ However, we do not yet know how early changes in bone vascularity during radiation correlate with subsequent development of ORN. To develop a predictive, imaging-based biomarker of ORN development, it is critical to identify DCE-MRI parameters in patients with existing ORN. This will facilitate the discrimination of the quantitative DCE-MRI parameters associated with injured versus healthy mandibular subregions. This characterization will ultimately serve as a guide to monitor temporal DCE-MRI changes after EBRT in an attempt to detect mandibular pathology before the development of symptoms.

Considering the existing clinical and preclinical data, we hypothesized that ORN is associated with critical changes in bone vascularity reflected in common DCE-MRI parameters (V_e and Ktrans) as a reflection of overall poor vascular flow and integrity. To this end, we sought to characterize the quantitative DCE-MRI parameters associated with the established diagnosis of advanced mandibular ORN compared with a healthy mandible in the context of a prospective clinical study with high intrinsic imaging acquisition consistency.

Methods and Materials

Patient selection

Patients with a confirmed diagnosis of advanced ORN that developed after curative-intent radiation treatment of head and neck cancer were prospectively enrolled in an observational imaging study (NCT03145077) after institutional review board approval and study-specific informed consent were obtained. Eligibility criteria included age >18 years, pathologic evidence of head and neck malignancy with history of curative-intent EBRT, patients with clinically confirmed high-grade ORN requiring surgical intervention, good performance status (Eastern Cooperative Oncology Group score, 0–2), and no contraindications to MRI. Clinical staging of ORN was conducted using the Common Terminology Criteria for Adverse Events version 4.0.

DCE-MRI imaging

DCE-MRI scans were obtained using GE 3.0-T Discovery MR750 scanners with a 6-channel Flex phased-array coil (GE Healthcare Technology, Milwaukee, WI). Before DCE-MRI, T1 mapping was performed using 6 variable flip angles (FAs) in a 3D spoiled gradient recalled echo sequence (FA = 2, 5, 10, 15, 20, and 25 degrees; TR/TE = 5.5/2.1 ms; FOV = 25.6 cm; slice number = 30; voxel size = 2 × 2 × 4 mm³). The DCE-MRI was acquired using a multiphase 3D Fast spoiled gradient recalled echo sequence to gain sufficient signal-to-noise ratio, contrast, and temporal resolution (FA = 15 degrees; TR/TE = 3.6/1 ms; voxel size = 2 × 2 × 4 mm³; temporal resolution = 5.5 seconds; number of repetitions = 56; pixel bandwidth = 326 Hz; ASSET acceleration = 2). Gadobutrol (Gadovist; Bayer Healthcare, Leverkusen, Germany) was administered at a dose of 0.1 mmol/kg of body weight at 3 mL/s, followed by the same amount of saline at 3 mL/s, via a power injector (Spectris MR Injector; MedRad, Pittsburgh, PA).

Computation of the kinetic model

Postprocessing of the DCE-MRI images was performed at a workstation running in-house MATLAB-based pipeline (MathWorks, Natick, MA). Before quantitative analysis, motion correction and noise suppression were applied using simultaneous spatial and temporal higher-order total variations (HOTVs) regularizations as described by our group and by Chan et al.^21,22 As shown in Figure 1, filtering with HOTVs demonstrated noise suppression and motion reduction.²² To quantify the physiologic parameters using DCE-MRI, the arterial input function (AIF) of the contrast agent (CA) entering the tissue was determined individually. AT1 map was calculated to convert the signal intensity into concentration time course. The extended Tofts model assumes that the CA resides in and exchanges between 2 compartments in the tissue: the vascular space and extracellular extravascular space (EES). When this model is used, the differential equation describing the kinetic behavior of the CA in the tissue of interest is given by:

$$\frac{{\text{dC}}_{\text{TOI}}(\text{t})}{\text{dt}}=\left(K^{trans}+v_p\cdot k_{ep}\right)\cdot {\text{C}}_p(\text{t})-k_{ep}\cdot {\text{C}}_{\text{TOI}}(\text{t})+v_p\cdot \frac{{\text{dC}}_p(\text{t})}{dt}$$ (1)

where C_TOI(t) and C_p(t) are the concentrations of the CA at time t in the tissue of interest (TOI) and blood plasma, respectively, K^trans are the transfer (CA permeability rate) constants between blood plasma and the EES of the TOI. k_ep = K^trans/V_e is the transfer (CA permeability rate) constant (min⁻¹) from the TOI back to the blood plasma, where V_e is the distribution volume fractions of CA in the EES per unit volume of tissue. When the kinetic model includes a vascular term, V_p, this is the capillary plasma volume fractions per unit volume of tissue. Otherwise, by ignoring vascular term, the extended Tofts model is reduced to the Tofts model as

$$\frac{{\text{dC}}_{\text{TOI}}(t)}{\text{dt}}=K^{trans}\cdot {\text{C}}_p(\text{t})-k_{ep}\cdot {\text{C}}_{\text{TOI}}(t)$$ (2)

Fig. 1.

Consecutive dynamic contrast-enhanced magnetic resonance imaging frame time series (from left to right) before (top row) and after (bottom row) filtering with higher-order total variations.

The AIF for the pharmacokinetic (PK) analysis was derived from selected arteries. The relevance of several analytical AIF models in DCE-MRI has been investigated extensively.²³ Computer simulations were performed recently to evaluate and compare a population of AIF models with the Parker model.²⁴ The results demonstrated that a 6-parameter linear function plus biexponential function AIF model was almost equivalent to the Parker AIF. It should be noted that the former is computationally faster and more reliable in functional fitting compared with the Parker AIF. However, predetermining the arrival time (AT) and time to peak (TTP) of upslope for each AIF time series is usually not accurate in using the 6-parameter model. Therefore, we extended the 6-parameter model to a biexponential and bilinear function where the AT and TTP of the upslope are included as parameters to be estimated in AIF fitting.

To acquire the corresponding AT and TTP time points of the upslope for each AIF time series, we designed a special cost function, where fitting is performed with global optimization on the AIF model function with 7 parameters. The new AIF model function (AIFM) is defined in Equation 3 as AIFM, where p1 to p7 are parameters to be determined by functional fitting,²³ min is minimal value operator, max is the maximal value operator, abs is absolute value operator, and p4 and p5 are AT and TTP to be determined; min(p4, p5) gives AT, max(p4, p5) is TTP, t is time points, and “uplimit” is a constant that is estimated by the maximal possible value of data to be fitted (eg, double the maximum value in the time course).

The new AIFM was defined as

$$\begin{gathered} \mathrm{AIFM}=\min \stackrel{part1}{\overbrace{(\max \left(p1,p1+(p2-p1)\cdot \frac{t-\min (p4,p5)}{abs(p5-p4)}+(p3-p1)\cdot \frac{t-\min (p4,p5)}{abs(p5-p4)}\right)}}, \\ \stackrel{part2}{\overbrace{\min (uplimit,p2\cdot \exp (-p6\cdot (t-\max (p5,p4)))+p3\cdot \exp (-p7\cdot (t-\max (p5,p4)))))}} \end{gathered}$$ (3)

For each given AIF time point, the function can be described in 2 separate parts, as indicated in Equation 3. Part 1 of the function determines the AT and TTP of the upslope and the bilinear functions through a maximization of operations containing parameters p1, p2, p3, p4, and p5. Here p1 (first term in the bracket of part 1) will fit the static signal, and the second term in the bracket in part 1 will represent the upslope approximated by bilinear functions; p2 and p3 represent the endpoints at TTP of each linear approximation. Moreover, the maximum value at each time point of part 1 represents the approximation of signal, and we therefore apply a max operator on the entire set of terms.

Part 2 of the function determines the signal approximation after TTP of the upslope with biexponential functions through a minimization of operations containing parameters p2, p3, p4, p5, p6, and p7, and the uplimit, which can be seen as a constant larger than the ceiling value the curve can take. Opposite to the implementation in part 1, the biexponential functions should always be below the uplimit, so a min operator is applied to the set of terms in part 2. Because of the min operator in part 2, we can efficiently restrict the parameter estimation of the biexponential functions.

Because part 1 is strictly increasing and part 2 is strictly decreasing, we applied a min operator on the combination of part 1 and part 2. Subsequently, the minimum of part 1 and part 2 is noted to be AIFM as in Equation 2. By enforcing these settings, the original complicated constrained (multiple) optimization problem in data fitting is changed into an unconstrained problem. Finally, the fitting cost function is implemented by optimization on ||DATA – AIFM||, where DATA could be DCE concentration or DCE signal. The fitting cost function is implemented by optimization on ||DATA – AIFM||, where DATA could be DCE concentration or DCE signal. Using this extended AIF model function, the AIF fitting can be completed in a more precise and reliable manner. Figure 2 shows the fitting process and the fitting results where a 56-time-point AIF time series is presented. The PK modeling was done on a pixel-by-pixel basis using a linearization equation of the models used (ie, Tofts and extended Tofts) as described by Murase.²⁵ Subsequently, we implemented the linear least squares method to acquire the PK parameters.^25,26 This method is preferred to the conventional nonlinear least squares method²⁷ because it is faster, does not require initial estimation, and has no local optima problems.

Fig. 2.

The fitting process and fitting results where a 56-time-point arterial input function time series (in cyan) is presented.

Image segmentation and registration

Manual segmentation of mandibular volumes harboring ORN was done by an expert radiation oncologist (A.S.R.M.) and reviewed by an expert neuroradiologist (J.M.J.). The segmentation was performed using the MRI anatomic sequences (T1, T2, and T1 with contrast) and coregistered contrast-enhanced CT scans (ie, acquired within 2 weeks of the MRI with no interval therapy) to create ORN volumes of interest (ORN-VOIs) for all included patients. The segmentation included abnormal signal intensity or irregular gadolinium enhancement of bone marrow and soft tissues seen on MRI^28,29 and cortical erosions, sequestrations, or fractures seen on CT.³⁰ Subsequently, normal mandibular VOIs were created on the contralateral healthy mandible of similar volume and anatomic location (ie, mirror image) to create self-control VOIs. Finally, the MRI anatomic sequences were coregistered to the DCE-MRI sequences, and then the contours were propagated to the respective quantitative parameter maps. This workflow is graphically summarized in Figure 3. For dosimetric correlation, the original planning CT scans and dose grids were retrieved when available to extract mandibular dose parameters (mean and maximum dose). In addition, ORN-depicting CT scans were coregistered to planning CT scans using a validated commercial image registration software (Velocity AI 3.0.1). Finally, ORN-VOIs were mapped to planning CT scans, and dose grid and dosimetric parameters were extracted for each VOI. The RT dose parameters included minimum, mean, dose to 95% volume (D95%), and maximum dose to ORN-VOIs in grays.

Fig. 3.

Workflow of advanced osteoradionecrosis analysis. Abbreviations: CT = computed tomography; DCE-MRI = dynamic contrast-enhanced magnetic resonance imaging; VOI = volume of interest.

Statistical analysis

Continuous data were presented as mean ± SD, and categorical data were presented as proportions. The comparison of quantitative DCE-MRI parameters between ORN and control VOIs was assessed using the nonparametric Wilcoxon signed-rank test. The effect size was calculated, and the magnitude of the effect size was determined using the Cohen criteria where an r of 0.1 = small effect, 0.3 = medium effect, and 0.5 = large effect.³¹ The nonparametric Spearman rho test was used to measure the correlation between radiation dose to ORV-VOIs and DCE-MRI parameters. P < .05 was deemed statistically significant. All statistical analyses were performed using JMP 14 Pro (SAS Institute, Cary, NC).

Results

Patients

Thirty patients with grade 3 ORN requiring surgery were included. Median age at diagnosis was 58 years (range, 19–78 years), and 83% were men. The site of tumor origin was in the oropharynx, oral cavity, salivary glands, and nasopharynx in 13, 9, 6, and 2 patients, respectively. IMRT was the radiation technique for all patients. The median IMRT prescription dose was 68 Gy in 32 fractions. The median time to ORN diagnosis after completion of IMRT was 38 months (range, 6–184 months), whereas the median time to ORN progression to advanced grade after initial diagnosis of ORN was 5.6 months (range, 0–128 months). Table 1 summarizes patient, disease, and treatment criteria.

Table 1.

Patient, disease, and treatment characteristics

Characteristic	Patients (N = 30)
Median age, (range), y	58 (19–78)
Sex, n (%)
Male	25 (83)
Female	5 (17)
Ethnicity, n (%)
White	29 (97)
African American	1 (3)
Smoking status, n (%)
Never	11 (37)
Former	13 (43)
Current	6 (20)
Smoking pack-years, mean (SD)	14.7 (24)
Disease subsite, n (%)
Nasopharynx	2 (7)
Oropharynx	13 (43)
Oral cavity	9 (30)
Salivary glands	6 (20)
T stage, n (%)
T1	7 (23.33)
T2	7 (23.33)
T3	9 (30)
T4	4 (13.33)
Recurrence	3 (10)
N stage, n (%)
N0	9 (30)
N1	3 (10)
N2	15 (50)
Recurrence	3 (10)
HPV status (p16 IHC), n (%)
Positive	8 (26.66)
Negative	2 (6.66)
Unknown	20 (66.66)
Pre-RT dental status, n (%)
No dental procedures	32 (47)
Dental extractions	35 (51.5)
Edentulous	1 (1.5)
Mean radiation dose (SD), Gy	66.1 (4.7)
Radiation fractions, mean (SD)	32 (2.8)
Chemotherapy, n (%)
None	11 (36.66)
Induction	1 (3.33)
Concurrent with RT	14 (46.66)
Induction + Concurrent	4 (13.33)
Surgery, n (%)
Yes	14 (46.66)
No	16 (53.33)

Abbreviations: HPV = human papilloma virus; RT = radiation therapy; SD = standard deviation.

Radiation dose

The RT dosimetric data were available for 21 patients (70%). The average of mean and maximum doses to the entire mandibular volumes were 51.4 Gy (range, 35–64 Gy) and 69.4 Gy (range, 52–76 Gy), respectively. The average of minimum doses to ORN VOIs (ie, the isodose line that covers 100% of the ORN volume) was 46.7 Gy (range, 26–66 Gy). The average of mean and maximum doses to ORN VOIs were 62.2 Gy (range, 44–75 Gy) and 67.9 Gy (range, 51–76 Gy), respectively; the average D95% was 55.6 Gy (range, 32–69 Gy).

DCE-MRI parameters

The median volume of segmented VOIs was 5.2 cm³ (range, 1.8–10.9 cm³). Using the extended Tofts model, the average K^trans values in ORN-VOIs were significantly higher compared with controls (0.23 ± 0.25 vs 0.07 ± 0.07 min⁻¹; P < .0001). The average relative increase of K^trans in ORN-VOIs was 3.2-fold those from the healthy mandibular control VOIs (range, 1.2–10.3). The effect size was large, with r = 0.52.

Likewise, the average V_e values in ORN-VOIs was significantly higher compared with controls (0.34 ± 0.27 vs 0.15 ± 0.15; P < .0001). The average relative increase of V_e in ORN-VOIs was 2.7-fold those in the healthy mandibular control VOIs (range, 1.1–6.9). The effect size was also large, with r = 0.69.

Using combined K^trans and V_e parameters, 27 patients (90%) displayed at least double the increase of either of the studied parameters in the ORN-VOIs compared with their healthy mandible control VOIs.

V_p was also significantly higher in ORN-VOIs compared with controls (0.17 ± 0.2 vs 0.07 ± 0.12; P < .0001). However, K_ep, as expected, did not show a significant difference between ORN-VOIs versus controls (0.5 ± 0.19 vs 0.46 ± 0.13 min^-1; P = .2) because of the increase in both K^trans and V_e parameters. Figure 4 depicts the comparison of K^trans, V_e, K_ep, and V_p values in ORN-VOIs compared with controls. Detailed histograms of patients’ ORN versus control VOI DCE-MRI parameters are presented for the entire cohort as supplemental material (Appendix E1).

Fig. 4.

Boxplots showing the comparison of dynamic contrast-enhanced magnetic resonance imaging parameters between osteoradionecrosis and control volumes of interest. *Statistically significant P values. Connected lines represent each patient parameter value changes in osteoradionecrosis versus control volumes of interest.

Using the Tofts model, the mean K^trans values in ORN-VOIs were likewise significantly higher compared with controls (0.27 ± 0.29 vs 0.08 ± 0.08 min⁻¹; P < .0001). The average relative increase of K^trans in ORN-VOIs was 3.4-fold that of the control VOIs (range, 1.5–12.3). In addition, the mean V_e values in ORN-VOIs was significantly higher compared with controls (0.34 ± 0.28 vs 0.13 ± 0.13; P < .0001). The average relative increase of V_e in ORN-VOIs was 4.04-fold that of the healthy mandibular control VOIs (range, 1.2–15.3). Using combined K^trans and V_e parameters also showed that 90% of patients had more than a 2-fold increase in either of the studied parameters in the ORN-VOIs compared with their healthy mandible control VOIs.

Using the Spearman rho test, there were no significant correlations between any of the dosimetric parameters of ORN-VOIs (minimum, mean, D95% and maximum doses) and any of the DCE-MRI parameters (P > .05 for all) for patients with available dose data (70%). The bivariate correlation between RT dose and DCE-MRI parameters is detailed in the supplemental material (Appendix E2).

Discussion

Nearly 4 decades ago, R. E. Marx postulated a theory for the development of ORN predicated in large part on altered bone vascularity, resulting in poor regenerative capacity and a decreased ability to resist mechanical and microbial insults.¹⁹ Although this remains the most likely mechanism for ORN, to date scientists and clinicians have lacked the means to study bone vascularity with subcentimeter spatial resolution in a noninvasive manner and have largely been forced to infer mechanisms of ORN development by combining limited preclinical studies, with static anatomic imaging and histologic evaluation of ORN specimens. For the first time, we now have the opportunity to leverage a clinically available imaging approach to provide real-time, noninvasive information about bone vascularity in the context of ORN (Fig. 5). This represents a breakthrough both in our ability to study this devastating disease and to begin to develop clinical trials designed to ameliorate the disease using objective, quantitative measures. In addition, implementation of this approach can generate utility in the context of surgical extractions after radiation therapy, and in the context of real-time image-guided surgical planning for resection of necrotic bone by distinguishing injured or poorly vascularized bone from viable bone.

Fig. 5.

Schematic cartoon of the suggested mechanistic changes of osteoradionecrosis (ORN). Right mandibular body with area of ORN and associated altered vascularity, and the contralateral body with normal body and vascularity. Normal blood vessels have intact walls (continuous red lines) with well-regulated fractional volume plasma (V_p) and appropriate contrast exchange (black dots) across the vessel wall (K^trans) to extracellular extravascular compartment in blue (V_e) and normal mandibular cellularity in yellow. Vessels associated with ORN demonstrate higher V_p with increased leakiness (K^trans) through fragile vessels (dotted red lines) that leads to increased V_e and hypocellularity. (A color version of this figure is available at https://doi.org/10.1016/j.ijrobp.2020.07.029.)

Our findings demonstrated a distinct profile of DCE-MRI parameter maps in mandibular volumes harboring ORN compared with the normal mandible. DCE-MRI parameters indicating vascular compromise showed a significantly higher degree of leakiness in mandibular vasculature as measured using K^trans and V_e of areas affected with advanced grade of ORN versus a healthy mandible. The fractional volume of plasma (V_p) was also higher in ORN-ROIs. We were able to measure significant increases in quantitative parameters with an average increase of approximately 3-fold that of both K^trans and V_e compared with values from the healthy mandibular bone. The vast majority of patients (90%) had at least doubling of the values of either K^trans or V_e for ORN-VOIs compared with control VOIs. We also demonstrated a clear separation of parameter histogram distribution for the majority of patients with higher median and interquartile range of K^trans, V_e, and V_p parameter values of ORN versus control VOIs as detailed in Appendix E1.

To our knowledge, the quantitative perfusion characteristics of mandibular injury after radiation treatment of head and neck cancer have not been assessed previously. One study has investigated the qualitative nature of contrast enhancement of DCE-MRI in patients with mandibular ORN.²⁸ That study showed that all patients with ORN had marked contrast enhancement of the osteoradionecrotic bone marrow, which was reduced after treatment with hyperbaric oxygen treatment.²⁸ DCE-MRI has also shown the ability to detect early-stage idiopathic osteonecrosis of the femur not otherwise visible on conventional MRI, as reported by Chan et al.¹⁵ In addition, DCE-MRI parameters were used to monitor bone healing secondary to trauma or fracture and chronic changes in bone health associated with age-related osteoporotic changes.^16–18 DCE-MRI was also reported to identify changes related to the development of bony metastasis and tumor response to treatment.^32,33

We have recently demonstrated that DCE-MRI can be used to detect radiation-induced changes in mandibular bone vascularity, and we showed dose-dependent changes in both K^trans and V_e in a subset of patients.²⁰ Unlike the findings of our previous study that showed variability in the dose-dependent changes of vascular parameters, where a percentage of patients had a decrease in the measured parameters after treatment, the current study demonstrated only an increase in these parameters. The increased fractional plasma and vascular leakiness in ORN areas reflects fragile vasculature that could be attributed to a process of neovascularization after the postradiation chronic hypoxia in high-dose areas. A serial imaging study of the natural history of vascular changes of the mandible is currently ongoing to determine at which time point this phenomenon of neovascularization begins to develop and to what extent this development could be correlated with early ORN symptom development. However, our findings suggest that a 2-fold increase in either K^trans or V_e parameters is an alarming sign of ORN development if detected in patients with an otherwise clinically apparent normal mandible after radiation treatment, especially in areas exposed to higher doses of radiation because of tumor proximity.

Our group and others have previously demonstrated the dependency of DCE-MRI quantitative output on the nature of PK models used for analysis.^34,35 Therefore, we used the most widely accepted models, such as the Tofts and extended Tofts models. We also used the patients’ contralateral mandible as an internal control to alleviate this model dependency. In addition, we have recently shown, using a multi-institutional comparison of patient-derived DCE-MRI data, that quantitative values might not be reliably compared across different patients because of the difference in patients’ specific imaging parameters and preprocessing and postprocessing factors.³⁴ Our results also confirmed that interpatient DCE-MRI parameter variations were independent of variation in the RT doses received by the ORN-VOIs, as shown in Appendix E2. Because of this limitation, in the current study we did not use the absolute values of the parameters associated with ORN, but instead compared the relative changes of these parameters to respective controls in the same image for each patient using the contralateral healthy mandibular VOIs. Therefore, our results might be more reproducible and generalizable because they represent the relative changes measured in the irradiated mandibular areas compared with the normal nonirradiated bony area. As a result, we avoided the intersubject variability of the parameter absolute values.

The 30 patients included in this study can be perceived as a limited sample size. However, to date this represents the largest prospective quantitative imaging study of ORN ever reported. Furthermore, this study represents the initial characterization of quantitative vascular parameters driven from DCE-MRI for patients with head and neck cancer treated with IMRT and affected by radiation-induced advanced ORN toxicity.

Conclusion

Our results confirm a higher degree of vascular leakiness in the mandibular areas of ORN as measured using DCE-MRI parameters compared with healthy mandible. Additional efforts will be required to develop the DCE-MRI parameter into viable noninvasive biomarkers that are useful for the early detection of subclinical cases of ORN.