Dynamic Contrast-enhanced Magnetic Resonance Imaging as a Predictor of Clinical Outcome in Canine Spontaneous Soft Tissue Sarcomas Treated with Thermoradiotherapy

Numerous phase III clinical trials have shown that thermoradiotherapy is superior to radiotherapy alone for achieving local tumor control and, in some cases, improving survival (1–6). It is well known that hyperthermia changes tumor oxygenation, inhibits DNA damage repair, and is directly cytotoxic (7, 8). The effects described above are strongly dependent on temperature and duration of heat exposure, but which effect dominates in improving tumor control with thermoradiotherapy is not known. In preclinical models, mild to moderate heating (41-43°C) leads to increased perfusion and oxygenation, whereas higher thermal exposures can cause vascular damage and hypoxia (8). Presumably, this latter effect would be deleterious to optimal thermoradiotherapy response, but there are no definitive clinical studies showing this.

Previous work with canine sarcomas have reported improved oxygenation following a single hyperthermia treatment when median temperatures were <44°C (9) and that changes in oxygenation can persist throughout a course of fractionated thermoradiotherapy (10). Milligan and Panjehpour (11) reported that perfusion was improved throughout a course of thermoradiotherapy in dogs with mast cell tumors. These data support the notion that part of the effect that mild hyperthermia has in improving radiotherapy response is due to increased oxygenation through improved perfusion. However, in neither of these reports was there a correlation made between changes in pO₂ or perfusion and treatment outcome.

Brizel et al. (12) reported that improvements in oxygenation 24 hours after the first hyperthermia treatment were associated with higher pathologic complete response rates after thermoradiotherapy of human soft tissue sarcomas. Jones et al. (13) presented similar data in a small series of 13 patients with locally advanced breast cancer who were treated with a combination of hyperthermia, paclitaxel, and radiotherapy. The response end points reported in both of these studies were relevant to short-term cell killing, but neither addressed the more important issue of the effects of reoxygenation on local tumor control.

Because of the importance of perfusion in cancer therapy (with or without hyperthermia), dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) has been gaining use to obtain physiologic information. This imaging method provides information related to perfusion and permeability-surface to volume product of a region/voxel of interest.

Recently, we reported results from a randomized phase II clinical trial involving thermoradiotherapy treatment of canine soft tissue sarcomas (3). Results of this trial indicated that higher thermal dose was associated with significant prolongation of progression-free survival. Extracellular pH (pHe) and ³¹P MR spectroscopy parameters were associated with metastasis-free and overall survival in a subset of patients from this same trial (14). In this particular study, we evaluate the potential of DCE-MRI–derived parameters, obtained before treatment and 24 to 72 hours after first hyperthermia treatment, to be prognostically predictive in subset of 37 patients who were evaluated with MRI as part of this trial.

Thirty-seven canine patients with spontaneously occurring soft tissue sarcomas were randomized to receive either a low (2-5) or high (20-50) thermal dose quantified as cumulative equivalent minutes that the T90 was equal to 43°C (CEM43°C T90), in combination with fractionated radiotherapy (56.25 Gy in 25 daily fractions of 2.25 Gy). T90 refers to the temperature reached or exceeded by 90% of measured temperature points. Tumor types included 20 hemangiopericytomas (54.1%), 9 fibrosarcomas (24.3%), 4 myxosarcomas (10.8%), 2 neurofibrosarcomas (5.4%), and 2 nonspecific sarcomas (5.4%). Classification of histologic subtype was done by a board-certified veterinary pathologist (15). Hyperthermia was induced using microwave applicators operating between 140 and 433 MHz. Thermal dose randomization was stratified by tumor volume and tumor grade. Approximately one hyperthermia treatment was given per week, after the daily radiation fraction. The duration of the hyperthermia treatment was adjusted so as to administer ∼20% of the prescribed hyperthermia total dose in each hyperthermia fraction. Occasionally, the prescribed thermal dose was administered in 4 weekly treatments, and in some other dogs, a sixth treatment had to be added to reach the prescribed thermal dose (3). MRI was done before and 24 to 72 h after the first hyperthermia treatment (Fig. 1).

The procedures were approved by the North Carolina State and Duke University Animal Care and Use Committees.

All studies were done on a GE 1.5T Signa System equipped with a surface coil. Dogs were anesthetized with isoflurane (Abbott Laboratories) for the MR study. Coronal and axial localizer images (fast spoiled gradient recalled echo pulse sequence) were acquired initially to define tumor position. Five baseline scans consisting of 15 T1-weighted images (repetition time = 400; echo time = 15) per volume were acquired before contrast injection (field of view = 256 × 128; slice thickness = 5-10 mm). Subsequently, an i.v. bolus injection (0.1 mmol/kg, 2 mL/s) of gadolinium dimeglumine (Gd-DTPA) was administered. Image volumes were acquired every 30 s for 20 min.

Relative enhancement images were fitted pixel by pixel with a double exponential Eq. A using a nonlinear iterative χ² minimization technique based on the marquadt algorithm.

Each fitted curve allowed the extraction of Smax (relative maximal enhancement constant), alpha (delivery rate constant), and beta (clearance rate constant) parameters. Negative values or values from the fit that never converged were not included for calculations of region of interest analysis. These parameters allowed the calculations of washin rate (Smax * α) and washout rate (Smax * β) along with area under the signal enhancement curve (AUC). With these parameters, descriptors of perfusion/permeability were obtained (16, 17). For evaluation of AUC, three different integration times were chosen for calculation, AUC60 (integration time from 0 to 60 s), AUC90 (integration time from 0 to 90 s), and AUCt-max (integration time from 0 to when the muscle reference signaled peaked, t-max). Time to maximal muscle enhancement was calculated by setting the derivative of Eq. A to zero and solving for time. Otherwise, this last value was used as the upper limit for calculating AUCt-max. Representative patient image data are shown in Fig. 2 for pre- and post-hyperthermia images and pseudo-false color for washin, washout, and AUCt-max data.

From the fitted data, region of interest for tumor (including any necrosis) and muscle was manually drawn on each slice to form volumes of interest. Tumor region of interest included the entire tumor and potential necrosis, and the muscle region of interest was drawn on muscle free of tumor in the field of view based on known normal anatomy. Median values were extracted from each volume of interest. Washin, washout, AUC60, AUC90, and AUCt-max parameters were analyzed individually. This yielded five different parameters relating to perfusion/permeability. All imaging analyses were made with a custom-made software running on MATLAB (The MathWorks, Inc.).

pHe measurements were made using needle microelectrodes (Microelectrode, Inc. and Agulian), as previously described. Multiple measurements were made and averaged to obtain an overall estimate of the pHe. Intracellular pH (pHi) was determined by the frequency difference between the Pi and PCr resonances, from ³¹P MR spectroscopy spectra. MR spectroscopy spectra were acquired with a surface dual frequency coil array, and triphenyl phosphite standard solutions were placed next to the tumor for pulse calibration. Pulse sequence protocol and data analysis methods have been previously reported in detail (2, 14, 18).

After treatment, dogs were reevaluated at 1, 2, 3, 5, 7, 9, and 12 mo and then at 3-mo intervals. Reevaluation consisted of physical examination, tumor measurement, thoracic radiographs, and assessment of regional lymph nodes. Tumors volumes were determined by direct physical measurements along 3 axis multiplied together with PI/6. Duration of local tumor control was defined as the time from date of the first hyperthermia treatment until local failure. Metastasis-free survival was defined as the time from date of the first hyperthermia treatment until metastasis or death, and overall survival was the time from date of the first hyperthermia treatment until death from any cause. Spread to regional lymph nodes was classified as positive for metastasis, as well as any spread to distant organs, such as the lung.

Values of AUC were log transformed due to the large range of calculated values; therefore, throughout the article, AUC values reported will correspond to log₁₀(AUC) values.

Each perfusion parameter was assessed for its predictive properties on metastasis-free survival, time to local failure, and overall survival, before and after treatment, as well as their difference post-pre (Δ) treatment, using Cox proportional hazard ratio and log-rank tests. The effect of the first hyperthermia treatment, as described by T50, T90, CEM43°C T50, and CEM43°C T90, along with administration time of hyperthermia on perfusion parameters obtained after the first hyperthermia, was assessed using a Wilcoxon rank sum test as well as a paired t test.

With one exception, multiregression was not attempted for correlation with treatment outcome because of the limited number of subjects. Univariable and multivariable analyses were done between perfusion-related parameters and other clinical variables as well as the pH parameters to determine whether these were related to each other. Corrections for multiple comparisons were not made.

Thirty-seven dogs (median age = 9.8 years; interquartile range = 4.3 years) were evaluated with DCE-MRI before starting treatment. Seventeen of the dogs had a second DCE-MRI 24 hours after the first thermoradiotherapy treatment. All subjects were planned to receive the second DCE-MRI study, but in 20 of these subjects, the data were not obtained either because of scheduling issues or because clean spectroscopic data could not be derived from the tissue for the second session, which was required before the DCE-MRI per protocol design. The descriptive statistics of the sample characteristics are listed by thermal dose group in Table 1.

The population was balanced with respect to thermal dose group (low dose = 56.7%; high dose = 43.3%). In the high thermal dose group, median temperature averaged T50 (median = 43.2°C; interquartile range = 1.9°C), T90 (median = 40.8°C; interquartile range = 0.7°C), CEM43°C T50 (median = 89.1 minutes; interquartile range = 144.8 minutes), and CEM43°C T90 (median = 8.2 minutes; interquartile range = 6.4 minutes) and the total duration time of heating (median = 214 minutes; interquartile range = 179.5 minutes). In the low thermal dose group, median temperature averaged T50 (median = 42.6°C; interquartile range =1.7°C), T90 (median = 39.9°C; interquartile range = 0.6°C), CEM43°C T50 (median = 17.5 minutes; interquartile range = 32.0 minutes), and CEM43°C T90 (median = 0.8 minute; interquartile range = 0.6 minute) and the total duration time of heating (median = 103 minutes; interquartile range = 58 minutes).

The T50 and T90 values are descriptions of the frequency distribution of temperatures that are measured during treatment. The T50 is the median temperature and the T90 describes the temperature value that is exceeded by 90% of the measured values. This is sometimes referred to as the 10th percentile, but in the hyperthermia literature, it has been traditionally referred to as the T90. CEM43°C is a conversion algorithm that converts any time-temperature history into an equivalent number of minutes at 43°C. Details of this algorithm are published elsewhere (19).

Cox proportional hazard ratio analyses showed that several imaging-based parameters obtained before and after the first hyperthermia treatment were found to be significant in predicting metastasis-free survival, time to local failure, and overall survival along with total duration of heating (Table 2).

AUCt-max (median = 5.1; interquartile range = 0.65) measured before therapy was predictive of time to local failure (P = 0.02; Fig. 3A). There was no significant correlation between AUCt-max and tumor volume or grade in univariable analysis (Table 3). However, because of competing risks of animals developing distant disease before reaching an end point for local tumor control, caution needs to be considered with respect to how to interpret this result.

The median time to metastasis was 24 months and the 2-year overall survival rate was 0.36. This included animals that died and at necroscopy metastasis were found. Pretreatment parameters that were found to be significant predictors of metastasis-free and overall survival included tumor grade (P ≤ 0.01; 70%; intermediate/high = 30%), volume (P < 0.05; median = 18.8 cm³; interquartile range = 55.7 cm³), and washout (P ≤ 0.02; median = 4.24; interquartile range = 2.72; shown in Fig. 3B). Better prognosis was associated with lower grade, smaller volume, and higher washout rate. The predictability of pHe parameters (obtained using microelectrodes) from Lora-Michiels et al. (14) is shown for comparison because these measurements were taken on the same subjects (Table 2).

AUCt-max, AUC60, AUC90, and washin parameter (Fig. 3C) obtained 24 hours following the first hyperthermia treatment were also significantly correlated with outcome. In all cases, higher values were associated with improved prognosis. These data suggest that improvements in perfusion and perhaps oxygenation after hyperthermia treatment are a good prognostic sign.

Multivariable analysis showed that pretreatment AUCt-max and washout were independent predictors of metastasis-free and overall survival (P < 0.05; Table 3). In contrast, univariable analysis showed that AUCt-max obtained after treatment was inversely related to tumor volume and tumor grade. Small and/or low-grade tumors had higher AUCt-max values (P < 0.05; Table 3). Washout and AUC values obtained after treatment also were negatively correlated with tumor volume in univariable analysis (P < 0.05; Table 3).

The total duration of hyperthermia was found to be prognostically important in our previous report (3). In this subset analysis, duration of heating remained an independent predictor of metastasis-free and overall survival with hazard ratios of 1.28 and 1.32 and P values of 0.007 and 0.004, respectively (Table 2). In this current study, heating duration was correlated with tumor grade in a univariable analysis (P < 0.05; Table 3).

Relationships between the MRI parameters measured here and pH parameters reported in our previous article (14) are shown in Table 4. univariable analysis showed that post-hyperthermia washout values were negatively correlated with pretreatment pHi (P < 0.05; Table 4) and positively correlated with the difference posttreatment (post-pre) pHe (P < 0.05; Table 4). Figure 3D depicts the relationship of post-washout with the change of pHe showing a positive relationship, consistent with the notion that improved perfusion after heating leads to more efficient clearance of acid from the tissue.

This study showed that DCE-MRI–based parameters may be useful in predicting overall and metastasis-free survival in canine patients with soft tissue sarcomas. The parameters from the signal enhancement curve allowed prediction of patient outcome regardless of being done before the first hyperthermia therapy or 24 hours after. To our knowledge, this is the first time that DCE-MRI parameters are predictive of clinical outcome for patients with soft tissue sarcomas. However, the results are based on a few patients, so larger clinical trials with adequate statistical power are needed to verify this initial result.

For data acquired before treatment, AUCt-max emerged as being related to local tumor control. This parameter integrates signal over time for the tumor, truncating it when the signal reaches a maximum value in the reference muscle tissue. Consequently, this parameter probably relates to both perfusion and permeability. Tumors with higher AUCt-max might be better perfused and therefore better oxygenated. In univariable analysis, this parameter was not associated with either tumor volume or grade, suggesting that it is an independent predictor of local tumor control (Table 3).

Higher washout values obtained before treatment were associated with greater metastasis-free and overall survival. This parameter relates to the reabsorption and clearance of contrast agent from the extravascular space. Thus, higher values maybe associated with a vasculature that is more efficient in transport. Factors that influence transport efficiency include vascular density, permeability, and perfusion rate of the microvessels (20–23).

In prior studies, we showed theoretically that adjustments in microvessel orientation can make quite dramatic differences in oxygen transport (20, 21). The same principle should apply in reverse for a MR contrast agent. Thus, the combination of vascular density and orientation of microvessels is likely to contribute significantly to the differences in washout between different tumors.

Vascular permeability is related to the size of the solute, the concentration gradient across the microvessel wall, and the physical characteristics of the vascular wall, due to the presence of large endothelial cell gaps and lack of basement membrane leading to increased permeability (24, 25). For large solutes, such as proteins, transport is strongly governed by convection generated by the pressure gradient across the vascular wall. With the presence of elevated tumor interstitial fluid pressure, these gradients are negligible (24, 25). However, small molecules also undergo diffusive transport, so the influence of convection on the transport of a small molecule such as a chelated Gd contrast agent is negligible. Consequently, the Gd would likely be reabsorbed into the vasculature down its own concentration gradient once the kidneys clear the Gd from the general circulation (24, 25).

It is difficult to know why animals with tumors with better transport characteristics would have longer metastasis-free and overall survival. It may be reflective of a more differentiated state, but washout was not significantly associated with either tumor size or grade, suggesting that it could be independently predictive of outcome (Table 3). Current studies by our group are investigating interrelationships between physiologic/metabolic and genomic parameters compared with DCE-MRI. From this, we may decipher more insight into this observation.

It was surprising to find that DCE-MRI values obtained after the first hyperthermia treatment were also associated with outcome. Increases in AUC and washin after the first heat treatment were associated with prolonged metastasis-free and overall survival (Table 3). This may relate back to the maturity of the vasculature. AUCt-max values (along with other parameters) for the posttreatment study were correlated with tumor volume and grade, suggesting that they may not be independent predictors of outcome. It is known that hyperthermia causes changes in permeability/perfusion but the distribution and amount may be dependent on the tumor volume/aggressiveness. Tumors with more mature vasculature may be able to respond more dynamically to thermal stress and therefore more readily increase perfusion in response to thermal stress. Presumably, tumors with more mature vasculature might be less likely to metastasize for a variety of reasons. They may be less hypoxic, which would tend to reduce up-regulation of transcription factors that influence metastatic propensity, such as hypoxia inducible factor-1 (26).

Only a few studies have been published describing the prognostic properties of DCE-MRI for clinical efficacy; studies involving gliomas, head and neck, breast, lung, and cervical cancers have been reported (Supplementary Table S1). However, to our knowledge, clinical studies involving prediction of outcome for soft tissue sarcomas using DCE-MRI have not been reported.

AUCt-max was the only perfusion parameter to predict local failure in our study (Table 2). Similar results have been reported by Mayr et al. (27, 28) in patients with cervical carcinomas at early stages of radiation therapy using relative signal intensity parameters (Supplementary Table S1).

Two studies have evaluated DCE-MRI parameters after treatment. Decreased tumor perfusion, at the end of a course of fractionated radiotherapy assessed by signal intensity perfusion indexes (initial slope, T_max), was related to low rates of local recurrence in patients with head and neck cancer (29). Decreased perfusion indexes using the two-compartment model, K^trans and V_e, had >65% tumor volume reduction after multimodality therapy in patients with breast cancer (30). Similar to the head and neck trial, these results are not surprising given they were obtained following therapy.

Although studies done at the end of primary therapy are of interest scientifically, as they suggest that therapy eventually leads to reduction in perfusion, such information would have little effect on therapeutic choices. A more desirable approach is to predict likelihood of local control before or during the early phases of treatment because such measurements could be used to alter the course of treatment and potentially lead to increased treatment efficacy.

Our results show that longer metastasis-free survival and increased overall survival occurred in dogs with highly perfused tumors before hyperthermia treatment measured by increased washout rate and after treatment with increased washin rate, AUC60, AUC90, and AUCt-max. Only one other study has been published that relates to these end points. An increased incidence of metastasis (and poorer survival) was observed in patients with Ewing's sarcomas displaying high percentage of necrosis (31). The percentage of necrosis was obtained by calculating nonenhanced tumor regions from pre– and post–contrast-enhanced images. Although this study did not specifically use dynamic MR parameters, lack of contrast agent is due to a lack of delivery (i.e., perfusion).

Our patients were part of a large thermoradiotherapy clinical trial that tested effects of high versus low thermal dose in dogs with spontaneous soft tissue sarcomas (3). We reported that increased cumulative thermal dose was related to increased time to local failure. One potential mechanism underlying the improvement in local tumor control might be increased oxygenation, resulting from increased perfusion. However, we did not find an association between higher thermal dose and increased DCE-MRI parameters in this study (Table 4). It may be that patient-related parameters, such as the maturity of the vasculature or rates of tumor cell killing after the first thermoradiotherapy session, had more of an influence on perfusion changes than thermal dose itself. Additionally, we did not measure changes in oxygenation of these tumors, so we cannot conclude whether the changes in DCE-MRI parameters were associated directly with improvements in tumor oxygenation. Studies of this type are ongoing now in our group.

Our prior analysis also showed that duration of heating time, independent of thermal dose, was related to metastasis-free and overall survival. This recapitulates what we saw in the larger cohort of animals in this study (3). A reasonable assumption might be that duration of heating would be related to perfusion in some way. For example, the time required to reach the predetermined thermal dose would be shorter for more poorly perfused tumors. In this subset analysis, we found that duration of heating was still important for overall and metastasis-free survival, but it was not related to any of the DCE-MRI parameters. There was a correlation between duration of heating and tumor grade. Higher-grade tumors tended to take longer to adequately heat. This may have to do with a greater propensity for preexisting necrosis in high-grade tumors. Such tumors may develop power-limiting hotspots in necrotic areas, which could lead to the need to prolong heating times to get the rest of the tumor exposed to the prescribed thermal dose.

Using the same animals in this study, we recently reported that low pHe is associated with increased likelihood for tumor metastasis and shortened overall survival (14). The acidic environment within tumors is thought to exist because of metabolic aberrations leading to propensity for anaerobic metabolism (e.g., the Warburg effect) and/or induction of anaerobic metabolism as a result of hypoxia (e.g., the Pasteur effect). However, the degree of acidosis will also be influenced by how efficiently the lactate is removed from the tissue by perfusion. In this study, there was no relationship between DCE-MRI parameters obtained before treatment and pH, but an increase in washout was associated with an increase in pH. This is consistent with washout being indicative of a more efficient vascular system for waste removal. The inverse relationship between pHi and washout is more difficult to explain. Generally, pHi is tightly regulated by a variety of proton pumps that move lactate and hydrogen ions out of the cell to maintain a neutral pHi (32). It is difficult to reconcile how pHi could have any influence on changes in DCE-MRI parameters after hyperthermia treatment. This may be a spurious observation.

There are two main models that have been used to derive physiologically based parameters from dynamic images. The first is a two-compartment model that includes the transfer coefficient between the intravascular and extravascular compartments (K^trans), the extracellular volume fraction, and vascular volume fraction (33, 34). The second model is an empirical fit of the signal enhancement data of a pixel or region of interest. An AUC can be calculated with this empirical fit. AUC calculations and the fitted values of the empirical equation can then be used to evaluate the dynamic data and referenced to normal tissue, such as muscle (17, 29, 35–38). This second technique has the advantage of not requiring an arterial input function (arterial concentration-time curve of the contrast agent) for analysis.

All of these methods attempt to ascertain the meaning of the dynamic signal from DCE-MRI into physiologically tangible properties (i.e., tumor perfusion and vascular permeability). Several prior reports have shown correlations between DCE-MRI parameters and treatment outcome (27, 29, 31, 39–41).

In this study, we used the second method, fitting the signal enhancement curve to an empirical equation. From this analysis, several DCE-MRI parameters were associated with treatment outcome in this study. In all cases, parameters that were associated with better overall perfusion indicated a better outcome. Being able to predict overall treatment outcome before, or shortly after treatment is initiated, potentially allows for treatment modification, which may improve overall efficacy.

Although the protocol called for scans to be taken before and 24 hours after the first hyperthermia treatment, only 17 of the original 37 animals actually had data at this second time point. There were multiple reasons for lack of data, but the two main ones related to lack of good data fits on the second session or scheduling conflicts. The scans were obtained on a human scanner and there were times when we were unable to schedule the canine patients for the second scan. We might have been able to avoid the poor data fitting on the second scan had we been able to analyze the results immediately after the scan was done. Unfortunately, this was not possible at the time that these studies were done. In fact, we did not know about the problem until years later when the data were post-analyzed. We have encountered similar problems previously in human (18) studies. This points to the need for more rapid evaluation of study-related parameters so that additional measurements could be made, thereby improving the data acquisition for studies that require multiple time points.

The canine sarcoma is an ideal model for human solid tumors in many respects. These advantages have been reviewed in detail but are briefly summarized here. The natural history of canine sarcomas is very similar to that of humans (42). They are locally invasive and a proportion of them metastasize, making the achievement of local control challenging. For the purposes of hyperthermia studies, they are of the same size range as human tumors, making them amenable to heating, using the same heating technologies that are used in humans. There are many advantages of the canine tumor model in general. The patients are outbred and they are typically older and therefore have concomitant diseases like human patients do. They are immune competent, yet their tumors are only weakly antigenic, similar to humans (42). We have used this model extensively over the last 25 years.

Although the number of tumors imaged was limited, the results of this study suggest that DCE-MRI perfusion parameters are associated with various indicators of prognosis. Because delivery of nutrients (oxygen) and removal of waste (lactate) are dependent on perfusion, imbalances in delivery of nutrients and waste removal may contribute to aggressiveness of soft tissue sarcomas.

No potential conflicts of interest were disclosed.