ABSTRACT

PURPOSE

This study aims to analyze the ability of quantitative dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) to distinguish between prostate cancer (PCa) and benign lesions in transition zone (TZ) and peripheral zone (PZ) using different methods for arterial input function (AIF) determination. Study endpoints are identification of a standard AIF method and optimal quantitative perfusion parameters for PCa detection.

METHODS

DCE image data of 50 consecutive patients with PCa who underwent multiparametric MRI were analyzed retrospectively with three different methods of AIF acquisition. First, a region of interest was manually defined in an artery (AIF_m); second, an automated algorithm was used (AIF_a); and third, a population-based AIF (AIF_p) was applied. Values of quantitative parameters after Tofts (K^trans, v_e, and k_ep) in PCa, PZ, and TZ in the three different AIFs were analyzed.

RESULTS

K^trans and k_ep were significantly higher in PCa than in benign tissue independent from the AIF method. Whereas in PZ, K^trans and k_ep could differentiate PCa (P < .001), in TZ only k_ep using AIF_p demonstrated a significant difference (P = .039). The correlations of the perfusion parameters that resulted from AIF_m and AIF_a were higher than those that resulted from AIF_p, and the absolute values of K^trans, k_ep, and ve were significantly lower when using AIF_p. The values of quantitative perfusion parameters for PC_a were similar regardless of whether PC_a was located in PZ or TZ.

CONCLUSION

K^trans and k_ep were able to differentiate PCa from benign PZ independent of the AIF method. AIF_a seems to be the most feasible method of AIF determination in clinical routine. For TZ, none of the quantitative perfusion parameters provided satisfying results.

Most malignant tumors, such as prostate cancers (PCa), demonstrate altered perfusion. Higher vascular permeability, neo-angiogenesis with greater micro-vessel density, and arteriovenous shunts represent the main reasons.¹ However, benign prostatic hyperplasia may also exhibit early perfusion and hinder accurate diagnosis of PCa especially in the transition zone (TZ) with magnetic resonance imaging (MRI) alone.² Dynamic contrast-enhanced MRI (DCE-MRI) shows differences of perfusion of a defined region of interest (ROI) in the examined tissue based on the change of signal intensity (SI). The repetitive image acquisition of the same slices requires a certain temporal resolution, so that in order to provide clinical feasibility, the chosen T1-weighted sequence usually represents a compromise between spatial and temporal resolution.³ The SI changes over time can be converted into a concentration-time-curve (CTC), which provides information about the dynamics and distribution of contrast medium in the examined tissue.

The semi-quantitative analysis of DCE, integrated in the multiparametric protocol of PIRADS version 1 (together with T2- and diffusion-weighted imaging), has been exchanged for a qualitative and therefore, subjective analysis of DCE in version 2.⁴ Quantitative DCE as in Tofts model⁵ assumes two compartments in the examined tissue, representing extravascular extracellular space and blood plasma in an effort to provide absolute and therefore more objective values for perfusion.³ In this model, the constants for the exchange rate of contrast medium between the two compartments (K^trans and k_ep) and the volume fraction of extracellular extravascular space within a voxel (v_e) are calculated. Although other methods have shown promising results,⁶ the most common way to determine the perfusion parameters K^trans, k_ep, and v_e requires the determination of the arterial input function (AIF). Since the initial intravenous bolus of contrast medium becomes dispersed and delayed through the circulation, the AIF, which describes the time course of arterial contrast delivery to the tissue, is needed to interpret SI changes in the examined ROI. Thus, the AIF gives crucial information about tissue perfusion.

Objectivity and reproducibility of quantitative DCE analysis may ultimately endorse its routine use to optimize the detection of PCa in spite of the more complicated and time-consuming assessment. There are different methods for the acquisition of AIF. One common method suitable for clinical practice consists of drawing a ROI in a suitable artery such as the femoral artery manually. This manually acquired AIF requires further involvement of the radiologist. As an alternative, the AIF may be acquired by using automated algorithms, thus combining the advantage of a user independent and a patient specific AIF. Especially when a suitable artery is not available, the average of a population may serve as a solution. However, this population-based AIF is not specific to the examined patient. Since different methods for the determination of AIF, a standardized choice of method would be desirable for clinical and research purposes.⁷

The objective of this study was to compare values of quantitative perfusion parameters using Tofts two compartment model for PCa and physiologic tissue in TZ and peripheral zone (PZ) generated by three different methods of AIF determination.

Methods

Study design

The study included 50 patients with prostate MRI and subsequently detected PCa (March 2013 to July 2014). The same consecutive patient collective with PCa was evaluated previously regarding the influence of AIF acquisition method on quantitative perfusion parameters and zonal prostate anatomy. Index lesions with biopsy proven PCa using targeted in-bore MRI-guided biopsy were included in the evaluation. All patients received additional systematic transrectal ultrasound-guided biopsy. This study was conducted in a university hospital setting and has been approved by the institutional ethical review board (protocol number of ethics committee approval: 3612). All patients provided written informed consent.

Study endpoints

The primary endpoint is to demonstrate a significant difference of quantitative perfusion parameters (K^trans, k_ep, and v_e) between PCa and benign prostatic tissue. Secondary endpoints focused on differences of performance between automated (AIF_a) and population-based (AIF_p) AIF in comparison to manual AIF (AIF_m) as a reference AIF.

Image acquisition, AIF determination, and quantitative perfusion parameter

The methods for image acquisition and for the determination of AIF_m, AIF_a, and AIF_p have been published before.^{7, 8} For AIF_m, a ROI was outlined in the common femoral artery. SI was converted into a CTC employing a linear approach by Mouridsen et al.⁹ and multi-flip-angle approach.¹⁰ For AIF_a, the same slice as in AIF_m was used. Cluster analysis with k-means function in MATLAB (MATLAB R R2011a, the Mathworks) was conducted. A ROI was created for AIF_a if the number of voxels in a cluster was between 10 and 80. The temporal resolution of our data was integrated in Parker’s population-based AIF for AIF_p.¹¹

Quantitative perfusion parameters after extended Tofts model (K^trans, v_e, and k_ep) were determined for each dataset in PZPCa and TZ-PCa as well as in physiological PZ and TZ tissues in the dynamic T1-vibesequence. K^trans and v_e, best fitting the CTC of each ROI, were determined using the MATLAB-function fminsearch._ep was then calculated as the quotient of K^trans and v_e.⁵

Statistical analysis

Statistical analysis was performed with MATLAB (MATLAB R R2011a, the Mathworks). Data were examined for normal distribution with the KolmogorovSmirnov test. Non-normally distributed data for independent samples were analyzed with non-parametric Mann-Whitney U test. Normally distributed data were tested with the Student t test. P < .05 was considered significant. Furthermore, receiver operating characteristic (ROC) analyses were conducted for every parameter of each AIF method. Performance was assessed by calculating the respective areas under the ROC curve (AUCs) and the maximum Youden index (J) with corresponding cutoff value, sensitivity, and specificity for each method and perfusion parameter. Correlations were determined with Spearman correlation coefficient with a value for correlationcoefficient ρ between 0.7 and 0.9 considered as strong and a value for ρ greater than 0.9 as very strong.¹² Descriptive statistics of the data are presented with n (%) and, for non-normalized variables, are shown as median (25-75 percentiles), and normally distributed data are shown as mean ± SD.

Results

Baseline characteristics with age, prostate-specific antigen (PSA), prostate volume, PSA density, and ISUP grade group separated by PCa localization (PZ or TZ) in the biopsied index lesions are shown in Table 1.

The absolute values of k_ep and K^trans in PCa were significantly higher for AIF_m, AIF_a, and AIF_p (P < .001), but not for v_e (P = .898 in AIF_m, P = .837 in AIF_a, and P = .651 in AIF_p). The Spearman correlation coefficient for the three different perfusion parameters K^trans, v_e, and k_ep between the different AIF methods AIFm, AIFa, and AIFp are given in Table 2 with a P <.001 for each correlation. Spearman correlation coefficient was highest for k_ep (ρ = 0.94 for AIF_m, AIF_a, ρ = 0.89 for AIF_m-AIF_p, and ρ = 0.86 for AIF_a-AIF_p). Both AIF_m and AIF_a display a similar CTC with a peak of approximately 1 mmol/L, whereas in AIF_p, highest contrast medium concentration was 6 mmol/L.

Mean K^trans, v_e, and k_ep in PCa and in benign tissue with all three AIF methods is shown in Table 3 with their corresponding P values. The ROC curves reveal that k_ep has a greater AUC in comparison to K^trans and v_e with the highest value in AIF_p (AUC_max = 0.859) (Figure 1a). Youden index was also generally greater for k_ep (J range: 0.52-0.61) in comparison to K^trans (J range: 0.4-0.48) and v_e (J range: 0.14-0.17) with the highest value in AIF_p (J_max = 0.61). Among the three methods, AIF_p demonstrates greater AUCs and Youden indexes in comparison to AIF_m and AIF_a for the parameters Ktrans and kep.

Mean K^trans, v_e, and k_ep in PCa of PZ, PCa of TZ and in benign tissue in PZ or TZ with all three AIF methods are shown in Table 4. Differences between PCa in PZ and PCa in TZ were not statistically significant. When comparing PCa of PZ with benign tissue in PZ, K^trans and k_ep demonstrate significantly different values in all three methods. The ROC curves reveal for PCa in PZ that K^trans and k_ep have a greater AUC (AUC range: 0.849-0.890) and Youden indexes (J range: 0.55-0.68 for K^trans; J range: 0.58-0.68 for k_ep) in comparison with v_e (AUC range: 0.603-0.611) (J range: 0.26-0.32) (Figure 1b). K^trans in AIF_p has the highest AUC (AUC_max = 0.889) and Youden index (J_max = 0.68), followed by k_ep  in AIF_p (AUC = 0.871, J = 0.67). For the differentiation of PCa in the TZ and benign TZ, P was only significant for k_ep in AIF_p with a P < .001, an AUC of 0.75, and a J of 0.50 (Figure 1c). Examples of colored K^trans, k_ep, and v_e maps of PCAs in TZ and PZ are shown in Figures 2 and 3.

Discussion

This study verified the ability of quantitative perfusion parameters K^trans and k_ep to distinguish PCa and benign prostate tissue in PZ, independent of the AIF method, while the performance in TZ was obviously inferior. This underlines a separate evaluation of each anatomical prostate zone.

Differences of absolute values of quantitative parameters gave rise to the question, in which AIF is best suited for PCa detection in clinical practice. The individualized nature of AIF_a and AIF_m, respecting patient’s specific physiology, constitutes a major advantage over AIF_p. The similarities and the good correlation of results in AIF_a and AIF_m suggest that the cluster-based automated AIF generates acceptable results with the advantage of a user independent method.

Other authors have identified K^trans and k_ep as best suited for differentiation of PZ and PCa.^{13, 14} Mehrabian et al.¹⁵ also demonstrated that K^trans is capable to separate PCa and physiologic tissue in PZ, whereas v_e failed in PCa detection in the study by Ocak et al.¹⁴ Bonekamp et al.¹⁶ have shown that DCE contributes to a higher sensitivity and specificity of PCa detection in PZ. Our data demonstrate good results for K^trans and k_ep in all methods, while for TZ only k_ep resulting from AIF_p-processing was significant. There is a great variety for absolute values of K^trans between 0.16 (min⁻¹)¹³ and 0.61 ± 0.83 (min⁻¹)¹⁷ in AIF_p in physiologic PZ and 0.75 (min⁻¹).¹³ and 186 ± 1.19 (min⁻¹)¹⁷ for PCa in current literature. Our K^trans values measured by AIF_a and AIF_m are within that range.^{13, 17} Our values for k_ep in all methods are within the range of results by SanzRaquena et al.¹³ in physiologic tissue, but for PCa and using AIF_p, our results are lower. Individual patient’s physiology affects the results in population-based AIF_p, as a major explanation for the different values and the relatively low correlation with AIF_m and AIF_a for perfusion parameters K^trans and v_e.¹¹We noticed a very high peak in the CTC of AIF_p in comparison with AIF_m and AIF_a. It has been demonstrated by Meng et al.¹⁷ that higher peaks result in smaller values for K^trans. Taking into account the relatively small sample size for TZ tumors and the deviating results for AIF_p (especially k_ep) in comparison with the reported values of other authors, the results may be caused by a sampling error. However, quantitative perfusion values resulting from AIF_m and AIF_a were in a realistic range, while the differentiation between PCa and PZ was equal to those measured by AIF_p. As AIF_a is user independent and resulted in nearly the same perfusion values as the AIF_m (as a standard method), this method seems to be most recommendable. The similar AUC values of K^trans and k_ep in PZ demonstrate that quantitative DCE analysis distinguishes PCa and benign tissue in PZ independently from the chosen AIF method. Brunelle et al.¹⁸ found that the diagnostic performance of quantitative DCE parameters remains unchanged using different AIFs and MR imagers.

Absolute values of K^trans and k_ep of PCa in TZ and physiologic TZ tissue generally differed. Although this finding was statistically not significant, it might promote the consideration of DCE not only in PZ but also in TZ. Thus, positive enhancement might support selection of index lesions or in case of circumscribed enhancement in correlating encapsulated hyperplasia nodules can rule out PCa. A major reason for the difficulty of PCa detection by DCE in TZ is probably the higher vascularization of TZ hyperplasia in comparison to normal PZ tissue.² Also, current PI-RADS version (v2.1) supports qualitative DCE evaluation, since scientific evidence shows no clear benefit of quantitative DCE in PCa detection.¹⁹ While some authors are completely against the standardized application of contrast medium due to similar performance in PCa detection of bi-parametric MRI (bpMRI) without DCE to prevent side effects, save costs and time,²⁰ a meta-analysis by Chen et al.²¹ further supports DCE.^{20, 21} DCE seems be especially useful in PI-RADS 3 and 4 category lesions patients.²² Therefore, PI-RADS v2.1 suggests that bpMRI is reserved for select clinical indications and multiparametric MRI with qualitative DCE analysis remains the standard approach.¹⁹ However, quantitative analysis of DCE promises comparability and reproducibility of results and may improve sensitivity or specificity in PCa detection not only in PZ but also in TZ.  Ullrich     and Schimmöller²³ pointed out that lower inter-observer agreement caused i.a. =inter alia caused among others by the subjective nature of scoring criteria like used for qualitative DCE analysis. Future iterations of PI-RADS might include more quantitative analysis, if PCa detection can be improved compared to current purely qualitative assessment of DCE in version 2.1.

It has been shown previously that perfusion parameters differ significantly between benign PZ and TZ.⁷ This does not seem to apply to PCa, since our study demonstrates similar characteristics of perfusion with no differences between PCa in PZ or in TZ.

Beyond the retrospective nature, this study has some limitations. Since there are numerous other methods for the acquisition of AIF, the chosen three methods in this study are exemplary, but not complete. The main question of whether quantitative evaluation after Tofts actually performs better than qualitative DCE evaluation, for example, as employed in version 2.1 of PIRADS analysis in PCa detection has not been investigated in this study since identification of index lesion was based on qualitative DCE analysis representing a potential bias. In addition, the advantage of additional DCE over a bi-parametric approach (with T2-weighted imaging and DWI) has not been explored. Validation of imaging findings is limited to MRI-guided biopsy and does not include surgical specimen. Also, added false positive lesions using quantitative DCE methods have not been analyzed. Furthermore, the sample size of TZ lesion subgroup was relatively small.

In conclusion, all AIF methods demonstrate satisfying results for the differentiation of PCa and PZ tissue. For absolute values and comparison between centers, an automated AIF that respects patient’s individual physiology represents an attractive solution that does not require manual drawing of AIF by the radiologist. Quantitative DCE analysis seems inferior to identify PCa in TZ, but absolute results differed from physiologic tissue. This invites to consider DCE also for TZ lesions. The fact that all AIF methods allowed a differentiation of PCa and benign lesions and the good correlation for k_ep and K^trans may support quantitative analysis in clinical practice. The objective approach could facilitate the comparison between different centers in the future. Moreover, quantitative analysis with K^trans and k_ep could play an important role for therapy monitoring.