Inter-Vendor Variability of Perfusion Parameters Derived from Dynamic Contrast-Enhanced MRI in Patients with Prostate Cancer Authors: Mingyu Kim, Seung Ho Kim, Joo Yeon Kim Published: 2026 ABSTRACT Simple SummaryDynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) has emerged as a potential surrogate marker in non-invasive evaluation of tumor angiogenesis. Prostate cancer (PCa) is one of the cancers for which treatment is individualized according to tumor aggressiveness. There have been attempts to assess PCa aggressiveness with DCE-MRI using various pharmacokinetic models and diverse post-processing solutions from different vendors. However, the results have been inconsistent and widely variable, requiring further verification. Therefore, the aim of this study was to investigate the inter-vendor variability of perfusion parameters derived from DCE-MRI in patients with PCa. The high variability of perfusion parameters across solutions should be considered when using quantitative perfusion parameters to evaluate treatment response. 1. Introduction Multiparametric magnetic resonance imaging (MRI) plays a central role in the evaluation of prostate cancer (PCa), and dynamic contrast-enhanced MRI (DCE-MRI) is an important component of mpMRI for the assessment of tumor location, extent, biologic behavior, and recurrent disease [1]. PCa typically demonstrates earlier and more intense enhancement than the surrounding normal prostate tissue on DCE-MRI [2]. This finding is thought to reflect tumor angiogenesis, in which increased production of angiogenic factors such as vascular endothelial growth factor leads to the development of abnormal micro-vessels that are more fragile and permeable than normal vessels [1,3,4]. Because DCE-MRI can depict these vascular characteristics of PCa, its role in the evaluation of tumor angiogenesis has been investigated [5,6,7,8]. Despite its potential, DCE-MRI has not yet been fully incorporated into routine clinical practice. Measurements derived from DCE-MRI can be influenced by multiple factors, including hemodynamic conditions, arterial input function (AIF) selection, pharmacokinetic modeling, and post-processing implementation, all of which may affect reproducibility [9]. Although attempts have been made to use DCE-MRI for evaluation of response to anti-angiogenic and anti-vascular therapies, quantitative parameters must first be shown to be reproducible before they can be used reliably as imaging biomarkers [9,10,11]. In addition, variability may arise from differences among vendor-specific solutions used for perfusion analysis. Previous studies have demonstrated inter-vendor variability of DCE-MRI-derived perfusion parameters in uterine leiomyoma and rectal cancer [4,12]. However, to our knowledge, such inter-vendor variability has not been sufficiently investigated in patients with prostate cancer. Therefore, the purpose of this study was to investigate the variability of perfusion parameters derived from DCE-MRI between two commercially available solutions in patients with prostate cancer. 2. Materials and Methods This retrospective study was approved by the institutional review board and informed consent was waived. 2.1. Patient Selection Criteria We retrospectively reviewed the medical records of patients who underwent radical prostatectomy for pathologically confirmed prostate cancer between December 2021 and September 2022. Patients were considered eligible if they had undergone preoperative prostate MRI, including DCE-MRI, at our institution; had available whole-mount pathologic topographic maps; and had DCE-MRI datasets that could be analyzed on both solutions. Cases were excluded when the tumor could not be confidently localized on MRI, when MRI findings could not be matched with the pathologic lesion, when the lesion was too small to allow reliable segmentation, or when image transfer to either solution was unsuccessful. 2.2. MRI Acquisition All examinations were performed using a single 3.0-T MRI scanner (Signa Architect, GE Healthcare, Waukesha, WI, USA) equipped with a body array coil, using the same acquisition protocol and reconstruction settings throughout the study period. The standard prostate MRI protocol included multiplanar T2-weighted imaging, axial diffusion-weighted imaging, pre-contrast T1 mapping, dynamic contrast-enhanced imaging, and axial post-contrast fat-suppressed T1-weighted imaging. Pre-contrast T1 mapping was obtained with a dual–flip angle method (5° and 15°). For dynamic contrast-enhanced imaging, a three-dimensional spoiled gradient-echo sequence was used. The acquisition parameters were repetition time/echo time of 3.7/1.6 msec, flip angle of 8°, field of view of 300 × 300 mm, section thickness of 2 mm, no inter-slice gap, matrix size of 224 × 160, and 32 sections. Dynamic imaging was repeated 60 times with a temporal resolution of 3.6 s, allowing coverage of the entire prostate gland. Contrast material was administered intravenously as gadoterate meglumine (Dotarem, Guerbet, Villepinte, France) at 0.1 mmol/kg (0.2 mL/kg) with an injection rate of 3.0 mL/s, followed by a 30 mL saline flush at the same rate. The full sequence parameters are presented in Table 1. 2.3. Measurements of Perfusion Parameters The DCE-MRI datasets were first organized in the PACS system (M6; INFINITT Healthcare, Seoul, Republic of Korea) and were subsequently transferred to workstations running the two commercially available solutions, MR permeability on IntelliSpace Portal version 10.1 (Philips, The Netherlands) and GenIQ on AW Server 3.2 (GE Healthcare, Waukesha, WI, USA) for pharmacokinetic analysis. Both solutions applied the extended Tofts model, and the population-based AIF was commonly selected for the initial analyses for both solutions. The extended Tofts model is a pharmacokinetic model used to describe the tissue contrast-agent concentration–time curve in DCE-MRI. The model assumes that the contrast agent distributes between the intravascular plasma compartment and the extravascular extracellular space, with the AIF serving as the vascular input [7,13]. Unlike the conventional Tofts model, the extended Tofts model additionally incorporates the plasma volume fraction, thereby accounting for the vascular contribution to tissue enhancement. Accordingly, it allows estimation of Ktrans, kep, ve, and vp. Ktrans represents the volume transfer constant from plasma to the extravascular extracellular space, kep represents the rate constant for contrast reflux from the extravascular extracellular space back to plasma, ve represents the volume fraction of the extravascular extracellular space, and vp represents the plasma volume fraction [7,13]. Specifically, AIF could be selected from three bi-exponential model presets according to the contrast injection duration, categorized as short (<5 s), medium (5–10 s), and long (>10 s) on the MR permeability application in ISP. In the present study, the medium preset was used. For exploratory analysis, individual AIF was applied to observe platform-specific differences in perfusion parameters between the individual and population-based AIF methods. The analyzed parameters included Ktrans, kep, ve, and vp. Ktrans reflects the transfer of contrast material from the intravascular plasma compartment into the extravascular extracellular space, whereas kep represents reflux from the extravascular extracellular space back into the plasma compartment. The fractional volumes of the extravascular extracellular space and plasma were expressed as ve and vp, respectively [13]. Two radiologists with 15 and 4 years of experience, respectively, reviewed the T2-weighted images, diffusion-weighted images, and corresponding apparent diffusion coefficient (ADC) maps. Both readers were blinded to the Gleason score of each prostate cancer and assigned the Prostate Imaging-Reporting and Data System (PI-RADS) score by consensus on the basis of PI-RADS version 2.1. The zonal location and tumor border were determined by consensus for each tumor-bearing section, after radiologic-pathologic correlation using the topographic maps. In patients with multiple lesions, the largest lesion was selected as the index tumor. Freehand regions of interest were then manually drawn along the tumor margin on the DCE images by the radiologist with 4 years of experience. Segmentation could not be transferred between the two solutions. Therefore, it was redone separately on both platforms. After segmentation, the perfusion values obtained from the tumor-bearing sections were averaged to derive representative whole-tumor values for each parameter. 2.4. Histology Experienced urologic surgeons performed radical prostatectomy in all study patients. After excision, the prostate specimens were submitted for standard histopathologic processing, including formalin fixation, paraffin embedding, and sequential sectioning from the base to the apex. Hematoxylin and eosin–stained slides were then reviewed by a dedicated pathologist, who established the Gleason score for each prostate cancer lesion [14]. Whole-mount topographic maps obtained from these sections served as the reference standard for tumor segmentation on MRI. 2.5. Statistical Analysis Statistical analysis was performed using MedCalc (MedCalc software version 22.032, Mariakerke, Belgium). For each perfusion parameter, paired measurements obtained from the two solutions were analyzed. Agreement between the two solutions was assessed using Bland–Altman analysis, from which the mean difference and the upper and lower limits of agreement were calculated together with their 95% confidence intervals (CIs). In addition, the null hypothesis that the mean difference between paired measurements is zero was tested for each parameter. Reproducibility between the two solutions was evaluated using the intraclass correlation coefficient (ICC), and 95% CIs were calculated for both single and average measures. ICC values were interpreted as poor (<0.40), fair (0.40–0.59), good (0.60–0.74), and excellent (0.75–1.00) [15]. An inter-observer reproducibility analysis was not performed for the segmentation process itself for the purposes of this study. A p-value less than 0.05 was considered to indicate statistical significance. 3. Results 3.1. Patient Demographics A total of 50 patients (mean age, 71.6; range, 56–86) with pathologically confirmed prostate cancer were included and analyzed. The mean prostate-specific antigen level was 20.8 ng/mL (range, 0.85–131). Tumor location was the peripheral zone in 23 patients (46%), the transition zone in 19 (38%), and diffuse in 8 (16%). PI-RADS version 2.1 scores were 3 in 4 patients (8%), 4 in 17 (34%), and 5 in 29 (58%). Their demographic information is provided in Table 2. 3.2. Inter-Vendor Variability of Perfusion Parameters Bland–Altman analysis revealed significant differences in all four perfusion parameters between the two solutions (Figure 1). The mean differences were −0.2102 (95% CI, −0.2687 to −0.1518) for Ktrans, −0.7632 (95% CI, −0.9005 to −0.6258) for kep, −0.1507 (95% CI, −0.2422 to −0.05907) for ve, and −0.02929 (95% CI, −0.03383 to −0.02476) for vp. These differences were statistically significant for all parameters (p < 0.0001 for Ktrans, kep, and vp; p = 0.0018 for ve). The corresponding limits of agreement ranged from −0.6136 to 0.1931 for Ktrans, −1.7105 to 0.1841 for kep, −0.7822 to 0.4809 for ve, and −0.05989 to 0.001304 for vp. The detailed results for all of the perfusion parameters are summarized in Table 3. 3.3. Reproducibility of Perfusion Parameters Between the Two Solutions Among the four parameters, kep showed good reproducibility between the two solutions according to the predefined ICC interpretation criteria, whereas Ktrans, ve, and vp showed poor reproducibility. The ICC showed variable reproducibility across the four perfusion parameters. For single measures, the ICCs were 0.1757 for Ktrans, 0.5248 for kep, −0.06239 for ve, and 0.1529 for vp. For average measures, the corresponding ICCs were 0.2989 (95% CI, −0.2355 to 0.6021) for Ktrans, 0.6883 (95% CI, 0.4507 to 0.8231) for kep, −0.1331 (95% CI, −0.9967 to 0.3570) for ve, and 0.2653 (95% CI, −0.3106 to 0.5881) for vp. The detailed results are summarized in Table 4. 4. Discussion Our results revealed that all perfusion parameters were significantly different between the two solutions. Only kep showed good reproducibility, whereas the other perfusion parameters derived from DCE-MRI showed poor reproducibility between the two vendors’ solutions. Our observation corresponds well with previous studies on rectal cancer and uterine myoma [4,12]. Beuzit et al. reported that poor reproducibility was found in the calculated perfusion parameters derived from different vendors’ perfusion analysis solutions in patients with rectal cancer [12]. Specifically, ICCs of Ktrans, kep and ve ranged from 0.1 and 0.16, which indicated very poor reproducibility. In addition, the within-subject coefficients of variation in Ktrans, kep, and ve were 67%, 53% and 47%, respectively, between the two vendors’ solutions. Those values are markedly beyond the upper limit of agreement of 20%, which was suggested by the Quantitative Imaging Biomarkers Alliance (QIBA) DCE-MRI technical committee. Heye et al. also reported that the ICCs for the analyzed perfusion analysis solutions showed 0.33–0.68 for Ktrans, 0.02–0.81 for kep, and −0.03–0.72 for ve, which indicated a very wide range from poor to moderate reproducibility [4]. The within-subject variation across four perfusion analysis solutions, which were not used in our study, was in a wide range of 48–69% for Ktrans, 37–60% for kep, and 28–74% for ve in patients with uterine myoma. In contrast, each vendor’s solution showed promising results in terms of intra-observer and inter-observer agreement. Specifically, Wang et al. observed an excellent intra- and inter-observer reproducibility (>0.9) for Ktrans, kep, and ve and a good reproducibility (0.84–0.89) for vp derived from DCE-MRI in patients with renal cell carcinoma by using the Omni-Kinetics workstation (GE Healthcare) [16]. In addition, Park et al. also reported an excellent inter-observer reproducibility for Ktrans (0.939, 95% CI, 0.913–0.959) and kep (0.920, 95% CI, 0.885–0.946) and good reproducibility for ve (0.834, 95% CI, 0.760–0.888) and vp (0.886, 95% CI, 0.823–0.926) in patients with prostate cancer by using ISP version 10 [17]. Those observations reflect that although the variability across vendors’ solutions is beyond the limits of QIBA recommendations, the reproducibility of a single vendor’s solution is acceptable for longitudinal comparison in each individual patient. We explicitly acknowledge that variability may differ across scanners, field strengths, or acquisition protocols, and results may not directly extrapolate to multi-center or multi-vendor acquisition settings. Our findings, together with the existing literature, suggest that inter-vendor variability may be related to differences in solution implementation and parameter fitting across vendors. Differences in population-based AIF presets between the two solutions may be one of the contributors to the high inter-vendor variability of the measured perfusion parameters. Although both solutions used the extended Tofts model, the AIF curves were not identical, and the parameter estimation in this model is highly dependent on the assumed AIF (Figure 2 and Figure 3). Based on our observation, the peak concentration of AIF on the short preset was 3.2 mM, whereas the medium preset showed a higher peak concentration of 4.4 mM. This indicates that the preset selection may influence the estimated Ktrans. Under the same tumor concentration at a fixed time point, a higher AIF concentration yields a lower estimated Ktrans value. In contrast, the population-based AIF peak concentration in GenIQ was 2.3 mM, which was the lowest. These observed AIF differences may have contributed, at least in part, to the tendency for Ktrans values obtained with ISP to be lower than those obtained with GenIQ. Our observation is supported by Huang et al., who showed that variation in AIF determination can substantially affect pharmacokinetic estimates in prostate DCE-MRI [18]. These findings have important practical implications. For longitudinal follow-up, quantitative comparison should be performed with the same vendor-specific solution whenever possible. If different vendors are used, numerical changes in perfusion parameters may reflect software-related variation rather than a true biologic change. This point is consistent with current recommendations for reproducibility in quantitative DCE-MRI. The QIBA DCE-MRI Profile was developed specifically to improve standardization, and it suggests that technical variance must be controlled before quantitative changes can be interpreted clinically [10,11]. In our analysis, the difference between population-based and individual AIF-derived parameters appeared to vary between the two solutions. Our results also support that an individual AIF may be preferable to a population-based AIF. Although a population-based AIF is convenient in perfusion analysis, it cannot fully reflect patient-specific hemodynamic factors such as cardiac output, circulation time, or vascular disease [19]. In contrast, an individual AIF is more likely to reflect the actual physiology of the patient. In our observation, the gap between population-based AIF and individual AIF appeared larger in ISP than in GenIQ (Figure 2 and Figure 3). This observation is assumed to be due to the different settings of the AIF curves in the two solutions. As mentioned above, in GenIQ, the population-based AIF showed the lowest peak concentration among the compared models, with a maximum peak of 2.3 mM. In the same solution, the values obtained with the individual AIF were relatively closer to those obtained with the population-based AIF, suggesting less variation according to AIF type. On the other hand, when the individual AIF was applied in ISP, Ktrans showed a much larger deviation from those obtained with the population-based AIF than in GenIQ. In several cases, this discrepancy was on the order of 10-fold to 100-fold, indicating a markedly high variability. These observations suggest solution-dependent differences depending on AIF type. However, this observation was derived from patients with prostate cancer; thus, it should be interpreted cautiously in other malignancies. We suggest that the variability of the estimated perfusion values can be attributed to the AIF type. In addition, measurement based on individual AIF rather than relying only on a population-based AIF may provide more physiologically relevant pharmacokinetic perfusion parameters in DCE-MRI. This study has several limitations. First, we compared only two commercial solutions that were available in our institution, although both are widely used. Therefore, our findings may not apply to all DCE-MRI perfusion analysis solutions. Second, though we meticulously searched for the pharmacokinetic model in detail in white papers and users’ manuals, the detailed pharmacokinetic algorithms were not fully accessible, which may limit our ability to identify other causes of the wide inter-vendor variability. Other hidden vendor-specific factors, including fitting techniques, signal-to-concentration conversion, T1 mapping implementation, and parameter limitations, may also have contributed to the observed variability but were not fully accessible. Based on our observations, we consider the AIF type and preset selection to be major potential contributors to high inter-vendor variability. Third, intra- or inter-reader variability was not assessed because this study focused on inter-vendor variability across solutions. Finally, the relatively wide CIs for some parameters may partly be due to the small cohort size and inherently high inter-vendor variability. 5. Conclusions In conclusion, all DCE-MRI-derived perfusion parameters obtained in patients with prostate cancer showed a wide range of inter-vendor variability. This variability is likely related, at least in part, to differences in vendor-specific AIF type and preset. Therefore, we suggest that comparison of perfusion parameters across different solutions is not recommended, and the same vendor-specific solution be used for longitudinal follow-up.