Early Prediction of Treatment Response of Neuroendocrine Hepatic Metastases after Peptide Receptor Radionuclide Therapy with 90Y-DOTATOC Using Diffusion Weighted and Dynamic Contrast-Enhanced MRI

The purpose of this study was to determine if parameters derived from diffusion-weighted (DW-) and dynamic contrast-enhanced (DCE-) magnetic resonance imaging (MRI) can help to assess early response to peptide receptor radionuclide therapy (PRRT) with 90Y-DOTATOC in neuroendocrine hepatic metastases (NET-HM). Twenty patients (10 male; 10 female; mean age: 59.2 years) with NET-HM were prospectively enrolled in this single-center imaging study. DW-MRI and DCE-MRI studies were performed just before and 48 hours after therapy with 90Y-DOTATOC. Abdominal SPECT/CT was performed 24 hours after therapy. This MRI imaging and therapy session was repeated after a mean interval of 10 weeks. Up to four lesions per patient were evaluated. Response to therapy was evaluated using metastasis sizes at the first and second therapy session as standard for comparison (regressive, stable, and progressive). DW-MRI analysis included the apparent diffusion coefficient (ADC) and parameters related to intravoxel incoherent motion (IVIM), namely, diffusion (D), perfusion fraction (f) and pseudo-diffusion (D∗). DCE-MRI analysis comprised Ktrans, ve and kep. For statistical analysis of group differences, one-way analysis of variance (ANOVA) and appropriate post hoc testing was performed. A total of 51 lesions were evaluated. Seven of 51 lesions (14%) showed size progression, 18/51 (35%) regression, and 26/51 (51%) remained stable. The lesion-to-spleen uptake ratio in SPECT showed a decrease between the two treatment sessions that was significantly stronger in regressive lesions compared with stable (p = 0.013) and progressive lesions (p = 0.021). ANOVA showed significant differences in mean ADC after 48 h (p = 0.026), with higher ADC values for regressive lesions. Regarding IVIM, highest values for D at baseline were seen in regressive lesions (p = 0.023). In DCE-MRI, a statistically significant increase in ve after 10 weeks (p = 0.046) was found in regressive lesions. No differences were observed for the transfer constants Ktrans and kep. Diffusion restriction quantified as ADC was able to differentiate regressive from progressive NET-HMs as early as 48 hours after PRRT. DW-MRI therefore may complement scintigraphy/SPECT for early assessment of response to PRRT. Assessment of perfusion parameters using IVIM and DCE-MRI did not show an additional benefit.


Introduction
Neuroendocrine tumors (NETs) are a heterogeneous group of mostly slowly growing malignancies originating from the cells of the diffuse neuroendocrine system and are most commonly located in the gastrointestinal tract [1]. e name is pointing at the common feature of these cells, which is the release of hormones upon neuronal input. NETs are a rare entity; however, their reported incidence has been steadily rising in the last decades, mainly due to improved diagnostic procedures [2,3]. ey represent a therapeutic challenge for three reasons: First, they stay clinically silent for a long time, which results in late diagnosis [4]. Second, for this reason, they often show hepatic metastases at the time of diagnosis and have therefore limited curative options [5,6].
ird, many neuroendocrine tumors reluctantly respond to standard therapeutic approaches like chemotherapy and show substantial recurrence rates after surgical resection [4,7]. Advanced therapeutic approaches are therefore warranted.
Particularly, the control of the hepatic metastases is crucial to mitigate symptoms, as the metastases can cause pain, compromise liver function, and release serotonine directly into the circulatory system, which results in carcinoid syndrome. An approach that has proven effective to control metastatic NETs is the peptide receptor radionuclide therapy (PRRT) [8][9][10]. It provides symptomatic relief and tumor control at the same time and is applied in more and more medical centers all over the world [11]. e mechanism of action is the selective binding of the peptide, such as DOTA(0)-D-Phe(1)-Tyr(3)-Octreotide (DOTATOC), to somatostatin receptors that are overexpressed at the surface of neuroendocrine tumor cells. e coupled β-emitter, e.g., Yttrium-90, finally exhibits the therapeutic effect that has to be monitored. e standard method for assessment of therapy response in oncology is RECIST 1.1 (Response evaluation Criteria In Solid Tumors) [12,13], which rates response based on the tumor lesions' changes in size. is approach is particularly limited in slowly growing tumors like NETs, where a robust assessment of treatment effectiveness with RECIST in accordance with the proposed frequency of tumor reevaluation in the original publication is performed after a period of several weeks after therapy [12]. is is too late for a prompt personalized adaption of therapeutic strategies and for patients with advanced and less differentiated NET grades 2 and 3. Early response assessment of those NETs within few days after PRRT can therefore be of high interest for these patients and their physicians. A currently used approach to estimate the effect of the PRRT is to measure tumor uptake and tumor-to-spleen ratio in planar posttreatment scintigraphy or SPECT/CTrun a few hours to days after injection of the radiotherapeutic agent. is predominantly reflects the radiopharmaceutical's biological distribution and the density of associated receptors.
Magnetic resonance imaging (MRI) provides additional indicators to assess tissue properties in an oncologic context, mainly using diffusion-weighted MRI (DW-MRI) and dynamic contrast-enhanced MRI (DCE-MRI) [14]. DW-MRI takes advantage of the Brownian motion of water molecules to make statements about the tissue microarchitecture that is subject to changes in the course of treatment. e apparent diffusion coefficient (ADC) provides a quantitative measurement of diffusion. e concept of intravoxel incoherent motion (IVIM) further differentiates the share of DW-MRI signal that can be attributed to diffusion (high b values) and perfusion (low b values), respectively [15]. DCE-MRI on the other hand tracks the signal variation of a tissue at multiple time points after intravenous injection of a contrast medium.
e widely applied Tofts model postulates two compartments (plasma space (PS) and extravascular extracellular space (EES)) and provides the quantitative parameters K trans (transfer constant), V e (fractional volume of EES), and k ep (flux rate constant) [16]. K trans (in 1 per minute) is the flow from PS to EES and physiologically represents plasma blood flow, vascular permeability, and surface area. V e is the volume of EES (in %). k ep is the reflux rate from EES into PS and equals K trans /v e (in 1 per minute). e model-independent initial area under the curve (iAUC) is the integral under the enhancement curve and more robust to noise. It reflects blood flow and vascular permeability, similar to K trans [17].

Patients.
is prospective study was performed at the Clinic of Radiology & Nuclear Medicine at the University Hospital Basel and approved by the local ethics committee (Ethikkommission beider Basel, case number 317/11). Informed consent was obtained from all participants. Twenty patients suffering from a neuroendocrine tumor with hepatic metastases were included. All of them presented for PRRT for the first time and were at least 18 years old. Exclusion criteria were MRI-incompatible foreign bodies (e.g., pacemakers, intracranial clips, and implants), a history of epilepsy, pregnancy, and a limited kidney function (creatinine clearance <50 ml/min).

SPECT/CT
Imaging. SPECT/CT imaging of the abdomen was performed 24 hours after the injection of the radiotherapeutic agent. e images were acquired with a Symbia T6 SPECT/CT system (Siemens Healthineers AG, Erlangen, Germany; matrix: 128 × 128; 64 views of 20 seconds). SPECT images were reconstructed with an OSEMbased Flash 3D algorithm (8 iterations, 4 subsets, and 8 mm Gaussian filtering). For 90 Y/ 111 In-DOTATOC, a scatter and attenuation correction was performed. Additionally, at 3 and 48 hours postinjection (p.i.), scintigraphy was performed to monitor radiation exposure of the kidneys and whole body assessment, respectively.

Magnetic Resonance
Imaging. MRI imaging was performed two times per cycle per patient on three different 1.5 Tesla scanners that were also used in the clinical routine (MAGNETOM Avanto/Symphony/Espree, all Siemens Healthineers AG, Erlangen, Germany). e baseline MRI scan (at T 1 and T 3 ) was performed shortly before the administration of the radiotherapeutic agent. A second scan (at T 2 and T 4 ) was performed 48 hours after the injection. A Gadolinium-based contrast agent (Dotarem ® , Guerbet AG, 0.1 mmol Gd/kg KG) was administered intravenously. e total examination time was 30 minutes. Transversal and coronal T2-weighted images were acquired with a 2D singleshot fast spin echo sequence before administration of the contrast agent (TR/TE: 1,000/89; echo train: 256; field of view: 360 × 360 mm; matrix size: 256 × 256; number of slices: 30; slice thickness: 6 mm; spacing between slices: 7.2 mm).

DCE-MRI.
During and after administration of the contrast agent, 17 serial transversal 3D T1-weighted datasets were acquired at multiple time points (first time frame before contrast agent injection; then every 10 seconds for the first minute p.i.; second and third minute p.i.: every 20 seconds; fourth minute until the end of acquisition at second 300 p.i.: every 30 seconds) with the following parameters: TR/TE: 2.23/0.7 ms; field of view: 360 × 315 mm; flip angle: 15°; matrix size: 128 × 112; number of slices: 30; slice thickness: 3.6 mm.

Image
Analysis. Image analysis was performed on a dedicated workstation separate from the scanner. A maximum of four representing lesions per patient were analyzed (51 lesions in total). e hepatic lesions were defined, and their maximum diameter was measured before and after the administration of the radiotherapeutic agent at each of the two cycles on transversal b700 DWI series using dedicated software (Mint Lesion, Mint Medical GmbH, Heidelberg, Germany). is was done by a dual board-certified radiologist and nuclear medicine physician with 11 years of professional experience (GS). e accuracy of the size measurement was determined by comparing the diameters of measurements at T 1 and T 2 as well as-where available-T 3 and T 4, assuming no change in diameter within 48 hours. e double standard deviation then served as cutoff for a significant increase or decrease in diameter between the first (average of T 1 and T 2 ) and the second treatment cycle (average of T 3 and T 4 , if available; otherwise: only T 3 ): lesions with an increase or decrease above the cutoff value were considered progressive (PD) or regressive (RD), respectively. Lesions with size changes below the cutoff value were considered stable (SD).

SPECT.
SPECT analysis was performed on a dedicated workstation (SyngoVia, Siemens Healthineers AG, Erlangen, Germany). ROIs containing the preselected liver metastases and the spleen were defined semi-automatically based on iso-contours by a dual board-certified radiologist and nuclear medicine physician (GS). Within each ROI, the peak signal intensity (highest number of counts per 1 ccm volume within the ROI) was recorded. e lesion-to-spleen uptake ratio was then calculated for each lesion and both therapy cycles.

DW-MRI.
Diffusion analysis of the preselected liver metastases was conducted by a dual board-certified radiologist and nuclear medicine physician (GS) using SyngoVia (Siemens Healthineers AG, Erlangen, Germany) for ADC and the Medical Imaging Interaction Toolkit (MITK, German Cancer Research Center, Heidelberg, Germany) for the IVIM parameters. Apparent diffusion coefficients (ADCs) were extracted from the ADC maps that were automatically generated by the MRI scanner (mono-exponential fitting) considering the complete tumor volume. Parameters for true diffusion (D), pseudo-diffusion (D * ), and perfusion fraction (f ) were calculated using a 3-parameter-fit according to the intravoxel incoherent motion model [33].

Statistical Analysis.
Statistical analysis was performed using SPSS 22 (IBM Corp., Armonk, NY). A p-value < 0.05 was determined to indicate statistical significance. One-way ANOVA between subjects was conducted to compare means of more than two groups. To test for homogeneity of variance, Levene's test was performed. Histograms, normal QQ-Plots, and the Shapiro-Wilk test were used to test for normal distribution of data. Post hoc comparisons were assessed with Tukey HSD test. In cases with violated assumption of homogeneity, Welch's F-test was used to compare means of groups and Games-Howell test was used for post hoc testing. For nonnormally distributed data, Kruskal-Wallis H test was applied to assess group differences.

Patients.
All 20 patients completed the MRI examinations at T 1 , T 2 , and T 3 . In one patient, DCE-MRI at T2 failed because of a technical error. Six patients (�30%) did not conduct the final MRI examination at T 4 . Patients' characteristics are listed in Table 1. e time distance between the first and the second treatment cycle was 10 weeks on average, with a standard deviation of ±2 weeks (range: 8-18 weeks).

Evaluation of Lesion
Size. 51 lesions were evaluated in total (see Table 2 for characteristics). When comparing the diameters measured at T 1 vs T 2 and (where available) T 3 vs T 4 , a mean test-retest variability of 2.6% with a standard deviation of 3.0% was found. Based on the double standard deviation, a cutoff value of 6% was defined as the threshold for lesion size progression or regression. Lesions with size changes less than ±6% were considered stable.
When comparing lesion sizes at the first and the second treatment cycle (average of T 1 and T 2 vs. average of T 3 and T 4 ; in four cases, T 4 was missing and diameter measurements at T 3 were used), 7/51 lesions showed size progression, 18/51 regression, and 26/51 remained stable. In 13 patients (33 lesions), all lesions behaved uniformly (7x SD, 5x RD, and 1x PD). Two lesions in one of the patients with PR were too small to measure at T 3 . Mixed behavior of lesions was seen in 7 patients (4x stable + regressive lesions,   as a whole rated as SD; 3x stable + progressive, as a whole rated as PD). In summary, 4 patients (20%) were rated as PD, 11 patients (55%) as SD and 5 patients (25%) as PR.

SPECT.
One patient (2 lesions) had to be excluded from the analysis of SPECT data because of splenectomy. e lesion-to-spleen uptake ratio measured at time point T 2 was lower in progressive lesions (1.6 ± 1.2) than in stable (3.1 ± 2.1) and regressive ones (3.4 ± 2.4; Figure 2(a)). However, these differences were not statistically significant (F � 1.8, p � 0.184). ere were also no statistically significant group differences at T 4 (F � 1.8, p � 0.176).
When analyzing the relative differences in lesion-tospleen uptake ratio between T 1 and T 3 , we found a strong decrease for regressive lesions (− 52% on average) and smaller decreases for stable (− 21%) and progressive lesions (− 11%). e difference of the means in the three groups was statistically significant (F � 5.8, p � 0.006; Figure 2(b)). Post hoc comparisons using the Tukey HSD test revealed statistically significant differences when comparing the groups of regressive vs. stable lesions (p � 0.013) and regressive vs. progressive lesions (p � 0.021). However, no statistically significant difference was found between the groups of stable vs. progressive lesions (p � 0.743).
When comparing the differences in ADC before treatment and after 10 weeks ADC(T 3 )-ADC(T 1 ), a higher increase in ADC was seen in regressive lesions (346 ± 322 × 10 − 6 mm 2 /s) than in stable (97 ± 192 × 10 − 6 mm 2 /s) and progressive lesions (98 ± 38 × 10 − 6 mm 2 /s). is difference was statistically significant (F � 6.2, p � 0.004). Post hoc testing showed that only the comparison of regressive vs. stable lesions was statistically significant (p � 0.040), while the comparison of regressive vs. progressive lesions showed a statistical trend (p � 0.058). Figure 3(d) displays the evolution of the average ADC values from time point T 1 , via T 2 to T 3 . Evidently, the abovementioned increase in ADC that is measured between T 1 and T 3 already manifest as a trend at T 2 , as early as 48 hours after treatment. e differences T 2 − T 1 between the three subgroups, however, were not yet statistically significant (F � 1.053; p � 0.357). A sample-case illustrating the abovementioned behavior of ADC as an early indicator of treatment response in a patient with NET-HM of pancreatic origin is shown in Figure 4. e values obtained for the diffusion coefficient D using the IVIM method corresponded well to the ADC values, yet they were associated with larger measurement errors (Figures 5(a) and 5(b)). At T 1 , D was highest for regressive lesions (1103 ± 217 × 10 − 6 mm 2 /s), followed by stable (983 ± 350 × 10 − 6 mm 2 /s) and progressive ones (728 ± 264 × 10 − 6 mm 2 /s). e group differences were statistically significant (F � 4.1, p � 0.023). A more detailed look with post hoc testing showed that only the comparison of regressive vs. progressive lesions (p � 0.017) was statistically significant. e evolution of D values over time from T 1 to T 4 was also comparable to that of ADC with a continuous increase seen for regressive lesions. However, when comparing the difference in D at T 3 minus T 1 and at T 2 minus T 1 , no statistically significant difference between the groups was found (F � 1.7, p � 0.198 and F � 1.1, p � 0.341, respectively).
e perfusion fraction f (Figures 5(c) and 5(d)) did not show any significant differences between the three groups of lesions at T 1 (F � 0.1, p � 0.912), T 2 (F � 1.8, p � 0.185), and T 3 (F � 0.9, p � 0.424). Measured values ranged from 0.08 in progressive lesions at T 4 to 0.16 in progressive lesions at T 1 and T 3 . Interestingly, a drop of perfusion fraction was observed for the progressive lesions at T 2 and T 4 ; however, the differences T 2 − T 1 (F � 0.1, p � 0.914) and T 3 − T 1 (F � 0.03, p � 0.968) were not statistically significant. e pseudo-diffusion coefficient D * (Figures 5(e) and 5(f)) did not show any significant variation between the three subgroups at any time point as assessed by the Kruskal-Wallis H test due to nonnormally distributed data (T 1 : χ 2 � 1.6, p � 0.4; T 2 : χ 2 � 0.3, p � 0.847; T 3 : χ 2 � 4.9, p � 0.088). Visually, the pattern of variation of D * over time was opposite to that of f, showing an increase in D * for progressive lesions at T 2 and T 4 . Yet these effects were not statistically significant (difference T 3 − T 1 : p � 0.872; T 2 − T 1 : p � 0.556). Measured values for D * ranged from 320 × 10 − 6 mm 2 /s for progressive lesions at T 3 to 1208 × 10 − 6 mm 2 /s for progressive lesions at T 4 . In general, the measurement values for f and D * were subject to a rather large variability in between subjects and lesions (see error bars in Figure 5).

Perfusion Weighted Imaging.
Because of the high sensitivity of DCE-MRI to breathing artifacts, five patients (15 lesions) had to be excluded from this part of the analysis. e values obtained for extracellular volume fraction (v e ) are displayed in Figures 6(a) and 6(b). Statistically significant group differences in v e were seen at T 3 (F � 3.8, p � 0.046) and for the difference T 3 − T 1 (F � 4.1, p � 0.025), i.e., between the first and second treatment cycles. Post hoc analyses revealed differences between v e at T 3 for regressive vs. progressive (p � 0.011) and regressive vs. stable lesions (p � 0.010) as well as for regressive vs. progressive (p � 0.036) and regressive vs. stable lesions (p � 0.049) regarding the difference T 3 − T 1 . is corresponds well to the increase in ADC and D seen in Figures 3(d) and 5(b), respectively. At time points T 1 and T 2 , v e was also higher for regressive lesions than for stable and regressive ones, but these  Figure 2: (a) Average lesion-to-spleen uptake of the radiopharmaceutical measured after the first treatment cycle in regressive (light grey), stable (dark grey) and progressive lesions (red). (b) Average difference in lesion-to-spleen uptake of the radiopharmaceutical from T 2 to T 4 indicating significantly higher decrease for regressive lesions (light grey) than for stable (dark grey) and progressive ones (red). e asterisks indicate statistical significance (comparison regressive vs. stable p � 0.013; regressive vs. progressive p � 0.021). (c) Evolution of ADC values from T 1 via T 2 to T 3 for the three subgroups of lesions. Evidently, the increase in ADC that is measured between T 1 and T 3 for all subgroups already manifest as a trend at T 2 , as early as 48 hours after treatment. It is most pronounced in regressive lesions. e asterisks indicate statistical significance.
differences were not statistically significant (p � 0.209 and p � 0.146, respectively). No statistically significant differences were seen when comparing T 1 and T 2 (p � 0.335). e measured values for the transfer constants K trans and k ep are displayed in Figures 6(c) and 6(d). e parameter K trans showed no statistically significant group differences at any point in time (T 1 : p � 0.835; T 2 : p � 0.868; T 3 : p � 0.327) and for the differences T 3 − T 1 (p � 0.611) and T 2 − T 1 (p � 0.921). e parameter k ep, which equals K trans /v e , showed no statistically significant group differences at time points T 1 (p � 0.477) and T 2 (p � 0.495), as well as for the differences T 3 − T 1 (p � 0.271) and T 2 − T 1 (p � 0.927). However, at T 3 , there was a statistically significant difference between the response groups (F � 9.4, p � 0.001) that was driven by the differences of regressive vs. progressive (p � 0.001) and stable vs. progressive lesions (p � 0.04), as revealed by post hoc analysis. For iAUC, there was no statistically significant variation at any point in time (T 1 : p � 0.687; T 2 : p � 0.816; T 3 : p � 0.413) and neither for the differences T 3 − T 1 (p � 0.539) and T 2 − T 1 (p � 0.600).

Discussion
e aim of this study was to investigate if diffusionweighted and contrast-enhanced magnetic resonance imaging can help to assess early tumor response to peptide receptor radionuclide therapy (PRRT) with 90 Y-DOTA-TOC in neuroendocrine hepatic metastasis (NET-HM). For this purpose, MRI was performed immediately before initiation of PRRT, as well as 48 h and 10 weeks after treatment. Results were compared with posttherapy 90 Y/ 111 In-DOTATOC SPECT. e most notable result of our study is that ADC can differentiate regressive from progressive lesions as early as 48 h after therapy. Lesions that were later found to be regressive in size had a significantly higher ADC at this time point than progressive lesions. e fact that apoptosis-inducing therapies like PRRT lead to an increase in ADC values due to the swelling of cells, tumor lysis, and necrosis is well known: Wulfert et al. found an increase in both responding and nonresponding lesions in their study on 38 hepatic NET after intra-arterial treatment with 90 Y/ 111 In -/ 177 Lu-DOTATOC [34], which is well in line with the increase in ADC that is observed after 10 weeks in our study for all groups of lesions. However, an effect on ADC that allows for predicting treatment response of PRRT in NET-HM as early as 48 h after therapy has not been reported yet. e ADC did show a continuous increase from regressive to progressive metastases (RD > SD > PD). e differences between RD/SD and SD/PD were not statistically significant, most likely due to the relatively small number of metastases analyzed.
As an extension to previously published studies on DWI in NET-HM, our study also included an analysis of perfusion effects in DW-MRI using the concept of IVIM. In this analysis, no additional effects to standard DWI were seen, except for the fact that the average D observed in regressive lesions before therapy was significantly higher than in  Figure 3. A statistically significant difference is seen only for D when comparing regressive vs. progressive lesions at T 1 (asterisk; p � 0.017).
progressive lesions. is effect, which is also seen as a trend for ADC in standard DW-MRI, is also in accordance with the results of Wulfert et al., who described a significant correlation between baseline ADC and decrease in lesion size after therapy [34]. Interestingly, a drop of perfusion fraction accompanied by an increase in D * was observed for the progressive lesions at T 2 and T 4 . However, this was not statistically significant and may be due to the considerable measurement variability present in our IVIM analysis. For DCE-MRI, the only significant observation was an increase in the extracellular volume fraction v e that occurred in regressive lesions 10 weeks after therapy, while for stable or progressive lesions, v e did not change statistically significantly.
is behavior of v e probably reflects the well-known effects of the treatment in terms of tumor lysis, necrosis, and fibrosis. It is closely related to the changes in ADC described above and is completely in line with previous results of Atuegwu and colleagues [35]. Regarding K trans and k ep , no significant effects of the treatment on these parameters was observed. is might be due to the low number of patients. Another reason might be the variability in quantitative DCE-MRI [36]. Our findings are in line with the quite heterogeneous results published on this topic. Zahra and colleagues reported that higher baseline K trans and k ep were positively correlated with tumor response of cervix cancer to radiation [37]. Higher baseline K trans and k ep in the responder group were also reported by Tao  for v e (a, b), K trans (c), k ep (d), and iAUC (e) and the three classes of lesions (light grey: regressive; dark grey: stable; red: progressive). e plots in (a) shows the variation of v e over 10 weeks after therapy (from T 1 to T 3 ) with significant differences for regressive vs stable (p � 0.049) and regressive vs progressive lesions (p � 0.036; asterisks). e graphs in (b-e) demonstrate the variation of the four parameters over time, indicating a significant increase of v e in regressive lesions as compared with the two other groups at T 3 (b), but no significant differences in behavior between the three classes of lesions for K trans (c), k ep (d), and iAUC (e). radiotherapy [38]. Gu et al. did not find baseline DCE-MRI parameters useful for the discrimination of responders vs. nonresponders; however, this study as many others operated on small sample sizes (n = 8; [39]).
Finally, the response of the NET-HM to PRRT was also assed with 90 Y-/ 111 In-DOTATOC SPECT, which displays the density of the somatostatin-receptor subtype 2 (SSR2) in tissue and is used as an indicator of expected treatment response of PRRT. As expected keeping in mind previous literature [40], a drop in lesion-to-spleen uptake ratio between the first and the second treatment cycles was seen that was significantly higher for regressive lesions than for stable or progressive ones. e uptake ratio after treatment cycle one, however, was not predictive of therapy response in our small group of patients, which is most likely a result of low statistical power.
Our results are affected by some limitations: First, the small sample size that is due to the exploratory nature of or study, which aimed at identifying parameters of potential prognostic value rather than carving out the exact prognostic value of each parameter. Second, the choice of the reference standard in terms of morphologic response criteria, which were calculated from the test-retest variation of size measurements over 48 h. We are well aware that the threshold of ±6% which resulted from our calculations is far below the thresholds for response assessment proposed in RECIST (12), mRECIST [41], or the EASL response criteria [42]. It was chosen to be more sensitive for changes in diameter, as it is well known that the usually applied response criteria for solid tumors are limited in slowly growing tumors like NET [40]. As the rates of treatment response in our study cohort (20% PD, 55% SD, 25% RD) are consistent with previous reports on PRRT [43], we are convinced that the ±6% threshold eventually represents a reasonable value in this setting. ird, no long-term follow-up imaging and patient survival data were available as reference standard due to the fact that patients underwent follow-up imaging mostly at other centers. Fourth, some series had to be excluded from DW-MRI and DCE-MRI analyses because of imaging artifacts. ese may be avoided by the use of modern sequences that apply respiratory triggering or self-gating technologies. Fifth, one of the assumptions of Tukey HSD test is independence of observations. However, one can argue that the observations are not fully independent due to the fact that in some cases, up to four lesions were measured in one liver. Due to the fact that NET is a rare entity, we did not want to exclude data by including only one metastasis per liver. erefore, we chose to accept this statistical limitation. Finally, no double reading was performed, so interrater variability is not accounted for in our analysis.

Conclusion
In conclusion, diffusion restriction quantified as ADC was the most sensitive MRI parameter to predict treatment response of NET after PRRT among those investigated in our study. ADC was able to differentiate regressive from progressive NET-HMs as early as 48 hours after PRRT. DW-MRI therefore may complement scintigraphy/SPECT for early prediction of treatment response in the framework of PRRT. Assessment of perfusion parameters using IVIM and DCE-MRI did not show an additional benefit in our study but may nevertheless be useful to investigate pathophysiological aspects of PRRT in a preclinical setting.

Data Availability
A large part of the data are patient data and thus confidential. Upon request, a minimal anonymized dataset will be available to interested researchers.

Conflicts of Interest
Markus Klarhöfer is an employee of Siemens Healthcare AG (Switzerland) and declares no conflicts of interest regarding this study. All other authors declare that they have no conflicts of interest.