3 Tesla multiparametric MRI for GTV-definition of Dominant Intraprostatic Lesions in patients with Prostate Cancer – an interobserver variability study

Radiotherapy (RT) of primary prostate cancer (PCa) has been modified in the past decade by using image-guided radiotherapy (IGRT) and intensity modulated radiotherapy (IMRT) techniques [1]. Whole gland dose escalation with IMRT proved to be safe in respect of acute and late toxicities [2–4]. Although prostate cancer is typically a multifocal disease, histopathologic studies revealed that most patients with prostate cancer have at least one or two dominant intraprostatic tumor lesions (DIPL) [5, 6]. For patients scheduled for primary radical radiotherapy obtaining high irradiation doses of the whole prostate are crucial to achieve high biochemical and clinical control rates [7]. However the risk of toxicity, especially in the rectal mucosa inevitably increases with dose escalatation [8], thus requiring highly precise and accurate radiation techniques. There is evidence that local prostate cancer recurrence after primary radiotherapy develops from the origination of the primary tumor or from the initial dominant intraprostatic tumor burden [9, 10]. Experience with IMRT has led to the concept of focal dose-escalation using simultaneous integrated boost of DIPL. Local dose escalation on DIPL may result in significant improved disease control without increasing normal tissue complication probability (mainly acute and chronic rectal mucositis/proctitis). This therapeutic approach has been calculated by Niyazi et al. in a mathematical model based on different assumptions of responsiveness of prostate cancer to irradiation and different sensitivities and specificities of an appropriate imaging method considering choline PET [11].

Many studies with histopathologic comparison on whole-mount sections as reference standard have shown that Magnetic Resonance Imaging (MRI) using anatomic and functional sequences like Magnetic Resonance Spectroscopy (MRS), Dynamic Contrast Enhanced MRI (DCE-MRI) and Diffusion weighted Imaging (DWI) results in high accuracies in detecting primary prostate cancer due to excellent spatial resolution with clear depiction of anatomy/pathoanatomy in combination with visualization of functional properties of prostatic lesions [12–23]. DWI-MRI in conjunction with T2-weighted showed accuracies of 81% and 89% at 1.5 Tesla respectively [17, 18]. DCE-MRI showed a sensitivity and specificity for identification of cancer foci > 0.5 mL of 86% and 94%, respectively [19]. Furthermore a combination of two functional sequences at 1.5 Tesla resulted in a significantly improved area under the receiver operating characteristic (ROC) curve compared to a single functional parameter when whole-mount sections with histologically defined tumor outlines were used as reference standard. Using the combination of apparent diffusion coefficient and initial area under the gadolinium plasma concentration-time curve for detection of cancer foci resulted in an area under the ROC curve of 0.94 reflecting high accuracy. Combination of all three functional parameters (DWI, DCE-MRI and MRS) showed no further improvement [20]. Using T2w sequences at 3 Tesla results in reported sensitivities and specificities of 80%–88% and 96%–100%, respectively [24]. Prostate imaging at 3 Tesla benefits from higher signal to noise ratio (SNR), enables higher quality imaging than obtained at 1.5 Tesla and moreover the use of an endorectal coil can be obviated with satisfying image quality [25] and without distortion of pelvic anatomy which is important for radiotherapy planning [26]. Recently the European Society of Urogenital Radiology (ESUR) published MR guidelines for imaging in prostate cancer and structured reporting [27].

MRI-Criteria to identify an intraprostatic tumor lesion are different throughout the MRI-sequences [27]. Few studies based on consensus reading of a radiologist and radiation oncologist using functional MRI sequences for definition of DIPL have shown that focal dose escalation results in low acute toxicities [28, 29] with better sparing of the rectal wall [30].

The strategy of focal dose escalation to DIPL within the prostate to improve local tumor control and outcome of primarily irradiated prostate cancer patients has gained increasing interest in the past decade [11, 28–30, 44]. A large multicentre randomized trial has been initiated that compares focal dose escalation based on multiparametric MRI findings vs. standard whole gland irradiation. In this trial GTV-delineation is performed by experts in the field of multiparametric prostate-MRI [45].

Other ongoing trials also use MRI to define the GTV for focal dose escalation (e.g. ‘Tumor TARGET Prostate Cancer’ (NCT01802242) or ‘The HEIGHT Trial’ (NCT01411332)). Many studies using a combination of anatomic with functional MRI sequences for detection of DIPL having whole-mount histopathologic as reference resulted in the definition of MRI guidelines by an expert panel [27]. However published anatomic and functional MRI criteria for DIPL have not yet been used in terms of GTV-delineation by different radiation oncologists to elucidate feasibility and potential confounding factors throughout application in clinical practice. To the best of our knowledge this is the first study that compares interobserver variability using multiparametric MRI for GTV-Definition of DIPL in patients with prostate cancer. The GTV-volumes were similar throughout the different MRI-sequences, although increased standard deviations indicate delineation difficulties in some sequences (Tables 3 and 4, Figure 2). We were able to show that a comprehensive but tailored teaching of radiation oncologists about published and widely accepted MRI criteria of DIPL results in substantial to partially excellent agreement compared to an experienced prostate MRI reader depending on the used MRI-sequences (Table 5). Mean KI at T2w and DCE was significantly higher than KI-DWI (Figure 3). Additionally we measured applicability with a 3-point rating score describing difficulties of the delineation process. We found that the degree of difficulty in contouring the GTV was significantly lower using T2w and DCE compared to DWI-sequences (p < 0.0001 for both, Figure 4).

We highlight some important aspects of the delineation process. First, it is important to have anatomic details provided by the T2w-sequence as an underlying dataset to fuse with the functional dataset (DWI, DCE). Complementary morphologic information is essential to avoid delineation errors exceeding organ contours like those that are described in Figures 8 and 9. Second, different signal characteristics of functional sequences should be critically compared to each other to check for possible non-specific findings like bilateral symmetric contrast enhancement described in Figure 5. Furthermore one has to keep in mind that the specificity of functional MRI-sequences is higher than the anatomic T2w sequences [12–23], but depends not only on the signal characteristic but also on the signal distribution in context with the surrounding anatomy [21, 22, 32]. Inadequate comparison may lead to delineation errors as described in Figures 6 and 9. T2w-sequences have a lower specificity for tumor detection than DWI or DCE-sequences [46–48]. However GTV-delineations done by DWI and DCE-MRI sometimes may not co-localize well in tumor-bearing prostate glands because both parameters reflect different tissue properties that are associated with the presence of tumor. To manage this problem Groenendal et al. suggested if DWI and DCE give consistent information, the delineation of a target can be straightforward, because there is a high probability that regions identified by both modalities contain tumor tissue. When the two imaging modalities give inconsistent information, the probability that tumor is present is smaller. A practical approach could be to treat the voxels on which the two modalities agree as the GTV. In case only one of the two modalities indicates a voxel as suspicious, the region could be considered a ‘high-risk CTV’. One could choose not to boost these regions, but in any case safe margins should be applied around these regions [49].

Our study has some limitations. We selected 5 consecutive patients that received functional MRI at 3 Tesla from our database with clearly visible DIPL. Depending on the type of cancer, its growth pattern and patient specific conditions (e.g. antiandrogen therapy prior to MRI [50]) visualization of DIPL may be hampered by difficulties to distinguish or by lacking distinct lesions [28]. 3 Tesla functional MRI is currently the imaging device with the highest accuracy in detection of DIPL due to different functional sequences offering important additional information about specific tissue characteristics. Magnetic resonance spectroscopy (MRS) – sequences were not available, adding MRS-sequences would have led to 15 min extra examination time and is not part of the routine diagnostic work up in our radiology department. Knowing the reported high specificity (but low sensitivity) of MRS to characterize prostate cancer nodules [51] and the limited spatial resolution, we prefer image characteristics of the two other functional MRI-sequences (DCE, DWI) for GTV-delineation and comparison metrics. However preselection of DCE-image-series, of iAUC60-derived maps and the ADC-maps by the reference radiologist may have introduced a bias in the image analysis. In fact according to the ESUR-guidelines further analyses of image data, e.g. comparison of ADC-maps with b-value images (at > b800) and generating DCE-enhancement curves in suspicious regions are useful to more precisely characterize image findings. Future studies may use the newly introduced PI-RADS scoring system to describe DIPL. Standard of reference was predetermined by a radiologist with thorough knowledge of imaging features of prostate cancer using functional MRI but we did not have a whole-mounted histopathologic reference standard. Although GTV delineation was performed with caution by the reference radiologist it cannot be ruled out that in the situation of low tumor to background contrast (e.g. Figure 7) the GTV was arbitrarily delineated to some extent and does not necessarily represent the true tumor extension. It is important to emphasize that the major goal in terms of dose escalation is to define the approximate volume of the dominant intraprostatic lesion, which will be irradiated with a certain safety margin that corrects for intrafraction organ movement and therefore submillimeter precision will not necessarily translate in altered planning target volumes (PTV). In fact it is always an individual decision whether dose escalation is feasible taking into account normal tissue dose constraints that may be influenced significantly by individual factors [30].

The teaching lectures and hand-outs (Table 1) comprised all currently available information to perform the required GTV-delineation. Our results do reflect that the attending radiation oncologists did successfully delineate GTV in some cases according to the MRI sequence. However our analysis also show that significant slips of the pen do occur while GTV-delineation in different MRI sequences and comparison to each other is challenging and therefore should not be used in a clinical setting without expert surveillance. Segmentation algorithms may be useful to reduce interobserver variability of prostate organ delineation [52]. In addition Groenendaal et al. described a logistic regression model that predicts tumor presence on a voxel level in the peripheral zone of the prostate gland based on ADC and K-trans values within a voxel. They found a high correspondence of model and pathologic findings at an AUC of 0.89 [53]. From the radiation-oncologists point of view an imaging device that offers objective and reliable detection of DIPL seems strongly desirable. For this purpose the proposed statistic model showing a high diagnostic performance may be a useful tool for the peripheral zone were most of the tumors occur [53].

MRI has been shown to be improve target delineation [54, 55] and isotropic voxels reduce delineation discrepancies [56]. But even using established MRI-sequences (T2w) for prostate organ delineation may result in significant variability as was recently shown in a multi-observer, -center and -sequence study based in T2w-sequences [57]. In this study Nyholm et al. found that the imaging sequence appears to have a large influence on the delineation variability. Interestingly they found that images with optimal quality were associated with the largest delineation variability. They concluded that increased amount of information increases the scope of interpretation and hence the importance of training and experience. Our results lead to a similar conclusion that a second observer (experienced radiologist) opinion is required until the skills of functional MRI delineation have been developed and trained by the radiation oncologists. Positron-Emission-Tomography in combination with computed tomography (PET/CT) may offer appropriate visualization of functional properties depending on the radiotracer, but experience with labelled choline in the untreated prostate with presence of PCa showed conflicting results with limited accuracy [58–60]. In this respect new and highly specific radiotracers for prostate cancer imaging are required, that are more appropriate for radiotherapy purposes [61].

Using T2w and DCE sequences at 3 Tesla for GTV-definition of DIPL in prostate cancer patients by radiation oncologists with knowledge of MRI features results in substantial agreement compared to an experienced MRI-radiologist, but for radiotherapy purposes higher KI are desirable. DWI sequences for GTV delineation were considered as difficult in application and resulted in only moderate interobserver agreement. From the radiation oncologists point of view GTV-delineation in different MRI sequences and comparison to each other is challenging and therefore should not be used in a clinical setting without expert surveillance.