Due to the increasing availability of multiparametric magnetic resonance imaging (mpMRI) in general, the improved image quality at 3 Tesla, and the increasing number of studies confirming the diagnostic reliability of mpMRI for prostate cancer (PCa) detection, mpMRI proceeds to become an important and widely used tool for PCa diagnosis [
The multiparametric approach using three different MRI techniques (T2-weighted MRI (T2W-MRI), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced MRI (DCE-MRI)) can improve the diagnostic accuracy. However its complexity and the sometimes contradictory findings of the different single modalities may result in a wide scope of possible interpretations of mpMRI findings leading to heterogeneities between different readers and different diagnostic centers [
To overcome these problems, the European Society of Urogenital Radiology (ESUR) has called a panel of experts and published a guideline providing recommendations for the performance of mpMRI investigations and a structured reporting scheme named Prostate Imaging Reporting and Data System (PI-RADS) in February 2012 [
The aim of this study was to evaluate the PI-RADS scoring system in our patient population and to find the best way of generating the overall PI-RADS score.
From October 2011 to May 2013, 143 consecutive patients with a history of at least one negative systematic prebiopsy, who underwent 3 Tesla mpMRI of the prostate because of ongoing tumor suspicion, were included in this retrospective single-center study in Innsbruck. None of the patients were under treatment with 5-alpha-reductase inhibitors at the time the MRI was performed. Of these 143 patients, 73 underwent systematic and targeted rebiopsy within 3 months after mpMRI. Patient characteristics are summarized in Table
Patient characteristics at the date of mpMRI.
All patients ( |
Patients with rebiopsy ( |
|
---|---|---|
Age (years), mean (s.d.) | 62 (7.8) | 62 (7.4) |
Prostate volume (cm³), median (interquartile range) | 45 (34–60) | 45 (34 to 61) |
Negative prebiopsies, |
||
1 | 51 (36%) | 17 (23%) |
2 | 50 (35%) | 31 (42%) |
3 | 23 (16%) | 15 (21%) |
4 | 13 (9%) | 6 (8%) |
|
6 (4%) | 4 (5%) |
PSA (ng/mL), median (IQR) | 6.4 (5.0–11.3) | 7.0 (5.1 to 12.9) |
Free PSA (%), median (IQR) | 13.8 % (11.0%–18.45%) | 13.4% (10%–18.6%) |
mpMRI was performed on a 3 Tesla whole body scanner (Magnetom Skyra, Siemens AG, Erlangen, Germany) using an 18-channel phased array body coil with 18 integrated preamplifiers. Examinations included 2D and 3D T2W-MRI, DWI, and DCE-MRI. MRI parameters are shown in Table
mpMRI parameters.
T2W-MRI | DWI | DCE-MRI | |
---|---|---|---|
Sequence | Fast spin echo | Spin echo EPI | T1w-3D FLASH |
TR (ms) | 4891 | 6800 | 2.89 |
TE (ms) | 101 | 67 | 1.12 |
Flip angle (°) | 160 | 90 | 2 |
FoV (mm²) |
|
|
|
Resolution |
|
|
|
Slice thickness (mm) | 3 | 3 | 4 |
|
— | 50/400/1000 | — |
TR: relaxation time, TE: emission time, FoV: field of view, and FLASH: fast low angle shot magnetic resonance imaging.
The mpMRI datasets were analyzed by two experienced uroradiologists with at least 6 years of experience in prostate MRI interpretation, who compared two different approaches to generate an overall PI-RADS score.
In a first step the three single-scores (1–5) for T2W-MRI, DWI, and DCE-MRI for each patient were defined according to the ESUR guidelines (Table
Single-modality scores according to the ESUR panel [
|
|
(1) Uniform high signal intensity | |
(2) Linear, wedge-shaped, or geographical areas of lower signal intensity, usually not well demarcated | |
(3) Intermediate appearances not in categories 1/2 or 4/5 | |
(4) Discrete, homogeneous low-signal focus/mass confined to the prostate | |
(5) Discrete, homogeneous low-signal-intensity focus with extracapsular extension/invasive behavior or mass effect on the capsule (bulging) or broad (>1.5 cm) contact with the surface | |
|
|
(1) Heterogeneous transition zone adenoma with well-defined margins: “organized chaos” | |
(2) Areas of more homogeneous low signal intensity, however, well marginated, originating from the transition zone/benign prostatic hyperplasia | |
(3) Intermediate appearances not in categories 1/2 or 4/5 | |
(4) Areas of more homogeneous low signal intensity, ill defined: “erased charcoal sign” | |
(5) Same as 4, but involving the anterior fibromuscular stroma or the anterior horn of the peripheral zone, usually lenticular or water-drop shaped | |
|
|
(1) No reduction in ADC compared with normal glandular tissue; no increase in signal intensity on any high- | |
(2) Diffuse, hyper signal intensity on ≥ | |
(3) Intermediate appearances not in categories 1/2 or 4/5 | |
(4) Focal area(s) of reduced ADC but isointense signal intensity on high- | |
(5) Focal area/mass of hyper signal intensity on the high- | |
|
|
(1) Type 1 enhancement curve | |
(2) Type 2 enhancement curve | |
(3) Type 3 enhancement curve | |
(+1) For focal enhancing lesion with curve types 2-3 | |
(+1) For asymmetric lesion or lesion at an unusual place with curve types 2-3 |
Subsequently, in a first algorithm based approach the first radiologist calculated a PI-RADS sum-score (scale from 3 to 15) by summation of the 3 single-scores. The overall PI-RADS score (1–5) was obtained by classifying the sum-score according to the algorithm proposed by Röthke et al. [
Calculation of the overall PI-RADS score according to the definitions of the ESUR panel compared to the algorithm presented by Röthke et al. [
Overall PI-RADS | Definition of the ESUR panel | Sum-score of T2W, DWI, and DCE |
---|---|---|
Score 1 | Clinically significant disease highly unlikely to be present |
3, 4 |
Score 2 | Clinically significant cancer is unlikely to be present |
5, 6 |
Score 3 | Clinically significant cancer is equivocal |
7–9 |
Score 4 | Clinically significant cancer is likely to be present |
10–12 |
Score 5 | Clinically significant cancer is highly likely to be present | 13–15 |
In a second more subjective approach the second radiologist independently generated an overall PI-RADS score by subjectively weighting the results of the single-scores according to the definitions of the ESUR panel (Table
Image interpretation and scoring were done before biopsy, so the radiologists were blinded to the histopathological outcomes. For reporting and localization of findings previous to targeted biopsies the prostate was divided into 27 regions as recommended by the ESUR guidelines according to a scheme presented by Röthke et al. [
Within 3 months after mpMRI 73 patients underwent re-biopsy, which was indicated in consideration of radiological and clinical findings by the attending urologist. Within one re-biopsy setting one of the uroradiologists, who interpreted the mpMRI images, took 5 targeted cores of those lesions that were suspicious on at least one single modality (PI-RADS sum-score
mpMRI-ultrasound image fusion: suspicious lesion (arrows) on T2W (a), on DWI with low ADC (b), and washout curve on DCE (c). Correlation of an anatomical landmark (cyst) for registration of
For histopathological analysis all biopsy specimens were numbered, reviewed by a pathologist with >10 years of experience in prostate characterization, and reported as PCa with an assigned Gleason score, prostatitis, adenomyomatosis, benign prostatic hyperplasia, or atrophy.
Summary statistics are provided using the appropriate measures of location and measures of variation for all 143 patients. The D'Agostino-Pearson test was used to test for normal distribution. Mean values ± standard deviations were given for normal distributed data and otherwise median with interquartile range. The different approaches to generate the overall PI-RADS score were compared regarding number and distribution of score levels for all patients within the collective.
Correlation of mpMRI findings and histopathological findings was performed only for the collective of 73 patients, who underwent re-biopsy: to assess a possible positive association between the number of biopsies conducted before the re-biopsy and the relative number of tumor cases, a Chi-squared test for trend was applied. A receiver operating characteristic (ROC) analysis was performed to evaluate sensitivity and specificity of the scoring system with regard to tumor incidence and tumor malignancy. For statistical analysis respective to tumor malignancy, histopathologic results were split into two groups (Gleason score level
After performing mpMRI, 39 (53%) out of 73 targeted rebiopsies were positive for prostate cancer. Of the 39 tumors, 22 (56%) were located in anterior parts of the prostate, and 17 (44%) in the transitional zone (TZ) while 17 tumors (44%) were located in the posterior parts and 22 (56%) in the peripheral zone (PZ). Regarding tumor malignancy, 29 (74%) were cancers with Gleason
After evaluating the 3 single modalities and adding the single-scores, the collective of 143 patients revealed sum-scores with a median of 8 (range 4–15, IQR 6 to 10). In the group of patients with targeted re-biopsy the PI-RADS sum-score was positively related to the number of cancer positive cores (
Distribution of tumor incidences for PI-RADS single-scores and sum-scores.
Receiver operation characteristic (ROC) curves for the PI-RADS sum-score, regarding thresholds for tumor incidence with a cutoff at 10 (a) and for tumor malignancy with a cutoff at 13 (b).
Suspicious lesions (arrows) on mpMRI with different PI-RADS sum-scores. Gleason 8 carcinoma: 5 points on T2W for hypointensity and bulging (a), 5 points on DWI for focal very low ADC (b), and 5 points on DCE-MRI for washout curve in a focal lesion (c, d) = sum-score of 15 points. Gleason 7 (
Overall PI-RADS score according to Röthke et al. [
Overall PI-RADS score level (1–5) | Score based on Röthke et al. | Score based on radiologist’s impression | ||
---|---|---|---|---|
Frequency of patients |
Tumor incidence (% of biopsies) | Frequency of patients |
Tumor incidence (% of biopsies) | |
1 | 1 (1%) | — | 0 (<1%) | — |
2 | 43 (30%) | 0% | 38 (27%) | 0% |
3 | 52 (36%) | 19% | 50 (35%) | 17% |
4 | 31 (22%) | 65% | 38 (27%) | 54% |
5 | 16 (11%) | 94% | 17 (12%) | 100% |
Both, the first approach based on the algorithm of Röthke et al. (PI-RADS scheme 1) and the second approach (PI-RADS scheme 2), based on the overall impression of the radiologist, revealed overall PI-RADS scores, which showed increasing tumor incidence with increasing score levels. When classified according to the algorithm of Röthke et al., it is noticeable that their cutoff between overall PI-RADS 3 and 4 corresponds to the calculated threshold for tumor incidence on the PI-RADS sum-score and their cutoff between 4 and 5 to our calculated threshold for higher tumor malignancy. According to this approach, the prostates of 47 (33%) patients revealed cancer suspicious lesions (PI-RADS scores of either 4 or 5) of which 35 (82%) proved to be cancer positive after targeted biopsy. When generating the overall PI-RADS score simply by the radiologist’s impression on the other hand 55 (38%) prostates revealed cancer suspicious lesions, but only 37 (67%) of these proved to be cancer positive after targeted biopsy. Regarding the frequency of PI-RADS 3 lesions, both approaches assigned a similar number of patients to this score level. Nevertheless with 19% compared to 17% biopsy proved tumor incidence in PI-RADS 3 patients was slightly higher for PI-RADS scheme 1. PI-RADS 1 and 2, which mean low suspicion for clinically relevant disease, were diagnosed in 44 (31%) patients when using PI-RADS scheme 1 and in only 38 (27%) patients with PI-RADS scheme 2. None of the biopsies taken from these patients revealed cancer positive cores. The very rare diagnosis of PI-RADS 1 in both approaches can be explained by the presence of multiple tissue alterations in this collective of patients with negative prebiopsies (Table
With this study we could demonstrate a good reliability of the PI-RADS risk stratification system for the interpretation of mpMRI in our patient population: all 3 single-scores and thus the calculated PI-RADS sum-score of 3–15 points showed a clear association with tumor incidence and tumor malignancy with large AUC in ROC curve analysis. In concordance with the other studies, which recently evaluated the PI-RADS classification system with slightly different approaches, this suggests high reliability for the use of a system with fixed criteria for mpMRI interpretation [
Regarding sensitivity and specificity levels of the PI-RADS sum-score on ROC analysis, our results suggest either ≥10 or ≥11 as possible thresholds for the increase of tumor incidence. The question of which of these two values should be used as a threshold to indicate distinct cancer suspicion was discussed with our clinicians, who clearly favored the threshold of ≥10 points for its very high sensitivity level of 90% with an acceptable specificity level of 62%. This goes along to the findings of Schimmöller et al., who also evaluated the sum-score level of ≥10 to be the threshold for tumor incidence and reported a sensitivity of 85.7% and a specificity of 67.6% [
The second goal of this study was to find a reliable approach to generate the overall PI-RADS score, which in the end shall be part of the clinical report as a simplified risk stratification system and which could provide recommendations for further diagnostic procedures. Regarding this issue, the ESUR guidelines lack a consistent instruction of how to generate the overall PI-RADS score [
Comparing the algorithm of Röthke et al. [
The second approach to generate an overall PI-RADS score, based on the radiologist’s impression, showed less association with the thresholds of the sum-score, and the evaluating radiologist assigned more prostates to PI-RADS 4 and 5, which lead to lower tumor incidences (67% compared to 82%) in these categories and thus less specificity. Regarding PI-RADS 4 and 5 as possible indications for re-biopsy this would have led to a higher number of interventions with a higher percentage of negative results. Therefore, according to our data, an algorithm based approach, which derives the overall PI-RADS score from the sum-score seems to be more reliable.
However, the overall PI-RADS score, recommended by Röthke et al. [
Recommendation to calculate an overall PI-RADS score, based on division from the sum-score, with tumor incidences derived from our data.
Overall PI-RADS score | Sum-score of T2W, DWI, and DCE-MRI | Number of patients (%) | Tumor incidence (% of biopsies) | Definition of the ESUR panel |
---|---|---|---|---|
1 | 3, 4 | 1 (1%) | — | Clinically significant disease highly unlikely to be present |
2 | 5, 6, |
59 (41%) | 11% | Clinically significant cancer unlikely to be present |
3 |
|
36 (25%) | 19% | Clinically significant cancer is equivocal |
4 | 10–12 | 31 (22%) | 65% | Clinically significant cancer likely to be present |
5 | 13–15 | 16 (11%) | 94% | Clinically significant cancer highly likely to be present |
Changes in comparison to the system of Röthke et al. [
This study is prone to some limitations. This study was designed as an evaluation of our clinical routine and not every patient underwent re-biopsy of the prostate. This might have led to a verification bias, since patients with few or no abnormalities on mpMRI less frequently underwent re-biopsy. Furthermore in patients without suspicious lesion on at least one single modality (sum-score <7) no targeted biopsies could be performed and systematic re-biopsy had to be used as a gold standard. Therefore all tumor incidences, calculated for low PI-RADS score levels (sum-score <7 or overall Pi-RADS 1 and 2), should be regarded as uncertain. Further studies with data based on a long followup will be necessary to evaluate reliable tumor incidences for these low suspicion groups. Additionally, since each of the evaluating radiologists used a different approach for scoring, we do not have data about interobserver variability within the same approach. For this we refer to a recent study of Schimmöller et al. [
The PI-RADS sum-score (3–15) shows a strong relation to tumor incidence and malignancy in our routine setting for PCa diagnosis. A score level of ≥10 seems to be an important threshold for a positive tumor diagnosis and of ≥13 for the existence of high Gleason scores (≥ 4+3). For generating the overall PI-RADS score, which is part of the clinical report, our results indicate a recommendation for a number based algorithm with a slightly elevated threshold between PI-RADS 2 and 3 compared to that of Röthke et al. [
The authors declare that they have no conflict of interests.