Introduction

Men who undergo radiotherapy for localized prostate cancer have a one in three to one in four chance of biochemical failure at 5–8 years follow-up.1 Biochemical failure is defined as either a PSA increase of ⩾2 ng dl−1 above the nadir (phoenix definition)2 or more than two consecutive rises of PSA above the nadir.3 Fifteen percent of low-risk and 67% of high-risk prostate cancer patients biochemically relapse within 5 years of radiotherapy.4 Yet the biochemical relapse does not always signify disease relapse and false-positive rates are reported as high as 32%.4, 5 Moreover, PSA elevation can represent recurrence of local disease or development of metastases. Therefore, stratification of patients between no therapy, salvage local therapy and systemic therapy remains to be a cause of concern.6

Conventional anatomical magnetic resonance imaging (MRI) using T2-weighted imaging has been evaluated for detection of local disease in this setting.7 However, glandular atrophy and fibrosis induced by radiation therapy severely limit the accuracy. Several studies have demonstrated that microstructural and functional MRI techniques such as diffusion-weighted imaging (DWI) and dynamic contrast-enhanced (DCE) imaging, either performed individually1, 8, 9, 10, 11 or as part of a multiparametric (mp) examination,12, 13, 14 can improve the detection of recurrent disease following the radiotherapy. However, the robustness of studies that use transrectal ultrasound-guided biopsy as a reference standard to evaluate MRI8, 11, 12, 13, 15 is limited, given the imperfections in the reference standard itself, that is, rates of false-negative findings on transrectal ultrasound-guided biopsy can be as high as 42%.16

Alternative and more robust reference standards are often difficult to obtain in the setting of potential radiorecurrent disease. Salvage radical prostatectomy is technically challenging and rarely performed. This means that 90% or more men who have radiorecurrent disease go onto androgen deprivation therapy and whole mount prostatectomy specimens are not readily available for direct correlation with MRI in this setting.12, 13, 15, 17, 18, 19 Even if they are, it is likely that large selection biases will exist and limit external validity to the biochemical failure group.

Template prostate mapping (TPM) provides an alternative to systematic biopsy. It can sample the entire gland. We have previously reported the outcomes of MRI versus TPM biopsies in a single-center study of 13 patients.18 Furthermore, there is little described on precisely which MRI sequences are necessary for the best diagnostic performance for detection of radiorecurrent disease.15, 17 Our aim was to evaluate the diagnostic accuracy of mp-MRI in a larger cohort of men and explore the added diagnostic value of functional sequences (DWI and DCE imaging) for the detection of radiorecurrent disease using TPM biopsy as the reference standard.

Materials and methods

The institutional review board approved the re-evaluation of patient data sets acquired for other research studies or during the routine clinical care (R&D No: 12/0195, date 16 July 2012).

Patient population

Our institutional TPM biopsy database of 509 patients was interrogated to identify patients presenting with biochemical failure following radiotherapy and who had undergone: (a) subsequent standardized mp-MRI (T2-weighted imaging, DWI (high b-value and apparent diffusion coefficient (ADC) map) and DCE imaging) and (b) TPM biopsy. The updated definition of biochemical recurrence as stated by Phoenix was used.2

Out of 49 patients eligible for inclusion; (a) five were excluded owing to history of brachytherapy; (b) one was excluded owing to limited TPM biopsy sampling, that is, less than 20 cores or non-systematic zonal sampling and (c) six were excluded owing to incomplete MRI arising from hip prosthesis artifact. This left a final study cohort of 37.

MRI protocol

Imaging was performed on two separate platforms, either 1.5 T (Siemens Avanto, n=34) (Siemens Healthcare, Erlangen, Germany) or 3.0 T (Philips Achieva, n=3) (Philips Healthecare, Best, Netherlands). In each case the manufacturers’ multichannel, only pelvic phased array coil was used for imaging. MRI comprised of T2-weighted imaging, DWI and DCE imaging. DWI consisted of multiple b-values, the highest b-value was b=1400 s mm−2 at 1.5T and b=2000 s mm−2 at 3.0 T.

Twenty milligrams of buscopan was administered intravenously before acquisition. Initially, T2-weighted images were acquired, followed by DWI and finally DCE imaging. For DCE imaging, 0.1 mmol kg−1 megluminegadoterate (Dotarem, Guerbet, Villepinte, France) was administered at 3 ml s−1 followed by 10 ml of saline chaser and T1-weighted imaging repeated through the gland volume with a temporal resolution of 13 s at 3 T and 17 s at 1.5 T. Full sequence parameters for 1.5-T and 3.0-T scans are given in Tables 1 and 2.

Table 1 1.5-T Siemens MRI scan parameters
Full size table
Table 2 3 T Philips MRI scan parameters
Full size table

Image viewing

Anonymous studies were independently reviewed on an OsiriX workstation (version 3.7.1 32-bit) (Pixmeo, Geneva, Switzerland) by two experienced radiologists (with 7 and 5 years of mp-MRI prostate experience). Both the radiologists were aware of the history of radiotherapy and biochemical relapse; but were unaware of other clinical details, PSA value or histology. Both the radiologists performed a locked sequential read in a single session with the following read order:

  1. 1

    T2-weighted images

  2. 2

    T2-weighted+high b-value images

  3. 3

    T2-weighted+high b-value+ADC images

  4. 4

    T2-weighted+high b-value+ADC+DCE images

For each read, the radiologists scored the left and right hemiglands for likelihood of cancer using mp-MRI parameters as recently published by the European Society of Urogenital Radiology20 and scored 1–5, where: 1=highly unlikely cancer, 2=unlikely cancer, 3=equivocal, 4=likely cancer and 5=highly likely cancer.

Transperineal TPM biopsy

Patients with biochemical failure following radiotherapy were routinely offered TPM biopsy at our institution. All patients within the study cohort underwent TPM biopsy under general anesthesia using a 5-mm sampling frame as in the method previously described by Barzell and Melamed.21 In all the included patients the full gland was systematically sampled. The mean time from mp-MRI to TPM biopsy was 2.1 months (interquartile range, 1–3 months).

Histopathology and mp-MRI matching

Pathologists were aware that the patients had biochemical failure after radiotherapy with or without neoadjuvant/adjuvant hormonal treatment at time of radiotherapy treatment. Only cores with no pronounced radiation/hormone effect were assigned a Gleason score. All the patients with positive TPM biopsy included in the study have been given Gleason scores. Primary and secondary Gleason patterns of tumor were recorded for each positive core, together with the cancer core length (length of cancer in each core excluding intervening normal areas).22 Histopathology matching was based on the presence of any cancer regardless of the clinical significance. We also performed the analysis according to the presence of clinically significant disease using different mp-MRI thresholds in the Appendix. mp-MRI reader scores were matched to the histopathology at the hemigland level.

Statistical analysis

Sensitivity, specificity, positive predictive values (PPV) and negative predictive values (NPV) together with positive and negative likelihood ratios (LR+ and LR−) were calculated for each reader at each locked sequential read step using MedCalc v13.0.0.0. (Medcalc, Ostend, Belgium) mp-MRI threshold of 3 was used to indicate positive recurrence of cancer. Receiver–operator characteristic (ROC) area under curve (AUC) analysis was performed for each reader at each locked sequential read step using SPSS v22 (IBM; Armonk, New York, USA) and the significant changes were statistically assessed as previously described23 with MedCalc. McNemar’s test was used to detect statistically significant differences between variable sensitivities and specificities of different mp-MRI data sets.

Interobserver agreement was evaluated using Cohen’s Kappa coefficient statistics. To test the agreement, the 1–5 scores were categorized into three groups on the basis of clinical interpretation: negative (score 1 and 2); equivocal (score 3) and positive (score 4 and 5) for cancer. Kappa values of 0.00–0.20 were considered as ‘slight’; 0.21–0.40, ‘fair’; 0.41–0.60, ‘moderate’; 0.61–0.80, ‘substantial’; and 0.81–1.00, ‘almost perfect’ agreement.24

Results

Forty-nine out of seventy-four (66%) hemiglands (thirty-three patients) showed positive TPM biopsy. Table 3 shows baseline demographics of patients included in the study. A typical mp-MRI data set for locked sequential read scores for readers 1 and 2 is presented in Figure 1.

Table 3 Patient demographics
Full size table
Figure 1
figure 1

Axial images from a 68-year-old patient with positive MRI (scoring 5/5), nine positive cores with a maximum cancer core length of 6 mm, 50%, Gleason 3+4 were present at template prostate mapping (TPM) biopsy in the right lateral zone. Both the readers scored the right hemigland as: 3/5 on T2 weighted (a); 4/5 on b1400 (b); on apparent diffusion coefficient (ADC) (c); and 5/5 on dynamic contrast-enhanced (DCE) imaging (d). The left hemigland was scored by readers 1 and 2 as: 3/5 and 3/5 on T2 weighted (a); 2/5 and 2/5 on b1400 (b); 2/5 and 2/5 on ADC (c); and 2/5 and 1/5 on DCE (d). No cancer was present in the left hemigland on TPM biopsy.

Full size image

Figures 2 and 3 show the ROC curves of AUC (ROC-AUC) for readers 1 and 2 at each locked sequential read step, respectively. Tables 4a and 4b show the accuracy figures (sensitivity, specificity, PPV, NPV, LR+ and LR−) for both the readers at different locked sequential read.

Figure 2
figure 2

Receiver–operator characteristic (ROC) curves of reader 1 for the four locked sequential read. ROC-AUC for T2, T2+high b-value, T2+high b-value+apparent diffusion coefficient (ADC) and T2+high b-value+ADC+dynamic contrast-enhanced (DCE) imaging were 0.67, 0.80, 0.77 and 0.80, respectively.

Full size image
Figure 3
figure 3

Receiver–operator characteristic (ROC) curves of reader 2 for the four locked sequential read. ROC-AUC for T2, T2+high b-value, T2+high b-value+apparent diffusion coefficient (ADC) and T2+high b-value+ADC+dynamic contrast-enhanced (DCE) imaging were 0.67, 0.80, 0.77 and 0.84, respectively.

Full size image
Table 4a Reader performance at the hemigland level with mp-MRI score ⩾3 as positive
Full size table
Table 4b P-values between different mp-MRI reads at threshold ⩾3 as positive
Full size table

First-read: T2-weighted imaging

Sensitivity/specificity and PPV/NPV were 76%/52% and 76%/52%, respectively, for reader 1, and 78%/44% and 73%/50%, respectively, for reader 2.

The ROC-AUC, for classification of hemigland status, was 0.67 (95% CI 0.55–0.80) for both readers 1 and 2 (Table 5).

Table 5 Receiver–operator characteristic area under curve (ROC-AUC)
Full size table

There was a ‘fair’ interobserver agreement (κ=0.39; 95% CI 0.20–0.58) between the two readers (Table 6).

Table 6 Interobserver agreement between readers of locked sequential mp-MRI reads
Full size table

Second-read: T2-weighted+high b-value DWI

Sensitivity/specificity and PPV/NPV were 69%/88% and 92%/59%, respectively, for reader 1, and 65%/92% and 94%/58%, respectively, for reader 2.

With addition of high b-value DWI, the ROC-AUC significantly improved for both the readers to 0.80 (95% CI, 0.70–0.92 and 0.70–0.90 for readers 1 and 2, respectively) (P=0.02 and 0.01 for readers 1 and 2, respectively).

Interobserver agreement was ‘moderate’ (κ=0.51; 95% CI 0.36–0.64) following the review of high b-value DWI.

Third-read: T2-weighted+high b-value DWI+ADC maps

Sensitivity/specificity and PPV/NPV were 63%/88% and 91%/55%, respectively, for reader 1, and 63%/92% and 94%/56%, respectively, for reader 2.

There was no statistically significant change in ROC-AUC for either reader on additional review of ADC maps (ROC-AUC=0.77, P=0.08 for reader 1; ROC-AUC=0.77, P=0.47 for reader 2).

Interobserver agreement was moderate (κ=0.48; 95% CI 0.31–0.64).

Final-read: T2-weighted+high b-value DWI+ADC maps+DCE imaging

Sensitivity/specificity and PPV/NPV were 80%/68% and 83%/63% respectively, for reader 1, and 76%/72% and 84%/60%, respectively, for reader 2.

ROC-AUC was marginally, but not significantly, higher after the inclusion of DCE images (0.80, 95% CI: 0.69–0.91 for reader 1 and 0.84, 95% CI: 0.76–0.93 for reader 2) compared with the T2+high b-value (P=0.90 and 0.27 for readers 1 and 2, respectively) and T2+high b-value+ADC reads (P=0.47 and 0.11 for readers 1 and 2, respectively).

Interobserver agreement was ‘substantial’ (κ=0.65; 95%-CI 0.51–0.79).

Comparison of diagnostic accuracy between locked sequential read steps

Sensitivity was highest for the full mp-MRI data set read (80% and 76% for readers 1 and 2, respectively). However, there was no statistically significant difference in sensitivity between full mp-MRI data set and T2+high b-value (P=0.13 and 0.23 for readers 1 and 2, respectively).

Specificity was highest at T2+high b-value and T2+high b-value+ADC (88% and 88% for reader 1 and 92% and 92% for reader 2, respectively). This was significantly higher than that of T2 alone (P=0.004 and P=0.002 for readers 1 and 2, respectively). There was no statistically significant difference on adding DCE imaging at the final read (P=0.063 for both the readers).

The highest LR+ was 5.78 and 8.16 for readers 1 and 2 at the locked sequential read of T2+high b-value. This was higher than that for the T2-weighted imaging alone (1.57 and 1.38 for readers 1 and 2, respectively) and full mp-MRI data set (2.49 and 2.70 for readers 1 and 2, respectively).

Discussion

Although evidence is starting to emerge that mp-MRI can aid in the detection of local disease following radiotherapy, study sizes remain small and what constitutes an optimal mp-MRI data set remains unknown.15, 18 In this work, we explored the incremental value of individual mp-MRI sequences for the detection of radiorecurrent prostate cancer validated against a robust TPM biopsy reference. We employed a locked sequential read paradigm using T2-weighted imaging, high b-value DWI, ADC maps and DCE imaging. Our sequence of reading mp-MRI pictures in the current study was T2-weighted≫high b-value≫ADC maps then DCE imaging. Our cohort was homogenous, all our patients had previous external beam radiotherapy only and we did not include patients with brachytherapy as in other studies.13, 15

Consistent with the work of others, our results show that addition of ‘functional’ MRI sequences to the anatomical T2-weighted imaging improves the performance of readers detecting radiorecurrent prostate cancer.13, 18, 19 However, we did not observe a statistically significant increase in ROC-AUC of readers between T2-weighted imaging+high b-value DWI and sequential addition of ADC and DCE MRI.

A single preliminary study of mp-MRI (validated by TPM biopsy) suggested a good performance for detection of radiorecurrent prostate cancer.18 AUC of the ROC curves for the two readers were reported as 0.77 and 0.89 for all the cancers and 0.86 and 0.93 for cancer core lengths ⩾3 mm (clinically significant). Our results of a larger cohort are in line with that previous study. Moreover, in the current study we looked at the added value of functional MRI sequences to the basic anatomical T2-weighted sequence. Akin et al.13 have also reported on mp-MRI performance in the radiorecurrent setting. Analysis was performed at patient and prostate sextant levels where transrectal ultrasound-guided biopsy (12–16 cores) was used as the reference standard. They did not perform a locked sequential read for each sequence. Yet, our results on the overall performance of mp-MRI are in agreement with their work.

Recent work shows that DWI, either using ADC and/or high b-value, has similar sensitivity and higher specificity compared with DCE imaging in patients with radiorecurrent prostate cancer.8, 10, 19, 25 There is little data on the incremental value of DCE imaging to other mp-MRI sequences for radiorecurrent disease. Donati et al.15 showed that DCE imaging did not add significant value in such patients. Our study compared the utility of high b-value DWI in combination with multi-b-value DWI-derived ADC maps, an area that was not covered by the work of Donati et al. We found no significant incremental value of reading ADC after high b-value DWI (P=0.08 and 0.47 for readers 1 and 2, respectively) and in keeping with their work the subsequent addition of DCE imaging. It is likely that DCE images do add value when used exclusively in conjunction with T2 imaging; Haider et al.11 found better performance of DCE imaging than T2-weighted imaging in localization of prostate cancer. However, no DWI was included in their study. Kim et al.17 found no statistically significant difference in sensitivity/specificity or accuracy among DWI, DCE imaging and both the sequences combined. However, they found statistically significant difference in ROC-AUC between combined DCE imaging+DWI and each isolated sequence.

Our inter-reader agreement data also demonstrate that agreement between readers improved with the addition of ‘functional’ imaging (κ=0.65, 95% CI 0.51–0.79), which is comparable to other studies.13, 18 The lowest agreement was for T2 (κ=0.39, 95% CI 0.20–0.58), which is also comparable to that in the literature.13 Our results confirm that, should a patient be considered for salvage treatment after biochemical recurrence, T2-weighted together with high b-value DWI are the minimum needed sequences for the detection and localization of local recurrence.

Our study has several limitations. We did not have access to salvage prostatectomy specimens as a reference standard. However, we were able to fully sample the prostate with TPM biopsy. Although no biopsy is free from sampling error,26 TPM addresses much of the systematic error that is inherent to the transrectal ultrasound-guided biopsy.27

mp-MRI was performed using pelvic phased array only without the use of endorectal coil. Although endorectal coil is known to increase the signal-to-noise ratio and potentially increase the sensitivity of T2-weighted imaging,7, 28, 29 however, to our knowledge no study has compared endorectal coil with pelvic phased array in the postradiotherapy setting.

Our retrospectively gathered patient cohort incorporates an unavoidable spectrum bias as in other similar studies;13 patients without a suspicious lesion clinically reported on prebiopsy mp-MRI often opted to avoid TPM and therefore would not have been included within the study cohort. To mitigate the effect of this, we based our analysis at hemigland level, thereby increasing true negative numbers and reducing the disease prevalence (66%) to a level consistent with that reported previously.9 We believe that the prevalence of disease where mp-MRI is routinely employed for detection of radiorecurrent disease will be lower.

We did not include patients with biochemical failure after brachytherapy13, 15 and therefore our results may not be applicable to this patient population. Brachytherapy seeds can potentially cause image artifact and the effect of this varies with the functional MRI method.15

We used more than one mp-MRI platform, 1.5 T and 3 T in our study; however, the impact of that is unknown and has been experienced in other studies.13 This is in addition to the fact that the value of using higher magnetic resonance strength on performance of MRI has not been clearly recognized.30

We did not perform separate locked sequential reads of T2-weighted+DCE imaging or DWI+DCE imaging as we opted to order the sequences based on the potential time/financial cost associated with each additional sequence. Specifically, high b-value DWI can be performed faster than the acquisition of data for a full ADC map, and performance of DCE imaging involves additional time and financial cost of the contrast agent. It should be noted, however, that DCE imaging could be beneficial if there is a motion artifact where DWI is difficult to interpret.

Conclusion

Establishing an optimal mp-MRI examination will help reduce examination cost, reduce scan time and improve patient comfort. Although the performance of mp-MRI (T2-weighted+DWI+DCE imaging) was the best, our study shows that a minimum examination of T2-weighted imaging along with high b-value DWI acquisition might be all that is necessary for the detection of radiorecurrent prostate cancer. Larger prospective studies are needed to confirm these findings.