Multiparametric whole-body 3.0-T MRI in newly diagnosed intermediate- and high-risk prostate cancer: diagnostic accuracy and interobserver agreement for nodal and metastatic staging

Since patient survival in intermediate- and high-risk prostate cancer depends heavily on TNM stage [1], accurate tumour staging should underpin all prognostication and management decisions. However, the mainstay of imaging-based staging decisions is still based on ^99mTc bone scintigraphy (BS) ± pelvic CT, as is still advised at least eight international guidelines [2]. Whilst these modalities are simple to implement, their diagnostic accuracy remains severely limited [3, 4], which has driven the development of a number of imaging methods for cancer staging,

While choline PET/CT offers improved sensitivity and specificity for both nodal [4, 5] and metastatic disease vs. BS and conventional CT [4], PET involves ionising radiation exposure and has a spatial resolution limited to 5 mm [6], poor contrast resolution, and financial and logistical difficulties which limit its use. Whole-body (WB)-MRI offers potential solutions to these problems, with improved spatial and contrast resolution [7], lack of ionising radiation, and comparable performance characteristics to choline PET/CT as shown by a number of early studies [8‐12]. Although prostate-specific membrane antigen (PSMA) PET/CT has demonstrated considerable early promise [13], its availability is limited and incurs considerable cost. Furthermore, since prostate cancer patients commonly undergo multiparametric (mp) prostate MRI, the possibility of a one-stop staging modality has been raised [12] whereby mp-WB approaches could also be applied. However, the interobserver concordance of WB-MRI remains uncertain, as does a definition of what constitutes an optimal acquisition. Further validation regarding diagnostic accuracy is also required.

The primary aim of the present study is to determine the diagnostic accuracy of WB-MRI vs. BS and ¹⁸fluoro-ethyl-choline (¹⁸F-choline) PET/CT for the primary staging of intermediate- and high-risk prostate cancer, using a multiparametric vertex-to-feet acquisition protocol and a locked sequential read (LSR) paradigm to determine the additive value of each MRI sequence. Secondary aims include assessment of lesion distribution, interobserver concordance, and intermodality concordance with BS and ¹⁸F-choline PET/CT. We hypothesise that (i) WB-MRI has a higher diagnostic accuracy than BS, (ii) WB-MRI has good interobserver concordance, and (iii) a multiparametric whole-body acquisition has a greater diagnostic accuracy than T1-weighted imaging plus DWI.

Our institutional review board approved this prospective single-centre study. Informed written consent was obtained from each participant, whereby 56 consecutive men (mean age 67.9 years, range 51.9–84.4) were identified at Multidisciplinary Tumour Board (MTB) meetings and recruited to the study between July 2012 and November 2015. Inclusion criteria were (i) men aged 18 or over and (ii) new diagnosis of intermediate- or high-risk prostate cancer according to the D’Amico criteria [14]. Exclusion criteria were (i) contraindications to MRI, e.g. severe claustrophobia or MR unsafe device, (ii) prior therapy for prostate cancer, and (iii) men unable to provide informed consent. A recruitment flow diagram is shown in Fig. 1.

Standard imaging comprised BS in all patients ± ¹⁸F-choline PET/CT, in 33 patients. The decision to perform a ¹⁸F-choline PET/CT was made on a case-by-case basis whereby the risk of extraprostatic disease was considered to be high at MTB discussion; however, the result of the WB-MRI was blinded to the MTB members, so it did not influence the decision to perform PET/CT. WB-MRI was performed within a mean of 15.9 days (range 0–49) of BS.

A panel comprising two board-certified radiologists (SP and EJ with 12 and 6 years of experience), and an oncologist with 8 years of experience (RD) reviewed baseline and follow-up WB-MRIs, in combination with all available clinical and radiological information at least 1 year from baseline imaging to carry out a patient-based analysis, and assign patients into the following categories using the definitions below for all modalities. Patients were included in the M1a sensitivity/specificity analysis if:

Similarly, patients were included in the M1b sensitivity/specificity analysis if:

The reference standard was subsequently derived using the following definitions:

(i) Lesion on WB-MRI (defined as suspicion level 4/5/6) which is BS and/or ¹⁸F-choline PET/CT positive (if performed). Follow-up WB-MRI (if performed) also demonstrates lesion progression without systemic therapy, decrease with systemic therapy, or new lesions. (ii) Lesion on WB-MRI which is negative on conventional imaging but progresses on WB-MRI follow-up without systemic therapy, or new lesions appear on WB-MRI follow-up.

A flow diagram of the statistical methods and reference standard used in this study is provided in Fig. 2.

A similar reference standard was used to compare BS and ¹⁸F-choline PET/CT, whereby positive scans concordant with WB-MRI were considered as true positive, with follow-up WB-MRI used for arbitration of discordant and negative BS and ¹⁸F-choline PET/CT findings. The details for this reference standard are provided in the supplementary materials.

Fifty-six patients (mean age 67.9 years, range 51.9–84.4 years), median PSA 20.05 (IQR 10.07–61.20). Fifty patients were ‘high-risk’ and 6 patients ‘intermediate-risk’. Maximum Gleason score was 3 + 3 for two patients, 3 + 4 for nineteen patients, 4 + 3 for fourteen patients, 4 + 4 for five patients, 4 + 5 for thirteen patients, and 5 + 5 for one patient.

The mean time of radiologists to report each component of the LSR was 15 min for mDixon + DWI, and an additional 6.5 min for T2W and 4 min for post-contrast scans.

No suspicious lesion (scoring 4, 5, or 6) was identified below the mid-thigh level on any imaging modality. Two cases had suspicious lesions in the cervical and thoracic spine; otherwise, no disease was identified above the diaphragm. The review panel also found that all sites of positive disease on both BS, ¹⁸F-choline PET/CT, and WB-MRI were anatomically matched.

The distribution of N/M disease for each imaging modality (BS, ¹⁸F-choline PET/CT, and WB-MRI) is presented in Table 2.

Concordance statistics (κ) between WB-MRI readers, between WB-MRI consensus and BS, and between WB-MRI consensus and ¹⁸F-choline PET/CT are presented in Table 3.

ROC-AUC statistics for ‘TNM’-based nodal and metastatic status following each part of the LSR are presented in Table 4 against the follow-up based reference standard.

No significant differences were detected between the mean ROC-AUC for each component of the LSR (p < 0.05), so the simplest WB-MRI combination was chosen for further analysis (DWI + mDixon). Youden’s index confirmed the optimal cutoff of the ROC-AUC was ≥ 4 in all cases. The sensitivity and specificity for BS, ¹⁸F-choline PET/CT, and WB-MRI were therefore calculated using a threshold of ≥ 4 as positive against the follow-up reference standard. Results are displayed in Table 5, along with their numerators and denominators.

Typical examples of lesions missed on BS which are detected by WB-MRI are provided in Figs. 3 and 4.

WB-MRI is gaining momentum as a staging modality in prostate cancer, but requires further validation prior to being introduced into clinical practice. Firstly, our results show that WB-MRI detected more positive bony metastatic (M1b) disease than ¹⁸F-choline PET/CT and BS with 8, 6, and 3 positive lesions for WB-MRI, ¹⁸F-choline PET/CT, and BS respectively. We then confirmed that WB-MRI had the highest sensitivity of all modalities for detecting metastatic bone disease: 0.90 vs. 0.80 for ¹⁸F-choline PET/CT and 0.60 for BS for specificities of 0.88, 0.92, and 1.00 respectively. This finding is in accordance with a meta-analysis which compared the diagnostic accuracy of BS, ¹⁸F-choline PET/CT, and WB-MRI and gave pooled sensitivities and specificities of 0.97/0.95, 0.91/0.99, and 0.79/0.82 for WB-MRI, ¹⁸F-choline PET/CT, and BS [4] respectively. High diagnostic sensitivity could reflect the fact that DWI sequences are designed to probe small changes in tissue microstructure, as found in the early cellular phase of a metastasis, before a sclerotic reaction has been effected in bone [18].

High and very similar sensitivities/specificities were also shown for WB-MRI and ¹⁸F-choline PET/CT for nodal disease, with values of 1.00/0.96 and 1.00/0.82 for N1 disease and 0.75/0.93 and 0.75/0.92 for M1a disease respectively. Both of the cross-sectional modalities therefore appear more accurate than conventional CT, which is again in accordance with a meta-analysis which reported pooled sensitivities and specificities of 0.42/0.82 for CT [3], vs. 0.49/0.95 for choline PET/CT [5]. MRI studies which incorporate DWI into their scanning protocols report a heterogenous sensitivity for lymph nodes which ranges from 0.17 [19] to 0.73 [20]. Whilst both of these studies used extended pelvic lymph node dissection as the reference standard, the lower sensitivity reported by Pinaquy and colleagues [19] could relate to their chosen b-values of 0 and 100 s/mm², which contravenes the recommendations of international consensus guidelines [21], and emphasises the need for optimised scanning technique. In concordance with the findings of our study, the specificity of MRI for nodal detection is thought to be high, with a limited number of studies quoting values ranging from 86% [20] to 98% [22].

A potential strength of our study was the use of a vertex-to-feet protocol which enabled direct comparison with BS and could assess potential lesions outside of the field of view for ¹⁸F-choline PET/CT. Whole-body cross-sectional studies regarding disease distribution in the PSA screening era are welcome since strongest data regarding disease distribution is provided by an autopsy study prior to PSA screening era, which did not routinely examine the peripheral skeleton [23]. Complete body coverage has been both suggested [24, 25] and deemed unnecessary [9], which is perhaps could partially be due to the uncertainty regarding disease distribution in the PSA screening era. Since no lesions were detected below the knee or extravertebral lesions above the diaphragm, our data suggests that scanning below the knee may indeed be unnecessary, and a cervical and thoracic spine MRI may be a reasonable compromise for detecting disease above the diaphragm, and is in keeping with the findings of another study [9], which reported all patients with peripheral metastases occurring in high-risk prostate cancer (60 in total) also had vertebral metastases, and no metastases occurred below the knee. With further confirmatory work, scanning the abdomen, pelvis, and femora using pre-contrast mDixon and DWI at 2 b-values paired with a whole spine MRI as a routine staging examination could be applied and would have approximately 700 images, vs. 12,000 images per patient in the present study. Reducing the number of images may further improve the interobserver concordance by reducing the complexity of imaging datasets. We found interobserver concordance to be ‘substantial’ for N1 and M1a disease (κ = 0.79 and 0.68 respectively), and ‘moderate’ for M1b (0.58), whereby the lower concordance in bone metastases could be explained by the non-specific features of bone lesions on MRI, and the fact that acquisitions were tailored for WB cancer staging rather than bone lesion characterisation. Furthermore, the more subjective criteria applied for assessing bone lesions vs. nodal size measurements may have given rise to further heterogeneity in the data and thus lower levels of concordance.

The LSR paradigm allowed the incremental value of additional sequences to be assessed, whereby adding T2W and post-contrast mDixon sequences did not improve ROC-AUC significantly. These results could be used to streamline WB-MRI scanning protocols in research and clinical practice. For example, performing pre-contrast mDixon + DWI alone could save 10-min reporting time and 20-min scan time and avoid the need for cannulation and gadolinium administration. Furthermore, as suggested by the MET-RADS-P consensus guidelines [25], the use of 2 b-values rather than 4 could be sufficient—especially for primary staging purposes, which would reduce scan time by a further 25 min. Whilst the MET-RADS-P guidelines were based on expert opinion, WB-Dixon and DWI were recommended in combination with whole spine T1 and short tau inversion recovery (STIR), meaning our findings provide evidence to support a similar simple scanning protocol when characterising oncological burden in prostate cancer.

Another potential strength of our study was the choice of a reference standard based on follow-up MRI rather than on best value comparator (BVC) alternatives [8] which rely upon imaging tests such as BS and plain radiographs with limited performance characteristics. Whilst TP was assigned without follow-up imaging when BS and MRI were concordant due to the high specificity of BS in the context of prostate cancer, we did not assign TN without MRI follow-up, since genuine lack of sensitivity, i.e. FN results on both modalities, is also possible.

The limitations of this study include patient number, its single-centre nature, and a relatively low number of positive cases. Furthermore, not all patients underwent ¹⁸F-choline PET/CT, meaning patients with a negative BS in whom the suspicion for metastatic disease remained high may have been more likely to be selected for ¹⁸F-choline PET/CT vs. patients with clear evidence of metastases on BS, leading to lower apparent levels of diagnostic accuracy for ¹⁸F-choline PET/CT.

In addition, not all patients underwent WB-MRI follow-up at 1 year, which could lead to selection bias, e.g. patients who were feeling well or unwell at the time of follow-up may be more likely to refuse the second scan. However, the most common reason provided at telephone consultation was that they had incurred too many imaging tests and therefore declined further participation. Whilst incorporation bias likely gave rise to the high values of sensitivity and specificity (e.g. vs. pelvic lymph node dissection (PLND) as a nodal reference standard), it would not have been practical or ethically acceptable to perform nodal dissection for the purposes of the study, and selecting patients who are undergoing PLND would incur spectrum bias.

Further work could include performing a lesion-based analysis in the same cohort of patients, whereby the number of lesions detected and their anatomical sites could be established. Further validation of WB-MRI could also be carried out in multicentre trials, where economic and clinical utility could also be considered. The findings of the present study are not limited to WB-MRI and can be used to inform rational PET-MRI protocols, e.g. in combination with prostate-specific membrane antigen (PSMA) PET tracer.

WB-MRI provides high levels of diagnostic accuracy for both nodal and metastatic bone disease, with higher levels of sensitivity than BS for metastatic disease, and similar performance to ¹⁸F-choline PET/CT. T2 and post-contrast mDixon had no significant additive value above a protocol comprising mDixon and DWI alone.