Defining the incremental value of 3D T2-weighted imaging in the assessment of prostate cancer extracapsular extension

Prostate cancer (PCa) is the commonest malignancy in men and the second leading cause of male cancer-related mortality [1]. Accurate local staging is essential for treatment planning and prognosis in patients where the aim is for curative treatment [2]. The principal aim is to differentiate between organ-confined disease and extracapsular extension (ECE) which is associated with higher rate of positive surgical margins, biochemical recurrence and micro-metastatic disease [3, 4]. The presence of ECE by itself does not preclude radical treatment; however, preoperative identification is important to allow for wider surgical margins on the affected side [5], or for more aggressive management to be delivered to patients undergoing external beam radiotherapy [6].

Magnetic resonance imaging (MRI) has been shown to have higher accuracy for prediction of ECE than clinical nomograms, and staging with digital rectal examination (DRE), PSA and biopsy Gleason score [7, 8, 9, 10]. However, the true sensitivity and specificity are unknown as gross ECE on MRI often precludes surgery, denying the opportunity for gold standard histopathological confirmation. A recent meta-analysis has shown only moderate overall performance using a standard subjective MRI interpretation, mainly due to a limited sensitivity [11]. Several studies have therefore evaluated different approaches to improve sensitivity, such as use of endorectal coils (ERC) [12, 13], risk-probability scoring using ESUR criteria [14], adding functional MRI sequences [15], or measuring tumour contact with the capsule [16, 17, 18]. Capsular contact length appears to be a promising predictive marker, but the reported threshold value varies significantly between studies from 6 to 20 mm [16, 17, 18], whilst PI-RADS 2.0 guideline suggest a cutoff of 10 mm for assessing ECE [19].

3D fast spin echo (FSE) with variable flip angle (“CUBE”) MR imaging allows for reconstruction in multiple planes using voxels acquired isotropically with no gap. The partial volume effect is thereby minimised, resulting in improved spatial resolution and potentially better diagnostic performance as previously shown in the liver, uterus and knee [20, 21, 22]. Measurement in 3D planes may thus allow for more accurate delineation of prostate capsular contact, and reformatting may help assess for ECE, analogous to the axial oblique imaging acquired for rectal and cervical cancer MRI staging [23, 24]. Rosenkrantz et al [25] and Itatani et al [26] have both compared the diagnostic performance of isotropic 3D T2 sequences. Whilst the first group found no significant difference between the 2D and 3D sequences in diagnosing T3a disease at 3 T, the latter reported improved sensitivity of 3D T2 at 1.5-T imaging [25, 26]. In addition, good performance of 3D imaging in staging of prostate cancer has been described by Cornud et al [27] and Matsuoka et al [28]; however, neither study directly compared the results to conventional 2D T2WI.

The aims of this study were therefore to provide further evidence on the performance of 3-T MRI for local prostate cancer staging and to assess the added value of 3D T2W imaging.

This single-institution retrospective study was approved by the local ethics committee (HBREC.2016.08) with the need for obtaining informed consent waived. Seventy-five consecutive patients who underwent preoperative MRI and subsequent radical prostatectomy for prostate cancer between September 2014 and January 2017 were included. Inclusion criteria included consecutive patients undergoing 3-T MRI at our institution prior to prostatectomy. Pre-determined exclusion criteria were applied for total hip replacement (n = 2) and any previous treatment for prostate cancer (n = 0).

Reader’s assessments of ECE were compared with the pathologic reference on a lesion-per-lesion basis. Standard performance statistics of sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated for ECE, based on a Likert score ≥ 3 being a positive result. McNemar’s test was used to compare sensitivities and specificities and Kosinski test to compare PPVs and NPVs. Receiver operating characteristic (ROC) analysis was then used to assess the diagnostic utility associated with the measured length of capsular contact from 2D and 3D for the detection of ECE. In particular, the ROC analysis was performed to identify the optimal threshold for the length and surface of different capsular contact types. The Youden index was used to define a single optimal threshold maximising the average of sensitivity and specificity. The DeLong test was used to compare two AUC curves. Likert scores for ECE-positive and ECE-negative lesions between the 2D and 3D reads were compared with exact the Wilcoxon-Pratt signed-rank test. For the purposes of analysing contact thresholds separately, we considered a low-grade group (LGG) where the Gleason score was ≤ 3 + 4 (grade groups 1–2) and a high-grade group (HGG) with Gleason score ≥ 4 + 3 (grade groups 3–5) [31]. Prevalence of ECE between the two groups was assessed with Fischer’s exact test. Intraclass correlation coefficient (ICC) was computed using a two-way mixed effects model to separately test reliability of 3D contact and 3D surface measurements. ICC values were defined as indicating poor (< 0.5), moderate (0.5–0.75), good (0.75–0.90) and excellent reliability (> 0.90) [32]. In addition, simple kappa (k) coefficients were used to separately assess the agreement between the 3D contact and 3D surface measurements at the derived optimal ROC thresholds, where 0–0.20 = slight, 0.21–0.40 fair, 0.41–0.60 moderate, 0.61–0.80 substantial and 0.81–1 almost perfect agreement [33].

A value of p < 0.05 was considered statistically significant. Statistical analyses were performed using IBM SPSS software, version 24.0.

The overall results of this study demonstrate the added value of 3D T2 imaging in evaluation of ECE. Our suggested quantitative parameters of 3D contact and 3D surface were both superior to previously reported tumour contact length measured in axial plane on 2D T2. 3D reads also increased sensitivity whilst maintaining high specificity by means of subjective scoring, but this did not reach significance. However, 3D reads significantly increased the mean Likert score in cases positive for ECE and showed a trend towards lower mean score in negative cases, reflecting increased reader confidence, which may also be important from a reporting standpoint. Finally, the optimal threshold for capsular contact predicting ECE differed between low- and high-grade tumours, which has not been reported previously and is potential of key clinical importance.

Our results for capsular contact are in agreement with the results of previous work which showed a strong association between the MRI-determined contact length and ECE. However, previous studies have measured contact only in the axial plane on conventional T2 axial images [16, 17, 18]. Our results showed that contact length measured with 3D T2 significantly improved sensitivity from 59.6 to 73.7% whilst maintaining a high specificity at 87.8%. This confirmed our hypothesis that 3D would allow us to measure the true maximal length of capsular contact, as this is not necessarily the longest in the axial plane. The sensitivity for 3D-measured surface area contact was even higher at 82.5% (2D = 59.6%), but with a lower specificity of 71.4% (2D = 87.8%). 3D contact and 3D surface showed excellent reliability (ICC > 0.95) and substantial to almost perfect agreement (k = 0.79 and 0.92, respectively) which is concordant with the previous work reporting substantial agreement for linear contact measurements on axial T2 imaging (k = 0.70) [18]. 3D reformats take a few seconds to load and finding the longest contact/line generally takes under a minute. 3D parameters may only be needed as a problem solver in borderline cases, meaning the overall impact on workflow would be negligible, especially in the light of significantly increased reader confidence. Although 3D-measured surface requires some additional time due to calculations, it may prove particularly useful when high-sensitivity readings are required.

Regarding contact length thresholds, our numbers of 10.5 mm for 3D contact and 13.5 mm for axial 2D contact are between the axial 2D thresholds previously reported by Rosenkrantz et al (6 mm) and Baco et al (20 mm) [17, 18] but similar to the results of Woo et al (14 mm) [16]. The reasons may relate to study design; for instance, the Baco et al [17] study was performed at 1.5 T and Rosenkrantz et al [18] made radio-pathological correlations on a side-specific basis whereas we made these on a lesion-specific basis. Moreover, the latter created a single straight line between the two outer margins of the lesion’s capsular contact rather than total (curved) contact, which likely underestimated contact and may explain their lower threshold. Interestingly, we noticed a significant difference between contact lengths and thresholds in the axial and the vertical planes (coronal or sagittal) at 13.5 mm versus 5.5 mm, which suggests that tumours tend to spread along the capsule more in the transverse than vertical plane, in accordance with earlier work by McNeal et al [34]. Case mix may further explain the variations in literature-reported capsular contact required for ECE.

Our finding that the optimal capsular contact threshold differs between high-grade (grade group 3–5) and low-grade (grade group 1–2) tumours is clinically important. HGG tumours overall showed significantly more ECE than LGG (88% vs 44%). Although HGG tumours had a longer mean capsular contact than LGG (15.3 mm vs 9.0 mm), the finding appears to be independent of tumour size, with a marked difference also observed when corrected for a contact length of ≤ 10 mm (41.6% biopsy-confirmed HGG index lesions showing established ECE vs. 3% LGG). Consequently, in HGG tumours, a PPV of 90.9% was reached at ≥ 5 mm, whereas a PPV of 90.4% was only achieved at a cutoff at ≥ 12.5 mm in LGG tumours. Although not previously reported in MRI studies, the finding that aggressive cancers require less capsular contact for extracapsular spread is perhaps expected: Partin nomograms incorporate biopsy Gleason score as a predictor for ECE at radical prostatectomy [35]. Similar conclusions can also be deduced from the recent study of Matsuoka et al where 73.3–74.1% of patients with a biopsy Gleason score of ≤ 7 had over-staged tumours in comparison with 35.7–44.4% of those with a biopsy Gleason score of ≥ 8 when a PI-RADS 2.0 cutoff at 10 mm was used as staging criteria [19]. This is an important consideration in the context of performing MRI pre-biopsy, when pathological information is yet to be obtained; thus, it may be appropriate to re-evaluate for the likelihood of ECE when biopsy results become available.

In terms of subjective scoring, two previous studies have compared 2D T2W to 3D T2W for predicting ECE. The results of the study by Itatani et al, who reconstructed using multiple planes at 1.5 T (slice thickness/gap 4/0 mm), are in concordance with ours [26]. Conversely, Rosenkrantz et al reported only a non-significant trend to increased sensitivity with the 3D method [25]. However, reformats were only performed in the three standard planes and using a slice thickness of 3 mm to match 2D T2, possibly underutilising the benefits of 3D imaging. We used the advantage of 3D imaging and reconstructed a plane perpendicular to the tumour-capsule interface similarly as in evaluating rectal or cervical cancer, where an oblique axial plane is recommended in order to avoid partial volume–related staging inaccuracy [23, 24] (Fig. 4). Additionally, the axial plane cube reformats afforded thinner axial slice (1 mm vs 3 mm) which resulted in less partial voluming and for more accurate delineation of capsular contact (Figs. 5 and 6).

Open image in new window

Open image in new window

Open image in new window

Our study has a number of limitations, including its single-centre and retrospective design. Our cohort is limited by its relatively small numbers, particularly for the sub-analysis of HGG versus LGG at a capsular contact of ≤ 10 mm. Additionally, the reader may have been biased by knowledge of the existing literature on capsular contact [16, 17, 18, 19], which may explain the limited additional benefit of measuring capsular contact over subjective scoring alone, with the latter being biased by the former. However, of note, previous work by Outwater et al [36] and Yu et al [37] did not show improvement when adding contact length to morphological criteria. In addition, readers in the earlier reported studies [16, 17, 18] are also likely to have been aware of the PI-RADS version 1 guidelines, which suggested broad capsular contact (> 15 mm) as a criterion for lesion scoring [38]. We did not compare signal-to-noise ratio (SNR) and overall image quality between the 2D and 3D imaging as this was not the principal aim of our work. However, we did demonstrate added value of 3D imaging in evaluating ECE, which together with other studies [26, 27, 28, 39] form a basis for consideration of using abbreviated MRI prostate protocols. Moreover, there is emerging data that a single acquisition of 3D T2W rather than time-consuming conventional 2D T2 in three dedicated planes may produce results of equal diagnostic quality [25, 40].

In conclusion, our study showed significantly increased sensitivity and confidence in calling ECE with 3D T2 over conventional 2D T2. The capsular contact length threshold differed between low- and high-grade cancers, suggesting that MRI restaging should be considered in light of the histological information when MRI is performed pre-biopsy.