Non-invasive estimation of prostate cancer aggressiveness using diffusion-weighted MRI and 3D proton MR spectroscopy at 3.0 T

Abstract

Background

Clinical management of prostate cancer increasingly aims to distinguish aggressive types that require immediate and radical treatment from indolent tumors that are candidates for watchful waiting. This requires reliable and reproducible parameters to effectively control potential cancer progression. Magnetic resonance imaging (MRI) may provide a non-invasive means for this purpose.

Purpose

To assess the value of diffusion-weighted imaging and proton MR spectroscopy for the prediction of prostate cancer (PCa) aggressiveness.

Material and Methods

In 39 of 64 consecutive patients who underwent endorectal 3-T MRI prior to radical prostatectomy, prostate specimens were analyzed as whole-mount step sections. Apparent diffusion coefficient (ADC), normalized ADC (nADC: tumor/healthy tissue), choline/citrate (CC), and (choline + creatine)/citrate (CCC) ratios were correlated with Gleason scores (GS) from histopathological results. The power to discriminate low (GS ≤ 6) from higher-risk (GS ≥ 7) tumors was assessed with receiver operating characteristics (area under the curve [AUC]). Resulting threshold values were used by a blinded reader to distinguish between aggressive and indolent tumors.

Results

Ninety lesions (1 × GS = 5, 41 × GS = 6, 36 × GS = 7, 12 × GS = 8) were considered. nADC (AUC = 0.90) showed a higher discriminatory power than ADC (AUC = 0.79). AUC for CC and CCC were 0.73 and 0.82, respectively. Using either nADC < 0.46 or CCC > 1.3, as well as both criteria for aggressive PCa, the reader correctly identified aggressive and indolent tumors in 31 (79%), 28 (72%), and 33 of 39 patients (85%), respectively. Predictions of tumor aggressiveness from TRUS-guided biopsies were correct in 27 of 36 patients (75%).

Conclusion

The combination of a highly sensitive normalized ADC with a highly specific CCC was found to be well suited to prospectively estimate PCa aggressiveness with a similar diagnostic accuracy as biopsy results.

Keywords

Magnetic resonance imaging (MRI)MR diffusion/perfusion MR spectroscopy prostate staging

Introduction

Prostate cancer (PCa) is the second most common cause of cancer-related death in Europe (1,2) and in the US (3). Disease progression, however, can be slow and tumors may not become symptomatic or metastatic for many years after initial diagnosis such that the patient may even die with prostate cancer but not from it (4). The adoption of therapeutic strategies like watchful waiting or active surveillance (AS) heavily depends on reliable parameters to properly detect and monitor progressing cancer (5). Parameters like the prostate specific antigen (PSA) doubling time or PSA density are easy to assess but rather unspecific (6). To date, the best prognostic factor for the biological malignancy of prostate cancer is the histological grade determined by the Gleason scoring (GS) system (7). Gleason himself has originally reported that prostate cancers show relatively fixed degrees of malignancy and growth rates rather than a steady increase in malignancy with time (7). The large variation in biological cancer aggressiveness seems to be closely linked to histological tumor appearance. Histological grading, however, requires the extraction of tissue, typically sampled by core needle biopsies under transrectal ultrasound (TRUS) guidance. An invasive biopsy, however, may not be the best option for monitoring cancer aggressiveness every 1 to 2 years because of the side-effects and limited patient compliance. Non-invasive methods before treatment decision, and particularly during follow-up of patients under AS, may therefore be a promising alternative.

Recent studies have aimed to determine the value of magnetic resonance imaging (MRI) correlates of cellular density (8,9), metabolite concentration (10), and tumor vascularization (11) for the prediction of tissue reorganization and cancer aggressiveness.

While the value of dynamic contrast-enhanced imaging appears to be limited in that respect (11), both diffusion-weighted imaging (DWI) as well as MR spectroscopy (MRS) have shown good correlation with GS (9,10). The aim of this work was therefore to determine to what extent DWI, MRS, and their combination may predict PCa aggressiveness.

Material and Methods

Patients and MRI

This study was approved by the institutional review board and written informed consent was obtained from all patients. Our retrospective analysis included all patients who underwent both 3-T MR examination of the prostate between May 2011 and May 2012 as well as radical prostatectomy with subsequent whole-mount step-section analysis of the prostate specimens.

MRI was performed in a 3-T system (Magnetom Trio, Siemens Healthcare, Erlangen, Germany) using the combination of a pelvic phased-array coil with an endorectal coil (eCoil, Medrad, Pittsburg, PA, USA) for signal acquisition. The endorectal coil was filled with 30–40 mL of perfluorocarbon solution (Perfluorooctyl bromide, ABCR GmbH, Karlsruhe, Germany). Bowel peristalsis was suppressed by intravenous injection of 1 mg glucagon (Glucagen, NOVO Nordisk Pharma AG, Gentofte, Denmark) before examination.

Fast T1-weighted localizer images were used to verify correct placement of the endorectal coil and to adjust the axial slices to be perpendicular to the dorsolateral prostate wall. Morphological images of the gland and seminal vesicles were obtained with a T2-weighted (T2W) fast spin-echo sequence (in-plane resolution [IPR], 0.57 × 0.57 mm²; repetition and echo time [TR/TE], 4400–4600/126 ms; slice thickness [ST], 3 mm; sections, 19–22; field of view [FOV], 110 × 110 mm²; flip angle [FA], 120–135°) in transverse, coronal, and sagittal planes. DWI relied on a transverse, single-shot echo planar imaging sequence (IPR, 1.02 × 1.02 mm²; TR/TE, 3000/85 ms; ST, 3 mm; sections, 19–22; FOV, 250 × 250 mm²) with b-values of 50, 500, 800, and 1500 s/mm². Maps of the ADC parameter were calculated from the raw data of the first three b-values as described elsewhere (12).

Three-dimensional ¹H chemical shift imaging (CSI) was performed by two spectroscopists (HB and GT, with 7 and 4 years of experience, respectively) according to a previously reported protocol (point-resolved spectroscopy [PRESS], TR/TE, 750/145 ms) (13). CSI data acquisition of the entire prostate (8 slices, nominal voxel size 6 × 6 × 6 mm³) took 12 min. Postprocessing included baseline and phase corrections, metabolite peak areas were then determined after automated Gaussian curve fitting (Syngo; Siemens Healthcare, Erlangen, Germany) to the resulting spectra.

Histological work-up

Immediately after surgical resection, prostates were submitted to the Pathology Department. Specimens were fixed in 10% neutral-buffered formalin for about 1 week and seminal vesicles were separated. The prostate was cut into transverse, 4–5-mm thick step sections perpendicular to the dorsorectal surface of the gland yielding about 10 slices per specimen. After paraffin embedding and staining, all slices were evaluated by a senior pathologist (L-CH) with 14 years of experience in urogenital pathology who was blinded to the MRI findings. Tumor foci were outlined on the microscope slides and primary through tertiary lesions were labeled with the corresponding Gleason grade. Following the seventh edition of the AJCC cancer staging manual (14), qualitative grade groups G1–G3 were reported indicating low (GS ≤ 6), intermediate (GS = 7) and high-risk cancers (GS 8), respectively. All processed slides were also digitized.

Correlation with MRI and data analysis

Histological step sections were visually matched to an optimum T2W image for correlation between pathological and multiparametric MRI findings (Fig. 1). Apparent diffusion coefficient (ADC), normalized ADC (nADC: tumor ADC/ADC in mirror region of healthy tissue), choline/citrate (CC), and (choline + creatine)/citrate (CCC) ratios were determined for all lesions suspicious on DWI. Lesions with diameters <6 mm were excluded. MR spectra of individual voxels were rejected in the case of poor signal-to-noise ratios, presence of lipid signals, or baseline distortions.

Fig. 1.

Example of a 68-year-old patient with preoperative PSA level of 4.8 ng/mL. (a) The lesion outlined in the whole-mount step section was classified a GS 4 + 3 tumor. Considering the morphology of central gland, peripheral zone, apex, and base of the prostate as well as landmarks like cysts, calcifications, and urethra, the most similar T2W image was selected (b). If the ADC image showed a corresponding signal loss (c), a freehand circular region of interest (ROI) was drawn within the outline of the respective lesion, avoiding tumor edges, healthy tissue, prostate capsule, and urethra. A mirror ROI was drawn on the contralateral side as healthy tissue reference. Care was taken to avoid any prostate cancer on the corresponding whole-mount step section. The ADC value was normalized by dividing the average ADC of the tumor region by that in the reference tissue. In the next step, both CC and CCC ratios were determined for the CSI voxel that best matched the relative coordinates of the lesion (d).

Statistical analysis

Statistical analyses were performed with SPSS 18 (SPSS Inc., Chicago, IL, USA) and a significance level of 5%. Areas under the curve (AUC) of the receiver operating characteristic (ROC) were calculated to assess the power of tumor ADC, nADC, CC ratio, and CCC ratio in discriminating between low and combined intermediate and high-grade tumors. The two parameters with the highest discriminatory power were chosen for further analysis using optimized threshold values.

Prospective estimation of tumor aggressiveness

One physician (JO, with 3 years of experience in MRI diagnostics of prostate cancer, 160 endorectal MRI cases read at 3.0 T) prospectively analyzed all T2W, DW, and spectroscopic image data. Based on the previously defined thresholds (see above), the reader was asked to allocate each patient to one of two qualitative grade options, either low risk or combined intermediate and high risk. The reader was blinded to all clinical and paraclinical information.

Results

Out of 64 patients examined with endorectal 3.0-T MRI during the above period at our institution, 39 met the inclusion criteria. The results of previous TRUS-guided biopsies were available for 36 of these patients. Patient age was in the range of 48–74 years (average, 65 years). The time between diagnostic MRI and radical prostatectomy was <3 days in 36 patients. In three patients, this time amounted to 42, 66, and 152 days because PCa was diagnosed after later MRI-guided biopsies. Median PSA level and average GSs were 7.3 ng/mL (range, 1.2–56.0 ng/mL) and 6.6 (range, 5–8). A detailed summary of patients and results is given in Table 1.

Table 1.

Patient characteristics.

Parameter	Value
Patients (n)	39
Clinical characteristics
PSA value (ng/mL) (median, range)	7.3, 1.2–56.0
Age (years) (average, range)	65, 48–74
Pathological characteristics
Stage T2a	4
Stage T2b	2
Stage T2c	27
Stage T3a	4
Stage T3b	2
Included lesions	97
Gleason grade 3 + 2	1
Gleason grade 3 + 3	41
Gleason grade 3 + 4	26
Gleason grade 4 + 3	10
Gleason grade 4 + 4	12

PSA, prostate specific antigen.

Based on DWI, a total of 90 malignant lesions in 39 patients were included for quantitative analysis. Forty-one foci corresponded to low-risk (G1), 36 to intermediate-risk (G2), and 12 to high-risk cancer (G3). MR spectra of sufficient quality were available for 68 foci only. Fig. 2 shows the relation between ADC, nADC, CC ratio, and CCC ratio and the corresponding grade group per lesion. The nADC demonstrated the best correlation (Pearson ρ = –0.68, P < 0.001) with dedifferentiation of tumor cells. Inspection of the ROC curves (Fig. 3) shows that the highest power to distinguish low from combined intermediate and high-grade cancers was observed for nADC (0.90 ± 0.03), followed by CCC ratio (0.82 ± 0.06), absolute ADC (0.79 ± 0.05), and CC ratio (0.73 ± 0.07).

Fig. 2.

Box plot diagrams showing (a) tumor ADC, (b) normalized ADC, (c) choline/citrate (CC) ratio, and (d) (choline + creatine)/citrate (CCC) ratio as a function of qualitative grade groups. The normalized ADC shows the best correlation with cancer aggressiveness (Pearson ρ = −0.68, P < 0.001).

Fig. 3.

ROC curves of absolute and normalized ADC values (a) as well as CC and CCC ratios (b) for differentiation between low and combined intermediate and high-grade lesions.

Based on these results, nADC values below 0.46 and CCC ratios above 1.3 were defined as thresholds for the presence of tumors with intermediate or high aggressiveness. Using the nADC criterion only, the reader correctly identified aggressive and indolent tumors in 31 of 39 patients (79%) (Fig. 4). The respective number for the CCC ratio was 28 of 39 patients (72%). By combining both criteria, 33 of 39 patients (85%) were correctly assigned. In comparison, TRUS-guided biopsies predicted the postoperative grade in 27 of 36 patients only (75%).

Fig. 4.

Prospective PCa evaluation by a blinded reader using (a) T2W images (left), ADC parameter maps (right), and MR spectroscopic data (b) in a 64-year-old patient with a preoperative PSA level of 4.8 ng/mL. The most suspicious region had an average nADC of 0.51 and a CCC ratio of 0.75, which were above and below the given thresholds for aggressive tumor growth, respectively. The patient was therefore assigned to the low grade group. (c) The pathologist had assigned a Gleason score of 3 + 3 to this prostate region.

Discussion

Clinical decision-making often relies on the simple detection or exclusion of low-grade carcinoma rather than the prediction of the exact GS. ROC analysis revealed that the nADC value had the highest discriminatory power to distinguish low from combined intermediate and high-grade cancers. As previously discussed (12,15), the concept of ADC normalization may compensate for equipment-related variations and improve discriminatory performance over absolute ADC values, which is also in line with the results presented here.

Significant differences in the absolute ADC values between low-, intermediate-, and high-grade tumors have also been observed in other 3-T MRI studies (8,9,15,16) (Table 2). In comparison, the median ADC values determined here were generally lower than previous ones (8,9,15). This is most likely due to the omission of DWI data acquired at b = 0 s/mm², which will reduce adverse perfusion effects and result in lower absolute ADC values (17). In an earlier comparison, for example, the relative ADC difference between acquisitions with b = [50, 800] and [0, 50, 400, 800] s/mm² was in the range of 20–25% (12).

Table 2.

Relationship between tumor ADC and Gleason grade groups.

Study	Patients	B₀ (T)	b-values (s/mm²)	Median ADC (×10⁻³ mm²/s)
Study	Patients	B₀ (T)	b-values (s/mm²)	Low grade	Intermediate grade	High grade
(9)	51	3.0	0, 50, 500, 800	1.30 ± 0.30	1.07 ± 0.30	0.94 ± 0.30
(8)	48	3.0	0, 188, 375, 563, 750	1.23 ± 0.32	1.08 ± 0.23	0.82 ± 0.14
(15)	51	3.0	0, 700 or 0, 1000	1.21	0.99*	0.69
(16)	54	3.0	0, 50, 500, 800	1.13 (0.95)^†	1.04 (0.51)^†	0.76 (0.61)^†
Current study	46	3.0	50, 500, 800	0.95 ± 0.23	0.77 ± 0.14	0.69 ± 0.17

Average of GS 3 + 4 and 4 + 3 subgroups.

†

Minimum 25th percentile of the ADC values for peripheral (transitional) zone tumors.

ADC, apparent diffusion coefficient.

Previous studies have determined mean ADC values for cancer in transition and peripheral zones individually, however, with inconclusive results (16,18,19). Separate zones were therefore not considered here, also because of the reduced statistical power in both subgroups. In addition, the concept of ADC normalization with values from the contralateral side might partially account for zonal differences.

MRS results showed a positive correlation with postoperative GS group. More aggressive, high-GS tumors are generally associated with increased cellularity and rapid proliferation and will therefore exhibit higher choline levels. At the same time, progressive loss of healthy glandular function will result in decreased citrate concentrations (20). In practice, there is still a considerable overlap of the individual values between different grade groups.

Our median CCC ratios for low-, intermediate-, and high-grade cancers are in good agreement with independent 3-T values reported by two studies by Kobus et al. (10,16). As expected, two earlier assessments at 1.5 T (20,21) yielded systematically higher CCC ratios (Table 3).

Table 3.

Relationship between tumor CCC and Gleason grade groups.

Study	Patients	B₀ [T]	Median CCC ratio
Study	Patients	B₀ [T]	Low grade	Intermediate grade	High grade
(17)	38	1.5	3.5*	3.8*	6.4*
(21)	94	1.5	1.5 ± 3.5*	2.4 ± 2.0*	4.1 ± 5.0*
(10)	43	3.0	0.57	0.55	1.01
(16)	54	3.0	0.44 (0.58)^†	0.52 (1.58)^†	0.89 (1.62)^†
Current study	46	3.0	0.76 ± 0.81	1.34 ± 0.54	1.42 ± 0.62

Data were estimated visually from the plots provided in the respective studies.

†

Median of the maximum CCC ratio for peripheral (transitional) zone tumors.

CCC, (Choline + creatine)/citrate.

Low-grade cancer prediction by 3-T MRS was previously found to yield AUCs of 0.74 (CC) and 0.70 (CCC) (10). For the subgroup of transition-zone cancers, AUCs of 0.90 (CC) and 0.82 (CCC) were reported (16). In the present work, the AUCs of CC and CCC metabolite ratios were very similar (0.73 and 0.82, respectively). With generally limited sample sizes and inconsistent results for different metabolite ratios, a clear preference for a specific ratio (CCC or CC) remains difficult unless more reliable data of prospective studies becomes available.

The key aspect of the present study was to test the predictive power of predefined thresholds between indolent and aggressive tumors in a prospective setting. For that purpose, we have chosen those two parameters (nADC and CCC ratio) that showed the highest AUC values. The underlying CSI threshold (CCC ratio of 1.3) was deliberately set somewhat higher than that in a previous study (22) to reduce the number of false positives. The highest prediction rate (85%) was obtained when both criteria were used. When using the normalized ADC criterion only, 79% of all patients were allocated to the correct qualitative grade group. In comparison, TRUS-guided biopsy correctly predicted indolent and aggressive tumors in 75% of all cases.

It should be noted that MRS is generally time consuming and requires special skills in data acquisition and analysis. The quality of the underlying MR spectra was sufficient for only 68 of all 90 lesions (75%) considered, for example. This amount agrees well with the previously reported figure of 77% (10). Although here, MRS information was found to slightly improve the predictive power over nADC thresholds alone, one needs to balance this incremental value against the extra time and efforts required for MRS examination.

While the clinical application of MRS is still discussed controversially, DWI has become a widely used part of prostate MRI diagnostics (23). On the other hand, the lack of standardized imaging protocols with corresponding b-values makes it difficult to define an appropriate ADC threshold for tumor aggressiveness. Further work should also explore the extent to which normalization of ADC values may contribute to the robustness of such parameters.

Our study had several limitations. It was carried out at a single institution only and the number of patients was relatively low. The spatial mapping of findings between whole-mount step sections and MR images is generally prone to errors because of inherent difficulties in histopathological sample preparation and variability introduced by subjective visual inspection. As mentioned above, peripheral-zone and central-gland cancers were not analyzed separately to avoid a further reduction of the number of cases in the subgroups. Another limitation is the relative absence of high Gleason scores here, which is attributed to the selection of prostatectomy patients, a common population of other prostate MRI studies as well.

In conclusion, both DWI and MRS suggest great potential for a non-invasive assessment of PCa aggressiveness revealing a discriminatory power – in a prospective setting – that is comparable to that of TRUS-guided biopsies. While additional MRS information slightly increased the predictive power, normalized ADC values alone were still sufficiently accurate and are generally much easier to assess in clinical routine. DWI should therefore be further considered as a prognostic tool in patients undergoing active surveillance.

Footnotes

Funding

Grant support from the German Federal Ministry of Education and Research under BMBF 13N10360 is gratefully acknowledged.

References

Haberland

Bertz

Wolf

. German cancer statistics 2004. BMC Cancer 2010; 10: 52–52.

Dupré

Niederlaender

Jougla

. Mortality in the EU 1997–1999: Statistics in focus. Eurostat 2004; 3: 1–7.

Jemal

Siegel

. Cancer statistics, 2010. CA Cancer J Clin 2010; 60: 277–300.

Bratt

. Watching the face of Janus – active surveillance as a strategy to reduce overtreatment for localised prostate cancer. Eur Urol 2006; 50: 410–412.

Klotz

Zhang

Lam

. Clinical results of long-term follow-up of a large, active surveillance cohort with localized prostate cancer. J Clin Oncol 2010; 28: 126–131.

Oesterling

. Prostate specific antigen: a critical assessment of the most useful tumor marker for adenocarcinoma of the prostate. J Urol 2006; 145: 907–923.

Gleason

. Histologic grading of prostate cancer: a perspective. Hum Pathol 1992; 23: 273–279.

Turkbey

Shah

Pang

. Is apparent diffusion coefficient associated with clinical risk scores for prostate cancers that are visible on 3-T MR images? Radiology 2011; 258: 488–495.

Hambrock

Somford

Huisman

. Relationship between apparent diffusion coefficients at 3.0-T MR imaging and gleason grade in peripheral zone prostate cancer. Radiology 2011; 259: 453–461.

10.

Kobus

Hambrock

Hulsbergen-van de Kaa

. In vivo assessment of prostate cancer aggressiveness using magnetic resonance spectroscopic imaging at 3 T with an endorectal coil. Eur Urol 2011; 60: 1074–1080.

11.

Franiel

Lüdemann

Taupitz

. [Pharmacokinetic MRI of the prostate: parameters for differentiating low-grade and high-grade prostate cancer]. Rofo 2009; 181: 536–542.

12.

Thörmer

Otto

Reiss-Zimmermann

. Diagnostic value of ADC in patients with prostate cancer: influence of the choice of b values. Eur Radiol 2012; 22: 1820–1828.

13.

Verma

Rajesh

Fütterer

. Prostate MRI and 3D MR spectroscopy: how we do it. Am J Roentgenol 2010; 194: 1414–1426.

14.

Edge

Compton

. The American Joint Committee on Cancer: the 7th edition of the AJCC cancer staging manual and the future of TNM. Ann Surg Oncol 2010; 17: 1471–1474.

15.

Vargas

Akin

Franiel

. Diffusion-weighted endorectal MR imaging at 3 T for prostate cancer: tumor detection and assessment of aggressiveness. Radiology 2011; 259: 775–784.

16.

Kobus

Vos

Hambrock

. Prostate cancer aggressiveness: in vivo assessment of MR spectroscopy and diffusion-weighted imaging at 3 T. Radiology 2012; 265: 457–467.

17.

Riches

Hawtin

Charles-Edwards

. Diffusion-weighted imaging of the prostate and rectal wall: comparison of biexponential and monoexponential modelled diffusion and associated perfusion coefficients. NMR Biomed 2009; 22: 318–325.

18.

Sato

Naganawa

Nakamura

. Differentiation of noncancerous tissue and cancer lesions by apparent diffusion coefficient values in transition and peripheral zones of the prostate. J Magn Reson Imaging 2005; 21: 258–262.

19.

Tamada

Sone

. Apparent diffusion coefficient values in peripheral and transition zones of the prostate: comparison between normal and malignant prostatic tissues and correlation with histologic grade. J Magn Reson Imaging 2008; 28: 720–726.

20.

Wang

Gao

. 1H-MRSI of prostate cancer: the relationship between metabolite ratio and tumor proliferation. Eur J Radiol 2010; 73: 345–351.

21.

Zakian

Sircar

Hricak

. Correlation of proton MR spectroscopic imaging with gleason score based on step-section pathologic analysis after radical prostatectomy. Radiology 2005; 234: 804–814.

22.

Fütterer

Scheenen

TWJ

Heijmink

SWTPJ

. Standardized threshold approach using three-dimensional proton magnetic resonance spectroscopic imaging in prostate cancer localization of the entire prostate. Invest Radiol 2007; 42: 116–122.

23.

Dickinson

Ahmed

Allen

. Magnetic resonance imaging for the detection, localisation, and characterisation of prostate cancer: recommendations from a European consensus meeting. Eur Urol 2011; 59: 477–494.