Compressed Sensing Radial Sampling MRI of Prostate Perfusion

Abstract

Purpose:

To investigate the diagnostic performance of a dual-parameter approach by combining either volumetric interpolated breath-hold examination (VIBE)- or golden-angle radial sparse parallel (GRASP)–derived dynamic contrast agent–enhanced (DCE) MRI with established diffusion-weighted imaging (DWI) compared with traditional single-parameter evaluations on the basis of DWI alone.

Materials and Methods:

Ninety-four male participants (66 years ± 7 [standard deviation]) were prospectively evaluated at 3.0-T MRI for clinical suspicion of prostate cancer. Included were 101 peripheral zone prostate cancer lesions. Histopathologic confirmation at MRI transrectal US fusion biopsy was matched with normal contralateral prostate parenchyma. MRI was performed with diffusion weighting and DCE by using GRASP (temporal resolution, 2.5 seconds) or VIBE (temporal resolution, 10 seconds). Perfusion (influx forward volume transfer constant [K^trans] and rate constant [K_ep]) and apparent diffusion coefficient (ADC) parameters were determined by tumor volume analysis. Areas under the receiver operating characteristic curve were compared for both sequences.

Results:

Evaluated were 101 prostate cancer lesions (GRASP, 61 lesions; VIBE, 40 lesions). In a combined analysis, diffusion and perfusion parameters ADC with K^trans or K_ep acquired with GRASP had higher diagnostic performance compared with diffusion characteristics alone (area under the curve, 0.97 ± 0.02 [standard error] vs 0.93 ± 0.03; P < .006 and .021, respectively), whereas ADC with perfusion parameters acquired with VIBE had no additional benefit (area under the curve, 0.94 ± 0.03 vs 0.93 ± 0.04; P = .18and .50, respectively, for combination of ADC with K^trans and K_ep).

Conclusion:

If used in a dual-parameter model, incorporating diffusion and perfusion characteristics, the golden-angle radial sparse parallel acquisition technique improves the diagnostic performance of multiparametric MRI examinations of the prostate. This effect could not be observed combining diffusing with perfusion parameters acquired with volumetric interpolated breath-hold examination.

Summary

Dynamic contrast agent–enhanced MRI with compressed sensing, parallel imaging, and golden-angle radial sparse parallel improves the diagnostic performance of multiparametric MRI examinations of the prostate.

Multiparametric MRI examinations of the prostate combine a morphologic assessment by using T2-weighted imaging sequences with functional and dynamic imaging by using either diffusion-weighted imaging (DWI) and/or dynamic contrast agent–enhanced (DCE) MRI, respectively. Whereas DWI has been established in the evaluation process of prostate cancer, DCE MRI has shown less reproducible results (1–3). A challenge for DCE MRI is the inherent trade-off between competing spatial and temporal resolution. Because the prostate is a vascularized organ and a substantial subset of prostate cancer lesions are smaller than 1 cm in cross-sectional diameter, both high temporal and spatial resolution are necessary to discriminate between benign processes and malignant lesions (4).

As a promising solution to address both requirements, DCE MRI with high spatiotemporal resolution by using compressed sensing, parallel imaging, and golden-angle radial sampling (golden-angle radial sparse parallel [GRASP]; StarVibe prototype sequence, fl 3d_star_vibe_991, VE11C; Siemens Healthineers, Erlangen, Germany) has been proposed. The golden-angle radial sampling scheme increases the angle of consecutive spokes by 111.25° and results in approximately uniform k-space coverage throughout the acquisition. Its feasibility and similar clinical performance compared with standard-of-care perfusion sequences for DCE MRI of the prostate was recently demonstrated (5).

Additionally, reliable extraction of quantitative parameters represents an equally important challenge. Because no standard has been established, volumetric and planimetric segmentation methods have been used concurrently and inconsistently. However, the superiority of volumetric quantification of tumor characteristics over single-section approaches has been demonstrated in other tumor entities (6,7).

The aim of our study was therefore to investigate whether a dual-parameter approach by combining either volumetric interpolated breath-hold examination (VIBE)- or GRASP-derived DCE MRI with established DWI yields better diagnostic performance for volumetric assessments of prostate MR examinations than does a traditional single-parameter evaluation on the basis of DWI alone.

Materials and Methods

Participants

Our institutional review board approved this study and all participants gave informed consent. Participants were prospectively enrolled and included into the final study population when the following inclusion criteria were met: clinical indication for a 3.0-T MR examination of the prostate performed at our institution because of a prostate-specific antigen level of 4 ng/mL or more or an increased prostate-specific antigen velocity (>0.5 ng/mL per year) biochemically determined within 30 days before the MRI examination and prostate cancer proven at histopathologic examination, as well as normal contralateral prostate parenchyma with biopsies performed within 30 days of imaging. Participants were eligible for inclusion during a 30-month observation period from January 2015 to June 2017.

GRASP DCE MRI Technique

The GRASP technique (Siemens Healthineers) used in this study is based on continuous data acquisition with a fat-saturated T1-weighted radial three-dimensional gradient-recalled-echo sequence (8). This sequence samples k-space with a stack-of-stars scheme in which radial views or so-called spokes are stacked along the section direction and then continuously rotated, resulting in a cylindrical spoke wheel–like trajectory. For GRASP acquisitions, the angular increment of the rotations is selected according to the golden-angle scheme, which increases the angle of consecutive spokes by 111.25° and results in approximately uniform k-space coverage throughout the acquisition (9). Images with arbitrary temporal resolution can be generated by grouping a respective number of consecutive spokes into frames. If mainly high temporal resolution is desired, severe streaking artifacts because of undersampled data would be encountered; to counteract, GRASP uses an iterative compressed-sensing reconstruction that exploits temporal correlations between successive frames to suppress undersampling artifacts (10). Additionally, GRASP takes advantage of parallel imaging by incorporating knowledge of the coil-sensitivity profiles. In this way, temporal resolutions of up to 1 second per frame can be achieved without compromising spatial resolution (8); for this evaluation, we used 2.5-second temporal resolution based on 21-spoke reconstruction. A detailed technical description of the sequence and imaging properties is provided in Block et al study (11).

MR Examination

MR examinations were performed with 3.0-T imagers (Siemens Skyra; Siemens Healthineers). Sequences for this study included the following: T2-weighted fast-spin-echo sequence, T1 mapping with fat-saturated three-dimensional gradient-recalled-echo precontrast sequence, DWI with consecutive generation of apparent diffusion coefficient (ADC) maps, and T1-weighted DCE MRI with either VIBE or GRASP after administration of 0.01 mmol/kg gadoterate meglumine (Dotarem; Guerbet, Villepint, France) as the underlying acquisition scheme (Table 1). Participants were examined on the same imager infrastructure with the same software version of VIBE from January 2015 until the implementation of GRASP in February 2016 at our institution. Since its implementation, GRASP represents the only acquisition scheme for prostate imaging, and during the observation period we used the same version of GRASP.

Table 1:

Multiparametric Examination Protocol

Parameter	T2-weighted Fast-Spin-Echo Sequence	T1-weighted Fat-saturated 3D GRE Precontrast Sequence	DWI*	DCE MRI†
				VIBE	GRASP
Repetition time/echo time (msec)	7.5/101	4.13/2.06	4.9/59	5/1.8	4/2
Section thickness (mm)	3.5	3	3.675	3.5	2.5
Matrix	320 × 320	288 × 202	118 × 118	192 × 154	288 × 288
In-plane resolution (mm)	0.56 × 0.56	0.82 × 1.20	1.69 × 1.69	1.25 × 1.25	0.56 × 0.56
Flip angle (degree)		2, 15		15	12

Note.— T1 maps were generated on the basis of dual-flip angle method. 3D = three-dimensional, DCE = dynamic contrast enhanced, DWI = diffusion-weighted imaging, GRASP = golden-angle radial sparse parallel, GRE = gradient-recalled echo, VIBE = volumetric interpolated breath-hold examination.

^*Diffusion-weighted imaging had a b value of 0/800 sec/mm².

^†Temporal resolution for volumetric interpolated breath-hold examination was 10 seconds; for golden-angle radial sparse parallel it was 2.5 seconds with 21 spokes per frame.

Transrectal Prostate Biopsy

Transrectal biopsies were performed by one of three board-certified urologists (C.W., with more than 3 years of subspecialty experience). Initially, MRI transrectal US fusion-guided biopsies were performed after lesions suspicious for cancer (identified by either T.J.H., M.R.B., or D.T.B., with more than 3 years of subspecialty experience) were centrally marked on axial T2-weighted MR images by using a crosshair fiducial marker; three cores per lesion were obtained. Secondarily, 12–18 conventional transrectal US–guided cluster biopsies were performed.

MRI Perfusion Processing and Histopathologic Correlation

Image processing was performed by using a commercially available software application (Syngo.via VB30, MR Prostate and MR Tissue4D; Siemens Healthineers). A volume of interest was defined that encompassed the entire prostate. Within this volume, the perfusion maps were generated by using a pixel-wise method and T1-weighted fitting with restriction to pixel values above noise level (threshold value, >20 IU). For Tofts modeling, a population-based arterial input function was used; a population-based arterial input function in this comparative setting allows a standardized evaluation because the temporal resolution of the VIBE sequence alone is insufficient to deduce a meaningful input function. Calculation of perfusion maps for influx forward volume transfer constant (K^trans) and rate constant (K_ep) completed the processing workflow.

In concordance with the histopathologic analysis–confirmed location in the peripheral anterior gland, the lesions and normal parenchyma were identified on the DWI series and corresponding ADC maps by using the T2-weighted images as morphologic reference. Volumes of interest were defined as follows: the area of visually marked diffusion restriction was outlined manually by placing the volume-of-interest border within the tissue suspicious for cancer (lesions only); next, a larger volume-of-interest border was drawn eccentrically incorporating visually nonsuspicious tissue, eventually creating two nested volume-of-interest borders; last, the software application (syngo.via VB30A, MR Prostate and MR Tissue4D; Siemens Healthineers) automatically extruded the lesion by using region-growing segmentation methods. Resulting volumetric segmentations were cloned onto all morphologic and perfusion image series.

ADC, K^trans, and K_ep values were subsequently extracted and the location of each tumorous lesion and corresponding normal tissue was encoded in the standardized Prostate Imaging and Reporting Data System scheme (12) (Table 2).

Table 2:

Quantitative Results Differentiating Analysis-proven Prostate Carcinoma from Normal Volumetrically Extracted Reference Tissue

Study Group	Normal Prostate Tissue	Prostate Cancer Tissue	P Value
VIBE (n = 40)
ADC (mm2/sec)	1.63 ± 0.33	1.01 ± 0.23	<.001
Kep (min−1)	0.34 ± 0.14	0.6 ± 0.53	<.001
Ktans (min−1)	0.15 ± 0.07	0.34 ± 0.31	<.001
GRASP (n = 61)
ADC (mm2/sec)	1.81 ± 0.28	1.16 ± 0.30	<.001
Kep (min−1)	0.39 ± 0.17	0.81 ± 0.53	<.001
Ktans (min−1)	0.13 ± 0.06	0.35 ± 0.30	<.001

Note.—Data are mean ± standard deviation unless otherwise indicated. P values less than .05 were considered to indicate statistical significance. ADC = apparent diffusion coefficient, GRASP = golden-angle radial sparse parallel, K^trans = influx forward volume transfer constant, K_ep = rate constant, VIBE = volumetric interpolated breath-hold examination.

Statistical Analysis

All quantitative ADC, K^trans, and K_ep parameters are presented as mean ± standard deviation with range.

Receiver operating characteristic evaluations were performed to assess diagnostic performance; receiver operating characteristic analyses resulted in areas under the curve and were compared by using Z statistics (13). Statistical analyses assessed whether combined diffusion and perfusion parameters evaluations can achieve higher diagnostic accuracies compared with a single-parameter evaluation on the basis of DWI alone. The following parameters were evaluated: ADC combined with K^trans and ADC combined with K_ep versus ADC alone. To adjust for multiple testing and to control the type I error in our study, we performed a Bonferroni correction of the significance level of the individual test with the following formula: $p^{*}<\frac{\alpha }{m}$ (14), where p * is the adjusted significance level, a is the critical P value, and m is the number of comparisons. The cutoff level with the optimal combination of true-positive and true-negative test outcomes was extracted from the receiver operating characteristic curve by using the Youden index (sensitivity + specificity −1) and determined the single-factor discriminator.

Accuracy was defined as:

$$\frac{\left(\mathit{\text{true\hspace{0.17em}positive}}\right)+\left(\mathit{\text{true\hspace{0.17em}negative}}\right)}{\left(\mathit{\text{true\hspace{0.17em}positive}}\right)+\left(\mathit{\text{false\hspace{0.17em}positive}}\right)+\left(\mathit{\text{true\hspace{0.17em}negative}}\right)+(\mathit{\text{false\hspace{0.17em}negative}}).}$$

To assess for differences in quantitative parameters (ADC, K^trans, and K_ep) in normal prostate tissue versus prostate cancer tissue, one-way analyses of variance evaluations were performed.

Statistical analysis was performed by using software (SPSS Statistics 22; SPSS, Chicago, Ill). The adjusted critical level of significance by using Bonferroni correction for pairwise testing with a maximum of two hypotheses was indicated by P values less than .025 for comparisons of areas under the curve. A P value less than .05 was considered to indicate statistical significance for the analysis-of-variance evaluations.

Results

Among the initial 263 participants, prostate cancer was found in 101 prostate lesions of the peripheral or anterior zones in 94 participants who also had histopathologic analysis–confirmed normal contralateral prostate parenchyma (Fig 1). Our inclusion criteria led to the exclusion of 169 participants. This selection approach allowed for the ability of intraindividual assessment to differentiate between pathologic and normal prostate parenchyma on the basis of the sequence design.

Figure 1:

Flowchart outlines the selection of the final study population with inclusion and exclusion criteria within the defined observation window. DCE = dynamic contrast-agent enhanced, GRASP = golden-angle radial sparse parallel, mpMRI = multiparametric MRI, PCa = prostate cancer, PSA = prostate-specific antigen, VIBE = volumetric interpolated breath-hold examination.

The overall mean prostate-specific antigen value was 6.8 ng/mL ± 4.3. Of the 101 lesions, 40 lesions were examined by using the VIBE technique (Fig 2) and 61 lesions were examined by using GRASP (Fig 3). For the participants who underwent VIBE, the mean age was 64 years ± 6 and the mean prostate-specific antigen value was 6.6 ng/mL ± 3.9; for the GRASP group, the mean age was 67 years ± 9 and the mean prostate-specific antigen value was 7.0 ng/mL ± 4.4. Gleason scores and anatomic location were also evenly distributed between groups (VIBE Gleason scores, 7, 30% [12 of 40]; VIBE Gleason scores ≥ 7, 70% [28 of 40]; GRASP Gleason scores, 7, 28% [17 of 61]; and GRASP Gleason scores ≥ 7, 72% [44 of 61]).

Figure 2:

Perfusion data set acquired at volumetric interpolated breath-hold examination. Images in a 68-year-old man with a Gleason score of 8 (4 + 4) and prostate-specific antigen level of 7.4 (mg/L). (a) T2-weighted MR image shows hypointense signal of the prostate cancer lesion. (b) Apparent diffusion coefficient (ADC) map shows hypointense signal of the prostate cancer lesion with an ADC value of 0.91 mm²/sec ± 0.29. (c) Perfusion map shows influx forward volume transfer constant (K^trans) value of 0.26 min²¹ ± 0.15. (d) Perfusion map shows rate constant (K_ep) value of 0.46 min⁻¹ ± 0.30. Volume of interest = 2.13 cm³. Note the extension of volume-of-interest boundaries (green outlines) defined on ADC series into interface regions between normal and pathologic parenchyma on dynamic contrast-enhanced MR images.

Figure 3:

Perfusion data set acquired by using golden-angle radial sparse parallel method. Images in a 67-year-old man with a Gleason score of 8 (4 + 4) and a prostate-specific antigen level of 18.15 mg/L. (a) T2-weighted MR image shows hypointense signal of the prostate cancer lesion. (b) Apparent diffusion coefficient (ADC) map shows hypointense signal in the prostate cancer lesion; ADC = 1.12 mm²/sec ± 0.39. (c) Perfusion map shows influx forward volume transfer constant (K^trans; 0.25 min⁻¹ ± 0.21). (d) Perfusion map shows rate constant (K_ep; 1.79 min⁻¹ ± 0.97). Lesion volume = 2.32 cm³. There is an extension of volume-of-interest boundaries (green outlines) defined on ADC series into interface regions between normal and pathologic parenchyma on dynamic contrast-enhanced MR images.

Assessment of the Diagnostic Performance of DCE MRI and DWI

The diagnostic performance of both DCE MRI techniques and DWI to help detect prostate cancer by using each MRI parameter individually is summarized in Table 3 and Figure 4. ADC values (cutoff value, 1.51 mm²/sec) showed higher accuracies to detect prostate cancer than both K^trans (cutoff value, VIBE, 0.155 min⁻¹; cutoff value, GRASP, 0.165 min⁻¹) and K_ep (cutoff value, VIBE, 0.325 min²¹; cutoff value, GRASP, 0.445 min⁻¹; Table 3).

Table 3:

Performance of Models for Prediction of Prostate Cancer

A: Performance of Single-Parameter Models
	ADC (mm2/sec)		Ktrans (min−1)		Kep (min−1)
Study Group	Accuracy (%)	AUC	Accuracy (%)	AUC	Accuracy	AUC
VIBE (n = 40)	93 (37/40)	0.93 ± 0.02	90 (36/40)	0.81 ± 0.05	85 (34/40)	0.70 ± 0.03
GRASP (n = 61)	93 (57/61)	0.93 ± 0.02	93 (57/61)	0.79 ± 0.04	90 (55/61)	0.78 ± 0.04
B: Performance of Dual-Parameter Models
	ADC with Ktrans		ADC with Kep		P Value for AUC Value Comparison
Study Group	Accuracy (%)	AUC	Accuracy (%)	AUC	ADC versus ADC with Ktrans	ADC vs ADC with Kep
VIBE (n = 40)	93 (37/40)	0.94 ± 0.03	91 (36/40)	0.94 ± 0.04	.18	.50
GRASP (n = 61)	95 (58/61)	0.97 ± 0.02	95 (58/61)	0.97 ± 0.03	.006	.021

Note.—Data in parentheses are numerator/denominator. Areas under the curve are ± standard error. Perfusion parameters derived from VIBE and GRASP MRI study participant groups are shown. P values less than .025 were considered to indicate statistical significance. ADC = apparent diffusion coefficient, AUC = area under the curve, GRASP = golden-angle radial sparse parallel, K_ep = rate constant, K^trans = influx forward volume transfer constant, ROC = receiver operating characteristic, VIBE = volumetric interpolated breath-hold examination.

Figure 4:

Quantitative assessment of combined diffusion MRI and dynamic contrast agent–enhanced MRI. Scatterplots display the data pairs of apparent diffusion coefficients (ADCs) with (a) influx forward volume transfer constant and (b) ADC with rate constant values. ADC values are on the x-axis and the perfusion parameters are on the y-axis. Horizontal and vertical lines represent the dual-parameter cutoff levels. Note that the cutoff levels on the basis of golden-angle radial sparse parallel (GRASP)–derived perfusion parameters combined with ADC allows a better differentiation (ie, less overlap) between normal and prostate parenchyma compared to cutoff levels on the basis of volumetric interpolated breath-hold examination (VIBE)–derived perfusion parameters combined with ADC.

Assessment of Single- versus Dual-Parameter Models

The dual-parameter analyses showed that the combination of cutoff levels defined by ADC and GRASP-based K^trans or K_ep showed significantly better diagnostic performance for discrimination of normal prostate tissue from prostate cancer lesions compared with a single-factor evaluation on the basis of ADC cutoff levels alone (ADC with K^trans vs ADC, P<, .006; ADC with K_ep vs ADC, P<, .02); ADC represented the best single-parameter evaluation technique. This statistical level of improvement was not observed when ADC was combined with perfusion cutoff levels on the basis of VIBE (Fig 4, Table 3).

Discussion

Multiparametric MRI of the prostate is a well-established method for the detection, localization, and staging of prostate cancer; to improve diagnostic performance, morphologic sequences are combined with functional, quantitative imaging methods such as diffusion-weighted and perfusion sequences. In the literature, DCE examinations are reported with temporal resolution of 2–15.8 seconds. In our study, temporal resolution with GRASP was 2.5 seconds. For spatial resolution, previously reported dimensions range from 3.0 × 0.5 × 0.6 mm (15) to 4.0 × 2.8 × 2.8 mm (16); the GRASP sequence showed spatial resolution of 2.5 × 0.56 × 0.56 mm.

Whereas technically it is possible to accelerate and enhance the VIBE sequence yielding comparable temporal and spatial resolution similar to the GRASP sequence by employing aggressive parallel imaging methods (integrated parallel acquisition technique factor of six, 24 reduced reference lines), a mathematically determined loss of voxel contrast approximating 80% has been predicted by the scanner software (Siemens Skyra 3.0 T, version syngo MR E11, software package N4_VE11C; Siemens Healthineers).

A recent study (17) showed that improving temporal resolution alone without increasing the spatial resolution did not lead to an improved diagnostic performance. In our study, both spatial and temporal resolutions were increased concurrently compared with the standard-of-care DCE MRI sequences. Improvements of individual characteristics of DCE MRI sequences alone may not lead to detectable improvements in diagnostic performance compared with a combination of these components because prostate cancer lesions show an especially heterogeneous tumor biology.

In our study, the dual-parameter analyses, which combined ADC values with either K^trans or K_ep, showed generally better diagnostic performance in discriminating normal prostate tissue and prostate cancer lesions compared with ADC single-factor evaluation. In particular, for the dual-parameter analyses incorporating the GRASP sequence, significantly more tumors that were not correctly characterized by the ADC component alone were identified as suspicious for tumor by the corresponding perfusion components. This statistically significant difference in tumor detection was only observed when the GRASP was used to calculate perfusion maps; therefore, it can be hypothesized that the combined increased spatial and temporal resolution of the former acquisition method benefits from differentiation between normal and low-grade parenchyma.

Our study had several limitations. First, participants were not evaluated with both acquisition schemes at consecutive imaging settings. For consecutive imaging, a second application of contrast medium would have been necessary, thereby doubling the dose of contrast material without a direct clinical benefit. This increased dose of contrast material would not have been justifiable. Second, the distribution of Gleason scores for the two different MRI pulse sequences was not identical. The relative frequencies of the Gleason scores within the groups, however, showed a similar distribution. Another limitation is that because of a lack of an independent validation set, diagnostic accuracies were potentially overestimated. The authors acknowledge that whole-mount histopathologic analyses remain the reference standard for assessing the entire gland. Last, a bias was introduced by using the same data to determine the optimal cutoff and subsequently assessing performance of that cutoff.

In conclusion, we demonstrated that if used in a dual-parameter model by incorporating diffusion and perfusion characteristics, the GRASP acquisition technique significantly improved the diagnostic performance of multiparametric MR examinations of the prostate. This effect was not observed with perfusion parameters acquired with VIBE.

Implications for Patient Care.

• Rapid perfusion imaging of the prostate by using a compressed-sensing radial MRI acquisition combined with diffusion-weighted MRI showed improved diagnostic performance compared with diffusion-only MRI.

• Compressed-sensing radial MRI with diffusion-weighted MRI showed superior diagnostic performance compared with standard contrast-enhanced MRI with diffusion-weighted MRI.

Abbreviations

ADC

apparent diffusion coefficient

DCE

dynamic contrast agent enhanced

DWI

diffusion-weighted imaging

GRASP

golden-angle radial sparse parallel

K_ep

rate constant

K^trans

influx forward volume transfer constant

VIBE

volumetric interpolated breath-hold examination