Introduction

Multiparametric MRI (mpMRI), the current standard imaging test for prostate cancer diagnosis, improves detection of clinically significant cancer¹. However, the standardized qualitative scoring system, PI-RADS, yields inconsistent results^2–4. The standard quantitative DWI metric, conventional apparent diffusion coefficient (ADC), depends on multiple scan parameters and also suffers from poor reliability⁵.

Restriction spectrum imaging (RSI) is an advanced diffusion modeling framework that allows for more complex tissue microstructure. A recent study found that male pelvic tissues can be optimally modeled with four diffusion compartments⁶. Prostate tumor conspicuity was highest using the model coefficient for the slowest diffusion compartment, RSI-C₁. Voxel-level prostate cancer detection was also more accurate using RSI-C₁ than ADC⁷.

Here, we compared RSI-C₁ to ADC and PI-RADS for detection of higher-grade prostate cancer. We hypothesized that RSI-C₁ would be a superior quantitative marker for higher-grade prostate cancer than conventional ADC.

Methods

Study Population

This retrospective study included men who underwent RSI-MRI for suspected prostate cancer at UC San Diego between January 2017 and December 2019 and had a prostate biopsy within 180 days of MRI. RSI-MRI exams included standard mpMRI, as well as an RSI series with four b-values in a single acquisition. This study was approved by the Institutional Review Board.

MRI Acquisition and Processing

Scans were collected on a 3T clinical MRI scanner (Discovery MR750, GE Healthcare, Waukesha, WI, USA) using a 32-channel phased-array body coil. A T2-weighted sequence was acquired with identical scan coverage to the RSI series (TR: 3850 ms, TE: 100 ms, resolution: 0.39x0.39 mm, acquisition matrix: 384x320, slice thickness: 3 mm). The RSI acquisition used field-of-view optimized and constrained undistorted single shot imaging and sampled b-values of 0, 500, 1000, and 2000 s/mm² at 2, 6, 6, and 12 unique gradient directions, respectively (TR: 4500 ms, TE: 68.7 ms, resolution: 2.5x2.5 mm, acquisition matrix: 96x48, slice thickness: 6 mm). The b=0 s/mm² volumes were acquired using forward and reverse phase encoding to allow for correction of B0-inhomogeneity distortions. RSI acquisition time was approximately 2 minutes.

Post-processing of MRI data was completed using in-house programs written in MATLAB (MathWorks, Natick, MA, USA). Diffusion data were corrected for distortions arising from B0 inhomogeneity, gradient nonlinearity, and eddy currents^8,9. Conventional ADC was calculated for each voxel using distortion-corrected DWI sequences performed with b-values of 0, 500, and 1000 s/mm².

RSI calculations were as previously described^6,7. Diffusion signals were corrected for noise and distortion, then scaled by the median b=0 signal within each patient’s prostate. Corrected signal intensity for each b-value, S_corr(b), was modeled as a linear combination of exponential decays representing four diffusion compartments (i=1-4) with fixed diffusion coefficients D_i and voxel-wise estimated signal contributions C_i.
$$S_{corr}(b)=C_{1}e^{-bD_{1}}+C_{2}e^{-bD_{2}}+C_{3}e^{-bD_{3}}+C_{4}e^{-bD_{4}}$$
Optimal D_i values for each compartment were previously determined for pelvic and prostate tissues: 1.0 e-4, 1.8 e-3, 3.6 e-3, and >>3.0 e-3 mm²/s, approximately representing restricted, hindered, free diffusion, and flow, respectively. The prostate gland was manually segmented on T2-weighted imaging and verified on DWI volumes using MIM (MIM Software Inc, Cleveland, OH, USA).

Clinical Data

Clinical records were reviewed to obtain histopathology results (highest grade group¹⁰ on biopsy or prostatectomy, if applicable) and imaging results (highest PI-RADS category reported). MRI exams were read by board-certified and subspecialty-fellowship-trained radiologists, using all available clinical images and standard PI-RADS criteria (transition to v2.1 from v2 occurred in 2019^2,3).

Patient-Level Analysis

Receiver operating characteristic (ROC) curves were generated for detection of higher-grade prostate cancer (grade group ≥2) using ADC, RSI-C₁, and PI-RADS. For ADC and RSI-C₁, the most suspicious voxel-level value within the prostate was used (i.e., smallest ADC and largest RSI-C₁). Gleason ≤6 cancers or fully benign biopsies were considered negative results. Performance was assessed by area under the ROC curve (AUC). Statistical comparisons were made via 10,000 bootstrap samples to calculate 95% confidence intervals and bootstrap p-values for the difference between the performance of ADC, RSI-C₁, and PI-RADS. Significance was set at two-sided α=0.05.

Results

151 patients were included. The median age was 66 years (IQR 59-72). The median time from MRI to biopsy was 16 days (IQR 1-35). PI-RADS and pathology characteristics are summarized in Table 1. 104 patients were scored using PI-RADS v2, 47 using v2.1.

The AUC values for ADC, RSI-C₁, and PI-RADS were 0.58 [0.51,0.67], 0.76 [0.68,0.83], and 0.78 [0.71,0.85], respectively. RSI-C₁ was superior to ADC (p=0.003) as a patient-level classifier of higher-grade prostate cancer. The performance of RSI-C₁ was comparable to that of PI-RADS (p=0.59).

Discussion

RSI-C₁ performed well as a quantitative, automated detector of higher-grade prostate cancer. The performance of RSI-C₁ was superior to that of ADC. RSI-C₁, based only on a two-minute diffusion MRI acquisition, also had performance comparable to PI-RADS categories assigned by experts using all standard imaging in a clinical multi-parametric MRI exam.

Limitations include retrospective design, single institution, lack of prostatectomy specimen in all patients, and inclusion of only patients who underwent biopsy with results available in our institutional patient records. Future work will evaluate the incorporation of spatial characteristics and other mpMRI modalities, as well as separate analyses for lesions in the peripheral and transition zones.

Acknowledgements

No acknowledgement found.