A Pilot Study of Multidimensional Diffusion MRI for Assessment of Tissue Heterogeneity in Prostate Cancer

Abstract

Objectives: The objectives of this exploratory study were to investigate the feasibility of multidimensional diffusion magnetic resonance imaging (MddMRI) in assessing diffusion heterogeneity at both a macroscopic and microscopic level in prostate cancer (PCa).

Materials and methods: Informed consent was obtained from 46 subjects who underwent 3.0-T prostate multiparametric MRI, complemented with a prototype spin echo-based MddMRI sequence in this institutional review board-approved study. Prostate cancer tumors and comparative normal tissue from each patient were contoured on both apparent diffusion coefficient and MddMRI-derived mean diffusivity (MD) maps (from which microscopic diffusion heterogeneity [MKi] and microscopic diffusion anisotropy were derived) using 3D Slicer. The discriminative ability of MddMRI-derived parameters to differentiate PCa from normal tissue was determined using the Friedman test. To determine if tumor diffusion heterogeneity is similar on macroscopic and microscopic scales, the linear association between SD of MD and mean MKi was estimated using robust regression (bisquare weighting). Hypothesis testing was 2 tailed; P values less than 0.05 were considered statistically significant.

Results: All MddMRI-derived parameters could distinguish tumor from normal tissue in the fixed-effects analysis (P < 0.0001). Tumor MKi was higher (P < 0.05) compared with normal tissue (median, 0.40; interquartile range, 0.29-0.52 vs 0.20-0.18; 0.25), as was tumor microscopic diffusion anisotropy (0.55; 0.36-0.81 vs 0.20-0.15; 0.28). The MKi could not be predicted (no significant association) by SD of MD. There was a significant correlation between tumor volume and SD of MD (R2 = 0.50, slope = 0.008 μm2/ms per millimeter, P < 0.001) but not between tumor volume and MKi.

Conclusions: This explorative study demonstrates that MddMRI provides novel information on MKi and microscopic anisotropy, which differ from measures at the macroscopic level. MddMRI has the potential to characterize tumor tissue heterogeneity at different spatial scales.

Conflict of interest statement

Conflicts of interest and sources of funding: F.M.F., C.M.T., and S.E.M. are funded through NIH P41EB 015898, NIH P41EB 028741, and NIH R01CA241817. F.S. and C.-F.W. are funded through NIH P41EB 015902. For the remaining authors, no conflicts or sources of funding were declared.

Copyright © 2021 Wolters Kluwer Health, Inc. All rights reserved.

Figures

FIGURE 1.

Example of prostate segmentation on MRI of tumor (1), nPZ (2), and nTZ (3) using 3D Slicer in a 67-year-old man. Radical prostatectomy 5 months post-mpMRI examination revealed Gleason grade group 3 (Gleason score 4 + 3 = 7), with 60% Gleason pattern 4. Top row (A–C) shows axial MD map slices through different axial levels, and bottom row (D and E) shows different axial slices through ADC maps. Image F is a T2-weighted image at the same axial level as image D, with the tumor lesion indicated by an asterisk.

FIGURE 2.

Flowchart outlining the total patient number in whom informed consent for study participation was obtained, number of patients excluded and reason for exclusion from the study, and number of patients in whom a pathological diagnosis of prostate cancer was made.

FIGURE 3.

Prostate MRI in a 66-year-old man with confirmed Gleason grade group 2 (Gleason score 3 + 4) on transrectal ultrasound–guided biopsy. Top row: (A) DWI (b = 1400 s/mm2), (B) ADC map, (C) T2-weighted image; and bottom row: (D) MKa, (E) MKi, and (F) MD, through the same axial slice. These maps show a well-defined focal lesion in the left anterior prostate (contoured on 3D Slicer on image A) demonstrating high signal on A, restricted diffusion on B, and is ill-defined on T2 weighted images (C). Likewise, there is diffuse low signal on F, but heterogeneous increased microscopic tissue diffusion heterogeneity (E) and, to a lesser degree, slightly increased microscopic tissue anisotropy (D). This is an example of how measures of macroscopic and microscopic diffusion heterogeneity appear grossly dissimilar, with the addition of microscopic diffusion heterogeneity likely providing novel information.

FIGURE 4.

Graphs show correlation between ADC (μm2/ms) and MD (μm2/ms) for tumor (ρs, 0.89; CI, 0.80–0.96), nPZ (ρs, 0.59; CI, 0.35–0.75), and nTZ (ρs, 0.63; CI, 0.41–0.78) (all P’s < 0.0001). Lines are lines of best fit.

FIGURE 5.

ROC analysis using AUC for both ADC and MddMRI parameters (MD, MKa, MKi) in differentiating tumor ROI versus normal ROI. ROC indicates receiver operating characteristic curve; AUC, area under the curve; and ROI, tumor region of interest.

FIGURE 6.

ROC analysis using AUC for both ADC and MddMRI parameters (MD, MKa, MKi) in differentiating nonclinically significant (Gleason score 6, Gleason grade group 1) from clinically significant (Gleason score ≥7, Gleason grade group 2–4) prostate cancer. ROC indicates receiver operating characteristic curve; AUC, area under the curve; and ROI, normal region of interest.

FIGURE 7.

Top row (A and B) outlines associations between tumor volume and microscopic diffusion heterogeneity and macroscopic diffusion heterogeneity. There was no significant correlation between tumor volume and microscopic diffusion heterogeneity (A, P = 0.2), whereas there was a significant correlation between tumor volume and macroscopic heterogeneity (B, R2 = 0.50, slope = 0.008 μm2/ms per millimeter, P < 0.001). Bottom row (C and D) outlines association between macroscopic diffusion heterogeneity and microscopic diffusion heterogeneity for all tumors (C), and when dichotomized by median volume into larger (red) and smaller (blue) tumors (D). No significant correlations were found between macroscopic and microscopic diffusion heterogeneity for all tumors evaluated together (C, P = 0.9) or when evaluated separately for small and big tumors (D, P = 0.7 and P = 0.2, respectively). MKi indicates microscopic diffusion heterogeneity; SD of MD indicates macroscopic diffusion heterogeneity.

FIGURE 8.

Pictorial example of how the morphological components on histology in 2 different tumors of the same grade can be reflected in quantitative MddMRI measurements. Two RP cases with similar MKi, elevated and slightly different MKa, and markedly different FA (indicative of different levels of macroscopic anisotropy), and their corresponding morphological findings. Top 2 rows are hematoxylin-eosin–stained sections. Their corresponding Gleason score, FA, MKa, and MKi values are displayed in the table below. Case A is a mix of multidirectional, ill-defined stroma (black arrow) and neoplastic glandular structures. At the voxel level, the anisotropic microscopic features are randomly oriented, which may explain the low FA (ie, diffusion being close to isotropic). However, at the level of individual cells (ie, the microscopic level), it is possible that diffusion may occur predominantly along 1 direction, with a resultant high MKa. The moderately elevated MKi reflects the presence of nonuniformity of tissue diffusion at the microscopic level, that is, a mixture of luminal space with almost unhindered diffusion, glandular tissue with moderately hindered diffusion, and large amounts of stromal tissue with markedly hindered diffusion. Case B demonstrates a more moderate amount of stroma, which appears more well organized over longer distances along 1 direction, that is, fascicular (red arrows), thus both FA and MKa are higher. The higher MKa in case B may indicate slightly different architecture at the cellular level as well. As in case A, the tissue architecture variety appears relatively evenly distributed between luminal space, glandular cells, and stromal cells, thus MKi is moderately elevated. Abbreviations: GP, Gleason pattern; FA, fractional anisotropy; MKa, microscopic tissue anisotropy; MKi, microscopic diffusion heterogeneity. Note: Final Gleason score is based on evaluation of the entire RP, but only select hematoxylineosin–stained images are shown and may not represent the final Gleason score.