VERDICT (Vascular, Extracellular, and Restricted Diffusion for Cytometry in Tumours) is a microstructural imaging technique that combines a detailed diffusion MRI acquisition with a mathematical model to map and measure microstructural tissue parameters. The technique has shown significant promise in the preclinical setting¹ and in a pilot study in prostate cancer², but to develop the technique for translational and clinical research it must demonstrate technical validity.

This study seeks to evaluate the short-term repeatability of VERDICT MRI in normal and cancerous prostate tissue.

5 men awaiting biopsy for suspected prostate cancer were identified and recalled for VERDICT MRI, with a median of 120 days (range 43- 151) from their original multiparametric prostate MRI.

VERDICT DW-MRI was performed using a 3T scanner (Achieva, Philips Healthcare, Netherlands) using a series of pulse-gradient spin-echo sequences with various diffusion gradient strengths and timings³. The series of scans was repeated following a 2-minute interval. Imaging parameters are shown in Figure 1.

Diffusion model

VERDICT is a three-compartment model that characterises diffusion in the vascular, extracellular-extravascular space (EES) and intracellular (IC) compartments in tumours. The prostate model has five parameters: fEES (EES volume fraction), fIC (IC volume fraction), cell radius R diffusivity D and pseudo-diffusion P. Vascular volume fraction can be determined from fVASC=1- fIC- fEES. Cellularity maps are calculated by dividing fIC by the cell radius estimate cubed.

Image analysis

MR datasets were analysed with Osirix Version 7.0 (Bernex, Switzerland). A board certified Radiologist (EJ) manually contoured a region of interest (ROI) on each prostate lesion, using the slice at the epicentre of the tumour. ROIs were drawn on the b=2000s/mm² image, using the previous mpMRI for further guidance. An ROI of equal size was then drawn in a normal region of the prostate, on the same slice in the same zone. ROIs were copied onto the subsequent acquisition and manually refined accordingly.

We fitted the VERDICT model to the data using a similar iterative optimization procedure to Panagiotaki et al.^1,2 that accounts for local minima and Rician noise. Fitting consisted in two steps, first the model was fitted to data averaged over all voxels of the prostate (tumor and benign regions) and then the fitting was performed in each voxel. Fitting was performed using the open source Camino toolkit⁴. Apparent diffusion coefficient (ADC) was also fitted for comparison.

An example of a multi-parametric (mp)MRI and the subsequent generated VERDICT maps are provided in Figure 2.

Statistical analysis

Median ROI values were used for all parameters, as data was not normally distributed. Bland Altman plots were constructed and intraclass correlation coefficients (ICCs) (3,1) single measures, with absolute agreement calculated.

5 patients had a median age of 67.4 (range 58.0 – 74.6), a median prostate specific antigen (PSA) of 8.5 (range 3.3 – 18.0). Subsequent biopsy performed within a week of the scan confirmed prostate cancer in the peripheral zone (n=4) and transition zone (n=1), with Gleason scores of 3+3(n =1) and 3+4 (n=4).

Bland-Altman plots were constructed and are interpreted in Figure 3.

Intraclass correlation coefficients (ICCs) are provided in Figure 4. In accordance with Landis and Koch⁵, the following ICC interpretation scale was used: poor to fair (below 0.4), moderate (0.41–0.60), excellent (0.61–0.80), and almost perfect (0.81–1).

To be clinically useful, any quantitative imaging technique must demonstrate accurate estimates of the measured biological parameter (accuracy) and be repeatable (precision)⁶.

Broadly, more complex biophysical models such as VERDICT provide a better fit to data than simpler models with fewer parameters, such as ADC, and our previous work has shown this to be the case^1,2. However, more complex models also tend to ‘overfit’ data, and become sensitive to noise, resulting in poor repeatability.

In this study, we were able to compare the repeatability of parametric maps generated from VERDICT MRI to those of ADC. As expected, ADC demonstrated almost perfect agreement.

Cellularity was the most reliable parameter with almost perfect agreement in both normal and cancerous prostate tissue. fIC and fEES were the next best performing, demonstrating almost perfect agreement in the normal prostate and excellent agreement in cancerous tissue.

However, cell radius and fvasc demonstrated poor to moderate repeatability, with the latter likely due to relatively low contribution to MR signal.

Cellularity, fEES and fIC are repeatable VERDICT parameters, but fvasc and cell radius show poor repeatability. Further work is required to establish the medium and long-term repeatability of VERDICT, and for biological validation to determine what constitutes a clinically useful measurement.