Diffusion-weighted imaging as a treatment response biomarker evaluating bone metastases in prostate cancer: a pilot study

Study design

The primary endpoint of the TOPARP-A trial was response rate, with response defined as any of the following: a response of soft tissue/visceral disease according to Response Evaluation Criteria In Solid Tumors (RECIST) 1.1(4); a confirmed reduction in the prostate specific antigen (PSA) level of ≥50%; or a conversion in the circulating tumor cell (CTC) count, with a reduction in the number of CTC from ≥5/7.5 ml of blood at baseline to <5/7.5 ml of blood during treatment, with a confirmatory assessment at least 4 weeks later (22). Detailed information of the inclusion and exclusion criteria and the results of the TOPARP-A trial have been published showing a response rate to olaparib in mCRPC of 33% (95% confidence interval [CI], 20 to 48%) (23).

Participation in the optional MRI sub-study was offered to those patients without contraindication for MRI at The Royal Marsden NHS Foundation Trust. WB-MRI was performed at baseline (within 28-days prior to starting treatment) and at Cycle 4 Day 1 (corresponding to 12-weeks after starting treatment) and every 12-weeks subsequently. The primary endpoint of the MRI sub-study was to assess the association between changes in parameters derived from WB-DWI (volume of bone metastases and median ADC) and response to olaparib. For MRI sub-study purposes, patients were classified as responders if they met the definition of the primary endpoint of the TOPARP-A trial and if they had not experienced radiological progression by 12-weeks.

Patient population in the MRI sub-study

Patients were included in this study if: a) signed informed consent for the MRI sub-study in the setting of the TOPARP-A trial, b) bone metastases identified based on review of combined imaging modalities: MRI, CT and BS (in all the cases), c) a minimum of two paired WB-MRI studies performed at baseline and after 12-weeks of treatment. Patients with suboptimal quality WB-MRI or incomplete studies were considered un-evaluable for analysis and excluded from the study.

Clinical data collection

Data were collated into an anonymized database and analyzed by the Institute of Cancer Research Clinical Trials and Statistical Unit (ICR-CTSU; London, UK). PSA and CTC counts were collected at baseline and every 12-weeks while on treatment. CTC counts were also pursued at weeks 1, 2, 4 and 8. RECIST assessments were evaluated at baseline and every 12-weeks using CT.

Whole-body MRI parameters

WB-MRI was performed on a 1.5-T MRI scanner (Avanto Siemens Healthcare, Erlangen, Germany), using surface and body coils on patients positioned supine. Axial images were acquired using free breathing single-shot twice-refocused echo-planar DWI from the upper cervical spine to mid-thighs, sequentially across four imaging stations, each consisting of 50 slices respectively. In addition to WB-DWI, anatomical imaging was also acquired using breath-hold axial T1-weighted. The scan parameters are summarized in Appendix Table 1.

Imaging analysis

Images were processed and analyzed with open-access imaging assistant software (OsiriX v5.6). Evaluation of T1-weighted and DWI (b50, b900 and ADC maps) was performed in order to assess the presence of metastatic bone disease. Regions of interest (ROIs) including areas of signal abnormality on DWI b900, corresponding to high signal on DWI b900 and low signal on T1-weighted imaging, in keeping with metastatic bone disease were delineated. Different delineation techniques of the signal abnormality on DWI b900 corresponding to bone metastases were undertaken. Firstly, ROI analyses were performed including all areas of signal abnormality on DWI b900 and T1-weighted MRI corresponding to bone metastases observed in the axial skeleton (spine and pelvis, not including ribs) between C4 and mid-thighs, labeled as total diffusion volume (tDV). Secondly, a more limited analysis was performed using a RECIST approach of a maximum 5 target representative bone metastases chosen using the following criteria: maximum axial dimension >1cm, well-defined lesion border and representing different skeletal areas. For this analysis, ROIs including total volume of up to 5 target lesions and ROIs including the central slice of the same target lesions were chosen.

Additionally, a single radiologist (RPL) manually delineated the entire axial skeleton (spine and pelvis, not including ribs) enclosing normal and abnormal bone marrow from C4 to lesser trochanters. This delineation technique was included in view of its possible advantage for automated segmentation of the skeleton.

A semi-automated segmentation tool from the OsiriX software v.5.6 was used for delineating ROIs. All the delineation techniques in every WB-DWI were performed by a single radiologist (RPL) with 3 years of experience in WB-DWI; manual correction of the segmentation mask corresponding to the regions of interest (ROI) was performed by the radiologist where necessary (Figure 1). The volume of metastases was calculated as the number of voxels for all ROIs multiplied by the voxel volume in each case. The ADC value of every pixel was recorded and histogram representations of the ADC values of bone metastases for each patient were generated using Microsoft Excel 2010.

Statistical analysis

Distribution of PSA, CTC counts, median ADC (mADC), tDV, volume and diameter of the target lesions at baseline and percentage change after 12-weeks on treatment are presented using descriptive statistics. Baseline distributions and median changes during treatment in mADC, tDV, volume and diameter of the target lesions are compared between responders and non-responders using non-parametric Mann-Whitney tests and their association with response to treatment using univariate and multivariate (adjusting for known prognostic factors of baseline PSA, lactate dehydrogenase [LDH] and alkaline phosphatase [ALP]) logistic regression models. The correlation between baseline and changes in tDV after 12-weeks on treatment with baseline and best percentage change in PSA and CTC respectively were assessed using Spearman´s correlation coefficient (r), with 0.4≤|r|<0.6 indicating moderate correlation, 0.6≤|r|<0.8 strong correlation and |r|≥0.8 very strong correlation. A significance level of 0.05 and 95% CI have been used. No adjustment for reporting of multiple analyses has been performed; therefore significant results must be interpreted with caution. The analyses are based on a data snapshot taken on 24^th April 2015 and were performed with Stata version 13 (StataCorp).

Analysis of axial skeleton DWI b900 signal abnormality

When delineating all the areas of DWI signal abnormality in keeping with bone metastases in the axial skeleton, the median tDV in this population was 0.45 L (min, max: 0.01, 1.31 L) and mADC was 782 x10^-6 mm²/s (min, max: 684, 1121 x10^-6 mm²/s). These parameters grouped by responders and non-responders are summarized in Table 2; there were no statistically significant differences between the baseline distribution of tDV and mADC between the two groups (p=0.243 and p=0.312 respectively).

All six patients who responded to olaparib showed a decrease in tDV (median: -41.1%; min, max: -58.8%, -6.3%), but no decrease was observed in any of the 15 non-responders (median: +20.7%; min, max: +0.0%, +76.9%); this difference between responders and non-responders was significant (p=0.001). (Table 3, Appendix figure 1). Patients who responded to olaparib showed a greater increase in mADC after 12-weeks of treatment (median: +35.4%; min, max: +1.3%, +59.5%), compared to non-responders (median: +7.5%; min, max: -9.0%, +32.7%, p=0.14); increases in mADC after 12-weeks of treatment were associated with increased odds of response (OR: 1.08, 95%CI 1.00, 1.15, p=0.037) (Table 4, Appendix figure 2). An example of a responding patient is represented in Figure 3.

The two patients with mixed osteoblastic/osteolytic pattern with predominantly lytic bone metastases were non-responders who had +55.5 % and +24.6 % increase of tDV and +3.40 %, + 15.6 % increase of mADC after 12 weeks on treatment respectively.

The correlation between PSA levels, CTC counts and DWI parameters was also explored; baseline PSA levels and CTC counts showed strong and moderate positive association with baseline tDV (r=0.64, 95% CI 0.29-0.84; and r=0.59, 95% CI 0.22-0.82 respectively) and there was a strong positive association between changes in tDV and best post-treatment percentage change in PSA and CTC (r=0.63, 95% CI 0.27-0.83; and r=0.77 95% CI 0.51, 0.90 respectively), indicating that changes in tDV correlate with response to therapy.

Of the six responding patients, four had a further evaluable WB-MRI at the time of radiological and/or PSA progression. In all four cases, we observed a decrease in tDV while responding to olaparib, followed by a later increase in tDV at the time of radiological and/or PSA progression. Three of these four responding patients also had an increase in mADC while responding to treatment, followed by a decrease in mADC at the time of radiological and/or PSA progression. The fourth patient experienced minimal mADC change at PSA nadir and disease progression. (Appendix figures 3 and 4).

Analysis of five target lesions (total volume and central slice)

With the aim of evaluating more limited radiological analyses, to decrease workload, we correlated changes in up to 5 target lesions per patient with treatment response. We evaluated 5 target lesions in 19 of the 21 patients (90%); the remaining two patients had only one and three evaluable bone lesions respectively. The median sum of total volumes corresponding to the target lesions in the population at baseline was 0.05 L (min, max: 0.01, 0.52 L), and the mADC when delineating total volume of the target lesions was 814 x10^-6 mm²/s (min, max: 606, 1712x10^-6 mm²/s). In patients with non-widespread bone disease (N=9) we also assessed the diameter of the target lesions in the central slice. When assessing only the central slice of the same target lesions, the median of the sum of diameters at baseline was 12.6mm (min, max: 2.8, 20.2 mm) and the mADC was 835 x10^-6 mm²/s (min, max: 554.5, 1263 x10^-6 mm/s). These parameters grouped by responders and non-responders are summarized in Table 2; there were no statistically significant differences between the baseline distribution of volume, diameter and mADC (central slice and volume) of the target lesions between the two groups (p=0.876, p=0.143, p=0.312 and, p=0.073 respectively).

Then, we assessed the same target lesions for each patient in their follow-up WB-MRI after 12-weeks of treatment; the percentage change of these parameters after 12-weeks of treatment is summarized by response status in Table 3, Figure 2. Changes in entire lesion volume of the target bone metastases also inversely associated with response (OR: 0.89, 95%CI 0.80, 0.99, p=0.037) (Table 4). The mADC change at 12-weeks, when analyzing the target bone metastases (total volume and central slice), also associated with response, although these associations did not reach statistical significance (p=0.056 and p=0.082 respectively) (Table 4). Results from the multivariate logistic regression analyses showed similar trends (Table 4).

Analysis of the axial skeleton (enclosing normal and abnormal bone marrow)

The baseline median of the mADC in our population when delineating the entire axial skeleton including both normal and abnormal bone marrow was 805 x10^-6 mm²/s (min, max: 614, 1182 x10^-6 mm²/s). mADC at baseline grouped by responders and non-responders are summarized in Table 2; there were no statistically significant differences between the baseline distributions of mADC between the two groups (p=0.94). The percentage change of mADC after 12-weeks of treatment is summarized by response status in Table 3, Figure 2. When comparing the mADC of the entire axial skeleton (normal and abnormal bone marrow) pre- and-post treatment with olaparib, changes in mADC did not associate with response to treatment (p=0.518). (Table 4)