This study showed that fast MRF acquisitions yield quantitative T1 and T2 measurements with good repeatability in bone metastasis, normal-appearing bone, and muscle in prostate cancer patients. The relative repeatabilities of T1 were 6.9% (bone metastasis), 32.6% (normal-appearing bone), and 5.8% (muscle), whilst for T2 were 21.8%, 32.2% and 16.1%, respectively. r% values for treated, mixed, or active metastases were similar to those reported for the overall metastasis group. In addition, all corresponding ICCs were > 0.82, suggesting good/excellent agreement between the two measurements. The r% values imply that, within the context of a future treatment efficacity study, proportional changes less than these values could be attributed to measurement error, while excess changes would likely be treatment-related. The relative repeatability of T1 (6.9%) and T2 parameters (21.8%) is better or of a similar order of magnitude to ADC estimates (12.5%) [32], which is widely used for MRI assessment of bone metastasis.

The highest r% values found in normal bone ROIs can be explained by the inherent heterogeneity of bone microstructure (in particular, its honeycomb-like network of trabecular bone), suggesting that this is a particularly challenging tissue for MRF relaxometry. By contrast, the muscle (the most homogenous tissue assessed here) had the lowest r% values for both T1 and T2 parameters. Moreover, the T1 and T2 repeatability of both bone metastasis and muscle tissues were similar, with T1 being always more repeatable than T2. Overall, our finding agrees with previous repeatability studies for various healthy tissues [12,13,14,15,16] that also reported better repeatability for T1 than T2. The improvement of both T1 and T2 repeatability in bone metastasis versus normal bone may be partially explained by the fact that prostate bone metastases are predominantly osteoblastic (sclerotic) [33], i.e., generating increased bone density compared to normal bone.

T1 measurements allowed clear differentiation between bone metastasis and normal-appearing bone (p-value < 0.001) within our study, in agreement with two previous MRF reports [10, 11] Note, however, that our study included prostate patients that were under treatment or have had previous lines of treatment; therefore, even if their MRI assessment did not detect any active disease, one cannot consider their bone as completely normal, as would be the case for a treatment-naïve patient. In addition, the osteoblastic changes seen in prostate cancers are expected to lead to reduced T1 values, which may explain the lower T1 found in normal-appearing bone in our study. Within the metastases sub-cohort, the T1 values could not separate between active/mixed/treated disease (p-value = 0.78), although an existing study [24] suggests that early response to therapy manifests as decreased T1 values. Compared with the two existing studies, our estimates for bone metastasis T1 are within the previously reported range (1301 ms vs 1195 ms [10] and 1675 ms [11]), whilst much lower values were found for bone in our study (140 ms vs 461 ms [10] or 447 ms for fatty bone [11]). Similar pelvic bone metastases were assessed for our study and [10], whilst [11] derived their findings from the vertebrae of patients undergoing MRF for diagnosis purposes, i.e. prior to any treatment. As mentioned before, we evaluated normal-appearing bone in the same cohort of metastatic prostate cancer patients, whilst studies [10] and [11] evaluated normal bone in non-malignant patients. By comparison, our muscle T1 measurements are in line (1145 vs 1100 ms) with a previous report [4].

The T2 parameter was less useful for bone metastasis versus normal-appearing bone differentiation as measured values overlapped across these two groups (p-value = 0.50). Conversely, within the bone metastases sub-cohort, the T2 parameter of the treated metastasis was significantly higher than those of mixed or active classes (p-values = 0.009, < 0.001) supporting the use of T2 as a treatment response biomarker. Similar behaviour of an increase in T2 values for treated versus untreated metastasis was reported in [10]. Our T2 measurements are similar to previous reports in bone metastases (70 vs. 58 ms), normal-appearing bone (65 vs. 78 ms) [11] and muscle (33 vs. 44 ms) [4]. Note that, the reported T2 in bone metastasis (averaged across three subclasses) of 70 ms was biased upwards by the contribution of an increased T2 of treated lesions (88.4 ms). Therefore, the active bone metastases value (60.5 ms) should be considered instead for further literature comparison.

A recent study [34] using SyMRI (an alternative technique to MRF and capable of generating T1, T2 and PD quantitative maps) highlighted the utility of the PD maps as this metric was shown to be able to differentiate bone lesion changes with treatment more sensitively than T1 or T2. Our study did not assess the PD parameter, so future work exploring the use of PD maps could help develop bone response biomarkers. Moreover, a comparative study of SyMRI and MRF [35] found similar repeatability for both methods and results comparable with standard references.

Unlike a recent paper [4], our MRF acquisitions did not include fat suppression pulses. Given the presence of both fat and water signals in bone metastases, additional MRF studies looking to separate T1 and T2 based on fat and water signals could be explored to allow further characterisation of such tissues.

To enable clinical adoption of MRF, three elements are required: (1) a prototype MRF sequence compatible with the available MR scanner; (2) a direct connection with a high-specification computer (to store the MRF dictionary and the pattern reconstruction algorithm allowing fast and reliable MRF-reconstructed maps); and (3) a standardised test-object to enable initial calibration of the sequence on the given scanner. For the prototype sequence and reconstruction used in this work, less than 3 minutes were required to automatically send raw MRF acquisitions to the external computer to generate the MRF maps and return them to the scanner console.

This study had several limitations. First, our results were derived from a small cohort of 20 patients recruited at a single institution on a 1.5-T scanner. However, an increased sample (n = 44) for bone metastasis assessment was available, due to most patients having multiple sites of disease. Second, the reported T1 and T2 values across bone metastasis, normal bone and muscle are derived only from patients undergoing treatment, and these T1, T2 values (including those of normal-appearing tissues) should be interpreted within this non-naïve treatment context. Nevertheless, the measurement repeatability of T1 and T2 of all bone metastases appears sufficiently robust for clinical evaluation. Third, this study did not assess the inter-observer variability, as the main aim of the study was to test the repeatability of the MRF measurements while keeping other factors, such as the observers, the same. The additional variability that could arise from differences in observer ROI delineation was beyond the scope of this study. Further studies on larger cohorts from multiple centres with multiple operators (performing lesion segmentation) using a standardised MRF sequence would help establish quantitative biomarkers that can differentiate normal from disease tissue or inform on treatment response [36]. For these purposes, recommendations on standardisation and validation strategies of MRF [37] are important and should be further developed.

In conclusion, fast MRF allows repeatable quantitative T1 and T2 measurements in bone metastasis, normal-appearing bone, and muscle in patients with primary prostate cancer. T1 was found to be able to differentiate between bone metastasis and normal-appearing bone, whilst T2 can separate active versus treated metastasis. Further longitudinal study is ongoing to assess the magnitude of treatment-related changes in T1 and T2 that occur in bone metastases with reference to the repeatability limits reported here and compared to standard quantitative MRI. Further validation of MRF measurement is required to develop new robust response biomarkers for malignant bone disease.