Clinical outcome of hyperthermia depends on the achieved target temperature, therefore target conformal heating is essential. Currently, invasive temperature probe measurements are the gold standard for temperature monitoring, however, they only provide limited sparse data. In contrast, magnetic resonance thermometry (MRT) provides unique capabilities to non-invasively measure the 3D-temperature. This study investigates MRT accuracy for MR-hyperthermia hybrid systems located at five European institutions while heating a centric or eccentric target in anthropomorphic phantoms with pelvic and spine structures. Scatter plots, root mean square error (RMSE) and Bland–Altman analysis were used to quantify accuracy of MRT compared to high resistance thermistor probe measurements. For all institutions, a linear relation between MRT and thermistor probes measurements was found with R² (mean ± standard deviation) of 0.97 ± 0.03 and 0.97 ± 0.02, respectively for centric and eccentric heating targets. The RMSE was found to be 0.52 ± 0.31 °C and 0.30 ± 0.20 °C, respectively. The Bland-Altman evaluation showed a mean difference of 0.46 ± 0.20 °C and 0.13 ± 0.08 °C, respectively. This first multi-institutional evaluation of MR-hyperthermia hybrid systems indicates comparable device performance and good agreement between MRT and thermistor probes measurements. This forms the basis to standardize treatments in multi-institution studies of MR-guided hyperthermia and to elucidate thermal dose-effect relations.

Background: Temperature distributions resulting from hyperthermia treatment of patients with high-risk soft-tissue sarcoma (STS) were quantitatively evaluated and globally compared with thermal simulations performed by a treatment planning system. The aim was to test whether the treatment planning system was able to predict correct temperature distributions. Methods: Five patients underwent computed tomography (CT) fluoroscopy-guided placement of tumor catheters used for the interstitial temperature measurements. For the simulations, five 3 D patient models were reconstructed by segmenting the patient CT datasets into different tissues. The measured and simulated data were evaluated by calculating the temperature change ( $\Delta \mathrm{T}$), T90, T50, T20, T_mean, T_min and T_max, as well as the 90th percentile thermal dose (CEM43T90). In order to measure the agreement between both methods quantitatively, the Bland–Altman analysis was applied. Results: The absolute difference between measured and simulated temperatures were found to be 2°, 6°, 1°, 4°, 5° and 4 °C on average for T_min, T_max, T90, T50, T20 and T_mean, respectively. Furthermore, the thermal simulations exhibited relatively higher thermal dose compared to those that were measured. Finally, the results of the Bland–Altman analysis showed that the mean difference between both methods was above 2 °C which is considered to be clinically unacceptable. Conclusion: Given the current practical limitations on resolution of calculation grid, tissue properties, and perfusion information, the software SigmaHyperPlan™ is incapable to produce thermal simulations with sufficient correlation to typically heterogeneous tissue temperatures to be useful for clinical treatment planning.