Proton resonance frequency (PRF) shift thermometry is a robust method to measure temperature changes in tissue. In PRF, phase images are used to measure the PRF change with temperature, which for water-based tissues is 0.01 ppm/°C. Unfortunately, signal phase is susceptible to spatial and temporal magnetic field changes which can mimic temperature changes. For example, magnetic field gradients can heat up over time which leads to a temporal field drift (at a typical clinical 3T-system, the field drift can be up to 0.1 ppm/h corresponding to 10 °C/h). In contrast to short ablative procedures, hypo- or hyperthermia treatments can last several hours, and the therapeutic temperature change can be on the order of several degrees only¹. If large volumes are treated (e.g., cerebral hypothermia²), reference-less thermometry^{3, 4} cannot be used as the borders of the treated area are not precisely known, fat cannot be used as reference due to potential susceptibility changes⁵, and MRS of metabolites⁶ is time consuming.

Similar to previous work from Svedin et al⁷, in this work we propose to use FID navigators to correct for general field drifts to accurately detect temperature changes on the scale of 1°C.

For FID-corrected temperature measurements, a 3D spoiled gradient echo sequence (FLASH) was developed with two FID navigators before and after the imaging acquisition (Fig. 1). The navigators are applied in sections of the sequence without spatial encoding, so that the complex overall signal within the excited volume is acquired. The following MRI parameters were used: TE/TR = 10/35 ms, spatial resolution 2×2×2 mm³ and matrix 256×98×24, acquisition time 80 s, and TE_nav,1 = 5 ms and TE_nav,2 = 15 ms. The acquisition was repeated 60 times so that the whole measurement lasted for 80 min.

In order to reconstruct the corrected images, the phase data from two FID navigators was subtracted. Then, the remaining phase difference was divided by the time difference between navigator readouts to find the drift-induced phase slope during each TR. Finally, a correction phase was calculated for each sampling point during the imaging readout and subsequently subtracted. A more detailed description of the correction can be found in Svedin et al⁷.

A hypothermia experiment was conducted in a 2%-agar phantom which was cooled through a plastic tube with constant flow of ice water. Ten reference data sets were acquired while water flowing at room temperature before ice was added to the water reservoir. The experiment was continued until all ice melted. Temperature images were calculated from the phase difference of the averaged reference data and the dynamic phase values. In the temperature images, 4 different ROIs were placed near the locations of embedded fiber-optical (FO) temperature probes (FOTEMP 4, Optocon AG, Dresden, Germany).

Figure 2 compares temperature changes from fiber-optical and MRI temperature measurements. Before FID correction, MRI deviates as much as 6°C for all probes, and the temperature error increases with time. After correction the maximum temperature error is 0.5°C (probes 2-4), and 0.7°C (probe 1 close to tube). The Bland-Altman plot (Fig. 3) shows a systematic but small mean deviation of 0.1°C during the cooling phase and a larger deviation of up to 0.7°C during the transient re-heating phase while the ice was melting. In total the mean difference is +0.2°C (confidence limits: -0.2°C to 0.6°C) which reflects an over-correction of MR temperature data.

The results show that FID navigator correction can substantially increase the accuracy and precision of PRF temperature measurements by up to one order of magnitude. As both FID navigators integrate the signal over the entire measured volume, the phase change used for correction is influenced by both, field drift and temperature. When the temperature-related phase changes are local, the dominant phase change is caused by the field drift, whereas if temperature effects become more global both influences impact the FID phase and cannot be discriminated. This is reflected in the Bland-Altman plot where temperature errors remain small during (local) cooling, whereas systematic errors increase during the re-heating phase when temperatures start to distribute over larger portions of the phantom. To overcome this limitation, a dummy load could be added with stabilized temperature. The volume of the load could be optimized depending on the expected heated or cooled volume.

In summary, FID navigator correction can be used to improve MR thermometry in thermal therapies that apply heating or cooling over a long time (several hours) and are intended to induce small temperature differences (several °C per hour), for which a high temperature accuracy and precision (≤0.5°C) is needed.