Synopsis

Keywords: Safety, Safety

We explored T₁-based magnetic resonance fingerprinting thermometry (MRFT) and compared it with proton resonance frequency shift (PRFS) thermometry in ex vivo bovine muscle data. We conducted temperature sensitivity calibration experiments for both methods. We compared MRFT and PRFS thermometry predicted temperatures in bovine muscles, validated by fluoroptic probe measurements. We observed strong correlation in temperature sensitivity calibration experiments (R²>0.958). Both methods predict temperature accurately (RMSE<1.37 ℃ and RMSE<0.93 ℃ for MRFT and PRFS, over a range of 25 ℃). We observed strong correlation between PRFS and MRFT predicted temperature in heated bovine muscle (R²>0.93).

Introduction

MR thermometry provides noninvasive temperature measurements in near real-time. Proton resonance frequency shift (PRFS) MR thermometry is the most commonly used method clinically but is vulnerable to challenges such as inter-scan motions, temperature underestimation in adipose tissues, and requirements of reference scans ¹.
In this work, we demonstrate magnetic resonance fingerprinting thermometry (MRFT) by exploiting T₁ relaxation time’s dependence on temperature and compare it with PRFS thermometry by comparing predicted temperatures in ex-vivo bovine muscles.

Methods

MRF implementation
The TR/FA scheme of the MRF sequence used parameters reported in previous literature ^{2, 3}. The raw data were reconstructed using singular value decomposition (SVD) based dictionary matching in MATLAB (MathWorks, Nattick, MA).
MRFT temperature sensitivity calibration experiment
Figure 1(a) shows the setup for the temperature sensitivity calibration experiment. The MR protocol for the calibration experiment consisted of a turbo spin echo (TSE) structure scan, followed by an MRF sequence every five minutes for two hours. The linear relationship between temperature (T) and T₁ can be expressed as in ref ¹:
$$T1 = m*T+b$$
Where m and b are tissue-specific and can be determined using calibration methods.
PRFS temperature sensitivity calibration experiment
Figure 1(b) shows the setup for the temperature sensitivity calibration experiment. We performed the calibration experiment using a SPoiled Gradient-Recalled (SPGR) sequence with the following parameters: TR=100 ms, TE=4.2 ms, flip angle=36°, FOV of 25 cm × 25 cm, slice thickness=5 mm, matrix size=225×225, excitation bandwidth=1.3 kHz, readout bandwidth=0.5 kHz/pixel, acquisition time=16 sec. The MR protocol for the calibration experiment consisted of a turbo spin echo (TSE) structure scan, followed by an SPGR sequence every five minutes for two hours.
We calculated Δs based on the equations presented in ref ^{4, 5}. The linear relationship between change in temperature (ΔT) and change in phase (Δɸ) can be expressed as:
$$Δs=⍺ΔT+b$$
Where ⍺ is the thermal coefficient in ppm/°C and b is a constant.
MRFT/PRFS thermometry experiment
We repeated the same setup in Figure 1, but with different bovine muscle samples. The experiment protocol included a turbo spin echo (TSE) structure scan, followed by an MRF sequence and an SPGR sequence every five minutes for two hours. We produced the temperature maps using the calibration equations described above. The results from both methods were compared with the gold standard fluoroptic probes measurements.

Results/Discussion

MRF temperature sensitivity calibration experiment
Figure 2(a) shows that the T₁ decreased as the heated bovine muscle (bottom) cooled down and T₁ increased as the cold bovine muscle (top) heated up for the first 110 minutes of the calibration experiment. The cold muscle showed longer T₁ than the heated muscle because the heated muscle lost water during the heating process which we drained. Figure 3(a) presents the best fit line for the center of cold muscle and the center and periphery of heated muscle and their respective R² statistics using simple linear regression. The R² values indicate a strong linear relationship (R²>0.9883). The slopes for probe 3 and probe 4 are very similar (5 ms/°C) because they are from the same heated bovine muscle. This linear relationship between temperature and T₁ agrees with previous literature ⁶.
PRFS temperature sensitivity calibration experiment
Figure 2(b) shows that the Δs increased as the heated bovine muscle (bottom) cooled down and Δs decreased as the cold bovine muscle (top) heated up for the first 110 minutes of the calibration experiment. Figure 3(b) presents the best fit line for the center of cold muscle and the center and periphery of heated muscle and their respective R² statistics using simple linear regression. The R² values indicate a strong linear relationship (R²>0.9583). The thermal coefficient ⍺ of cold bovine muscle is larger than heating muscle because heated muscle lost water during the heating process. Our values of ⍺ agree with previous literature ¹.
MRFT/PRFS thermometry experiment
Figure 4 shows the comparison of PRFS thermometry predicted temperature, MRFT predicted temperature, and fluoroptic probe measurements. The RMSE of PRFS and MRFT are presented in the figure. Errors may be attributed to temporal averaging during data processing. PRFS thermometry produced a smaller RMSE compared to MRFT. Heated muscle produced larger errors compared to cold muscle due to inhomogeneous heating for both methods.
Figure 5 presents the correlation plots between PRFS thermometry and MRFT predicted temperature for probe 3 and probe 4. We observed a strong correlation for probe 3 and probe 4 (R²>0.926). No correlation was performed for probe 1 because the change in temperature was too small. No meaningful correlation could be observed between PRFS thermometry and MRFT predicted temperature for probe 1.

Conclusion

In this study, we investigated the feasibility of measuring temperature using T₁-based MRF thermometry. The calibration experiments show strong correlations for both methods (R² > 0.958). We observed MRFT is able to predict temperature accurately (RMSE<1.3688 °C), similar to PRFS thermometry predicted temperature (RMSE<0.934 °C). PRFS thermometry and MRFT predicted temperature show strong correlations (R² > 0.926) for heated bovine muscle center and periphery. The T₁-based MRFT can mitigate challenges in PRFS thermometry such as motion and inaccuracies in adipose tissues.

Acknowledgements

This work was supported, in part, by GE-Columbia research partnership grant and also performed at Zuckerman Mind Brain Behavior Institute MRI Platform, a shared resource, and Columbia MR Research Center site. The subject matter, content, and views presented do not represent the views of the Department of Health and Human Services (HHS), US Food and Drug Administration (FDA), and/or the United States.