Purpose

To evaluate the possibility of performing a temperature-based safety assessment with a simplified body model.

Introduction

Absorption of radiofrequency power causes tissue heating, which can be source of discomfort or even tissue damage in an MRI patient. For this reason safety guidelines limit the maximum whole-body and local power absorbed (SAR), and the maximum core and local temperature reached during an MRI exam. SAR and temperature values can be predicted through numerical simulations where the geometry of the system, including patient and coil geometries, is accurately represented. Different human subjects result in different field distributions in the patient, and consequently different SAR and temperature distributions: therefore, it is important to use body models as similar as possible to the shape of the individual patient to have accurate estimations of SAR and temperature. Although a selection of body models of different shapes and ages is available, an actual representation of the body of the specific patient should provide the optimal model for accurate simulations. It was previously shown that automatic segmentation tools can be used with fast MRI sequences to provide a simplified segmented model of the patient where all the tissues have been assigned to three different categories: muscle, fat and lungs¹. In the same work, it was also shown that with the simplified segmentation, the SAR distribution was still similar to the one obtained with a model representing many more tissues of the body¹. In this work we compare the temperature distributions when a detailed body model and when a simplified body model are simulate within a birdcage coil.

Methods

Temperature distributions were computed with a combination of Pennes’ bioheat equation² with consideration of effects on core body temperature³ and including various physiological responses⁴. Core body temperature was allowed to depend on many parameters such as whole-body SAR, respiration, perspiration, heat lost by radiation, conduction and convection. Blood perfusion was allowed to vary according to local temperature.Temperature distributions were computed for three different body models: 1) the original body model where all 35 tissues are represented; 2) the simplified model where only 3 tissues are present: muscles, fat and lungs; 3) the simplified model where 4 tissues are present: muscles, fat, lungs and skin (Figure 1). For all the three different models, first an equilibrium temperature with no SAR exposure was computed, and then the SAR from a 16 legs birdcage coil, normalized so that the whole-body SAR is equal to 2 W/kg (maximum allowed normal mode value in safety guidelines), was applied for 1 hour.

Results and Discussion

Figure 2 shows the baseline temperature, the temperature increase, and the maximum temperature at the location where the maximum temperature occurs for the three body models. In all three cases, the maximum temperature occured in the shoulder of the patient. Temperature distributions before and after the SAR application are reported in Figure 3. Although temperature distributions are still similar, the simplification of the model affects temperature more than it does SAR. This is mainly due to the impact of physiological parameters on temperature. In fact, while SAR is affected only by electrical properties of the tissues, temperature is also conditioned by thermal properties. In particular, blood perfusion has a significant impact on temperature distribution, especially in brain and other highly-perfused tissues. We found that automated addition of a thin layer of skin helps to increase temperature in the extremities to bring them closer to the values in the original model. Although baseline value is lower in the simplified models, lower average perfusion causes faster temperature rise in the simplified model, especially in the brain.

Conclusions

Simplified models can be used for patient-specific MRI safety assessment based on temperature. In future work, incorporation of many readily-available tools for identifying and segmenting brain should increase accuracy of automatically-segmented models further.

Acknowledgements

This work has benefitted from funding by the National Institutes of Health through NIH P41 EB017183 (CAI2R)