Feasibility study of subject-specific, brain specific-absorption-rate maps retrieved from MRI data

Abstract

Introduction: Specific absorption rate (SAR) is crucial for monitoring radiofrequency power absorption during MRI. Although local SAR distribution is usually calculated through numerical simulations, they are impractical during exams, limiting real-time patient-specific SAR assessment. This study confirms the feasibility of deriving in vivo, subject-specific, image-based SAR and 10-g SAR maps directly from MRI data.

Methods: Complex B₁ ⁺ maps were derived by combining a B₁ ⁺ product (XFL) magnitude sequence with balanced steady-state free precession phase. Anatomical information and tissue masking were obtained from a T₁ magnetization-prepared rapid gradient echo sequence. Electrical conductivity maps were generated from balanced steady-state free precession phase. Whole-brain SAR maps were created from MRI data acquired at 3 T using a 32-channel head coil on 2 healthy volunteers. A correction factor was applied to account for underestimation due to reliance on measurable B₁ ⁺ data. Numerical simulations compared image-based SAR with simulation-based SAR distributions.

Results: A multi-slice image-based brain SAR map was obtained in 12 min (9-min acquisition, 3-min SAR reconstruction). In vitro experiments validated B₁ ⁺ distribution and electrical conductivity values. Calculated electrical conductivities for in vitro and in vivo experiments were within reference ranges. Image-based SAR and 10-g SAR maps showed a distribution similar to simulation-based maps (r = 0.5) after correction.

Conclusions: This study shows the feasibility of inline, subject-specific SAR and 10-g SAR maps from standard brain clinical sequences. Image-based SAR maps can be a practical alternative during MRI exams when simulations are not feasible.

Keywords: EPT; RF heating; SAR.

FIGURE 1

Workflow for deriving image‐based, subject‐specific specific absorption rate (SAR) maps. (A) A magnetization‐prepared rapid gradient echo (MPRAGE) sequence was used as an anatomical reference due to its high tissue contrast. (B) The phase of a balanced steady‐state free precession (bSSFP) sequence was used to retrieve electrical conductivity using a phase‐based Helmholtz–electrical properties tomography (EPT) approach. Mean electrical properties (conductivity and permittivity) maps per tissue were used to mitigate noise in subsequent steps. (C) The electric field was derived from the complex B₁ field, constructed from the bSSFP and a B₁ ⁺ magnitude sequence (in this case, B₁ ⁺ product [XFL]) data along with the mean electrical properties. (D) SAR and 10‐g SAR were calculated assuming a tissue density of 1000 kg/m³. *The permittivity was obtained from literature sources. EP, electrical property.

FIGURE 2

In vitro B₁ ⁺ magnitude (A) and phase (B) distribution plots for the image‐based and simulation‐based method. (C) Electrical conductivity distribution plot per vial within the region of interest (orange denotes values obtained with a dielectric probe kit). Qualitative maps are shown below the plots.

FIGURE 3

(A) In silico virtual model database (ground‐truth) and Helmholtz–electrical properties tomography (EPT) conductivity reconstruction from simulated B₁ ⁺. (B) Specific absorption rate (SAR) maps obtained from the simulated B₁ ⁺ and ground‐truth simulated SAR maps. The B₁ ⁺ SAR maps were corrected using a correction factor. (C) 10‐g SAR maps obtained from the simulated B₁ ⁺, and ground‐truth 10‐g SAR maps derived from the simulated SAR maps. (D) Difference SAR and 10‐g SAR maps and two‐dimensional voxel wise heatmap for the B₁ ⁺ corrected and the simulation‐based method.

FIGURE 4

In vivo singe central slice qualitative image‐based b ₁ ⁺ phase, Helmholtz–electrical properties tomography (EPT) electrical conductivity, non‐corrected and corrected image‐based specific absorption rate (SAR) maps, and simulation‐based SAR map for the healthy volunteers. SAR maps were calculated using the Helmholtz‐EPT electrical conductivity. Image‐based SAR maps present a similar distribution compared with the simulation‐based SAR maps.

FIGURE 5

Single‐slice b ₁ ⁺ magnitude maps obtained from image‐based and simulation‐based methods for both volunteers, alongside two‐dimensional heatmaps comparing the two methods on a voxel‐wise basis.

FIGURE 6

(A,C) Distribution plots for the image‐based and simulation‐based specific absorption rate (SAR) maps. Corrected image‐based SAR maps align more closely with the distribution of the simulation‐based SAR maps than the uncorrected image‐based SAR maps. (B,D) Two‐dimensional voxel‐wise heatmap for the image‐based corrected SAR versus simulation‐based SAR. Median [interquartile range] values for the uncorrected image–based, corrected image–based, and simulation‐based SAR maps were 1.2 [0.8] W/kg, 2.5 [1.9] W/kg, and 2.6 [1.0] W/kg for Volunteer 1, and 1.1 [0.7] W/kg, 2.3 [1.7] W/kg, and 2.2 [1.5] W/kg for Volunteer 2, respectively.

FIGURE 7

The 10‐g specific absorption rate (SAR) maps and projections in the center, front, back, right, and left views for the uncorrected and corrected image‐based and the simulation‐based methods for the healthy volunteers.

FIGURE 8

(A,C) Distribution plot for the image‐based and simulation‐based 10‐g specific absorption rate (SAR) maps. (B,D) Two‐dimensional voxel wise heatmap for the image‐based corrected 10‐g SAR versus simulation 10‐g SAR. Peak 10‐g SAR values were 5.22 W/kg (image‐based) and 5.04 W/kg (simulation‐based) for Volunteer 1 and 6.11 W/kg (image‐based) and 6.15 W/kg (simulation‐based) for Volunteer 2.

FIGURE 9

Method comparisons for specific absorption rate (SAR) and 10‐g SAR mapping. (A,B) Image‐based method comparison: simulation‐based versus MRI image–based SAR (A) and 10‐g SAR (B) maps. (C,D) Simulation‐based method comparison: complete B₁ field versus B₁ ⁺‐only simulations for SAR (C) and 10‐g SAR (D) maps. The simulation‐based approach uses the complete B₁ field (including magnitude and phase information), whereas the image‐based method (A,B) uses B₁ ⁺ values retrieved from MRI data, and the B₁ ⁺‐based simulation method (C,D) uses only the B₁ ⁺ component from simulations. Note that the colormap limits for SAR and 10‐g SAR are different.