Uncertainty Quantification by Generalized Polynomial Chaos for MR-Guided Laser Induced Thermal Therapy
S Fahrenholtz1,2*, D Fuentes1, R Stafford1,2, J Hazle1,2,(1) The University of Texas MD Anderson Cancer Center, Houston, TX,(2) The University of Texas Graduate School of Biomedical Sciences at Houston, Houston, TXSU-F-BRB-8 Sunday 4:00:00 PM - 6:00:00 PM Room: Ballroom B
Purpose: Magnetic resonance-guided laser induced thermal therapy (MRgLITT) is a minimally invasive thermal treatment for metastatic brain lesions, offering an alternative to conventional surgery. The purpose of this investigation is to incorporate uncertainty quantification (UQ) into the biothermal parameters used in the Pennes bioheat transfer equation (BHT), in order to account for imprecise values available in the literature. The BHT is a partial differential equation commonly used in thermal therapy models.
Methods: MRgLITT was performed on an in vivo canine brain in a previous investigation. The canine MRgLITT was modeled using the BHT. The BHT has four parameters'" microperfusion, conductivity, optical absorption, and optical scattering'"which lack precise measurements in living brain and tumor. The uncertainties in the parameters were expressed as probability distribution functions derived from literature values. A univariate generalized polynomial chaos (gPC) expansion was applied to the stochastic BHT. The gPC approach to UQ provides a novel methodology to calculate spatio-temporal voxel-wise means and variances of the predicted temperature distributions. The performance of the gPC predictions were evaluated retrospectively by comparison with MR thermal imaging (MRTI) acquired during the MRgLITT procedure in the canine model. The comparison was evaluated with root mean square difference (RMSD), isotherm contours, spatial profiles, and z-tests.
Results: The peak RMSD was ~1.5 standard deviations for microperfusion, conductivity, and optical absorption, while optical scattering was ~2.2 standard deviations. Isotherm contours and spatial profiles of the simulation's predicted mean plus or minus two standard deviations demonstrate the MRTI temperature was enclosed by the model's isotherm confidence interval predictions. An a = 0.01 z-test demonstrates agreement.
Conclusions: The application of gPC for UQ is a potentially powerful means for providing predictive simulations despite poorly known input parameters. gPC provides an output that represents the probable distribution of outcomes for MRgLITT.