Meeting Banner
Abstract #4302

Predictive Magnetic Resonance Temperature Imaging with Machine Learning

Joshua P. Yung1, 2, Christopher J. MacLellan1, 2, Anil Shetty3, Roger J. McNichols3, John D. Hazle1, 2, R. Jason Stafford1, 2, David T. A. Fuentes1, 2

1Imaging Physics, The University of Texas M.D. Anderson Cancer Center, Houston, TX, United States; 2The University of Texas Graduate School of Biomedical Sciences, Houston, TX, United States; 3Visualase, Inc, Houston, TX, United States

In previous studies, the Pennes bioheat transfer equation was used to predict temperature heating during thermal therapy. In this work, a non-physical model that takes a priori temperature measurements to predict future heating is used. The method was tested on in vivo data of laser induced interstitial thermal therapy of human brain. The prediction was run using two a priori time steps and three a priori time steps. In conclusion, the proposed method predicted future temperature values with uncertainty values allowing for confidence intervals. The mean and standard deviation values can offer additional information for the procedure.

Keywords

ability academic accelerating accuracy actual adjacent agree amount artifact artifacts assess background biomedical blue brain calculating cancer comparing compute conductivity confidence consecutive contaminated contaminations cooling correction corrupted covariance dependent deviation distribution dynamically easily equation estimating examined exceeding facilitating fall fiber field free fully function functions future gives good gradient gradients graduate guided heating human identifying immediate includes induced inputs instead interstitial interval investigated john laser learning limited limits located look loss machine magnitude maps materials maximizing measuring mesh minimize missing model modeled modeling models monitoring motion necessary needed neurosurgery noise optical optimize optimized outside perturbed physics plotted portion posterior predict predicted predicting predictions predictive presence press prevent previous prior priori probable procedure process processing profile profiles providing real relies roger safety scattering school sciences seem simulate static statistics steepest steps studies subsequent susceptibility system temperature temporal therapies therapy thermal tissue train training transfer treatment treatments undergoing unlike update useful variance variances