Jasmine Vu1,2, Bach T Nguyen2, Bhumi Bhusal2, Justin Baraboo1,2, Joshua Rosenow3, Ulas Bagci 2, Molly G Bright1,4, and Laleh Golestanirad1,2
1Biomedical Engineering, McCormick School of Engineering, Northwestern University, Evanston, IL, United States, 2Radiology, Feinberg School of Medicine, Northwestern University, Chicago, IL, United States, 3Neurosurgery, Feinberg School of Medicine, Northwestern University, Chicago, IL, United States, 4Physical Therapy and Human Movement Sciences, Feinberg School of Medicine, Northwestern University, Chicago, IL, United States
1Biomedical Engineering, McCormick School of Engineering, Northwestern University, Evanston, IL, United States, 2Radiology, Feinberg School of Medicine, Northwestern University, Chicago, IL, United States, 3Neurosurgery, Feinberg School of Medicine, Northwestern University, Chicago, IL, United States, 4Physical Therapy and Human Movement Sciences, Feinberg School of Medicine, Northwestern University, Chicago, IL, United States
The presented deep learning-based framework was
capable of predicting maximum SAR at the tips of implanted conductive leads with
only knowledge of the lead trajectory and the RF coil’s background field in
hand.
Figure 2: (A) Superposition of Etan
distributions along different lead trajectories. Similar Etan values
were observed along the intracranial regions across lead trajectories. (B)
Distribution of Etan (color field) and the incident electric field
(arrows) for lead trajectories that demonstrate low and high 1gSARmax values.
(C) A DNN was created to predict 1gSARmax using the extracranial Etan
values as inputs into the algorithm due to similarities in Etan values
in the intracranial regions.
Figure 4: (A) Performance of network training, as
indicated by the mean squared error, for all folds of five-fold
cross-validation. (B)
Predicted 1gSARmax values from testing
of the DNN plotted against the ground-truth 1gSARmax values from
EM simulations.
