Automated Volumetric Skull Segmentation Utilizing Ultra Short Echo Time Magnetic Resonance Imaging for MRgFUS
Singh, Samarth, Biomedical Engineering - School of Engineering and Applied Science, University of Virginia
Miller, Wilson, MD-RADL Rad Research, University of Virginia
Transcranial Magnetic Resonance guided Focused Ultrasound (tcMRgFUS) is a noninvasive treatment method which involves deposition of sonic energy using ultrasonic beams in the neurocranium region to achieve thermal effects (e.g. tissue ablation, hyperthermia), mechanical effects(e.g. blood-brain barrier permeabilization,clot dissolution, tissue homogenization), and physiological effects (e.g. neuromodulation). The skull, made out of predominantly cancellous and cortical bones, presents itself as an acoustic barrier to this ultrasonic energy. The skull absorbs ultrasound to an extent, heating up and also causing the sonic beams to scatter from the intended target. It is extremely critical from a treatment standpoint to ensure the focal point decided upon for thermal energy deposition is the only region which is sonicated, since the treatment efficacy and outcome depends on the same. Current skull extraction methods involve a prior computed tomography(CT) of the patient undergoing treatment, which provides a bone map for correcting the Focal Ultrasound transducers' phases to ensure accurate energy deposition. Previous work has shown Ultrashort Echo Time(UTE) Magnetic Resonance(MR) imaging allows us to visualize the skull in a patient, and sonication efficacy was improved for tcMRgFUS procedures with accurate, albeit ad-hoc estimation of the skull. In this work, we propose a completely automated volumetric skull extraction algorithm that operates on UTE MR scans of a patient, and extracts the skull utilizing stochastic image modeling and segmentation techniques such as Gaussian mixture modeling, expectation maximization, Markov random fields and least squares optimization.UTE MR volumes were compared with co-registered CT ground truth for 7 subjects who underwent essential tremor (ET) treatment. Dice coefficients varying from 0.721 to 0.817 were obtained.This work has the potential of integration in a clinical workflow for tcMRgFUS treatment, saving costs both in money and time for the patient, while involving zero harmful ionizing X-Ray radiation.
MS (Master of Science)
Transcranial MRgFUS, Stochastic Image Processing, Bias Field Correction, Ultra Short Echo Time (UTE)
A Thesis submitted for an MS in Biomedical Engineering by Samarth Singh. This research focused on skull segmentation utilizing only ultra short echo time (UTE) MRI image volumes. Retrospective bias field correction, stochastic image processing concepts such as expectation maximization algorithm, Gaussian mixture models and Markov random field models were combined to yield promising skull segmentation results.