PROJECT SUMMARY / ABSTRACT
C-arm cone-beam CT (CBCT) plays an increasing role in guidance of interventional radiology (IR) procedures in the abdo-
men, with special emphasis in embolization procedures, such as transarterial chemoembolization (TACE) for treatment of
hepatocellular carcinoma (HCC) or transarterial embolization (TAE) for control of internal hemorrhage. However, relatively
long scan time of CBCT results in artifacts arising from organ motion (respiratory and cardiac motion and peristalsis). This
poses a significant challenge to guidance in interventional radiology: for example, motion artifacts were found to render
up to 25% of CBCT images un-interpretable in image-guided TACE, and 18% in CBCT-guided emergency TAE. The impact
of motion is most significant in cases of single or isolated lesions treated with selective embolization that requires visual-
ization of very small vascular structures. Existing motion correction methods often invoke assumption of periodicity, lim-
iting their applicability outside of cardiac and respiratory motions, or rely on fiducial tracking or gated acquisition that
disrupt IR workflow and/or increase radiation dose. Therefore, the application of CBCT in image-guided interventional
procedures in the abdomen would significantly benefit from new methods that estimate complex deformable motion
directly from image data. “Autofocus” techniques based on maximization of a regularized image sharpness criterion were
shown to yield effective patient motion compensation in extremity, head and cardiac CBCT. However, current applications
of such methods are limited to rigid motions. We hypothesize that deformable organ motion compensation in interven-
tional soft-tissue CBCT can be achieved with advanced autofocus techniques using multiple locally rigid regions of in-
terest, preconditioned with basic motion characteristics obtained through a machine learning decision framework. The
following aims will be pursued: 1) Develop a joint multi-region autofocus optimization method to compensate deforma-
ble organ motion. This includes incorporation into a comprehensive artifacts correction and image reconstruction pipe-
line, design of multi-stage optimization schedules for convergence acceleration, and performance evaluation in deforma-
ble phantoms, and cadaver and animal experiments. 2) Develop a decision framework for preconditioning of the motion
compensation method through a combination of projection-based approaches for physiological signal estimations (res-
piratory cycle) and a multi-input, multi-branch, deep learning architecture trained on extremely realistic simulated data
that will estimate basic properties of motion (spatial distribution of amplitude, direction, and frequency) from an initial
motion-contaminated image and its associated raw projection data. 3) Evaluate deformable motion compensation in
animal experiments and in a clinical study in 50 cases of CBCT-guided TACE and assess image quality via expert observer
evaluation of satisfaction and utility. The proposed work will yield a robust, practical method for compensation of deform-
able soft-tissue motion in CBCT, removing a critical impediment to 3D guidance in IR. The deformable autofocus frame-
work will be applicable to other interventions in which soft-tissue motion diminishes CBCT guidance, such as image-guided
radiation therapy.