DE Seminar: David Stonko (Hopkins)

Title: An Obstacle Problem in the Operating Room: Predicting Stiff Guidewire-Induced Aortoiliac Deformation on SE(3); Can mathematical first principles inform a new surgical paradigm?

Abstract: An aortic stent is delivered over a guidewire that is stiffer than the artery into which it is placed. The operation, however, was planned on a three-dimensional CT scan taken before the wire went in. The procedure is then also performed in the new anatomy that the wire has changed, with a single two-dimensional X-ray projection as the intraoperative view. Can we model the wire-artery interaction from first principles, and where does that lead?

The vessel centerline and the wire are modeled as curves of frames in SE(3), with the wire modeled as a Cosserat rod. The vessel rests on an elastic foundation with a piecewise-constant anatomic anchoring stiffness. The wire and the vessel are coupled through a unilateral inequality keeping the wire inside the lumen. The model is an obstacle problem in which both curves deform such that the overall potential energy of the wire, vessel, and surrounding tissue is minimized. Can we predict this new, energy-minimized intraoperative anatomy from preoperative imaging plus knowledge of this physics? This is the setting for a proposed class of anatomy-informed neural networks, in which anatomic knowledge enters as a hard constraint in the state representation or as a soft penalty in the loss, analogous to how physics enters the loss function of a physics informed neural network. Supervision of this model on our application is a Wasserstein-2 loss against a 2-D fluoroscopic projection. 

Speaker: David Stonko (M.D. and M.S.) is a former UMBC Math&Stat BS/MS student who is now a surgeon at JHU.  He focus on vasculature, and is working on creating a PINN like ``Anatomy-Informed" neural network to assist with autonomous robotic navigation in the repair of blood vessels. He has a potential interest in developing collaboration at UMBC.