Samuel Feemster
BME MS Thesis Defense Presentation
Date: 2026-11-23
Time: 12:30 pm - 2pm
Location / Meeting Link: https://teams.microsoft.com/meet/2538186996707723?p=IqrKZN4tFqfIe65sca

Committee Members:
Dr. Amir Pourmorteza (advisor), Dr. John Oshinski, Dr. Greeshma Agasthya, Dr. Sara Fridovich-Keil


Title: CT Guide-wire Localization from 2 Projections

Abstract:
Fluoroscopy provides fast and familiar catheter guidance, but physicians must relate the device’s position to disease characterized on a separate diagnostic CT or MRI. In embolization, this means connecting the lesion to its feeding arteries and to the tissue reached from the current catheter position. Guidance on a diagnostic CT scanner could show devices, vessels, and CT-visible targets in one scene. However, acquiring a new CT volume for every device update adds radiation dose and reconstruction time. To address this, a pipeline is developed to localize the guide wire directly from two selected projections without image reconstruction. This involves x-ray pulse programming, allowing for selection of discrete projections at chosen gantry angles. The guide wire is then segmented in each projection and triangulated using the known scanner geometry, producing a set of (x, y, z) coordinates that give the device’s location in 3D. Localization accuracy is then evaluated in scanner-specific simulations and in real CT data as a function of starting gantry angle and the angular separation between the two views, with the goal to find the angular separation that balances localization speed and accuracy. Current results are promising, finding submillimeter accuracy with a minimum separation angle of 9.9 degrees in simulation. Spectral imaging is also explored as a way to distinguish the guide wire from surrounding tissue using the energy information provided by simulated photon-counting CT. This work is a step toward rapid, lower-dose device tracking with CT, and future work will combine this localization pipeline with periodic CT scans allowing for clear and near continuous real-time guidance.