Premium
Poster — Thur Eve — 45: Three‐Dimensional US Probe Localization by Single Perspective Pose Estimation
Author(s) -
Lang P,
Seslija P,
Habets D,
Chu M,
Holdsworth D,
Peters T
Publication year - 2010
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.3476150
Subject(s) - fluoroscopy , image registration , computer vision , computer science , artificial intelligence , rotation (mathematics) , tracking (education) , gold standard (test) , medicine , radiology , image (mathematics) , psychology , pedagogy
Purpose: Intra‐operative fluoroscopy and transesophageal (TEE) ultrasound are imaging modalities commonly used in many cardiac procedures, including trans‐catheter aortic valve replacement. Fluoroscopy‐to‐ultrasound registration would enhance conventional image guidance by providing a common frame of reference in which both modalities can be viewed. An important component of this registration is 3D localization of the TEE probe with respect to the fluoroscopic image. Traditional approaches to this problem employed magnetic tracking systems, however these systems are hindered by metallic distortions and restrictive patient access within the operating room. Methods: Two 2D‐to‐3D registration techniques, a point‐based and intensity‐based technique, were implemented. These registration techniques determine the 3D pose of the TEE probe directly from single‐perspective fluoroscopy images, which facilitates the localization of both the probe and fluoroscopic image in a common frame of reference. In vitro experiments were performed to assess the accuracy of each registration technique. Measured displacements were compared against mechanical translation/rotation tables, utilized to provide a gold standard. Results: Maximum root‐mean‐square displacement and rotation errors were found to be 0.58mm, 0.32° and 2.29mm, 3.76° for point‐based and intensity‐based tracking techniques, respectively. The accuracy of the point‐based registration technique is significantly higher than the intensity‐based technique, but requires the use of a rigid tracking attachment. Conclusion: Localization of the TEE from single‐perspective fluoroscopy images provides an accurate means of intra‐operative fluoroscopy‐to‐ultrasound registration, and does not significantly interrupt the regular workflow within the operating room.