Efficient automatic 2D/3D registration of cardiac ultrasound and CT images

ABSTRACT

Hybrid ablations are a promising but difficult intervention for the treatment of atrial fibrillation. With the ultimate goal of providing navigation support for such procedures, we investigate a registration algorithm for routinely available preoperative and intraoperative images. We propose a fully automatic segmentation algorithm for the boundaries of cardiac chambers in intraoperative TEE ultrasound using a generic heart model. The resulting ultrasound segmentations are initially registered to the preoperative CT model using a frame-to-slice search, which is then refined using an efficient continuous optimisation. Results are presented for data sets from three patients who underwent hybrid ablations at our institution. The mean time to process a single ultrasound image from segmentation to registration with CT was 1.5 s, with the mean RMS error across the sequence being 4.8 mm. With further validation, these results show promise for surgical navigation in hybrid ablation procedures.

KEYWORDS:

• Image registration
• echocardiography
• cardiac CT
• surgical navigation
• cardiac surgery

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. The R-R interval is the time between peaks of an ECG signal.

Additional information

Funding

This work was supported in part by the Natural Sciences and Engineering Research Council of Canada under Grant [RGPIN-2018-04430].

Notes on contributors

Katy Scott

Katy Scott completed her Bachelor's of Computing Honours with a Specialization in Biomedical Computing at Queen's University, Kingston, Ontario, Canada in 2020. She is currently a Masters candidate at Queen's University in Computing and the Field of Study in Artificial Intelligence program. She is also a Vector Scholarship in Artificial Intelligence Recipient 2020-21.

Duncan Stuart

Duncan Stuartgraduated from the Queen's University School of Computing with a Bachelor's degree in computing in 2020. Currently, they are a Master's candidate in the Artificial Intelligence program at the School of Computing and hold a Vector Scholarship in Artificial Intelligence for 2020-2021.

Jacob J. Peoples

Jacob Peoples received his Bachelor of Science (Honours) with Specialization in Mathematical Physics from Queen's University in 2014 and his doctoral degree in Computing in 2020.  He is now a post-doctoral fellow at Queen's University, specializing in artificial intelligence and image analysis.

Gianluigi Bisleri

Gianluigi Bislerigraduated from University of Brescia Medical School, Brescia, Italy summa cum laude in 2001. Following residency in cardiac surgery at University of Padua, he completed a research fellowship in arrhythmia and minimally invasive/robotic surgery at the Presbyterian Medical Center, New York. From 2006-2016 he was a cardiac surgeon at University of Brescia Medical School, where he became an Associate Professor of Surgery. Dr. Bisleri joined Queen’s University and the Kingston Health Sciences Centre in 2016 where he continues his interest in cardiac arrhythmia and minimally invasive cardiac surgery.

Randy E. Ellis

Randy Ellis received his doctorate in Computer Science from the University of Massachusetts at Amherst.  He is a Professor in the School of Computing at Queen's University, with cross-appointments to the Department of Surgery, the Department of Mechanical and Materials Engineering, and the Department of Biomedical and Molecular Sciences. He is the Queen's Research Chair in Computer Assisted Surgery.