Real-time guiding catheter and guidewire detection for congenital cardiovascular interventions

TY - CHAP

T1 - Real-time guiding catheter and guidewire detection for congenital cardiovascular interventions

AU - Ma, Ying Liang

AU - Alhrishy, Mazen

AU - Panayiotou, Maria

AU - Ananth Narayan, Siri

AU - Fazili, Ansab

AU - Mountney, Peter

AU - Rhode, Kawal

PY - 2017/5/23

Y1 - 2017/5/23

N2 - Guiding catheters and guidewires are used extensively in pediatric cardiac catheterization procedures for congenital heart diseases (CHD). Detecting their positions in fluoroscopic X-ray images is important for several clinical applications, such as visibility enhancement for low dose X-ray images, and co-registration between 2D and 3D imaging modalities. As guiding catheters are made from thin plastic tubes, they can be deformed by cardiac and breathing motions. Therefore, detection is the essential step before automatic tracking of guiding catheters in live X-ray fluoroscopic images. However, there are several wire-like artifacts existing in X-ray images, which makes developing a real-time robust detection method very challenging. To solve those challenges in real-time, a localized machine learning algorithm is built to distinguish between guiding catheters and artifacts. As the machine learning algorithm is only applied to potential wire-like objects, which are obtained from vessel enhancement filters, the detection method is fast enough to be used in real-time applications. The other challenge is the low contrast between guiding catheters and background, as the majority of X-ray images are low dose. Therefore, the guiding catheter might be detected as a discontinuous curve object, such as a few disconnected line blocks from the vessel enhancement filter. A minimum energy method is developed to trace the whole wire object. Finally, the proposed methods are tested on 1102 images which are from 8 image sequences acquired from 3 clinical cases. Results show an accuracy of 0.87 ± 0.53 mm which is measured as the error distances between the detected object and the manually annotated object. The success rate of detection is 83.4%.

AB - Guiding catheters and guidewires are used extensively in pediatric cardiac catheterization procedures for congenital heart diseases (CHD). Detecting their positions in fluoroscopic X-ray images is important for several clinical applications, such as visibility enhancement for low dose X-ray images, and co-registration between 2D and 3D imaging modalities. As guiding catheters are made from thin plastic tubes, they can be deformed by cardiac and breathing motions. Therefore, detection is the essential step before automatic tracking of guiding catheters in live X-ray fluoroscopic images. However, there are several wire-like artifacts existing in X-ray images, which makes developing a real-time robust detection method very challenging. To solve those challenges in real-time, a localized machine learning algorithm is built to distinguish between guiding catheters and artifacts. As the machine learning algorithm is only applied to potential wire-like objects, which are obtained from vessel enhancement filters, the detection method is fast enough to be used in real-time applications. The other challenge is the low contrast between guiding catheters and background, as the majority of X-ray images are low dose. Therefore, the guiding catheter might be detected as a discontinuous curve object, such as a few disconnected line blocks from the vessel enhancement filter. A minimum energy method is developed to trace the whole wire object. Finally, the proposed methods are tested on 1102 images which are from 8 image sequences acquired from 3 clinical cases. Results show an accuracy of 0.87 ± 0.53 mm which is measured as the error distances between the detected object and the manually annotated object. The success rate of detection is 83.4%.

UR - http://www.scopus.com/inward/record.url?scp=85020416008&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-59448-4_17

DO - 10.1007/978-3-319-59448-4_17

M3 - Other chapter contribution

AN - SCOPUS:85020416008

SN - 9783319594477

VL - 10263 LNCS

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 172

EP - 182

BT - Functional Imaging and Modelling of the Heart - 9th International Conference, FIMH 2017, Proceedings

PB - Springer Verlag

T2 - 9th International Conference on Functional Imaging and Modelling of the Heart, FIMH 2017

Y2 - 11 June 2017 through 13 June 2017

ER -