Share Email Print

Proceedings Paper

Deep learning based guidewire segmentation in x-ray images
Author(s): Martin G. Wagner; Paul Laeseke; Michael A. Speidel
Format Member Price Non-Member Price
PDF $17.00 $21.00

Paper Abstract

X-ray fluoroscopy is commonly used during liver embolization procedures to guide intravascular devices (e.g. guidewire and catheter) to the branches of the hepatic arteries feeding tumors. A vascular roadmap can be created to provide a reference for the device position. Recently, techniques have been developed to create dynamic vessel masks to compensate for respiratory motion. In order to superimpose the intravascular guidewire onto the vessel mask, robust segmentation is required. Commonly used techniques often use mask subtraction to isolate the device in x-ray images. However, this is not suitable due to the motion in liver applications. The proposed method uses a deep convolutional neural network to segment the guidewire in native (unsubtracted) x-ray images. The neural network uses an encoder / decoder structure, which is based on the VGG-16 network. To create a large dataset of annotated images, simulated images were created based on 3D digital subtraction angiography acquisitions of hepatic arteries in porcine studies. Random guidewire shapes were generated within the vascular volume and superimposed on the original non-contrast projection images. The network was trained using a set of 56,768 images created from 10 acquisitions. The segmentation results of the trained network were compared to a mask-subtraction-based algorithm for an independent validation data set. The deep learning algorithm (Dice = 58.1%, false negative rate (FNR) = 9.6%) outperformed the subtraction technique (Dice = 23.7%, FNR = 40.8%). This study shows that the deep learning approach is suitable for robust segmentation of curvilinear structures such as guidewires and could be used to superimpose the segmented device on dynamic roadmaps.

Paper Details

Date Published: 1 March 2019
PDF: 7 pages
Proc. SPIE 10948, Medical Imaging 2019: Physics of Medical Imaging, 1094844 (1 March 2019); doi: 10.1117/12.2512820
Show Author Affiliations
Martin G. Wagner, Univ. of Wisconsin-Madison (United States)
Paul Laeseke, Univ. of Wisconsin-Madison (United States)
Michael A. Speidel, Univ. of Wisconsin-Madison (United States)

Published in SPIE Proceedings Vol. 10948:
Medical Imaging 2019: Physics of Medical Imaging
Taly Gilat Schmidt; Guang-Hong Chen; Hilde Bosmans, Editor(s)

© SPIE. Terms of Use
Back to Top
Sign in to read the full article
Create a free SPIE account to get access to
premium articles and original research
Forgot your username?