Cardiovasc Imaging Asia.  2019 Apr;3(2):35-46. 10.22468/cvia.2018.00248.

Ideal Bolus Geometry Predicted from In vitro Pulsatile Flow Phantom and Artificial Neural Networks for the Optimization of Image Acquisition Protocols for Aortic Contrast-Enhanced Computed Tomography Angiography

Affiliations
  • 1Department of Radiology, Catholic University of Daegu Medical Center, School of Medicine, Catholic University of Daegu, Daegu, Korea.
  • 2Department of Biomedical Engineering, School of Medicine, Kyungpook National University, Daegu, Korea. jonglee@knu.ac.kr
  • 3Department of Radiology, Kyungpook National University Hospital, School of Medicine, Kyungpook National University, Daegu, Korea.

Abstract


OBJECTIVE
This study sought to explore the novel use of artificial neural networks (ANNs) to develop a contrast-enhanced computed tomography (CT) angiographic (CECTA) protocol based on ideal bolus geometry.
MATERIALS AND METHODS
An aortic phantom connected to a closed-circuit pulsatile flow system was developed to simulate the bolus geometry of the human abdominal aorta. A total of 135 CECTA datasets were obtained using a 16-row multidetector CT scanner, and timeenhancement curves (TECs) were generated using varying input conditions including heart rate (HR), iodine delivery rate (IDR) and concentration (IC), and tube potential (kVP). Time points and density values including peak enhancement (PE) and time-to-peak (TTP) were assessed as a function of injection and scan protocols. Statistical analysis was performed using correlation and linear regression analyses. By using data from phantom experiments, machine learning produced networks between four input (HR, IC, IDR, and kilovoltage) and five output [TTP-time-to-foot (TTF), PE, (PE)/(TTT-TTF), maximal-upslope-gradient (MUG), and peak-plateau-length (PPL)] conditions. The bolus geometry index was defined as (TTT-TTF)/Σ(PE, (PE)/(TTT-TTF), MUG, PPL). The lowest bolus geometry index value was considered ideal in ANN testing.
RESULTS
The geometric changes on TECs were observed based on changes in HR, IDR, IC, and kilovoltage value. PE was closely related to IDR (B=17.471) and kVp (B=−0.208) (corrected R²=0.919; all p<0.001). TTF, TTP, and PPL were related to HR and IDR, respectively. HR and IDR remained contributing factors after multiple linear regression analysis (corrected R²=0.901, 0.815, and 0.363; all p<0.001). Among 39690 total datasets produced following ANN training, the combination of IC, HR, tube potential, and IDR in the 38010th dataset resulted in the lowest bolus geometry index. Tables of input variables are presented after modification to clinically acceptable ranges.
CONCLUSION
ANN of phantom experiments showed the potential to determine optimal CECTA parameters for ideal bolus geometry individualized for each subject.

Keyword

CT angiography; Neural network; CT protocol; Imaging phantom; Aorta

MeSH Terms

Angiography*
Aorta
Aorta, Abdominal
Dataset
Heart Rate
Humans
In Vitro Techniques*
Iodine
Linear Models
Machine Learning
Phantoms, Imaging
Pulsatile Flow*
Iodine
Full Text Links
  • CVIA
Actions
Cited
CITED
export Copy
Close
Share
  • Twitter
  • Facebook
Similar articles
Copyright © 2024 by Korean Association of Medical Journal Editors. All rights reserved.     E-mail: koreamed@kamje.or.kr