Generalized prediction framework for reconstructed image properties using neural networks

Jianan Gang, Kailun Cheng, Xueqi Guo, Joseph Webster Stayman

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Model-based reconstruction (MBR) algorithms in CT have demonstrated superior dose-image quality tradeoffs compared to traditional analytical methods. However, the nonlinear and data-dependent nature of these al- gorithms pose significant challenges for performance evaluation and parameter optimization. To address these challenges, this work presents an analysis framework for quantitative and predictive modeling of image proper- ties in general nonlinear MBR algorithms. We propose to characterize the reconstructed appearance of arbitrary stimuli by the generalized system response function that accounts for dependence on the imaging conditions, reconstruction parameters, object, and the stimulus itself (size, contrast, location). We estimate this nonlinear function using a multilayer perceptron neural network by providing input and output pairs that samples the range of imaging parameters of interest. The feasibility of this approach was demonstrated for predicting the appearance of a spiculated lesion reconstructed by a penalized-likelihood objective with a Huber penalty in a physical phantom as a function of its location and reconstruction parameters β and I. The generalized system response functions predicted from the trained neural network show good agreement with those computed from mean reconstructions, proving the ability of the framework in mapping out the nonlinear function for combinations of imaging parameters not present in the training data. We demonstrated utility of the framework to achieve desirable (e.g., non-blocky) lesion appearance in arbitrary locations in the phantom without the need for performing actual reconstructions. The proposed prediction framework permits efficient and quantifiable performance evaluations to provide robust control and understanding of image properties for general classes of nonlinear MBR algorithms.

Original languageEnglish (US)
Title of host publicationMedical Imaging 2019
Subtitle of host publicationPhysics of Medical Imaging
EditorsHilde Bosmans, Guang-Hong Chen, Taly Gilat Schmidt
PublisherSPIE
ISBN (Electronic)9781510625433
DOIs
StatePublished - Jan 1 2019
EventMedical Imaging 2019: Physics of Medical Imaging - San Diego, United States
Duration: Feb 17 2019Feb 20 2019

Publication series

NameProgress in Biomedical Optics and Imaging - Proceedings of SPIE
Volume10948
ISSN (Print)1605-7422

Conference

ConferenceMedical Imaging 2019: Physics of Medical Imaging
CountryUnited States
CitySan Diego
Period2/17/192/20/19

Fingerprint

Neural networks
Nonlinear Dynamics
predictions
Imaging techniques
Neural Networks (Computer)
stimuli
lesions
Multilayer neural networks
Robust control
Image quality
self organizing systems
hubs
evaluation
tradeoffs
penalties
education
dosage
optimization
output
estimates

ASJC Scopus subject areas

  • Electronic, Optical and Magnetic Materials
  • Atomic and Molecular Physics, and Optics
  • Biomaterials
  • Radiology Nuclear Medicine and imaging

Cite this

Gang, J., Cheng, K., Guo, X., & Stayman, J. W. (2019). Generalized prediction framework for reconstructed image properties using neural networks. In H. Bosmans, G-H. Chen, & T. G. Schmidt (Eds.), Medical Imaging 2019: Physics of Medical Imaging [109480L] (Progress in Biomedical Optics and Imaging - Proceedings of SPIE; Vol. 10948). SPIE. https://doi.org/10.1117/12.2513485

Generalized prediction framework for reconstructed image properties using neural networks. / Gang, Jianan; Cheng, Kailun; Guo, Xueqi; Stayman, Joseph Webster.

Medical Imaging 2019: Physics of Medical Imaging. ed. / Hilde Bosmans; Guang-Hong Chen; Taly Gilat Schmidt. SPIE, 2019. 109480L (Progress in Biomedical Optics and Imaging - Proceedings of SPIE; Vol. 10948).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Gang, J, Cheng, K, Guo, X & Stayman, JW 2019, Generalized prediction framework for reconstructed image properties using neural networks. in H Bosmans, G-H Chen & TG Schmidt (eds), Medical Imaging 2019: Physics of Medical Imaging., 109480L, Progress in Biomedical Optics and Imaging - Proceedings of SPIE, vol. 10948, SPIE, Medical Imaging 2019: Physics of Medical Imaging, San Diego, United States, 2/17/19. https://doi.org/10.1117/12.2513485
Gang J, Cheng K, Guo X, Stayman JW. Generalized prediction framework for reconstructed image properties using neural networks. In Bosmans H, Chen G-H, Schmidt TG, editors, Medical Imaging 2019: Physics of Medical Imaging. SPIE. 2019. 109480L. (Progress in Biomedical Optics and Imaging - Proceedings of SPIE). https://doi.org/10.1117/12.2513485
Gang, Jianan ; Cheng, Kailun ; Guo, Xueqi ; Stayman, Joseph Webster. / Generalized prediction framework for reconstructed image properties using neural networks. Medical Imaging 2019: Physics of Medical Imaging. editor / Hilde Bosmans ; Guang-Hong Chen ; Taly Gilat Schmidt. SPIE, 2019. (Progress in Biomedical Optics and Imaging - Proceedings of SPIE).
@inproceedings{b4bb58f045954971a242ad5867a010f1,
title = "Generalized prediction framework for reconstructed image properties using neural networks",
abstract = "Model-based reconstruction (MBR) algorithms in CT have demonstrated superior dose-image quality tradeoffs compared to traditional analytical methods. However, the nonlinear and data-dependent nature of these al- gorithms pose significant challenges for performance evaluation and parameter optimization. To address these challenges, this work presents an analysis framework for quantitative and predictive modeling of image proper- ties in general nonlinear MBR algorithms. We propose to characterize the reconstructed appearance of arbitrary stimuli by the generalized system response function that accounts for dependence on the imaging conditions, reconstruction parameters, object, and the stimulus itself (size, contrast, location). We estimate this nonlinear function using a multilayer perceptron neural network by providing input and output pairs that samples the range of imaging parameters of interest. The feasibility of this approach was demonstrated for predicting the appearance of a spiculated lesion reconstructed by a penalized-likelihood objective with a Huber penalty in a physical phantom as a function of its location and reconstruction parameters β and I. The generalized system response functions predicted from the trained neural network show good agreement with those computed from mean reconstructions, proving the ability of the framework in mapping out the nonlinear function for combinations of imaging parameters not present in the training data. We demonstrated utility of the framework to achieve desirable (e.g., non-blocky) lesion appearance in arbitrary locations in the phantom without the need for performing actual reconstructions. The proposed prediction framework permits efficient and quantifiable performance evaluations to provide robust control and understanding of image properties for general classes of nonlinear MBR algorithms.",
author = "Jianan Gang and Kailun Cheng and Xueqi Guo and Stayman, {Joseph Webster}",
year = "2019",
month = "1",
day = "1",
doi = "10.1117/12.2513485",
language = "English (US)",
series = "Progress in Biomedical Optics and Imaging - Proceedings of SPIE",
publisher = "SPIE",
editor = "Hilde Bosmans and Guang-Hong Chen and Schmidt, {Taly Gilat}",
booktitle = "Medical Imaging 2019",

}

TY - GEN

T1 - Generalized prediction framework for reconstructed image properties using neural networks

AU - Gang, Jianan

AU - Cheng, Kailun

AU - Guo, Xueqi

AU - Stayman, Joseph Webster

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Model-based reconstruction (MBR) algorithms in CT have demonstrated superior dose-image quality tradeoffs compared to traditional analytical methods. However, the nonlinear and data-dependent nature of these al- gorithms pose significant challenges for performance evaluation and parameter optimization. To address these challenges, this work presents an analysis framework for quantitative and predictive modeling of image proper- ties in general nonlinear MBR algorithms. We propose to characterize the reconstructed appearance of arbitrary stimuli by the generalized system response function that accounts for dependence on the imaging conditions, reconstruction parameters, object, and the stimulus itself (size, contrast, location). We estimate this nonlinear function using a multilayer perceptron neural network by providing input and output pairs that samples the range of imaging parameters of interest. The feasibility of this approach was demonstrated for predicting the appearance of a spiculated lesion reconstructed by a penalized-likelihood objective with a Huber penalty in a physical phantom as a function of its location and reconstruction parameters β and I. The generalized system response functions predicted from the trained neural network show good agreement with those computed from mean reconstructions, proving the ability of the framework in mapping out the nonlinear function for combinations of imaging parameters not present in the training data. We demonstrated utility of the framework to achieve desirable (e.g., non-blocky) lesion appearance in arbitrary locations in the phantom without the need for performing actual reconstructions. The proposed prediction framework permits efficient and quantifiable performance evaluations to provide robust control and understanding of image properties for general classes of nonlinear MBR algorithms.

AB - Model-based reconstruction (MBR) algorithms in CT have demonstrated superior dose-image quality tradeoffs compared to traditional analytical methods. However, the nonlinear and data-dependent nature of these al- gorithms pose significant challenges for performance evaluation and parameter optimization. To address these challenges, this work presents an analysis framework for quantitative and predictive modeling of image proper- ties in general nonlinear MBR algorithms. We propose to characterize the reconstructed appearance of arbitrary stimuli by the generalized system response function that accounts for dependence on the imaging conditions, reconstruction parameters, object, and the stimulus itself (size, contrast, location). We estimate this nonlinear function using a multilayer perceptron neural network by providing input and output pairs that samples the range of imaging parameters of interest. The feasibility of this approach was demonstrated for predicting the appearance of a spiculated lesion reconstructed by a penalized-likelihood objective with a Huber penalty in a physical phantom as a function of its location and reconstruction parameters β and I. The generalized system response functions predicted from the trained neural network show good agreement with those computed from mean reconstructions, proving the ability of the framework in mapping out the nonlinear function for combinations of imaging parameters not present in the training data. We demonstrated utility of the framework to achieve desirable (e.g., non-blocky) lesion appearance in arbitrary locations in the phantom without the need for performing actual reconstructions. The proposed prediction framework permits efficient and quantifiable performance evaluations to provide robust control and understanding of image properties for general classes of nonlinear MBR algorithms.

UR - http://www.scopus.com/inward/record.url?scp=85068372071&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85068372071&partnerID=8YFLogxK

U2 - 10.1117/12.2513485

DO - 10.1117/12.2513485

M3 - Conference contribution

C2 - 31007339

AN - SCOPUS:85068372071

T3 - Progress in Biomedical Optics and Imaging - Proceedings of SPIE

BT - Medical Imaging 2019

A2 - Bosmans, Hilde

A2 - Chen, Guang-Hong

A2 - Schmidt, Taly Gilat

PB - SPIE

ER -