Accelerated statistical image reconstruction methods for X-ray CT X-ray CT provides valuable imaging information for numerous medical applications. Increasing use of CT scans con- tributes significantly to population radiation dose. Decreasing X-ray CT dose will require advances in data acquisition, image reconstruction, post processing, changes in acquisition protocols, and elimination of unneeded scans. Statistical image re- construction (SIR) methods have been used routinely in nuclear medicine imaging for over a decade and have begun to be applied to reduced dose clinical X-ray CT scans. The types of SIR methods that are likely to be most suitable for sub-mSv CT scans use accurate models for the physics and statistics of X-ray CT systems;these methods have required very long computation times that impede their routine clinical use. The primary goal of this project is to develop significantly faster SIR algorithms that will enable routine use of SIR methods for all types of CT scans, particularly sub-mSv scans. The methods developed will be applicable to a wide variety of image acquisition geometries and statistical cost functions, and thus will complement advances in these components by other researchers. Achieving significant acceleration of SIR requires much more than simply waiting for advances in computer technology. Clock frequencies are no longer increasing, and advances in CT technology (e.g., dual energy, dual sources, and wider cone angles) continue to increase CT data sizes. Traditional SIR algorithms have been designed mathematically with little consideration of modern computing architectures. In this project, imaging scientists and computer scientists will collaborate to develop, implement, and evaluate SIR algorithms that are tailored to modern many-core computing systems that provide parallelism at multiple scales (instruction level, core level, and node level). The goal is to achieve SIR compute times of less than 5 minutes for routine helical chest CT scans at sub-mSv doses, to enable universal use of SIR methods. The methods developed will benefit all anatomic imaging sites, clinical applications, and patient populations. This project's investigation will focus on a large collection of patient lung nodule C scans that were acquired at the University of Michigan at both 80% and 20% of the usual 40-80 mA (depending on patient size) tube currents for such scans. The 80% dose scans are typically in the 1-1.5 mSv range whereas the 20% dose scans are in the 0.25-0.4 mSv range. The archived sinograms in this collection provide a unique and valuable resource for investigating sub-mSv chest CT and for comparing to regular dose chest CT scans. Both numerical observer studies and radiologist observer studies will evaluate detection performance and morphological characterization of lung nodules with advanced SIR methods at sub-mSv doses. These lung studies are particularly timely in light of the recent USPSTF draft recommendation [1, 2] that is likely to lead to substantial increased use of annual CT exams for patients at risk of lung cancer.

National Institute of Health (NIH)
National Institute of Biomedical Imaging and Bioengineering (NIBIB)
Research Project--Cooperative Agreements (U01)
Project #
Application #
Study Section
Special Emphasis Panel (ZEB1-OSR-D (M1))
Program Officer
Sastre, Antonio
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
University of Michigan Ann Arbor
Engineering (All Types)
Schools of Engineering
Ann Arbor
United States
Zip Code
Zheng, Xuehang; Ravishankar, Saiprasad; Long, Yong et al. (2018) PWLS-ULTRA: An Efficient Clustering and Learning-Based Approach for Low-Dose 3D CT Image Reconstruction. IEEE Trans Med Imaging 37:1498-1510
Kim, Donghwan; Fessler, Jeffrey A (2017) On the Convergence Analysis of the Optimized Gradient Method. J Optim Theory Appl 172:187-205
Ravishankar, Saiprasad; Nadakuditi, Raj Rao; Fessler, Jeffrey A (2017) Efficient Sum of Outer Products Dictionary Learning (SOUP-DIL) and Its Application to Inverse Problems. IEEE Trans Comput Imaging 3:694-709
Schmitt, Stephen M; Goodsitt, Mitchell M; Fessler, Jeffrey A (2017) Fast Variance Prediction for Iteratively Reconstructed CT Images With Locally Quadratic Regularization. IEEE Trans Med Imaging 36:17-26
Nien, Hung; Fessler, Jeffrey A (2016) Relaxed Linearized Algorithms for Faster X-Ray CT Image Reconstruction. IEEE Trans Med Imaging 35:1090-8
Kim, Donghwan; Fessler, Jeffrey A (2016) Optimized first-order methods for smooth convex minimization. Math Program 159:81-107
Weller, Daniel S; Pnueli, Ayelet; Divon, Gilad et al. (2015) Undersampled Phase Retrieval with Outliers. IEEE Trans Comput Imaging 1:247-258
McGaffin, Madison Gray; Fessler, Jeffrey A (2015) Edge-preserving image denoising via group coordinate descent on the GPU. IEEE Trans Image Process 24:1273-81
Cho, Jang Hwan; Fessler, Jeffrey A (2015) Regularization designs for uniform spatial resolution and noise properties in statistical image reconstruction for 3-D X-ray CT. IEEE Trans Med Imaging 34:678-89

Showing the most recent 10 out of 14 publications