Research

 

Medical Imaging

My current research in computed tomography (CT) involves the development of iterative algorithms for CT image reconstruction, with particular focus on reconstruction of polyenergetic, undersampled, or very noisy data. These cases may arise from attempts to reduce dose to the patient, increase scan time, or when objects with high attenuation, such as metal, are present. Iterative methods have the power to address these issues in a comprehensive way. In particular, my recent research has considered the application of the recently proposed superiorization methodology to problems in CT. I am also interested in the application of techniques from machine learning, such as convolutional neural networks, to imaging problems.

My earlier research on SPECT imaging focused on image reconstruction in dynamic SPECT, where we attempt to reconstruct a 4D, time-varying distribution of tracer in the body using the same amount of data that is typically used for conventional (static) 3D SPECT reconstruction. This results in a severely underdetermined inverse problem. My research in dynamic SPECT has consisted of investigating regularization techniques to improve image quality, and identifying specific artifacts that arise as a result of the highly underdetermined nature of the problem.

Code:

I have made some Matlab code for reconstruction of polyenergetic CT data available on github. The code is my own implementation of a superiorized version of the polyenergetic SART (pSART) algorithm proposed by Lin & Samei (paper here). In addition to reconstruction code, it also contains some examples with simulated data that have been published in some of the papers listed below.

To download the code, please visit my GitHub Repository.

Related publications:

  • Y. Jia, N. McMichael, P. Mokarzel, B. Thompson, D. Si and T. Humphries. Superiorization-inspired unrolled SART algorithm with U-Net generated perturbations for sparse-view and limited-angle CT reconstruction. Phys. Med. Biol. 67 (24), p. 245004. 2022. (preprint) (Online Journal)
  • T. Humphries and B. Wang. Superiorized method for metal artifact reduction. Medical Physics. 47(9), pp. 3984-3995. 2020. (preprint) (Online Journal)
  • T. Humphries, M. Loreto, B. Halter, W. O'Keeffe, L. Ramirez. Comparison of Regularized and Superiorized Methods for Tomographic Image Reconstruction. Journal of Applied and Numerical Optimization. 2(1), pp. 77-99. 2020. (Preprint) (Online Journal)
  • T. Humphries, D. Si, S. Coulter, M. Simms and R. Xing. Comparison of deep learning approaches to low dose CT using low intensity and sparse view data. SPIE Medical Imaging . 2019. (Preprint) (Online Journal)
  • T. Humphries and A. Gibali. Superiorized polyenergetic algorithm for reduction of metal artifacts in CT images. IEEE Nuclear Science Symposium Conference Record. 2017. (Preprint) (Online Journal)
  • T. Humphries, R. McGarity and K. Uy. Comparison of image and data domain methods for three-material decomposition in dual-energy CT. IEEE Nuclear Science Symposium Conference Record. 2017. (Preprint) (Online Journal)
  • T. Humphries, J. Winn and A. Faridani. Superiorized algorithm for reconstruction of CT images from sparse-view and limited-angle polyenergetic data. Phys. Med. Biol. 62(16), p. 6762. 2017. (Preprint) (Online Journal)
  • T. Humphries and A. Faridani. Reconstruction of CT Images from Sparse-View Polyenergetic Data Using Total Variation Minimization. IEEE Nuclear Science Symposium Conference Record. 2015. (Preprint) (Online Journal)
  • T. Humphries. Technical Note: Convergence analysis of a polyenergetic SART algorithm. Medical Physics. 42(7), pp. 1407-1404. 2015. (preprint) (Online journal)
  • T. Humphries and A. Faridani. Segmentation-free quasi-Newton method for polyenergetic CT reconstruction. Poster presentation at the 2014 IEEE Nuclear Science Symposium/Medical Imaging Conference, November 2014. (Preprint) (Online Journal)
  • T. Humphries, A. Celler and M. Trummer. Effects of attenuation in single slow-rotation dynamic SPECT. Phys. Med. Biol., 57(14), pp. N253-N265. 2012. (preprint) (Online journal)
  • T. Humphries, A. Celler and M. Trummer. Slow-rotation dynamic SPECT with a temporal second derivative constraint, Medical Physics, 38(8), pp. 4489-4497. 2011. (preprint) (Online journal)
  • T. Humphries, A. Saad, A. Celler, G. Hamarneh, T. Moeller and M. Trummer. Segmentation-Based Regularization of Dynamic SPECT Reconstruction. IEEE Nuclear Science Symposium Conference Record, pp. 2849-2852, 2009. (preprint) (Online Journal)
  • S. Shcherbinin, A. Celler, M. Trummer and T. Humphries. An APD-based iterative reconstruction method for simultaneous technetium-99m/iodine-123 SPECT imaging". Physica Medica: European Journal of Medical Physics. 2009. (Online journal)
  • T. Humphries. Temporal Regularization and Artifact Correction in Single Slow-Rotation Dynamic SPECT. Ph.D. Thesis, Simon Fraser University, Summer 2011. (link)
  • T. Humphries. Improved Numerical Integration for Analytical Photon Distribution Calculation in SPECT. M.Sc. Thesis, Simon Fraser University, Summer 2007. (link)

Optimization of oil field operations

Maximizing production from an oil field requires making a number of critical decisions, such as how many injection and production wells to drill, where to drill them, what flow rates to maintain at these wells, and how to schedule drilling operations. Manually exploring even a relatively small number of production scenarios using a reservoir simulator is an extremely time-consuming task, which has led to interest in the use of automated, mathematical optimization algorithms. Our work on production optimization has been focused on determining optimal well locations and well flow control strategies using a simultaneous methodology, which is a problem that has been largely unexplored to this point. This is a challenging optimization problem which involves expensive function evaluations, a lack of readily available derivative information, and a highly nonconvex objective function. Our approach has involved the use of several optimization techniques, particularly particle swarm optimization (PSO) and pattern search (GPS). These algorithms are well-suited to the problem due the fact that they do not require gradient information and are readily parallelizable.

Related documents:
  • G .L. C. Carosio, T. D. Humphries, R.D. Haynes and C.G. Farquharson. A closer look at differential evolution for the optimal well placement problem. Genetic and Evolutionary Computation Conference, Madrid, Spain. July 2015. (preprint)
  • T. Humphries and R. Haynes. Joint optimization of well placement and control for nonconventional well types. Journal of Petroleum Science and Engineering , 126, pp. 242--253. February 2015. (preprint) (online journal)
  • A. Butler, R.D. Haynes. T.D. Humphries, P. Ranjan. Efficient Optimization of the Likelihood Function in Gaussian Process Modelling. Computational Statistics & Data Analysis, 73, pp. 40--52. 2014. (online journal)
  • T. Humphries, R. Haynes and L. James. Simultaneous and sequential approaches to joint optimization of well placement and control. Computational Geosciences. 18(3-4), pp. 433-448. 2014. (preprint) (online journal)

A full list of publications is included in my CV. See also my Google Scholar page.