John Baxter, BSE, PhD


John Baxter

Program

PhD

Project

Continuous Max-Flow Theory in Medical Image Processing

Research Summary

What happens when you take graph-cut theory to its extremes? What happens when you take a cut through a graph with an infinite number of infinitely connected vertices? Continuous max-flow theory addresses these questions concerning complex probabilistic graphical models in these extreme cases. With the help of the principle of duality and variational optimization, many of these problems are computationally tractable, leading to a wealth of new image processing algorithms.

All of this must be made computationally feasible in order to use. General purpose graphics processing unit (GPGPU) programming has the accelerate algorithms addressing these problems, previously disregarded as too cumbersome for visualization, and them to real-time speeds. One aspect of my research is accelerating algorithms using nVidia’s CUDA architecture, providing a framework for continuous max-flow theory in medical image processing. (See  http://asets.github.io/asetsMatlabMaxFlow  and  https://github.com/VASST/RobartsVTK  for my open-source contributions.) It is my belief that having research work available open-source to the community is crucial to ensuring reproducibility and integrity.

My research is concerned with applying continuous max-flow theory to medical image processing, creating new algorithms for addressing medical image enhancement and segmentation. My research has been guided by two key principles: topology and interactivity.

Topology:

Understanding the topology of a medical image processing problem is key to creating faster and more accurate enhancement and segmentation algorithms. These topological considerations can be as simple as enforcing one object of interest lies in a particular region of interest, to as complicated as warping the intensity of images into a non-linear domain for easier processing as in MRI phase processing (shown below).
MRI Phase SmoothingMRI Phase Smoothing can be performed via continuous max-flow by using a cyclic topology, avoiding the use of complex filtering or phase unwrapping in this third, new paradigm. The image on the left is the raw MR phase, the central image the max-flow processed low-pass image, the right is the high-pass image which isolates the clinically relevant contrast. This research can dramatically improve the robustness of phase-sensitive MRI techniques such as susceptibility weighted imaging, which is now being used to investigate neurodegenerative disorders.

Interactivity:

Medical image processing must fundamentally rely on the clinician. By creating algorithms that focus on the interactivity between the clinician and the computer, medical image processing can become more transparent and more robust. My research has been oriented around creating frameworks in which the clinician can readily encode their anatomical knowledge in a computer-understandable format in run-time, creating a common language between clinicians and computers. (For more on this, see my interview in the Robarts Discovery Newsletter!)

 Interactivity in medical image processing
Certain kinds of anatomical knowledge, such as part-whole relationships, are inherently general purpose. By allowing the clinician to specify these relationships in a hierarchy (left), the computer can understand the imaging problem better and segment more accurately, regardless of image modality (right).

Research Interests

  • Markov Random Field Theory
  • Continuous Max-Flow Theory
  • Optimization-Based Image Enhancement and Segmentation
  • Human Computer Interaction in Medical Image Processing
  • GPGPU Acceleration of Medical Image Processing

Key Questions

  • What clinical knowledge can be incorporated into a medical image processing framework without losing generality? How can clinical knowledge be expressed meaningfully to a computer in run-time?
  • What families of label orderings can be solved for efficiently in segmentation? How can these be specified by the user without constraining the user?
  • Can numeric parameters be eliminated from medical image processing by determining them automatically in an image-agnostic process? What effect will this have on accuracy and interactivity?
  • Can shape constraints be applied efficiently with any regularization scheme, or are they constrained to only the case of (an)istotropic L2 norm regularization?
  • What higher-order clique terms from Markov Random Field theory have analogues in the continuous domain? How can continuous max-flow be extended to address these terms?

Education

Doctor of Philosophy in Biomedical Engineering, University of Western Ontario

Bachelor of Software Engineering (Honours) with a Cognitive Science Option and a Management Sciences Option, Graduated with Distinction on the Dean's List, University of Waterloo

Awards

Doctoral Excellence in Research Award, University of Western Ontario, 2016-2017

Best Paper Award, MICCAI Workshop on Bayesian and Probabilistic Graphical Modeling in Biomedical Imaging (BAMBI), October 2015

Alexander Graham Bell Canada Graduate Scholarship (CGS-D), Natural Sciences and Engineering Research Council of Canada, September 2015

Ontario Graduate Scholarship, September 2014

Ontario Graduate Scholarship, September 2013

Alexander Graham Bell Canada Graduate Scholarship (CGS-M), Natural Sciences and Engineering Research Council of Canada, September 2012

Sanford Fleming Foundation Award for Academic Excellence, University of Waterloo, Waterloo, Ontario, April 2012

Finalist for the 2011 Sunnybrook Prize, Sunnybrook Health Sciences Centre, Toronto, Ontario, January 2011

Sanford Fleming Foundation Work Term Report Award, University of Waterloo, Waterloo, Ontario, May 2010

Dean's Honour List (Faculties of Mathematics and Engineering), University of Waterloo, 2007-2012

Publications

  1. Baxter, J.S.H., Rajchl, M., McLeod, A.J., Yuan, J. & Peters, T.M. (2017) "Directed Acyclic Graph Continuous Max-Flow Image Segmentation for Unconstrained Label Orderings" International Journal of Computer Vision. (accepted)
  2. Baxter, J.S.H., Inoue, J., Drangova, M., & Peters, T.M., (2016) "Shape Complexes: the intersection of label orderings and star convexity constraints in continuous max-flow medical image segmentation", Journal of Medical Imaging, 2016. (accepted)
  3. Baxter, J.S.H., Hosseini, Z., Liu, J., Drangova, M. & Peters, T.M. (May 2016) "Cyclic Contious Max-Flow: Phase Processing Using the Inherent Topology of Phase", International Society for Magnetic Resonance in Medicine (ISMRM)
  4. Baxter, J.S.H., Rajchl, M., Peters, T.M., & Chen, E.C.S., (2016) "Optimization-based interactive segmentation interface for multiregion problems" SPIE Journal of Medical Imaging (JMI)
  5. Baxter, J.S.H., Yuan, J., Drangova, M., Peters, T.M., & Inoue, J. (Feb 2016) "Shape Complexes". In SPIE Medical Imaging International Society for Optics and Photonics.
  6. Pardasani, U., Baxter, J.S.H., Peters, T.M. & Khan, A.R. (Feb 2016) "Single slice US-MRI registration for neurosurgical MRI-guided US", In SPIE Medical Imaging International Society for Optics and Photonics.
  7. Rajchl, M., Baxter, J.S.H., McLeod, A.J., Yuan, J., Qiu, W., Peters, T.M., & Khan, A.R. (2016) "Hierarchical Max-Flow Segmentation Framework For Multi-Atlas Segmentation with Kohonen Self-Organizing Map Based Gaussian Mixture Modeling" MedIA 27 (2016): 45-56.
  8. Cantor-Rivera, D., Baxter, J.S.H., De Ribaupierre, S., Lau, J.C., Mirsattari, S., Goubran, M., Burneo, J.G., Steven, D.A., Peters, T.M. & Khan A.R. (2015) "Individual feature maps: a patient-specific analysis tool with applications in temporal lobe epilepsy". International Journal of Computer-Assisted Radiology and Surgery (IJCARS)
  9. Baxter, J.S.H., Rajchl, M., Yuan, J. & Peters, T.M. (Oct 2015) "Directed Acyclic Graphical Continuous Max-Flow Image Segmentation". MICCAI Workshop on Bayesian and Probabilistic Graphical Modeling in Biomedical Imaging (BAMBI)
  10. Chen, E.C.S, McLeod, A.J., Baxter, J.S.H., Peters, T.M., (Oct 2015) "An Iterative Closest Point Framework for Ultrasound Calibration", MICCAI Workshop on Augmented Environments in Computer Assisted Interventions (AECAI)
  11. Ameri, G., Baxter, J.S.H., McLeod, A.J., & Chen, E.C.S. (Oct 2015) "Augmented Reality Ultrasound Guidance for Central Line Procedures: Preliminary Results", MICCAI Workshop on Augmented Environments in Computer Assisted Interventions (AECAI)
  12. Inoue, J., Baxter, J.S.H. & Drangova, M. (Oct 2015) "Left Atrial Wall Segmentation from CT for Radiofrequency Catheter Ablation Planning." MICCAI Workshop on Clinical Image-Based Procedures (CLIP))
  13. McLeod, A.J., Baxter, J.S.H., Ameri, G., Ganapathy, S., Peters, T.M., & Chen, E.C.S., (2015) "Detection and visualization of dural pulsation for spine needle interventions", International Journal of Computer Assisted Radiology and Surgery (IJCARS) 10(6)
  14. Chen, E.C.S., McLeod, A.J., Baxter, J.S.H., & Peters, T.M. (2015) "Registration of 3D shapes under anisotropic scaling: Anisotropic-scaled iterative closest point algorithm", International Journal of Computer Assisted Radiology and Surgery (IJCARS) 10(6)
  15. Baxter, J.S.H., Rajchl, M., Peters, T.M., & Chen, E.C.S. (Feb 2015) "Optimization-based interactive segmentation interface for multi-region problems". In SPIE Medical Imaging International Society for Optics and Photonics.
  16. Ameri, G., McLeod, A.J., Baxter, J.S.H., Chen, E.C.S., & Peters, T.M., (Feb 2015) "Line fiducial material and thickness considerations for ultrasound calibration", In SPIE Medical Imaging International Society for Optics and Photonics.
  17. Abhari, K., Baxter, J.S.H., Khan, A.R., Peters, T.M., de Ribaupierre, S., Eagleson, R., (2015) "Visual Enhancement of MR Angiography Images to Facilitate Planning of Arteriovenous Malformation Interventions", ACM Transactions on Applied Perception 12(1)
  18. Rajchl, M., Baxter, J.S.H., Bae, E., Tai, X-C., Fenster, A., Peters, T.M., Yuan, J. (Jan 2015) "Variational time-implicit multiphase level-sets" International Workshop on Energy Minimization Methods in Computer Vision and Pattern Recognition
  19. Abhari, K., Baxter, J.S.H., Chen, E.C.S., Khan, A.R., Peters, T.M., de Ribaupierre, S., Eagleson, R., (2014) "Training for Planning Tumour Resection: Augmented Reality and Human Factors", IEEE Transactions on Biomedical Engineering (TBME) 62(6)
  20. Baxter, J.S.H., Rajchl, M., McLeod, A.J., Khan, A.R., Yuan, J., & Peters, T.M. (Feb. 2014). "Smoothness parameter tuning for generalized hierarchical continuous max-flow segmentation". In: SPIE Medical Imaging. International Society for Optics and Photonics.
  21. Ameri, G., Baxter, J.S.H., McLeod, A.J, Jayaranthe, U.L, Chen, E.C.S., & Peters, T.M. (Feb 2014). "Synthetic aperture imaging in ultrasound calibration". In: SPIE Medical Imaging. International Society for Optics and Photonics.
  22. Kayvanrad, M.H., McLeod, A.J., Baxter, J.S.H., McKenzie, C., & Peters, T.M., (2014) "Stationary wavelet transform for under-sampled MRI reconstruction", Magnetic Resonance Imaging 32(10)
  23. McLeod, A.J, Baxter, J.S.H., de Ribaupierre, S., & Peters, T.M. (Feb 2014). "Motion magnification for endoscopic surgery". In: SPIE Medical Imaging. International Society for Optics and Photonics.
  24. Rajchl, M., Baxter, J.S.H., Yuan, J., Peters, T.M, & Khan, A.R. (Sept 2013). "Multi-Atlas-based Segmentation with Hierarchical Max-Flow" MRBRAINS13 Segmentation at MICCAI 2013.
  25. Abhari, K., Baxter, J.S.H., Chen, E.C.S, Wedlake, C., Peters, T.M., Eagleson, R., & de Ribaupierre, S., (2013) "The Role of Augmented Reality in the Training and Planning of Brain Tumour Resection", Joint Medical Image and Augmented Reality (MIAR) & Augmented Environments in Computer-Assisted Interventions (AE-CAI) Workshop, Nagoya, Japan.
  26. Abhari, K., Baxter, J.S.H., Chen, E.C.S., Khan, A.R., Wedlake, C., Peters, T.M. de Ribaupierre, S., & Eagleson, R. (2013) "Use of a mixed-reality system to improve the planning of brain tumour resections: Preliminary results". Lecture Notes in Computer Science
  27. Abhari, K., Baxter, J.S.H., Chen, E.C.S, Wedlake, C., Peters, T.M., Eagleson, R., & de Ribaupierre, S., (2013) "Development and Evaluation of an Augmented-Reality Training System for Planning Brain Tumour Resection Interventions", Medical Image Understanding and Analysis, Birmingham, UK.
  28. Abhari, K., Baxter, J.S.H., Chen, E.C.S, Wedlake, C., Peters, T.M., Eagleson, R., & de Ribaupierre, S., (Oct 2012) "Development and Evaluation of an AR-based Surgical Training Platform", AE-CAI, Nice, France
  29. Abhari, K., Baxter, J.S.H., de Ribaupierre S., Peters T. M., & Eagleson R., (2012), "Perceptual Improvement of Volume-Rendered MR Angiography Data using a Contour enhancement Technique,” SPIE Medical Imaging", San Diego, CA
  30. Chen, E.C.S, Sarkar, K., Baxter, J.S.H., Moore, J., Wedlake, C., & Peters, T.M., (2012) "An augmented reality platform for planning of minimally invasive cardiac surgies", SPIE Medical Imaging, San Diego, CA
  31. Buchanan, S., Moore, J., Lammers, D., Baxter, J.S.H., & Peters, T.M., (2012) "Characterization of tissue-simulating phantom materials for ultrasound-guided needle procedures", SPIE Medical Imaging, San Diego, CA
  32. Kayvanrad, M.H., McLeod A.J., Baxter, J.S.H., McKenzie C.A, & Peters, T.M., (2012), “T1 Map Reconstruction from Under-sampled k-Space Data using a Similarity Constraint,” ISMRM, Melbourne, Australia
  33. Baxter, J.S.H., Peters, T.M., & Chen, E.C.S. (2011) "A unified framework for voxel classification and triangulation", SPIE Medical Imaging, Orlando, FL

Curriculum Vitae

Email

jbaxter -AT- robarts.ca


© 2017 Robarts Research - Imaging - All Rights Reserved | Privacy Policy | Terms of Use | Contact Us