Advanced Imaging Algorithms & Instrumentation
Laboratory

The Advanced Imaging Algorithms and Instrumentation Lab is in the Biomedical Engineering Department at Johns Hopkins University and is under the direction of J. Webster Stayman, PhD. The lab is a proud member of the I-STAR Laboratory Consortium.

The AIAI Lab is dedicated to high-fidelity modeling and design of medical imaging systems and the application of strong priors and knowledge of the class of objects being imaged to drive customized data acquisitions, data processing, and image formation algorithms that are optimized for specific imaging tasks. The lab’s mission to personalize medical imaging takes many forms including the design of dedicated imaging systems for specific anatomical sites, novel acquisition methods that permit adaptation of the medical imaging equipment to specific sites or diagnostic tasks, and novel image formation methods that accommodate non-standard or low-fidelity data. 

Specific imaging modalities investigated by the AIAI Lab have included dedicated cone-beam CT systems, diagnostic x-ray CT, image-guided radiotherapy systems, mobile and robotic C-arms, tomosynthesis devices, and phase-contrast CT.

News and Events

We’ll try to post regular news and events associated with the AIAI Laboratory in this section. This will include conference presentations, announcements, and major lab events.

June 1-4, 2015

AIAI and I-STAR lab members had a great time at the 13th International Meeting on Fully Three-Dimensional Image Reconstruction in Radiology and Nuclear Medicine in Newport, RI. Lab members gave three oral presentations and a poster at the meeting:

Amir Pourmorteza introduced the Reconstruction of Difference approach - a new strategy for integrating prior images into a model-based reconstruction method where only the difference between sequential images is reconstructed yielding opportunities for dose reduction and better visualization of anatomical change.

Paper Talk

Steven Tilley demonstrated the application of correlated noise models to high spatial resolution reconstruction of flat-panel cone-beam CT data. This work represents the first application of such high-fidelity noise models to physical projection data. Improved noise-resolution performance was observed suggesting potential application in high resolution CBCT applications (e.g. extremity imaging, CBCT mammography, etc.)

Paper Talk

June 1-4, 2015

AIAI and I-STAR lab members had a great time at the 13th International Meeting on Fully Three-Dimensional Image Reconstruction in Radiology and Nuclear Medicine in Newport, RI. Lab members gave three oral presentations and a poster at the meeting:

Web Stayman discussed task-based optimization of interventional CBCT acquisitions with arbitrary source-detector orbits. Nonconvex optimization of the acquisition trajectory for intracranial imaging and a multi-location bleed detection task near an embolized aneurysm showed improved visualization of bleeds near the embolization coil.

Paper Talk

Qian Cao presented a multi-resolution model-based reconstruction approach that permits user-defined standard and high-resolution regions in a reconstruction. This method permits very small voxels in a targeted region of interest which ordinarily might be computationally impractical. The approach was applied to targeted trabecular bone imaging where quantitative assessment of bone morphology and bone health demands the resolution of very fine bony details.

Paper 

August 2015

Congratulations to Hao Dang on his publication "Statistical Reconstruction for Cone-Beam CT with a Post-Artifact-Correction Noise Model: Application to High-Quality Head Imaging" in Physics and Medicine and Biology. This paper is a featured article in the August issue.

Hao's work is helping to bring better soft-tissue imaging capability to cone-beam CT through the integration of sophisticated physical modeling and artifact correction as well as rigorous noise modeling and model-based reconstruction to mitigate noise magnification.

Present

The AIAI Laboratory is recruiting new PhD students and/or postdoctoral fellows. The following list summarizes some of the skills and background that we are seeking in new researchers:

  • Strong background in the mathematics of imaging - especially tomography and CT. Particular strengths may include model-based reconstruction, large scale optimization methods, statistical estimation theory, regularization strategies, image quality metrics including human observer models.

  • Strong background in the physics of tomographic imaging including detection physics, imaging system modeling, matching theoretical models with empirical data, dose measurement.

  • Strong research programming skills including Matlab, Python, C/C++, CUDA. Experience with the prototyping and efficient implementation of research algorithms.

  • Experience with electromechanical, imaging sensors, and x-ray system construction and integration including embedded control, device and imaging pipeline programming, debugging electronics and synchronization issues.

  • Experience with experimental design methodology, hands-on data acquisition, phantom development and construction.

Anyone who is interested should please take a moment to read through the research topics listed below and consider which projects hold the most interest for you, and which projects to which you feel that you and your skills could contribute most. 
If you are a prospective postdoctoral fellow, please email me your CV and a brief email detailing what projects you feel are a good fit and interest you most.
If you are a prospective PhD student (and you are not currently enrolled at Johns Hopkins) you will need to apply to the PhD program and be accepted before you can join the AIAI Lab. However, please do contact web.stayman@jhu.edu to let me know your interest when you are applying.

Research Team

The following researchers and students are part of the core members of the AIAI lab. However, the lab works closely with a number of Johns Hopkins clinicians, corporate research teams, and other research labs. In particular, the AIAI lab is closely partnered with the I-STAR lab and its members.

J. Webster Stayman, PhD (Principle Investigator)

J. Webster Stayman, PhD (Principle Investigator)

Dr. Stayman is an Assistant Professor in the Biomedical Engineering Department and Principle Investigator of the AIAI Lab. His research interests have included tomographic reconstruction for emission and transmission tomography, statistical reconstruction approaches, regularization design, prior-image-based reconstruction,  cone-beam CT (CBCT) modeling and analysis, optical system modeling and analysis, test bench construction, instrumentation, physical experimentation, and task-driven and adaptive acquisitions. Google Scholar Publications

Grace Gang, PhD (Postdoctoral Fellow)

Grace Gang, PhD (Postdoctoral Fellow)

Grace's research is in task-based optimization to help design novel acquisition approaches and reconstruction methods for CT and CBCT; as well as advanced reconstruction methods that integrate prior image information. Google Scholar Publications

Aswin Mathews, PhD (Postdoctoral Fellow)

Aswin Mathews, PhD (Postdoctoral Fellow)

Aswin’s research involves novel instrumentation, system modeling, algorithm development for dynamic fluence field modulation in computed tomography.

Hao Dang (PhD Candidate)

Hao Dang (PhD Candidate)

Hao’s research is advanced reconstruction methods for CBCT including prior-image-based reconstruction and high-fidelity model-based reconstruction for artifact-corrected CBCT. Google Scholar Publications

Steven Tilley II (PhD Candidate)

Steven Tilley II (PhD Candidate)

Steve is working on advanced reconstruction methods for flat-panel CBCT that includes high-fidelity system models of system blur and noise correlations for ultra-high spatial resolution imaging. Google Scholar Publications

Esme Zhang (Master's Student)

Esme Zhang (Master's Student)

Esme is working on forward models for reconstruction of clinical CT data with particular interests in high-quality cardiac imaging.

Ang Li (PhD Rotation Student)

Ang Li (PhD Rotation Student)

Ang is helping to characterize a new exposure control system for an x-ray testbench.

AIAI Alumni

Amir Pourmorteza, PhD (Former Postdoctoral Fellow)

Amir Pourmorteza, PhD (Former Postdoctoral Fellow)

Amir conducted research involving the use of prior imaging studies, including cross-modality incorporation of images, to directly reconstruct anatomical change in CBCT. Amir is now a scientist at the National Institute of Health Clinical Center. Google Scholar Publications

Shiyu Xu, PhD (Former Postdoctoral Fellow)

Shiyu Xu, PhD (Former Postdoctoral Fellow)

Shiyu conducted research on the integration of prior knowledge of surgical tools and implants into tomographic reconstruction including cases with spectral and positional uncertainty. Shiyu is now a research scientist at Philips Healthcare. Google Scholar Publications

Research Topics

Click to view different topics

Novel Acquisition Methods

+ Non-circular and non-helical computed tomography

+ Fluence Field Modulation

Prior-Image-Based Reconstruction 

+ Prior Image Registration Penalized-Likelihood Estimation (PIRPLE)

+ Reconstruction of Difference

High-Fidelity Modeling

+ Correlated Noise and Blur Flat-Panel CBCT Reconstruction

Using General Knowledge in Image Formation

+ Known Component Reconstruction

+ Non-circular and non-helical computed tomography

Computed tomography has largely focused on circular and spiral trajectories since its invention. While such acquisitions have been well-suited to diagnostic imaging over the years, with the advent and use of robotic C-arms in interventional imaging, there is now the capability for largely arbitrary positioning of the x-ray source and detector yielding new opportunities to customize data collection to the patient and imaging task. Our multi-axis CBCT test bench (seen in the video below) can emulate such orbits. We are conducting ongoing research on how to leverage and prospectively control such acquisitions to be optimized to the patient and imaging task.

+ Fluence-field modulation and automatic exposure control

One strategy for reducing CT radiation dose is to provide additional flexibility in the specification and creation of x-ray beams - in effect putting the x-rays where they are needed. Typically CT achieves a static spatial modulation of the x-ray beam via a bowtie filter that variably attenuates the beam across detector locations. Additional intensity modulation can be achieved by varying tube current through automatic exposure control (AEC) methods. We are researching a novel beam modulation strategies that permit dynamic spatial and intensity modulations for acquisitions that are customized to the patient and imaging task. An illustration of dose reduction using beam modulation is shown below.

+ Prior Image Registration Penalized-Likelihood Estimation (PIRPLE)

Sequential and dynamic imaging studies are commonly applied throughout the course of diagnosis and treatment. Traditionally, each imaging study is treated in isolation, effectively throwing away an incredibly rich source of patient-specific prior information about anatomical features. We have developed a framework for including prior image studies into the image formation process of subsequent studies. The framework leverages both statistical noise model to account for measurement noise as well as registration of the prior image in a joint registration and reconstruction approach called PIRPLE. A version that accommodates deformable registration between the prior image and the current anatomy, called dPIRPLE, has also been developed to maximize the utility of the prior image data. The PIRPLE approach is remarkably tolerant of low-fidelity data as illustrated below in a pulmonary nodule surveillance scenario where the radiation exposure for a followup CT scan has been dropped by an order of magnitude while maintaining sufficient image quality to monitor tumor growth. The PIRPLE framework has potential application in many sequential imaging scenarios including interventional imaging, image-guided radiation therapy, cardiac imaging, longitudinal studies, etc.

+ Reconstruction of Difference

In some dynamic and sequential imaging studies, it is not the current anatomical state that is of greatest interest, but the change in anatomy or other aspects of the imaging volume over time. While most other prior-image-based reconstruction approaches incorporate the prior image through a penalty function or constraint, we have developed an alternate approach where the prior image is incorporated directly into the forward model and it is the change between images that is estimated. A flowchart is illustrated below. This framework may have several advantages over other prior image methods including improved estimation of the change volume, reduced field-of-requirements (e.g., only need the region-of-change), and joint reconstruction/registration where the registration step is operating on the measured data.

+ Correlated Noise and Blur Flat-Panel CBCT Reconstruction

The power of model-based reconstruction methods to provide improved trade-offs between radiation exposures and image quality is well-known and continues to find increasing use in clinical settings. Much of the advantage of model-based approaches arises from accurate modeling of noise in the measurements - effectively weighting each measurement by its information content. Nearly all statistical reconstruction methods make a fundamental assumption that the noise in measurements is uncorrelated. While this is often true in nuclear imaging and, perhaps, a valid presumption in traditional multi-row CT, data correlations in flat-panel-based cone-beam CT can be prominent (particularly in high-spatial resolution applications). We have developed statistical reconstruction methods that model such inherent correlations in flat-panel detection as well as detector and source blur effects to provide improved image quality - particularly for very high spatial resolution applications (e.g. extremity image of trabecular bone details, microcalcifications in CBCT mammography, etc.). An application to trabecular bone imaging of a wrist is shown below.

+ Known Component Reconstruction

In many imaging scenarios there is an untapped wealth of knowledge about elements in the field-of-view. For example, in interventional imaging, surgical tools and implants can often be found in the imaging field. In some imaging modalities these components can present a great deal of difficulty for high quality imaging. For example, metal tools and implants can produce heavy “metal artifacts” in CT images. We have developed techniques to integrate specific component knowledge directly in the reconstruction (e.g., via CAD models, generalized deformable models, etc.) to both improve imaging performance but to also obtain auxiliary information about the positioning of the tools and implants that is independently useful in image-guided procedures and as a quantitative estimate on implant placements. A sample joint reconstruction and registration of a two-component pedicle screw is shown in the video below. The process, called known component reconstruction (KCR) is iteratively solved - starting with a traditional filtered-backprojection image and initial guesses for screw positions. Successive iterations improve both the registration estimate as well as the image quality throughout the image volume and, in particular, in the region nearest the implant where image quality needs are greatest for detection of complications.

Selected Publications

Link to all publications at Google Scholar

Task-Driven Acquisition

J. W. Stayman, G. Gang, and J. H. Siewerdsen, "Task-based optimization of source-detector orbits in interventional cone-beam CT," Int'l Mtg. Fully 3D Image Recon. in Radiology and Nuc. Med., Newport, RI, (June 1-4, 2015). Full Proceedings PDF
G. Gang, J. W. Stayman, T. Ehtiati, J. H. Siewerdsen, “Task-driven image acquisition and reconstruction in cone-beam CT,” Physics in Medicine and Biology, 60 3129-3150 (March 2015). PubMed 
J. W. Stayman and J. H. Siewerdsen, “Task-Based Trajectories in Iteratively Reconstructed Interventional Cone-Beam CT,” Int'l Mtg. Fully 3D Image Recon. in Radiology and Nuc. Med., Lake Tahoe, (June 16-21, 2013). Full Proceedings

Integration of Prior Images in Reconstruction

A. Pourmorteza, H. Dang, J. H. Siewerdsen, and J. W. Stayman, "Reconstruction of difference using prior images and a penalized-likelihood framework," Int'l Mtg. Fully 3D Image Recon. in Radiology and Nuc. Med., Newport, RI (June 1-4, 2015). Full Proceedings PDF
H. Dang, A. S. Wang, M. S. Sussman, J. H. Siewerdsen, J. W. Stayman, "dPIRPLE: A joint estimation framework for deformable registration and penalized-likelihood CT image reconstruction using prior images," Physics in Medicine and Biology, 59(17) 4799-826 (September 2014). PubMed
J. W. Stayman, H. Dang, Y. Ding, and J. H. Siewerdsen, “PIRPLE: A Penalized-Likelihood Framework for Incorporation of Prior Images in CT Reconstruction,” Physics in Medicine and Biology, 58(21) 7563-7582 (November 2013). PubMed

High-Fidelity Modeling

H. Dang, J. W. Stayman, A. Sisniega, J. Xu, W. Zbijewski, X. Wang, D. H. Foos, N. Aygun, V. E. Koliatsos, J. H. Siewerdsen, "Statistical Reconstruction for Cone-Beam CT with a Post-Artifact-Correction Noise Model: Application to High-Quality Head Imaging," Phys. Med. Biol., August 2015. PMB
S. Tilley II, J. H. Siewerdsen, and J. W. Stayman, "Generalized Penalzied Weighted Least-Squares Reconstruction for Deblurred Flat-Panel CBCT," Int'l Mtg. Fully 3D Image Recon. in Radiology and Nuc. Med., Newport, RI (June 1-4, 2015). Full Proceedings PDF
A. S. Wang, J. W. Stayman, Y. Otake, G. Kleinszig, S. Vogt, G. L. Gallia, A. J. Khanna, J. H. Siewerdsen, “Soft-Tissue Imaging with C-arm Cone-Beam CT Using Statistical Reconstruction,” Physics in Medicine and Biology, 59(4) 1005-1026 (February 2014). PubMed
S. Tilley II, J. H. Siewerdsen, J. W. Stayman, "Iterative CT Reconstruction using Models of Source and Detector Blur and Correlated Noise," The Third International Conference on Image Formation in X-Ray Computed Tomography, Salt Lake City, 363-367, (June 2014). Full Proceedings PubMed
J. W. Stayman, W. Zbijewski, S. Tilley II, J. H. Siewerdsen, “Generalized least-squares CT reconstruction with detector blur and correlated noise models,” SPIE Medical Imaging, San Diego, (February 2014). PubMed

Known Component Reconstruction

J. W. Stayman, Y. Otake, J. L. Prince, J. H. Siewerdsen, "Model-based Tomographic Reconstruction of Objects containing Known Components," Trans. Medical Imaging, 31(10), 1837-1848 (October 2012). PubMed

Classes

Courses

Imaging Instrumentation - in development for Spring 2016.

Office Hours

Currently not holding fixed office hours. Please email me to arrange a time to speak.

Contact Us

A Note to Prospective Students and Postdocs

The AIAI laboratory is always looking for highly skilled and highly motivated individuals to add to the research team. In particular, we seek researchers that can span the range of in-depth theoretical analysis, efficient implementation of ideas including algorithm development and hands-on hardware integration, meticulous experimental method, and excellent communication skills. We welcome your interest.
Please do send us an email. However, please include a CV and take a moment to read a few of our publications and let us know specifically what you find interesting and how your skills would contribute. Thanks!

Mailing Address, Phone, Email

J. Webster Stayman, PhD
Assistant Professor
Biomedical Engineering Department
Johns Hopkins University
Traylor Building, Room #624A
720 Rutland Avenue
Baltimore MD 21205-2109
Phone: (410) 955-1314
Email: web.stayman@jhu.edu