Physics Medical Imaging Seminars - 2008


Wednesday, January 16, 2008, Time: TBD

SPIE BIOS Warm-Up

Sarah Patch and Hao Zhang, University of Wisconsin-Milwaukee


Wednesday, February 6, 2008, 4:30pm

Dual Energy Computed Tomography for Explosive Detection

Carl Crawford, Ph.D., Csuptwo, LLC

Single energy computed tomography (CT) scanners use measurements of densities to detect explosives in luggage. It is desirable to apply dual energy techniques to these CT scanners to obtain atomic number measurements to reduce false alarm rates. However, the direct application of existing dual energy techniques has practical problems, such as, approximation errors and lack of boundary constraints in dual energy decomposition, image artifacts, and x-ray spectral drifts. In this paper, we present methods to reduce these problems. The methods include constrained dual energy decomposition, adaptive scatter correction, nonlinear filtering of decomposed projections, and real-time image-based correction for x-ray spectral drifts. We demonstrate the effectiveness of the methods using simulated data and real data obtained from a commercial dual energy CT scanner.


Wednesday, 13 February, 2008, 4:30pm (Rescheduled from 6 Feb)

Dual Energy Computed Tomography for Explosive Detection

Carl Crawford, Ph.D., Csuptwo, LLC

Single energy computed tomography (CT) scanners use measurements of densities to detect explosives in luggage. It is desirable to apply dual energy techniques to these CT scanners to obtain atomic number measurements to reduce false alarm rates. However, the direct application of existing dual energy techniques has practical problems, such as, approximation errors and lack of boundary constraints in dual energy decomposition, image artifacts, and x-ray spectral drifts. In this paper, we present methods to reduce these problems. The methods include constrained dual energy decomposition, adaptive scatter correction, nonlinear filtering of decomposed projections, and real-time image-based correction for x-ray spectral drifts. We demonstrate the effectiveness of the methods using simulated data and real data obtained from a commercial dual energy CT scanner.


Wednesday, 27 February, 2008, 4:30pm

Adaptive Optics and Retina Imaging Application

Jungtae Rha, Ph.D., MCW - Dept. of Ophthalmology

The human eye suffers from optical aberrations that preclude imaging the retina with high-resolution. Nearly a decade ago, a technique called "adaptive optics" was combined with conventional ophthalmoscopy and since then there has been rapid evolution in system design and biological applications of such devices. Adaptive optics enables dynamic measurement and correction of the eye's aberrations, which results in the ability to resolve structures as small as 2 microns in diameter. These instruments have opened a new window of opportunity to non-invasively observe retinal structures and more recently, retinal function at the cellular level. In this talk, I will explore recent work for detecting novel intrinsic retinal signals in vivo.


Wednesday, 5 March, 2008, 4:30pm

Compressive Sensing With Tomographic Image Reconstruction

Emil Sidky, Univ. of Chicago, Dept. of Radiology

Medical scanners that yield tomographic images require a large amount of data to provide accurate 3D images. For example, a typical scan with computed tomography (CT) requires on the order of a thousand x-ray projections. As a result the radiation dose delivered to the patient for a diagnostic CT scan is a concern, particularly when patients receive multiple scans within a year. For magnetic resonance imaging (MRI), dose is not a big issue, but scan time can be. It is not unusual to have MRI scan protocols take up to an hour to obtain the required data to form the 3D tomographic images. One of the issues of interest in the development of image reconstruction algorithms is to reduce the amount of data required to form the desired images. In this talk, I will discuss recent developments in the field of "compressive sensing" that can potentially reduce scanning effort in tomographic imaging by a factor of 10 or more. Such advances obviously can have an impact on reducing x-ray dose in CT, or reducing scan time in imaging modalities such as MRI.


Wednesday, 12 March, 2008, 4:30pm

fMRI of Vision for Lab and Clinic: A New Approach for Assessing Brain-related Vision Deficits

Ted DeYoe, MCW

Functional magnetic resonance imaging (fMRI) provides a unique opportunity to integrate structural and functional data into a comprehensive system for the diagnosis and management of brain pathology. This exhibit will review recent advances in the use of fMRI to identify and map visual cortex in patients with brain-related vision deficits. Such fMRI-based maps of normal and impaired visual function in and around a site of pathology can be helpful in planning surgery or other treatment alternatives, especially for cases in which treatment may involve the risk of further visual impairment. A novel application of the technology is to explore potential surgical scenarios and to simulate expected visual side-effects for the patient. In such applications, the integrated use of structural, functional and pathophysiological data can provide a unique view of the relationship between a site of focal pathology and its effect on the function of surrounding brain tissue thereby allowing more informed decisions about treatment alternatives. Materials & Methods: The exhibit will review technological advances in the application of fMRI for the assessment of brain-related vision pathologies and will highlight case studies that illustrate the use of the technology. Results: Techniques for mapping the cortical representations of the visual field in human patients were first developed for basic science research but have now been used with clinical patients suffering from a variety of brain pathologies. Rather than simply activating visually responsive cortical regions, these techniques provide a detailed mapping of the layout of visual space within the brain. In cases of focal brain pathology, such retinotopic maps show the relationship between the site of pathology and the layout of visual space, especially the critical representations of foveal vs. peripheral vision. Since the exact topography of these maps is unique to each individual, fMRI makes it possible to map visual function near a site of pathology on a patient-by-patient basis. By identifying which portions of the visual field are represented adjacent to a site of pathology, it is possible to predict where surgical intervention is most likely to affect vision. Since different life activities can depend more on foveal than peripheral vision (e.g. reading vs. driving), the potential side-effects of invasive treatment on quality of life will also be unique to each individual (e.g. librarian vs. truck driver). However, the fMRI maps can be used to predict potential vision loss associated with different surgical scenarios and can be used to simulate the experience of a predicted scotoma for the patient. To date, FMRI visual field mapping has been used successfully with patients having a variety of pathologies including stroke, arteriovenous malformations and tumors. Conclusion: The technology and applications reviewed in this exhibit represent an initial "proof of concept" using a physiological system (vision) that is easy to manipulate for purposes of development. Though, the full clinical utility and prognostic capabilities of this approach have yet to be established, the concepts developed here can be extended to other systems and pathologies. Funded by NIH EB00843, EY13801, RR00058.


Thursday, 13 March 2008, 4:00pm -- SPECIAL PRESENTATION --

Development of Cone-Beam CT Methods

Dr. Hengyong Yu, Research Scientist, Biomedical Imaging Division, VT-WFU School of Biomedical

Because of the importance of the so-called long object problem, spiral cone-beam computed tomography (CT) has become a hot area in the CT field. As a main stream in the development of the next generation medical CT, spiral cone-beam CT has been greatly improved in terms of reconstruction methods since it was first proposed in 1991. Now, the state-of-the-art cone-beam algorithms can reconstruct images exactly from severely truncated data collected in rather flexible geometrical settings. Here we will first present an overview of this area. In our opinion, spiral cone-beam CT algorithms have been developed in four phases. In the practical reconstruction phase (1991-1996), various approximate algorithms were proposed, including generalized Feldkamp algorithms. In the multi-turn-based reconstruction phase (1997-2001), quasi-exact/exact helical cone-beam algorithms were developed based on a generalized Grangeat condition. In these algorithms, longitudinally truncated data are needed from multiple helical turns, which prevent these theoretically exact solutions from being practically applicable. In the PI-line-based reconstruction phase (2002-2004), the Katsevich helical cone-beam CT method was invented in the filtered backprojection format (FBP) to rely only on data associated with the PI-arc for exact reconstruction. Inspired by his formulism, a backprojected filtration (BPF) method was proposed by Zou and Pan, enabling the minimum data based reconstruction. In the general reconstruction phase (2004-2007), there has been a remarkable surge in research on exact image reconstruction in the case of general cone-beam scanning. Our group generalized helical cone-beam BPF and FBP algorithms for exact reconstruction from data collected along a general curve, in parallel to other groups' important work. Since 2006, we have been working on the quasi-short object problem. The motivation is to enable localized yet faithful reconstruction of fine details within a long or large object. However, a long-standing barrier in this direction has been the inability to reconstruct exactly an interior region of interest (ROI) from projection data acquired with x-rays only through the ROI, which is also referred to as the interior problem. While conventional wisdom states that the interior problem is not uniquely solvable, last year our research group demonstrated that the interior problem can be solved in a theoretically exact and numerically stable fashion if a small sub-region within the ROI is known, in contrast to the well-established area "lambda tomography" that targets localized but only edge-like reconstruction. Such a development will have major impacts on many fields of scientific research. In this talk, we will also present our latest results in this regards.


Wednesday, 26 March, 2008, 4:30pm

Monte Carlo simulations for investigating CT scatter and dose

Taly Gilat-Schmidt, Marquette Univ. - BME

Monte Carlo simulation is a powerful tool for modeling the stochastic nature of computed tomography (CT) imaging. Unlike simulation methods that calculate the net attenuated x-ray beam, Monte Carlo methods model the physics of x-ray transport and track the trajectory of x-ray photons through the imaged object. Monte Carlo simulations are useful for studying effects that depend on the individual photon interactions, for example scattered radiation and radiation dose deposition. We present two investigations performed with the GEANT4 Monte Carlo simulation toolkit. The first study compares the scatter-to-primary ratio and effects of scatter for inverse-geometry and cone-beam dedicated breast CT systems. The second study examines reduction of radiation dose to the breast during conventional CT scanning through anatomically-based half-scan acquisitions.


Thursday, 3 April 2008, 2:00pm

Compressive Magnetic Resonance Imaging

Ali Bilgin, Dept. of Electrical and Computer Engineering, Dept. of Radiology, University of Arizona

An emerging theory known as "compressive sampling" or "compressed sensing," demonstrates that a very large class of signals can be accurately (or under some conditions exactly) reconstructed from far fewer samples than suggested by the Nyquist-Shannon theory. While the Nyquist-Shannon theory describes sufficient sampling by exploiting the band-limitedness of signals, this new theory defines sufficient sampling conditions based on compressibility of a signal. This remarkable result is expected to have far reaching implications in many fields. In this talk, we provide a brief overview of the emerging compressive sampling theory and illustrate how this theory can be applied to magnetic resonance imaging.


Friday, 11 April, 2008, 10:00 - 11:30am

Photoacoustic Tomography: High-Resolution In Vivo Imaging at New Depths

Lihong Wang, Distinguished Professor, Optical Imaging Lab, Dept. of Biomedical Engineering, Washington University in St. Louis

In this talk, we will present the development of biophotonic technologies for functional and molecular imaging by physically combining non-ionizing electromagnetic and ultrasonic waves via energy transduction. Key applications include early-cancer detection and functional brain imaging. Unfortunately, electromagnetic waves in the non-ionizing spectral region do not penetrate biological tissue in straight paths as x-rays do. Consequently, high-resolution pure optical imaging technologies are limited to within about one optical transport mean free path (~1 mm) of the surface of biological tissue. Ultrasonic imaging, on the contrary, provides good image resolution but suffers from strong speckle artifacts as well as poor contrast in early-stage tumors. We have developed ultrasound-mediated imaging modalities by combining electromagnetic and ultrasonic waves synergistically to overcome the above problems. In photoacoustic tomography PAT), an expanded pulsed laser beam diffuses into the biological tissue and generates a small but rapid temperature rise, which causes the emission of ultrasonic waves as a result of thermoelastic expansion. The short-wavelength ultrasonic waves are then detected to form highresolution tomographic images. Thermoacoustic tomography (TAT) is similar to photoacoustic tomography except that low-energy radio-frequency pulses, instead of laser pulses, are used. Although the long-wavelength radio-frequency waves diffract rapidly in the tissue, the shortwavelength ultrasonic waves provide high spatial resolution.


Wednesday, 16 April, 2008, 4:30pm

Novel Methods to Address Data Truncation Artifacts & Streaking Artifacts in 4D CBCT

Shuai Leng, Research Assistant, Dept. of Medical Physics, UW-Madison School of Medicine and Public Health

Cone-beam computed tomography (CBCT) using an 'on-board' x-ray imaging device integrated into a radiation therapy system has recently been made available for patient positioning, target localization and adaptive treatment planning. Due to the limited detector size, projection data are truncated which will generate truncation artifacts in images reconstructed with traditional algorithms like FDK. This issue will be addressed using a new image reconstruction scheme, filtering a backprojection image of differentiated projection data (FBPD). To tackle motion artifacts due to respiratory motion in lung patient scan, 4D CBCT has been proposed. While the poor image quality, strong streak artifacts and low CNR, limits its wide application in clinic. Two novel methods were proposed to reduce artifacts and improve image quality in 4D CBCT. A simple scheme will be presented to significantly reduce the streak artifacts. Another new scheme, Prior Image Constrained Compressed Sensing (PICCS), was explored to simultaneously reduce streak artifacts and increase CNR of 4D CBCT images.


Wednesday, 23 April, 2008, 4:30pm

Cone-Beam Reconstruction Algorithm Development and Validation for C-Arm Based CT

Tingliang Zhuang, Research Assistant, Dept. of Medical Physics, UW-Madison School of Medicine and Public Health

CT systems with multi-row detectors or flat-panel imagers provide larger volume coverage in one single gantry rotation and have been widely used for both diagnostic radiology and image-guided therapeutic procedures. However, the cone-beam nature of the data acquisition with large area detectors poses challenges for image reconstruction. The first challenge is that mathematically exact image reconstruction algorithms have to be developed for each individual cone-beam scan configuration. There is no general analytical algorithm which may be applied to any CT scanning geometry. The second challenge is that new data acquisition geometries must be developed since a single circle/arc scan can not enable artifact free image reconstruction from cone-beam projection data. In this talk, we will present several novel cone-beam image reconstruction frameworks developed in our lab. These frameworks are used to generate mathematically exact image reconstruction algorithms for each different source trajectory. As an example we will present images reconstructed from a C-arm based cone-beam CT system.


Wednesday, 30 April, 2008, 4:30pm

Image Guided Interventions via Novel Cone-Beam Data Acquisition and Reconstruction Methods

Brian Nett, Research Assistant, Dept. of Medical Physics, UW-Madison School of Medicine and Public Health

C-arm based cine/fluoroscopic guidance provides simultaneous high spatial resolution (e.g. 200 micron pixel pitch) and high temporal resolution (e.g. 33 ms). These characteristics along with ease of patient access and wide availability of compatible interventional tools have positioned this technology as the defacto standard for image guidance for many interventional neurological procedures. However, these systems are subject to the limitations of projection imaging including reduced low contrast resolution and confusion due to overlapping anatomy. In order to overcome these limitations interventional computed tomography (CT) has been implemented by several of the major manufacturers. We will discuss several novel methods to improve upon the current state-of-the art including: a method for accurate surgical tool placement combining interventional CT and fluoroscopy, a technique to reduce patient dose by an order of magnitude for multiple interventional CT scans, and the potential for imaging contrast dynamics via limited angle tomography.


Wednesday, 7 May, 2008, 4:30pm

Iterative Reconstruction: The New Frontier to Improving CT Image Quality

J. B. Thibault, Ph.D., GE Healthcare, Applied Science Lab

With the advent of helical scanning, multi-slice geometry, fast acquisitions, and new tube/detector configurations, Computed Tomography (CT) has enabled a host of new non-invasive clinical diagnostic applications. While driven by new hardware technology, these improvements have given rise to many reconstruction algorithms adapted to the change in geometry and sampling characteristics. Analytical reconstruction algorithms have been the focus of much attention in recent years due to the development of so-called "exact" inversion formulae. Iterative reconstruction algorithms fall into a different category: they are primarily designed to conform to the statistics of the data. Combined with appropriate physics modeling of the data acquisition process, they promise unparalleled noise, resolution, and low-contrast performance compared to Fourier-based approaches. This talk will focus on modeling and optimization challenges for iterative reconstruction in general and CT in particular. To conclude, practical benefits will be illustrated on low-dose clinical patient cases.