Travis B. Smith, PhD

PhD in Electrical Engineering, May 2012
Pre-doctoral Fellow of the American Heart Association
TA Fellow of the USC Center for Excellence in Teaching

Research Assistant in the
Magnetic Resonance Engineering Laboratory
Signal and Image Processing Institute
Ming Hsieh Department of Electrical Engineering
University of Southern California

Contact Info:

3740 McClintock Ave, EEB 412
Los Angeles, CA 90089
phone: (213) 740-4652
fax: (213) 740-4651


Journal (peer-reviewed):
T Smith, K Nayak. Automatic off-resonance correction in spiral imaging with piecewise linear autofocus. Magn Reson Med (in press).
T Smith, K Nayak. Reduced field of view MRI with rapid, B1-robust outer volume suppression. Magn Reson Med 67(5):1316-1323, May 2012.
T Smith, K Nayak. An overview of MRI artifacts and correction strategies. Imag Med 2(4):445-457, August 2010.
T Smith, Z Zun, E Wong, K Nayak. Design and use of variable flip angle schedules in transient balanced SSFP subtractive imaging. Magn Reson Med 63(2):537-542, February 2010.

Conference (peer-reviewed):
T Smith, K Nayak. Automatic off-resonance correction with piecewise linear autofocus. Proc. 20th Annual ISMRM, Melbourne, May 2012, p. 218. (Summa Cum Laude Merit Award)
T Smith, K Nayak. Recovering fine-scale features in spiral imaging with piecewise linear off resonance correction. Proc. 19th Annual ISMRM, Montreal, May 2011, p. 4579.
T Smith, K Nayak. A novel B1-insensitive outer volume suppression pulse. Proc. 19th Annual ISMRM, Montreal, May 2011, p. 4414.
T Smith, K Nayak. Retrospective slice prescription compensation improves coronary cross-sectional area measurement by MRI. Proc. 14th Annual SCMR, Nice, February 2011, P110.
T Smith, K Nayak. Analysis of small dilation detection in coronary angiography. Proc. 18th Annual ISMRM, Stockholm, May 2010, p. 1242.
T Smith, K Nayak. Dynamic imaging motion artifact reduction using adaptive k-space polynomial interpolation. Proc. 18th Annual ISMRM, Stockholm, May 2010, p. 3065.
T Smith, Z Zun, E Wong, K Nayak. Variable flip angle schedules for detecting prepared longitudinal magnetization in snapshot balanced SSFP. Proc. 17th Annual ISMRM, Honolulu, April 2009, p. 2661.

Automatic deblurring (autofocus) in spiral MRI

Publication: conference abstract

I have developed a new method to correct blurring due to off-resonance phase errors in spiral and radial images without knowledge of the field map. Unlike current autofocus algorithms, this one does not use objective functions. The image is divided into blocks and linear field map estimation and correction are performed on each block. The local linear coefficients are estimated through a combination of k-space spectral analysis and mapdrift, an image-domain correlation technique. The deblurring performance is comparable to field map-based techniques. The proposed method requires only a blurry image (magnitude and phase) and a trajectory time map, and is suitable for fine-resolution (Fig. 1) and low-SNR (Fig. 2) images.

Figure 1   Figure 2

Reduced field of view imaging at 3T with outer volume suppression

Publication: journal paper

Videos of small tip angle spiral pulses: conventional tip-down with 1 spiral arm and tip-back with 2 spiral arms. (H.264) Shown from left to right are the waveforms, the spectrum, and the spatial profile.

I have developed a new outer volume suppression (OVS) pre-pulse for reduced field of view spiral imaging at 3 Tesla. The OVS pulse suppresses signal from the patient exterior to a cylindrical volume. This allows the imaging field of view to be decreased without an aliasing penalty. The reduced field of view enables shorter spiral readouts (to mitigate blurring artifacts), finer resolution (without the usual increase in scan time), or shorter breath-holds (Fig. 1).

The novel design of our pulse includes elements that a) impart robustness to RF field inhomogeneity for scanning at higher magnetic field strengths, and b) reduce the pulse duration (to 10.2 msec) for rapid application. The cylindrical OVS profile is obtained with a –90° 2-D spiral spatial sub-pulse. This tipback sub-pulse incorporates a unique solution to the Bloch equations that is exploited to achieve two-shot performance from the one-shot spiral pulse (Fig. 2).

Figure 1   Figure 2

Fast N-dim wavelet toolbox for MATLAB

Download the toolbox (coming soon)

We have developed an efficient and easy-to-use toolbox for fast N-dim wavelet transforms in MATLAB. The toolbox uses one function interface, regardless of the data dimensionality. Unlike other implementations, this toolbox places no restriction on data size, can accommodate highly asymmetric data volumes, and offers complete flexibility in the choise of decomposition level for each data dimension. Computationally, it supports double- and single-precision data (both real and complex), and is multi-threaded with Pthreads (currently available on linux and OSX platforms only). The toolbox currently supports up to 4-D data, but is easily scalable to process data with more than four dimensions. This toolbox was created in collaboration with R. Marc Lebel. Fig. 1 shows the speed performance of the proposed toolbox (Prop.) in comparison with other wavelet packages for 3-D asymmetric (N x N x 16) data.

Figure 1

Cross-sectional coronary artery imaging

I am researching improved methods of cross-sectional imaging of the coronary arteries. Here, images are formed by orienting the imaging plane perpendicular to the coronary artery (Fig. 1). Cross-sectional imaging is useful for testing the function of the endothelial cells that line the arteries. Endothelial dysfunction is considered to be an early marker of coronary artery disease, and is associated with all major cardiac risk factors. Current endothelial function testing with MRI requires measuring the small changes in vessel diameter that occur in response to non-invasive stimulators such as placing the hand in cold water (Fig. 2). This is a much safer alternative to X-ray angiographic methods, which require cardiac catheterization and exposure to ionizing radiation. I am collaborating with cardiologists at Stanford University Medical Center to develop MRI-based techniques which, in comparison, will be safe for early screening, serial examinations, and therapeutic monitoring.

Figure 1   Figure 2

Analysis of small dilation detection in coronary angiography

Publication: conference abstract

This work develops a statistical framework for analyzing the detection of subtle lumen dilations in cross-sectional images of coronary arteries (Fig. 1). I use this framework to relate detection performance to SNR requirements and minimum detectable dilation. I also discuss the use of a generalized likelihood ratio test to perform estimation and detection of the dilation. An example ROC curve from the analysis is shown in Fig. 2.

Figure 1   Figure 2

Efficient motion artifact reduction in dynamic imaging using adaptive polynomial interpolation (API)

Publication: conference abstract

This work discusses an algorithm to reduce motion artifacts in dynamic imaging. It seeks to balance the tradeoffs between low-pass filtering (good artifact suppression, but blurs motion) with linear interpolation (weak artifact suppression but good motion resolvability) by adaptively fitting polynomials to the data based on the degree of local temporal variation. Figs. 1 and 2 show results for various frames of a dynamic imaging sequences comparing reconstructions with sliding window (SW), low-pass filtering (LPF) with cutoff Π/13, linear interpolation (LI), and the proposed adaptive polynomial interpolation (API) algorithm. Frames 3, 7, and 12 are reconstructed from data containing no, moderate, and heavy motion, respectively. In effect, the algorithm selects the appropriate filter from a bank of low-pass filters (Fig. 3) for each frequency sample in the reconstructed frame. The API algorithm is inexpensive and suitable for real-time implementation.

Figure 1   Figure 2   Figure 3


Variable flip angle schedules in subtractive balanced steady-state free precession sequences

Publications: journal paper, conference abstract

This contribution introduces the design of variable flip angle schedules in subtractive balanced SSFP sequences using multi-stage numerical optimization. Schedules can be customized to optimize any property of the received signal using relaxation and off-resonance information. Fig. 1 displays example optimized schedules at each stage, with the final schedule shown in black. Fig. 2 demonstrates an application in myocardial tissue, where signal levels from a constant angle schedule (a) are compared with those from an optimized variable angle schedule (b).

Figure 1   Figure 2