Current Projects:	Accelerated MRI using Parallel Imaging Correction of artifacts in MR Bayesian techniques for optimal reconstruction Arterial segmentation from angiographs
Past Projects:	Image Interpolation using Regularized Least Squares Wavelet-based image denoising Design of wide-bandwidth microwave sensors using coplanar waveguides paper

1.  Motion Correction in MR:                                                                                                                                                                               

---

                                                                                                                                                                                                                                                                         Motion-corrupted angiograph                                                 After correction
         

Patient motion is quite likely during typical MR scans, lasting up to 2 minutes.  This causes extremely disturbing artifacts, especially in angiographs, as shown above (left).  We have designed an iterative correction method based on successive convex projections.  We apply a series of 4 projections (P1 to P4 shown on left) to the corrupted image, which nudge the bad image slowly towards an artifact-free image.

Download paper.                           








2.  Total Least SENSE : Optimal parallel reconstruction in presence of sensitivity errors

---

Conventional least-squares based SENSE methods for parallel imaging are prone to errors in obtaining sensitivity maps. Our new technique is insensitive to random Gaussian sensitivity noise.  A new total least squares algo was derived for this problem from a maximum-likelihood formulation.

Figure on left shows input vs output SNR for both conventional SENSE, and Total Least SENSE.  We obtained more than 20 dB SNR gain with our algorithm in noisy cases, as shown.
Download paper.






                                            
       

        
        Standard SENSE reconstruction                  Total Least SENSE reconstruction


3.  Bayesian techniques in MR:

---

Since reconstruction and processing problems in MR are freqiuently ill-posed or under-determined, a Bayesian framework is desirable for such problems.  Unfortunately, Bayesian priors are not as readily available in MR as in image processing.  This project aims to develop a set of Bayesian methods for MR problems.  As a first pass, we have developed a maximum a posteriori (MAP) method for MR angiography, whereby we use a jointly Gaussian signal model for MR data.  This allows us to use covariance statistics to obtain better angiographs compared to standard recosntructions.  As seen below, the conventional parallel reconstruction truly messes up the angiogram, since it suffers from excessive noise amplification, which is further worsened by mask subtraction used to obtain the angiogram.  Our method on the other hand manages to preserve diagnostic quality.
 










  

             Unaccelerated angiograph            Accelerated 3x, reconstructed by MAP   Reconstructed by conventional SENSE