Page 390 - AWSAR 2.0

366 || AWSAR Awarded Popular Science Stories - 2019
guidance from my research supervisor. So, I continued working on it and showed that PM diffusion can also run significantly faster than TV. Imagine having to recalibrate a weighing machine every time one has to weigh something. Likewise, reconstruction using TV required tuning its parameter for every new data. This was not needed with PM diffusion.
This time, though the
proposal was rejected by the
journal, the response said
“all is good, but a comparative
study showing some of the
basic theoretical advantages of PM diffusion over TV is not published yet.” The very same comparative study which I was trying to publish earlier. I felt like this was my moment and communicated the earlier work along with new findings to the same journal. Furthermore, to support the idea of reproducible research, all the codes needed to reproduce the results were also publicly shared. We, finally, got the work published! It eventually grabbed attention of a group working on magnetic resonance spectroscopy at the University of California, Los Angeles, and the advantages of PM diffusion were later demonstrated using spectroscopic data as well.
The job was not done yet. The
edge-preserving methods, like TV and PM diffusion, generally have a problem that they tend to introduce an artifact of their own, which looks like staircases and speckles. Obviously, we wouldn’t want that information in an MRI. In the case of TV- based reconstruction methods, these problems were already addressed. Therefore, when we suggest PM diffusion as an alternative to TV, it also becomes necessary to provide solutions for these artifacts as well. This led to the second phase of my work where we developed a method to
prevent these artifacts from appearing, rather than removing it after it has formed in the reconstructed image. When we finally had the results, publishing didn’t take that long. Thanks to the lessons learnt from all the earlier rejections.
Even after both these works, just like TV, the stand-alone performance of PM diffusion in terms of image quality was not comparable to some of the more recent methods. Soon, we realized that this was due to a factor common to both PM diffusion and TV. Both methods generally approximate the edges using computations along just the horizontal and vertical directions, while the actual orientation of the edges in an MR image
Much like the journey of any other researcher, mine also started with understanding the literature, studying all the different ways in which people have already tried to reconstruct the magnetic resonance (MR) image. The first task my guide assigned was to identify a missing link/knowledge gap in the literature, where I could make an original contribution.