week 5

I met with Dr. Gauthier on Tuesday in the lab. We discussed several challenges we were facing regarding MS lesion QSM analysis. One of the challenges was the thickness inconsistency between 2mm and 3mm QSM acquisitions. We needed these two acquisitions to have comparable susceptibility values, but for now the 2mm cases were slightly brighter than the 3mm one. The other challenge was to efficiently remove the veins from the white matter on QSM. For the first challenge, I  applied MRI sampling theorem to truncate one third part of the k-space data by using low pass fermi filter, and transform the k-space data back into image space to get resized 3mm volume. The result why we needed to do this was that only the baseline acquisition utilized 2mm acquisition parameters, and the following-up acquisitions should be registered to the baseline, which made the whole workflow redundant and inaccurate since registration procedure from 3mm to 2mm introduced uncertainty and took much time. Right now I'm working on eight patients cases, and after we verified our resizing procedure, we'll apply it to the whole MS lesion dataset and continue the following workflow.

The second challenge was more difficult. We wanted to estimate the general susceptibility value in the healthy white matter tissue to compare it with MS lesion tissue, but veins in white matter look bright because of large susceptibility value deoxygenated hemoglobin in veins have. Therefore, it's essential to get rid of the veins before calculating the average susceptibility values in white matter. I'm looking into some papers dealing with the similar issues in medical research, using tools from contour segmentation to state-of-art convolutional neural network. I feel excited because working on this project, I can learn a lot of knowledge including medical background in MS lesion research, physical principles and computational tools of QSM, and powerful artificial intelligent technologies to assist doctors accomplish complex and cumbersome work.