I am seeking consensus definitions for Preprocessing and Reconstruction in the context of MRS.
I have noticed these terminologies being used in various articles, softwares, and discussions, but I often find myself confused about their distinctions.
Based on my current understanding, I perceive preprocessing as encompassing all operations performed on the spectra prior to fitting, while reconstruction involves transforming the raw FID data into the spectra domain.
I kindly request your assistance in providing precise consensus definitions for these terms. Additionally, I would greatly appreciate it if you could recommend references that can enhance my understanding of these concepts.
Transformation of the time-domain FID into the spectral (frequency) is usually described by the standard method for doing this: Fourier transformation.
âReconstructionâ would more typically be used in imaging contexts, or in MRSI to recover spatial information (perhaps confusingly, this also often involves Fourier transforms from k-space).
Thank you very much for your attention and for sharing the paper on terminology and concepts for the characterization of in vivo MR spectroscopy methods and MR spectra. I will definitely take a look at it to gain further insights.
Regarding the term âReconstruction,â I appreciate your clarification that it is typically used in imaging contexts or in MRSI to recover spatial information. However, I came across an article which uses the expression âNMR Spectra Reconstructionâ (https://chemistry-europe.onlinelibrary.wiley.com/doi/full/10.1002/chem.202000246), which seems to discuss the topic of getting the MRS spectra from the FID. In Figure 6 of the article, they demonstrate this reconstruction process. Could you please provide some insights into this or confirm if my understanding aligns with the content of the mentioned article?
There are plenty of examples where these (and other) terms have been used inconsistently in the literature â indeed, one motivation for the consensus paper was to encourage greater consistency.
That said, the article you mention is discussing a slightly different application: deep learning in NMR spectroscopy. Often, you could think of âreconstructionâ as recovering information which has been âlostâ in the sampling (eg, through various acceleration techniques). In this case, they discuss (non-uniform) undersampled data, possibly in multiple dimensions. So they are reconstructing high-quality spectra from undersampled data, using a CNN. See Figure 6 of that paper: the Fourier transform from FID to spectral domain is still labelled as such.
Thank you for your response and clarification. I appreciate your insights on the proper usage of these terms in the literature.
Your explanation has been very helpful, and now I feel more confident in discerning the proper usage of these terminologies. Thank you once again for your insights.
I hope that this topic can be helpful for other people who have faced the same questions as I did.