Description: Description: Description: Description: Description: Description: Description: 75x75

MIDAS eNewsletter




New and Improved


Segmentation: The default segmentation method in MIDAS uses an older version of the FSL/FAST segmentation routine and we have found this to be prone to incorrect classifications of subcortical gray-matter structures. An alternative segmentation method has now been implemented using SPM, which appears to produce better results. This is included as part of the IDLSEG program. The SPM options are highlighted in the image shown to the right, and include options to run SPM using the standalone (i.e. compiled) or native installations. To use this function users will have to install SPM from their main site, and set the path to the installation in the ProcFile for IDLSEG.

Faster FITT:  Changes have been made to the FITT program to speed it up, primarily through use of an optimized FFT function. Thanks to Saumya Gurbani for this effort. Results indicate no differences in the resulting metabolite maps with the previous implementation, but to enable sites to continue using existing processing pipelines we have released this as a different version, 2.1. To use it in the BATCH processing pipeline you will have to change “FITT2” to “FITT21” and modify the processing file to enables the LOWESS and FFTW processing options, which are under the Baseline and Optimize tabs in the Editor.

The METAFIT program has also been modified to include an additional parameter “FittVersion” (see also the example given below), which sets whether it runs the old version of FITT, “2”, or the new “21”. The two versions of FITT cannot be run in the same IDL session as IDL flags an error from conflicting data structures.

Cramer-Rao Lower Bounds: A change has been made to the FITT program for calculation of Cramer-Rao Lower Bounds. Previously, the program would get an estimate of the noise, which is needed for this calculation, from the right most part of the spectrum; however, this could result in anomalous values, as for example in the spectrum shown to the right where the CRLB values for NAA, Cre, and Cho were 47%, 40% and 64%, even though the fit result, shown in orange, is excellent. An obvious feature in the spectrum, the large lipid signal between 0.0 and 1.0 ppm, gives a clue to what has happened. Even though the baseline is subtracted prior to making the noise measurement the resultant standard deviation value was clearly incorrect.

The FITT program has now been modified to allow definition of any spectral region for the noise measurement, with the preferred region being the leftmost (downfield) part of the spectrum. The “new “QA” tab for the parameter settings is shown to the right. To use this it is necessary to retain the full spectrum in FDFT, i.e. not select just the right half, which is currently the default processing. The CRLB values with this change are generally much smaller than with the previous method.

EPSIBugFixEPSI2: A fix has been made to the EPSI2 regridding program that gets rid of a shading artifact that was only seen with the 20- and 32-channel RF coils. This artifact can be most clearly seen on the water Reference image (not the Water_SI, which is intensity corrected), and an example of this shading is shown to the right (top). The reprocessed data does not show this shading (bottom).

The default EPSI processing deletes the volumized raw data to save space, therefore if you have data with similar shading it is necessary to reimport the raw data and redo the processing.

New EPSI sequence WIP: A new release of the EPSI WIP has been released by Siemens that include the spatial reconstruction and multichannel combination. This reduces the data sizes for export from ~26 Gb (for 20-channel coil) to 281 Mb. There are some limitations, however, in that large memory size is needed for the acquisition computer and the reconstruction can take from 10 min (for 16 channels) to 50 min (for 64 channels) after the scan, during which time no other image processing can take place. The spectral reconstruction and fitting is still done using MIDAS processing and a modified processing file is available for this on request.

One important change in the new version of the EPSI WIP is that the X phase-encoding direction has been flipped. This requires an additional processing step, ROTOR. An example processing pipeline for this version of the sequence will be available on the web site.


MIDAS Tips, Questions, and Answers


Viewing Spectral Fit Results: The FITT program has always included the Viewer function, which brings up a new widget that can be configured to display different combinations of the data, the spectral model, and fit results; however, this is not most convenient method of viewing fitted results. Here we describe a quick method of overlaying the fitted result on the data, which is built into the new SID v2.0. This new version of SID was introduced in the last newsletter.

The first requirement is that the fitted spectrum must be saved in FITT. This is done by setting the option to save the fitted result, which is done by going to the Output tab in the Processing File Editor. Options are available for the metabolite fit result only or the metabolite+baseline fit result, as shown in the figure below.

Of these, the most useful option is the combined FITT+BASELINE result, which gets saved as a new Spectral node under the SI series and labeled as “Spectral_FitBase”. Since these are smooth functions these datasets are saved using a compressed file format to save space.

To view this result in the SID program use the “File->Open/Close 2nd SI” option and then select the node labeled Spectral_FitBase under the SI series. The program will initially display the data and the fit result as separate plots, but these can be overlaid using the “Options->Dual Plot Overlay On/Off” option. This option and an example plot are shown below.

Making Slides With Spectral Plots: The plots displayed in SID can be copied in a variety of formats to include them in your papers and presentations. The easiest way to do this is to save the plot window as a bitmap, using the “File->Copy to Clipboard” function and then paste the image into your program, but this often does not display well in presentations. A better option is to use a vector format and use Insert/Pictures to put this in your PowerPoint presentation. The Spectral “File->Output” option in SID makes available three suitable formats, EPS (Encapsulated PostScript), CGM (Computer Graphics Metafile) and WMF (Windows Metafile). These can be selected by checking the “Change Output Format” or using the “Options->Preferences” on the main SID window. Unfortunately the support for these formats in PowerPoint has changed with the 2013 version, and currently only the WMF format is recognized; however, this still needs an additional step to make it work. When the WMF file is first inserted into PowerPoint the spectra are not shown. You first need to select “Ungroup” or “Edit Picture” to convert it to a drawing object, and then you can select the whole object or individual various sections to set the plot colors. The text on the axis also comes over as a “system” font and you may want to set the font type.


Developer’s Corner


METAFIT: Recommendations for spectral fitting include that the spectral model should be kept as simple as possible and that you need to have good starting values. The METAFIT program aims to follow these recommendations using multiple calls to the FITT program, with each taking on a different task. The first step determines the phase and frequency and applies these corrections to the data. For this step, the spectral model is kept very simple, using only the NAA, creatine, and choline singlets. In the second step the data is smoothed, then the fitting is carried out with a complete spectral model on what is now relatively high SNR data. Because B0 was corrected beforehand the linewidth is maintained after smoothing and tight constraints can be used for frequency and phase. The results of this fit are then used as the starting values for a final run of the FITT program using the unsmoothed version of the phase and frequency corrected data.

Two other processing options are available. Between the first and second fit a PCA denoising step can be applied (Abdoli et al, 2016). This has been found to improve SNR without increasing linewidth, although it is recommended that the degree of denoising be kept small to minimize loss of detailed spectral information. By applying this after the phase and frequency correction the performance is improved. The second option is that the baseline found in step two, or a scaled fraction of this signal, can be subtracted from the data. The aim of this is to simplify the baseline model for the third run.

The program also allows the results from the second analysis to be saved, which is the smoothed version of the data. This has been found to be useful for those metabolites that are most difficult to detect, namely Glx, mI, and lactate. The selection of images that get saved must be defined in the processing file and the program saves these with the extension “_Sm”, e.g. Glx_Sm.

An example processing file is shown below. Most of these are self-explanatory, but for further information please look at the help file.

The METAFIT program will take longer than a single run of the FITT program and its main benefit is for analysis of short TE data.


Abdoli A., Stoyanova R., and Maudsley A.A. Denoising of MR spectroscopic imaging data using statistical selection of principal components. Magn Reson Mater Phy, 29(6), 811-822 (2016).


Example Results


Reducing Lipid Artifacts Using SVD: The whole-brain EPSI acquisition uses inversion-nulling to reduce the signal from subcutaneous lipids and lipid extrapolation (the LITE program) to further reduce the spread of these signals within the brain. However, strong lipid artifacts can still occur, for example due to motion. In many cases the lipid signal may not be severely overlapping with the NAA signal, but nevertheless it can appear as a rapidly varying signal that cannot be described by the baseline model used in the spectral fitting, as shown in the example to the right, which causes errors in the metabolite fitting. In a recent study (Goryawala et al. ISMRM 5529, 2017) we showed that a frequency-selective fitting and subtraction of these lipid signals can reduce these baseline errors. The improvements gained from this processing will be most for datasets that have moderate contamination from motion effects, but in these cases it can increase the volume of the brain that meets the quality criteria.

An example of NAA images obtained using the standard processing (top) and following the lipid-selective subtraction (+SVD) is shown below. Reduced lipid contamination and effects of subject motion can be seen with the additional processing.

Limitations of the SVD lipid removal processing include that any lactate signal would also be removed and the additional time needed for this processing. The lipid subtraction method is available in the HILITE program, which will be included in the next release.


More Midas


Glitter Lips Midas Kiss2 

MIDAS Kiss, Undercover GlamorGlitter Lips - Midas KissGlitter Lips - Midas Kiss




Congratulations and many thanks to our collaborators for these recent reports:

·        Maudsley AA, Govind V, Saigal G, Gold SG, Harris L, Sheriff S. Longitudinal MR spectroscopy shows altered metabolism in traumatic brain injury. J Neuroimaging. 27(6):562-569 (2017).

·        Donadieu M, Le Fur Y, Maarouf A, Gherib S, Ridley B, Pini L, Rapacchi S, Confort-Gouny S, Guye M, Schad LR, Maudsley AA, Pelletier J, Audoin B, Zaaraoui W, Ranjeva JP. Metabolic counterparts of sodium accumulation in multiple sclerosis: A whole brain 23Na-MRI and fast 1H-MRSI study. Mult Scler. 2017 Oct 1:1352458517736146. [Epub ahead of print]

·        J. Mauler, A.A. Maudsley, K-J. Langen, O. Nikoubashman, G. Stoffels, S. Sheriff, P. Lohmann, C. Filss, N. Galldiks, E. Rota Kops, N.J. Shah. Spatial Relationship of Glioma Volume Derived from FET PET and Volumetric MRSI: a hybrid PET-MRI study. J. Nuc. Med. In Press (2017).

Please let me know of additional publications that make use of MIDAS.


Andrew Maudsley, December 2017

This newsletter provides information to the developers and users of the MIDAS software package. If you would prefer not to receive notification of these reports just let me know (