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

MIDAS eNewsletter




New and Improved


MINT:  This program was introduced in our last newsletter but since then several new features have been added that warrant another mention. The program provides ROI integration of spectra, where ROIs can be defined in an imported mask, e.g. as a user-defined region such as a tumor volume, or as part of a brain atlas. The first new features added to this program is the ability to separate component spectra from separate tissue types, typically being from gray-matter or white-matter.  This uses a matrix decomposition procedure, illustrated below, that includes information on the fractional tissue content at each voxel.

Gray and White Spectral Estimation

This procedure has been shown to considerably improve estimation of metabolites in gray matter, where tissue volumes are typically smaller than the MRSI resolution. More details on the method are described in the publication by Mandl et al.

The second option added to the MINT program has been the ability to integrate spectra over the volume sampled by a single-voxel acquisition. A report comparing results from MRSI using ROI integration and single voxel acquisition is in development. A further development of the MINT program is that user documentation has now been added.

The MINT program also includes some tools that simplify consolidating multiple results (from different ROIs and subjects) and for viewing the integrated spectrum and the fitted result. Viewing the result uses the VIEWSPECTRA program, which can also be run from the Tools section of the MIDAS toolbar. The figure shown to the right shows an example that was generated using MINT from a mask that identified voxels within an oligodendroglioma, and which resulted in integration of 276 (interpolated) voxels. The integrated spectrum result (blue) and the fit result for the metabolite+baseline (red) were then displayed using the VIEWSPECTRA program. Note that to get the fit result this option has to be enabled in the FITT program.

TMAPS: The TMAPS program creates temperature maps based on the frequency of the water resonance relative to a reference derived from the metabolite resonances. The processing required for the temperature measurement differs from the processing normally carried out, so it requires a modified processing pipeline. Some examples from this module are shown below and can be found in a recent publication.

QMAPS: The QMAPS program creates a “Quality Map” based on one or more criteria that indicate the quality of the spectral data. Up to now this has only used the spectral linewidth, but now additional options have been added to the QMAPS criteria, including Cramer Rao lower bounds (CRLB), confidence limits, measures from the water resonance, and outlier detection. The topic of robust spectral quality evaluation remains an active area of investigation and more experience and testing is still needed to find a robust quality evaluation method, but here are some observations from our current set of options:

·        The fitted spectral linewidth has been found to be a good “first pass”. It works for most cases, but this still gets some false results.

·        CRLB values are the most widely used for spectral quality evaluation. The CRLB for creatine was added since it was felt this would be more reliable than NAA, for example, which can at times be totally absent. However, this can at times be too restrictive and eliminate voxels for which the image appearance is otherwise acceptable, while it may be appropriate for quantitative analyses.

·        Using measures from the water resonance is of particular value for studies of brain lesions, where one or all metabolites can at times be absent. In contrast to a quantitative spectral analysis, for an image analysis a region with no metabolites should still be marked as good quality if measures such as the water linewidth and baseline/lipid signals are acceptable. Using any parameters derived from the metabolite fitting would produce a false negative.

·        A spectral outlier detection has been added using an idea from Kalyanam et al. This first evaluates the mean and standard deviation of every spectral point and then eliminates voxels that have spectral points greater than some number of SDs from the mean value. This is of value to eliminate voxels with large lipid or residual water signals. To allow for cases where a metabolite peak may be absent in a lesion, which could be identified as a negative outlier, there is also an option to “Exclude major peak regions”. This test typically removes less than 3% of voxels (after the linewidth test), but is felt to be worthwhile.

The program includes an option to set those voxels that do not pass the quality test to zero in all metabolite maps, with the idea that someone unfamiliar with the image features that commonly arise from artifacts in MRSI would not be misled. However, the down side of enabling this option is that metabolite maps can end up with having missing regions, and often look better if the voxels are not set to zero.

 Support for Single Voxel MRS:  MIDAS has supported spectral analysis for single voxel acquisitions (for Siemens format only) for a long time, but this used an intermediate data format and was not integrated within the MIDAS data management system. SVS support has now been fully integrated with the MIDAS, allowing spectra to be imported using the standard procedure (Volumizer), Fourier Transformed (FDFT), analyzed using the FITT program, and viewed in SID. It uses a “SVS” series node that is selectable from the MIDAS browser. Additional options for viewing the data and fit results are planned.

Metabolite Map Normalization Using SINorm: This program applies bias field correction, signal normalization using the tissue water signal, and creates metabolite ratio maps. A new option that has been added is to create an additional set of maps that are normalized by the mean parameter value in “Normal Appearing White Matter” (NAWM). The program seeks to find the largest contiguous white-matter volume and uses the mean value from this region to normalize each SI parameter map. For studies with brain lesions this will automatically identify the WM region contralateral to the lesion, and if the OTHER tissue map is defined, which encompasses the edema+lesion volume, then this will be included in the calculation. There are obvious limitations for cases with multiple lesions or very small lesions. An image of the detected volume is displayed (shown in the figure to the right), showing the initially identified WM at MRI resolution and the final volume at SI resolution. The normalized ratio maps are identified with the prefix “rel”, i.e. relCho/NAA. In addition, a frame is created with the binary mask used to calculate the mean white matter value, labeled NAWM_map.

A potential application of this “self normalization” approach (which is also used for several other types of image analysis) is to improve thresholding for lesion delineation in cases where the values in the normal tissue may vary, as for example, using Cho/NAA in studies of gliomas, where metabolite values in NAWM have been shown to change.


MIDAS Tips, Questions, and Answers


Processing with Sagittal-Acquired T1:  Two related problems can occur with a sagittal-acquired T1, the segmentation and registration programs can fail. This occurs because there is a large volume of the image covering the neck region that results in a failure of the BET program. There are two solutions to this problem.

1)      The “TRUNCATENECK” utility can be used, which simply zeros out a portion of the MRI covering the neck. This can only be run in batch mode, with the command line:

TruncateNeck, Subject_XML, Study_ID, Series_ID

2)     The IDLSEG program can be used with the option for “Run Iterative BET”. The repeated use of BET for this purpose was described by Fagiolo et al.

Gremlins and Other Strange Behavior: Programs don’t always do what they are supposed to do and with different operating systems and IDL versions we have noticed a few peculiarities. Here are two that crept in following the release of Windows 7:

·        When using the “Export to Analyze” function in the MIDAS Browser (see figure to the right) a dialog box will open up to select the location and name of the file to be saved. However, this dialog window always appears in the center of the screen, and will be hidden behind the Browser window. Therefore, before selecting this function just move the Browser window off to one side.

·        Another problem with the Browser is that sometimes it locks up. This seems to happen only after MIDAS has been used for a long time and is accessed through IDLDE (i.e. not from the toolbar which uses IDL.exe). The only option for this problem is to close the Browser window and IDL session.

A long-standing issue with the MIDAS Viewer is that it has problems with some images that have different FOVs. The program aims to use the largest FOV of all the image series that it reads in, but it can get confused with perfusion and diffusion maps, resulting in these appearing with different scaling in x and y and being mis-registered. The latest release of this program should have fixed these problems (we thank Sulaiman for taking this on), so we recommend downloading the latest release of MIDAS. If you don’t want to update your software, another option is to edit the subject.xml file and make sure the SI and SI_REF series nodes are placed at the end of the file.




Siemens VD Implementation: The EPSI sequence is now available for the Siemens VD software platform and is distributed as WIP # 814 (not as a C2P as is still done for the VB version). The new version includes prospective correction for frequency drift using a short navigator measurement incorporated into the water suppression module. We thank Sinyeob Ahn and Wes Gilson of Siemens for this collaborative effort.


Developer’s Corner


Spectral Basis Functions for FITT: The basis functions used in FITT are generated by spectral simulation using the VESPA program.  See the help files for that program for instructions on running the simulation.

To export the basis functions created in VESPA the “Third Party Export” to “MIDAS Prior XML” is used (see figure to the right). This then brings up a widget that allows combining metabolites (e.g. glutamate+glutamine to get Glx) and saving the results to an XML file.

The basis functions are saved as a list of frequency, amplitude and phase values for every resonance, not as a spectrum, so the same basis functions can be used for different sweep width or number of sample points.

To include the basis functions in the FITT processing file you must open the “Editor”, select the “Priors” tab, click the “Prior File\Browse” button, and navigate to the XML file that you created from the VESPA simulation and select it. This section of the widget is shown in the following figure:

You may want to check the PPM range, but it is recommended at this point to save the processing file, exit the program, and restart fit and reload the processing file before selecting the metabolites to be included from the list of metabolites available.

To check the fitting you can select the “Viewer” option from the main FITT widget and run the fitting on single voxels. Setting up new prior information can be an iterative process between testing in the Viewer and changing parameters in the Editor. At the end of this process, it is recommended to always exit the program and restart it to check that the settings are correctly saved to the processing file.

The spin parameters used in VESPA are derived from the report of Govind et al. 2000, for which an updated list is available in a recent publication.  A document showing simulated spectra for several brain metabolites at TE=17.6 can be found by clicking on the image shown on the right. This provides a useful reference when analyzing short-TE data.

Processing Files Location: The default location for the processing files is a subdirectory under each project directory, called “ProcessingFiles”, e.g.

C: \MIDAS_Data\MyProject\ProcessingFiles

Since this is a “local” subdirectory, the path shown in the Importer (and the project.xml file) is displayed as “.\ ProcessingFiles”. This means that if you move your project to another location you do not have to update the path for the processing files.

The Importer will only assign the name “ProcessingFiles” to this directory, but there are no restrictions on the name or location of this directory. If you have multiple projects that use the same processing files, then it is convenient to have all projects point to the same directory location, e.g.:


Where M: may be a common drive shared by multiple users. Note that in this case the full path has to be specified in the _project.xml file for each project that uses these files. The Importer will automatically do this when you select a non-local path when creating a project. Another situation may be where you may want to keep track of multiple variants of the processing files for the same project. In this case you can rename the processing files directory, e.g.




The requirement for this to work is that the full path must be defined in the _project.xml file, i.e. it cannot use the relative path “.\...”.

 MIDAS MRSI Group: A group has been set up where you can post questions to the MIDAS+EPSI development team. This link will take you to the group. To post you first have to get a Google account.


Example Results


GBM Post-Surgery: The following image shows the choline map for a volumetric MRSI study taken 23 days post-surgery and prior to treatment for treatment of glioblastoma. The spectra shown indicate metabolic changes seen outside of the contrast-enhancing volume of the residual tumor.

This example illustrates an advantage of obtaining volumetric MRSI as well as using the “whole-brain” approach to include regions near the cortical surface. Acknowledgements to Drs. R. Stoyanova and F. Ishkanian for this data.

Temperature and Susceptibility Mapping: As mentioned above, a module to create temperature maps from the volumetric MRSI data has been added and results of this analysis have recently been published in NeuroImage. Unfortunately, the quality from single studies provide little information on individual temperature variations, but by averaging over multiple subjects some interesting features can be seen. An example is shown below, which shows the mean apparent temperature averaged over 150 subjects and derived from the frequency difference between water and NAA:

It turns out that the features shown in this image largely reflect the effects of local magnetic susceptibility variations, which affect the water and NAA peaks differently. It is not possible to separate out the frequency shifts caused by temperature and susceptibility, hence the use of the term “apparent” temperature. Further investigations indicated that these localized susceptibility variations also cause frequency differences between the resonances from different metabolite, an example of which is shown in the following figure: 

This results shows that the metabolite resonance frequencies are not fixed, and therefore spectral fitting must allow for some variation between peaks (in addition to accounting for B0 variations), which is included in the spectral model used in FITT. In addition to showing another application of MIDAS and volumetric MRSI – for whole brain temperature mapping, these results illustrate the potential of signal averaging over multiple studies using the voxel-based image analyses methods provided by the registration and PRANA programs provided in the MIDAS package.

The frequency differences shown in the previous figure are most strongly seen with maps involving NAA and appear to be associated with the axonal orientation and reflect the intra-axonal compartmentation of NAA. We can provide additional evidence for this by comparing the frequency difference maps with maps of the axonal orientation derived from a DTI measurement in the same subject. The following figure shows for a coronal section through the corona radiate the mean diffusion tensor direction map obtained from 20 subjects, with axonal tracts aligned along the direction of the B0 field shown in dark blue; the frequency difference map for NAA-Choline; a map of the angle of the principal diffusion vector, where darker values indicate a smaller angle of the diffusion vector to the direction of the B0 field; and the MRI used for spatial normalization. DTI reconstruction and image registration was carried out using DTIStudio and DiffeoMap.

This result shows a correspondence between decreased NAA-choline frequency differences and the angle of larger axonal structures relative to B0. This association is seen only with larger axonal structures, which may reflect the limitations of the lower-resolution MRSI data. The result additionally demonstrates the easy integration of DTI with the MRSI data using the PRANA program. Additional details will be presented at the ISMRM 2017 meeting.




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

·        Maudsley AA, Goryawala MZ, Sheriff S. Effects of tissue susceptibility on brain temperature mapping. Neuroimage. 2016 Sep 27.

·        Cordova JS, Shu HK, Liang Z, Gurbani SS, Cooper LA, Holder CA, Olson JJ, Kairdolf B, Schreibmann E, Neill SG, Hadjipanayis CG, Shim H.  Whole-brain spectroscopic MRI biomarkers identify infiltrating margins in glioblastoma patients. Neuro Oncol. 2016 Aug;18(8):1180-9

·        Cordova JS, Gurbani SS, Olson JJ, Liang Z, Cooper LA, Shu HG, Schreibmann E, Neill SG, Hadjipanayis CG, Holder CA, Shim H. A systematic pipeline for the objective comparison of whole-brain spectroscopic MRI with histology in biopsy specimens from grade III glioma. Tomography. 2016 Jun;2(2):106-116.

·        Goryawala M.Z., Sheriff S., Maudsley A.A.. Regional distributions of brain glutamate and glutamine in normal subjects. NMR Biomed. 29(8):1108-1116 (2016).

·        Abdoli A, Stoyanova R, and Maudsley AA. Denoising of MR spectroscopic imaging data using statistical selection of principal components. Magn Reson Mater Phy, Jun 3. [Epub] (2016).

·        Widerström-Noga E, Govind V, Adcock JP, Levin BE, Maudsley AA. Subacute pain after traumatic brain injury is associated with lower insular N-acetylaspartate concentrations. J Neurotrauma. 2016 Jan 15. [Epub]

·        Ding XQ, Maudsley AA, Sabati M, Sheriff S, Schmitz B, Schütze M, Bronzlik P, Kahl KG, Lanfermann H. Physiological neuronal decline in healthy aging human brain - An in vivo study with MRI and short echo-time whole-brain 1H MR spectroscopic imaging. Neuroimage. 15:137:45-51 (2016).

·        M. Donadieu, Y. Le Fur, A.A. Maudsley, A. Lecocq, S. Gherib, E. Soulier, S. Confort-Gouny, F. Pariollaud, M-P. Ranjeva, J. Pelletier, M. Guye, W. Zaaraoui, B. Audoin, J-P. Ranjeva. Metabolic voxel-based analysis of the complete human brain using fast 3D-MRSI: Proof of concept on Multiple Sclerosis. JMRI. 44(2):411-9 (2016).

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


More Midas


Midas Star – The Reiki Symbol for Abundance and Prosperity

The Midas Star - Reiki Abundance and Prosperity Symbol


Andrew Maudsley, November 2016

This newsletter provides information to the developers and users of the MIDAS software package. Any questions or comments can be sent to me at amaudsley at