Software New and Improved
A new version of MIDAS has been released that includes several significant changes:
EPSI2: Replaces the EPSI3D regridding program with much faster processing and new options. Comparisons indicate that the metabolite images processed with the new EPSI2 are comparable with using the older EPSI3D program (on average 1% higher (maximum 2.0%) using mean amplitude for lobar-analysis than for the EPSI3D program). While it is preferable to process all data with the identical pipeline it is possible to combine data processed with the old and new versions.
SEGMENTATION: Now includes the FSL/FAST version 4.1 (www.fmrib.ox.ac.uk/fsl/fast4). This version was recompiled to run under Windows and Jiping is thanked for that monumental effort.
LITE: The lipid k-space extrapolation program has been optimized for volumetric SI data. With typical settings (e.g. Max. iterations=60 and termination=0.05%) the program is roughly five times faster than the previous version (with # iterations=40).
FITT2.0: Replaces FITT v1.7 with significant speed improvements for multi-core computer. A 12-iteration fit for NAA/Cre/Cho on a 4-core CPU takes only 45 minutes, down from several hours with the previous version. Thanks to Brian Soher for this effort. Processing and prior information files from the previous version will need modification. FITT2 uses IDL features that are only available in the most recent versions of IDL, and we will only be supporting this under IDL v8.
REGISTRATION: A new version of the non-linear spatial registration program has been provided by Colin Studholme that includes calculation of the inverse transform. It also makes use of multi-core processing.
VIEWER: Several new features, but notably you can access data via the web. To test this you can view data on our web server by setting the “Server” to “mrir.med.miami.edu:8080”. This is a Java program and so displaying results doesn’t need the full MIDAS installation or IDL license to run. Other details can be found in the viewer help document.
BATCH: Option to delete results of previous processing. This is recommended when redoing complete processing.
IDL8 and 64-bit: The latest version of IDL is fully supported and all external DLLs have been compiled to work under 64-bit Windows. Future development will be on 64-bit CPUs and it is strongly recommended that users switch to 64-bit versions of Windows, IDL, and MIDAS, and to IDL version 8.
Note – the MIDAS toolbar still contains buttons for older versions of EPSI3D and FITT in addition to the newer ones. This will be simplified in a future release.
The new release is now available on the web site but remains labeled as the “Unstable” version while it undergoes further testing.
Processing Files V2:
The new MIDAS modules include some changes to the processing parameters, but we’ve also used the introduction of MIDASv2 to reorganize and rename the processing files. There are now 3 processing pipelines supplied for 3T data acquisition:
· Batch_Processing_… -the normal processing for studies in humans. Includes spatial normalization to MNI space and TIMO processing.
· Batch_Processing_Phantom _… -includes spectral fitting and signal normalization for phantom data (also see below).
· Batch_Processing_Quickview_… -basic SI reconstruction only. Useful for quickly checking test data, human or phantom.
All processing files can be downloaded from the web site. They are also in SVN (under Example_ProcessingFiles and subdirectories for GE/Philips/Siemens) and the latest versions are periodically updated to the web site. Remember, these files are frequently changed, and are provided as a starting point for users to develop their own processing protocols.
A requirement to have the files located under a directory named “ProcessingFiles” has been removed. This helps to organize multiple sets of processing files in an environment where you may be testing new methods while also wanting to maintain common processing pipelines in a central location. For example, the files could be located on a shared drive, with a directory structure something like:
Multi-Select MIDAS Browser:
A new version of the MIDAS Browser has been implemented in the BATCH and PRANA programs that allows multiple Subject/Study selection using the SHIFT and CTRL keys.
Tips, Questions, and Answers
The MIDAS package uses a simple XML-based database management scheme to organize the data. It consists of three files, the first of which is a list of “Projects” that the user is associated with, which is defined in the file: \Users\login_name\MIDAS\projects.xml (Windows 7), or :\Documents and Settings\login_name\MIDAS\projects.xml (Windows XP). This “projects.xml” file includes the path to the second part of the system, which is a file with a name of the form: “xxxx_project.xml”, where “xxxx” is the name the user assigned to the project, such as “Normals” or “Epilepsy”. The third part of the system is the “subject.xml” file associated with every subject.
It is possible to map data on to multiple projects, so for example, our collection of 150 normal subject data (which is defined in the file Normals_30T_project.xml) could be mapped into two sub-projects “Normal_males_project.xml” and “Normal_females_project.xml”. Similarly, projects can be combined, and the Miami normals data actually comprises of data acquired under five different projects, each of which has its own xxxx_project.xml file.
These “Virtual projects” can be created using a text editor to create and change an existing xxxx_project.xml file, or by using functions of the Importer to a) create new projects, and b) import existing subject.
One important thing to note is the format of the paths in the xxxx_project.xml file to the subject.xml file. If the subject data is under the same directory that the xxxx_project.xml file is in, then a relative path will be used, e.g. as:
<param name="Subject_Directory" value="./TBI_023" />
However, if the data is in another location the full path must be used, e.g.:
<param name="Subject_Directory" value="M:/TBI_Project/TBI_023" />
If the first case, if the whole project is moved to another location the data organization will still work, once the new location has been made known to the system (using “Import Existing Project” in the Importer). However, in the second case if the data is moved then the link will break.
For more information on the XML structure see the document: C:\Midas\Documents\Help_files\MIDAS Project Description.pdf
Processing Phantom Data:
There are several differences for processing SI data from phantoms in comparison to that used for the human brain protocol. The segmentation and LITE steps no longer apply, and because LITE is not used this means that the spatial smoothing must be turned on in the FDFT reconstruction. Also, spatial registration to an atlas isn’t required. Another consideration is that the spectral analysis of your phantom data may contain different basis functions than those used for brain.
This has caused difficulties for some people when they tried to use the standard processing pipeline meant for human brain studies, so two processing pipeline files have been added that support processing of phantom data:
The first is the “_Quickview” processing, which provides a basic reconstruction with minimum options and is useful for just checking if the SI data is OK. It does not include B0 or ECC correction, so can be useful for checking shimming and eddy currents.
The second processing pipeline file has the “_Phantom” in the name. This includes spectral analysis and signal normalization, so it can be used to check the signal normalization. Our spectral analysis file is set for NAA, Cre, and Cho only, and uses fewer iterations to save time.. To do the signal normalization it is necessary to have a segmentation result, so in this case we generate a fake segmentation result that is equivalent to 100% CSF, and then the signal normalization program uses a ‘CSF’ water density factor of 100%, as opposed to the value of 98% assumed for brain.
Another note about using the whole-brain EPSI sequence with phantoms is that relative spectral signal intensities can be altered by the action of the lipid inversion-nulling pulse. Depending of the T1s of the compounds in the phantom, a typical observation is that the choline peak becomes negative. To avoid this you can turn off the inversion pulse. A longer TR may also be beneficial.
The EPSI pulse sequence is now running on 3T instruments from GE and Philips, in addition to the original implementation for Siemens, with identical RF pulses for all implementations. Studies are ongoing to examine equivalence between systems.
File Names: One benefit of the MIDAS data management system is that you never have to be concerned about file names and locations. However, there are occasions where it’s still useful to look at the actual data directories. For example, if you delete a result and redo the processing, some old files remain, resulting in a lot of clutter that needs to be cleaned out occasionally. The raw data files are imported from DICOM, for which multiple filename conventions get used, but after the VOLUMIZER program there is some standardization of filenames used in the MIDAS system. For the SI data, the filename convention is:
Series _ “Date _ Study# _ DataSet# _ Channel# _ Version# . extension
Series is: MRI; SI, SI_Ref, etc;
Date is: code as ddmmyy;
Study number is: “s0”, “s1”, etc, to allow for multiple studies of the same subject;
DataSet number is: “t1”, “t2”, “t3”, etc, to indicate the dataset number in a multi-acquisition (e.g. 2DJ) type study;
Channel number is: “c1”, “c2”, etc, for multi-channel (raw) data;
Version number is: “v1”, “v2”, etc, to allow for multiple versions of the processing pipeline. The version numbers will increase on reprocessing if the previous datafile is not deleted;
Extension is: This usually indicates the type of data, e.g. ‘vol’=MRI volume; ‘sid’=SI spectral data; ‘map’=parametric results (e.g. of fitting);
File extensions used include:
*.vol (volumizer output); *.epsi (regridded version);
*.sid (spectral data);
*.map (SI data result that has spatial dimensions only)
*.qmap, *.B0, *.T2S, *.ecc (various SI data results as indicated by the extension)
*_Norm.map (SI maps after spatial normalization)
B0, CRB, and Weight:
This image shows maps of the mean B0 values and mean Cramer-Rao bounds for fitting of the NAA area, in the brain for two groups of age-matched subjects, for low and high weight (mean 57 Kg and 95 Kg respectively). The top row shows the reference MNI MRI.
These images were obtained after spatial normalization of all the SI results and calculation of mean value images using the PRANA program. The subject selections were done using lists of Subject IDs. The B0 maps were calculated for all voxels within the brain, as defined by the total tissue volume fraction (grey-matter + white-matter + CSF) being >0.4, whereas the CRB values were only calculated for voxels with a water linewidth of less than 15 Hz. The images were then saved in Analyze format and displayed using MRICro (www.mricro.com).
Results indicate that quality of the spectral data and fitting can be altered by body weight. Note also the excellent CRB values for fitting NAA over a large volume of the brain.
The following figure shows the mean SNR calculated for the peak NAA to RMS noise, for white matter voxels within the atlas-defined lobar regions the brain, for 8-, 12-, and 32-channel coils, all taken on the Siemens 3T/Trio. The 12 and 32-channel results are the average of two subjects and the 8-channel results from 8 subjects (age matched).
The 12 and 32 channel data were acquired by Dr. Charlie Stagg, at the Oxford FMRIB Center, who also ran this analysis. We look forward to seeing lot’s more great data from this new site.
The results show the expected advantage of increasing the number of detection channels, as well as demonstrating the good SNR possible over most of the brain volume, with a spatial resolution of ~1ml, with the EPSI acquisition.
In August 2010 the developers assembled for a planning meeting:
From left to right: Sulaiman Sheriff, JC Norman, Juan Wei, Mohammed Sabati, Andrew Maudsley, Peter Barker, Dan Spielman, Jeff Alger, Meng Gu, Jeff Steinberg, and Brian Soher.
Over the last year there have been a few changes of personnel:
Jiping Zhan, Ph.D. has taken over from Rajesh Garugu as the primary MIDAS software developer and coordinator. Rajesh has been on the project since 2005 and we thank him for many significant contributions to the package, notably in the Java modules, the new web site, and development of automated release and testing procedures.
Xin Wang, Ph.D. and He (Henry) Zhu, Ph.D. are taking over the Philips sequence development from Juan Wei. Juan developed the EPSI sequence on the Philips and has returned to China. We wish her well in her new job.
Jeff Steinberg, who has worked with Brian Soher for a couple of years developing the new FITT program, got married and moved to Singapore, and we wish him well with both of those moves.
This newsletter is for the developers and users of the MIDAS software package. If you would prefer not to receive notification of these reports just let me know (email@example.com).