Interpretation of Chemical Shift Imaging (CSI) data is quite challenging, not only because a typical dataset consists of hundreds of spectra with relatively low SNR, but also because spectra from nearby locations are often related to one another, making it preferable to analyze an entire dataset as a whole. Furthermore, multiple tissue types often contribute to a given voxel, necessitating the use of spectral decomposition to identify the shapes of the underlying source spectra and their contributions to the observed data.

3D Interactive Chemical Shift Imaging (3DiCSI) is a software tool developed to address these challenges. It combines data visualization, spectral processing, spectral localization and estimation, spectral quantification together with multi-variate analytical procedures, such as Principal Component Analysis (PCA) and constrained Non-negative Matrix Factorization (cNMF). In particular, cNMF provides an efficient way to recover underlying spectral patterns that are physically meaningful.

3DiCSI is provided as a free-of-charge software program for research and clinical purposes. It is written in C++ and runs on Windows platform. It has been funded by U01CA062556, NIH DK070301.

• Data Input and Visualization:
  • Automatic importation of 1D/2D/3D CSI raw data from Philips, GE, Siemens and Bruker scanners.
  • Any unknown data type can be imported by manually entering parameters.
  • CSI data and anatomic reference images are registered and displayed in the same window.
  • Each voxel is displayed in a separate window, its spatial location clearly visualized in the three orthogonal image windows.
  • 3D image reconstruction from images in one orientation.
• Multi-dimensional Spectral Data Processing:
  • Constant phase and linear phase correction.
  • Temporal filtering ( Lorentzian, Gaussian, or combined).
  • Spatial smoothing ( Hamming, Cosine, Fermi).
  • Baseline correction.
  • Water suppression.
  • PPM origin adjustment and zero filling.
  • Lipid Suppression.
  • Peak alignment.
• Batch Processing:
  • Sequence of processing steps can be recorded in a macro file and applied to the CSI data set.
  • The same macro file can be applied to a series of exams in a batch mode.
• Spectral Localization:
  • Real-time mouse-facilitated interactive voxel shifting in all 3 dimensions.
  • Averaging of selected multiple voxels.
  • Spectral estimation of a user drawn 2D/3D region of interest.
• Multi-variate spectral analysis:
  • Principal Component Analysis (PCA) for user defined spectral region and ROI, to identify Principal Components (PCs) for the variation sources contained in the CSI data set.
  • PCA-based iterative peak align procedure to remove frequency and phase variations.
  • Constrained Non-Negative Matrix Factorization (cNMF) method to recover spectral patterns and their spatial distributions that are physically meaningful.
• Spectral Quantification:
  • Metabolic map generation using area integration and PCA.
  • Map generation for peak ratio and for user defined regions.
  • Absolute quantification of metabolite concentration using reference spectrum.
  • Map display scale adjustment to enable comparison between different sets.
  • Spectral integration with user-controlled baseline correction.
• Data Output and Program Extensibility:
  • Various result output options, such as HTML format report, CSI slice and image print out, display copy and paste functions, spectral ascii/binary exportation, metabolic image exportation.
  • Interface for users to extend the functionality on the basis of 3DiCSI.

You can also download a demo project here. The demo project is made from a 3D P31 CSI data set. First uncompress it, and then load 3dicsi_demo.tpj to 3DiCSI program using "open project" menu item. Then you can try out all the functionalities provided by 3DiCSI.

The program will expire six months after the first run on your computer. Please feel free to contact Qi Zhao to get a newer version. If you process your data using 3DiCSI, we would appreciate your acknowledgement of the software and sending us a copy of your publication.

3DiCSI is an ongoing project, with new features constantly added. Your feedback would be highly appreciated and beneficial to the software's continuous improvement. Please send your questions, comments or bug reports to Qi Zhao at qz2106@columbia.edu. We look forward to hearing from you!

• Update v1.9.9 ---- 11/27/2007
  Some of the new additions are (for more details, refer to the help document):
  1. Added importation support for Bruker image and CSI files.
  2. Added water suppression based on HLSVD method.
  3. Added lipid suppression based on singal space projection method.
  4. Added voxel-by-voxel normalization for absolute quantitation.
• Update v1.9.8 ---- 05/16/2007
  Some of the new additions are (for more details, refer to the help document):
  1. Added support for importation of more data formats, such as Philips Release 2, Siemens RDA file, etc.
  2. More flexible option for reading "unknown" data type.
  3. Added individual phasing on a voxel by voxel basis.
  4. Added manual peak alignment for each voxel or selected voxels.
  5. Bug fixes.
• Update v1.9.7 ---- 09/29/2006
  Some of the new additions are (for more details, refer to the help document):
  1. Added Constrained Non-Negative Matrix Factorization (cNMF), which is a spectral pattern recognition method to recover spectral patterns and their spatial distributions that are physically interpretable.
  2. Added spectral integration with user-controlled baseline correction.
  3. Added initial peak alignment functionality.
  4. Added peak finding functionality to create selection mask.
  5. Added easy browsing of the voxels using arrow keys.
• First release v1.8.6 ---- 03/05/2006