Images, images and still more images

1
Images, images and still more images

• Cinly Ooi

2
Presentation Outline

• Program imgutils-image_format_convert
• Data Format
• Endian problem
• Split/merging files
• Auxillary data (TR etc) recognition, insertion
  and checking
• Work to increase robustness of the programs
• Programming interface
• Some programming advices
• Current and future work

3
Users vs Programs (programmers?)

• What the user expects
• Program accepts all file formats
• Program deals with all ways of presenting data

• What the program expects
• User provides only one file format
• A very specific way of dealing with data

IT IS GOING TO BE A COMPROMISE BETWEEN THIS TWO
EXTREME
4
Image Data

• File Format
• What type of data it can holds? (integer, real
  number)
• Any auxillary data, such as TR?
• Machine Dependency
• Big Endian vs Little Endian
• Size of Data (sizeof(int)4??)
• How To Access These Data?

5
File Formats in Use

• Analyze Format (.hdr, .img)
• Most widely used. Mainly due to SPM dominance.
• Bruker-enhanced Analyze
• Used to supplement Analyze. Unique to Cambridge
  and perhaps Oxford.
• UNC Format
• Used by BAMM.
• DICOM
• Emerging standard. But with SPMs dominance, it
  is unlikely to be accepted in fMRI community.

6
Analyze Format

• Advantage
• Simple and easy to program access routine
• Since image data is in separate file, very easy
  to import it into other programs.
• The most supported format in fMRI programs
• Disadvantage
• Experience says that header and image file can be
  in different endian !
• It only carries structural Information
• Abused by SPM programmers because of
  misinterpretation of specification, SPM-specific,
  M-like, hijack of reserved fields.

7
Bruker-enhanced Analyze

• Advantage
• Carries detail information about fMRI
• As a text file, easily viewed by user
• Disadvantage
• Very complex structure, difficult to program
  access routine
• Cannot find any documentation

8
UNC Format

• Advantage
• Can hold any auxiliary data
• Disadvantage
• Only used (and abused) by BAMM.
• Unlike brkhdr, auxiliary data can only be seen
  if you use a hex editor
• There is completely no specification on what data
  must be there for fMRI, PET etc.

9
DICOM

• Advantage
• Very well documented Documentation run into 10
  volumes!!
• Have specific specification on what data must be
  there for MRI, PET etc
• Disadvantage
• Not easy to program access routines for auxiliary
  data
• Auxiliary data are coded and cannot be easily
  seen.

10
imgutils-image_format_convert

• Use it for
• File-format conversion
• Combining a set of N dimensional files to form a
  N1 dimensional file and vice-versa.
• Copying only a subset of the data
• Changing the endian of the data
• Inspecting, and if file format allows, insert new
  field into the file.
• Check and validate the auxiliary data
  carried/inserted.

11
Usage

• imgutils-image_format_convert options
  ltInputImageStringgt ltOutputImageStringgt
• All it does is to make available, in a technical
  user-friendly format, all functionalities
  available in Image File I/O, i.e., everything in
  FileManager.cpp and its associated files.

12
File Format Conversion
imgutils-image_format_convert input.brhdr
output.unc
13
Combining And Splitting Files
imgutils-image_format_convert input3d??.img
output4d.img
14
Endian Conversion
Solaris Big Endian
automatic
imgutils-image_format_convert
automatic
Intel Little Endian
--force-convert-endian --no-covert-endian
imgutils-image-format-convert input.hdr
output.hdr
15
Options for imgutils-image_format_convert

• --input-analyze-spm99-compatible,
  --output-analyze-spm99-compatible
• the input/output analyze image will be spm99
  complient
• --input-unc-fmri3d
• user guarantee that the input UNC 3D image is a
  fmMRI image
• Technically speaking, all programmer accessible
  options can be input as command line options

16
BammL XML-based Data Description

• Human/Machine readable
• lt? xml ?gt
• ltBAMM_IMAGE_INFORMATIONgt
• ltdatagt
• ltdata_datatypegt
• ltnamegtFLOATlt/namegt
• ltsizegt4lt/sizegt
• lt/data_datatypegt
• ltdata_dimensiongt
• ltnum_dimensiongt3lt/num_dimensiongt
• ltwidthgtltnum_pixelgt128lt/num_pixelgtlt/widthgt
• ltheightgtltnum_pixelgt128lt/num_pixelgtlt/heightgt
• ltvolumegtltnum_pixelgt21lt/num_pixelgtlt/volumegt
• lt/data_dimensiongt
• lt/datagt
• lt/BAMM_IMAGE_INFORMATIONgt

17
BammL Data Classification

• Classified into different classes.
• Each category contains guaranteed information.
• The guaranteed information are checked for
  accuracy whenever possible
• Simplify programming

18
BammL Class Hierarchy
BAMM_IMAGE_INFORMATION Any image data (1D data
treated as special case)
BAMM_SCAN_INFORMATION Any medical scan data (PET,
MRI, CT etc, 2D and above)
BAMM_fMRI_SCAN_INFORMATION fMRI data(3D and 4D
data)
It is possible to replace the class represented
in the image file using a separate BammL file
19
Programming Stuff BAMM

• Collection of Data Classes and algorithms to
  build application
• Aim to make program more robust by
• encouraging reuse of classes and algorithms
• automated background data test
• systematic catching and reporting of errors
• rigorous testing to prevent errors
• better documentation
• making program more readable
• Restructuring internal organization to better
  cope with possible future change

20
Programming Changes

• Current
• main()
• int Data4D, SIZE, Dim, DimSize
• File UNC
• if( (UNC imopen(a.unc)) NULL)
  //
• imdatainfo(UNC, DimSize, DType, Dim)
• if(DimSize ! 4) //
• if(DType ! INT) //
• SIZE Dim0Dim3
• Data4D new intSIZE
• imgetdata(UNC, Data4D)
• //there is still no guarantee that Data4D is
• //valid fMRI data
• //
• if(!imgetinfo(UNC,rep_time, TR)) //

• New
• main()
• fmri4dltintgt Data4D
• try
• FileManager FMgr
• FMgr.LoadFromFile(a.unc)
• FMgr.WriteDataTo(Data4D)
• catch(exception e)
• //
• //from here on Data4D is guaranteed
• //valid fMRI data
• //
• TR Data4D.GetRepetitionTime()

Could had been a.brkhdr
21
Programming Advice

• Simple steps that will save you a lot of grief in
  the long run.
• They are just Rule of Thumb, meaning they are
  useful at about 80 of the time

22
Programming Advice I

• SAVE MULTIPLE VERSIONS OF THE SAME PROGRAM
• The aim is the ability to go back to a previous
  version, if the version you are currently working
  on goes horribly wrong
• For larger projects, consider concurrent version
  system (CVS)

23
Programming Advice II

• Breakdown task to simple tasks, then code the
  simple task separately
• Example Converting BAMMs design matrix to SPM
  design matrix (dmc-bamm2spm.pl)
• Read in BAMM design matrix then
• Write SPM design matrix
• Dont read and write at the same time its
  difficult to debug

24
Programming Advice III

• Test, test and retest
• Testing should be done continuously on the whole
  design/coding cycle
• As soon as you finish writing a task, test that
  the program really does the task.
• Regression test If you find a possible failure
  route for your program, devise a test procedure,
  and then use the test procedure at regular
  intervals to make sure the same problem does not
  pop up again

25
Programming Advice IV

• Document as much as you can
• User documentation how to use it as a program
• Program documentation
• What your program/subprogram does
• Whats the assumption you made
• How to use them
• Possible failure/bugs

26
Current Work on BAMM

• Maintaining the already created classes and
  routine
• Making the programming documentation available on
  the website
• Write a user manual

27
Possible Future Works for BAMM

• Incorporate more image formats such as BMP, PCX
  and JPEG
• Anonymizing tools to make Data Protection Act
  compliance a easier task
• More diagnostic routines to find out whats wrong
  with your image
• Rework current f/g/x/sbamm to the more robust
  method if and only if it make sense to do so
• More numeric routines, possibly from boost.org

28
Other current works possible future work

• Current Work
• FTP server
• Allow upload/download of large chunks of data
• Analyzing Carolines new data
• Possible future work
• Incorporating spatial information when regressing
  the fMRI model