2017-10-12 version 1.8-0
  - slice-timing implemented in fmri2.stimulus, fmri2.design and fmri2.lm 
2015-07-24 version 1.7.1
  - new functions on group analysis

2014-10-30 version 1.6.1
  - code now assumes that spm$cbeta is a 3D array of scalars
    simplified R- and Fortran-code
    vaws3D and vaws3Dfull simplified and renamed to aws3D and aws3Dfull
  - removed unused code from function fmri.pvalue
  - removed arg delta (latency) from function fmri.pvalue
2014-09-25 version 1.6.0
  - rewrite of the fmri.lm(), fmri.design(), fmri.stimulus() functions
  - dropped some obsolet arguments, see documentation

2014-09-23 version 1.6.0
  - parallel version of the smoothing functions "aws", "awsfull"

2012-09-13 version 1.5.0
  - new plot method for fmrisegment class which takes into account that a two-sided test is performed and should be shown!

2012-09-06 version 1.5.0
  - bugfix in extract.data for residuals, which
    set wrong dimension (time as last index
    instead of first)

2012-06-06 version 1.5.0
  - final adjustments for kritical values in segm3D
    (old values were far to conservative)

2012-06-06 version 1.4.8
  - correcting problem in extract.data

2011-10-24 version 1.4.7
  - documenting S3 methods using \method markup
  - new CITATION entry with JSS article

2011-10-18 version 1.4.6
  - removed obsolete file src/fastica.f

2011-05-30 version 1.4.5
  - changed arg gamma to arg gammatype for make.image at those places
    where it has been forgotten (known from R CMD check NOTE on CRAN)
  - arg names for layout() are 'widths' and 'heights'

2011-02-28 version 1.4.4
  - referenced Segmentation paper Polzehl et al. NI 52, 515-523 (2010)
    in help pages.

2011-02-23 version 1.4.4
  - documented alpha and delta argument for fmri.smooth()
  - removed fov argument from the list of possible additional arguments to fmri.smooth()
    as the user should not set them

2011-01-26 version 1.4.3
  - fix for bug for usage of mri.colors() in plot.fmridata() which
    threw an error due to negative arguments in hsv() call.

2010-12-08 version 1.4.3
  - bugfix for read.DICOM() use readChar(con, n, TRUE) now to prevent from utf8 error

2010-10-27 version 1.4.2
  - enable color plots from plot() GUI
  - remove \code{adaptive} argument from \code{fmri.smooth()},
    use \code{adaptation="none") instead

2010-10-06 version 1.4.2
  - fmri.gui() includes possibility of evoking
    structural adaptive segmentation
  - corrected R CMD check error for invisible binding
    of global variables
  - initialized nbeta in fastica in ngca.f, otherwise
    compiler warning

2010-09-20 version 1.4.2
  - results of segmentation results can be
    plotted by plot.fmridata() now.

2010-04-06 version 1.4.1
  - new line in Makevars
  - removed unused Fortran code

2010-03-01 version 1.4.1
  - new line in Makevars
  - removed unused Fortran code

2010-03-01 version 1.4.1
  - CITATION file
  - documentation of cutroi() falsely referred to lists instead of
    vectors in argument description

2010-02-22 version 1.4.0
  - implemented new structural adaptive segmentation algorithm. use adaptation="segment".

2009-05-27 version 1.4.0
  - bugfix in show.slice() which ignored maxpvalue

2009-02-11 version 1.3
  - bugfix read.ANALYZE(), when picstart!=1

2008-11-18 version 1.3
  - bugfix for "progress bar" for fmri.smooth() which delivered NA%
  - new graphical user interface for single subject fmri analysis function fmri.gui()

2008-10-01 version 1.3
  - new logical argument setmask for read.AFNI(), read.ANALYZE(), read.NIFTI() with default TRUE
    to switch the determination of a mask based on the level argument on/off

2008-06-20 version 1.3
  - fixes to function ngca
  - new function detrend

2008-06-17 version 1.3
  - region growing for bandwidth iteration in adaptive smoothing is 25% more variance reduction
    at each step now. Thus, the bandwidth sequence is no longer geometric, and it leads to more
    balanced adaptation.
  - fmri.lm() has default actype="smooth" to stabilize the estimate for temporal autocorrelation.
    the bandwidth for this smoothing operation can no longer be set in order to provide general
    setting for the adaptation parameter in fmri.smooth()
  - The adaptation parameters have been re-tuned for most of the statistical kernels in fmri.smooth().
    This was missing in former version. It could have lead to adaptation to noise and thus signal detection
    with many false positives.
  - bugfix in plot.fmridata() with (real) anatomical underlay for no symmetric rotation matrices!
  - implemtented method 3 from NIFTI spec. in plot.fmridata() with (real) anatomical underlay

2008-05-06 version 1.2-6

  - new function cut.roi()
  - plot.fmridata() is now able to produce anatomical overlay for pvalue data for AFNI, and NIFTI data
  - read.AFNI() should be able to interpret filenames more flexible: with, without
    extensions, with dots in name.

2008-01-21 version 1.2-5

  - read.AFNI() can read BRICK_TYPES 3 and 5 now. Beware: all sub-bricks are assumed to have
    the same BRICK_TYPE
  - bugfix in write.AFNI() for writing BRICKTYPES 3 (float) and 5 (complex)
  - bugfix for numbered ANALYZE files in read.ANALYZE() which lead to reading only one file.
  - set level for mask in all read functions as argument. reset data to zero only
    for NA values not for !mask
  - bugfix for SQ in read.DICOM which caused incorrect reading of data sequences
    of unknown length
  - dimension setting bugfix in read.ANALYZE()
  - several corrections for ngca,
  - function ngca now works for matrix- and fmridata-objects
  - new fastICA function for fmri (fmriica)
  - fmriica allows to restrict data cube by arguments xind, yind, zind and tind

2007-12-07 version 1.2-4

  - if dataset contains NA, set mask at these voxels to FALSE, set data to zero.
  - to determine the (head) mask the quantile function in all read functions
    has na.rm = TRUE now. This avoids error for datasets with NaN in datacube.

2007-11-12 version 1.2-3

  - read.AFNI() has a new argument 'vol' which contains a vector of
    volumes to be read. This hopefully enables to read parts of large datastes
    without running into memory problems.
  - if in fmri.pvalue() the df is larger than 171, a Gaussian Random Field instead of t-field is used.
    gamma() is not defined otherwise. On the other df is large enough!
  - bugfix in read.NIFTI which contained a misinterpretation of the header entries
    (bitpix, dim)
  - new function write.NIFTI()

2007-09-05 version 1.2-2

  - write.AFNI() completely re-written, it has new arguments header (AFNI-header
    list with entries such as DATASET_RANK etc.) and logical taxis, which controls,
    whether sub-bricks contain time series data. All other (old) arguments are
    depreciated, but the function should behave as in former version when called
    as such. This will be removed in some future version!
  - read.AFNI() now does not include ' and ~ in string-attributes in header

2007-07-02 version 1.2-1

  - bugfix in fmri.lm() which was unable to calculate spatial correlation for
    some datasets with optimize().

2007-04-23 version 1.2-0

  - to save memory fmri datasets are now stored in in the $ttt list element in
    raw() format. This allows for handling much larger time series. To directly
    access the data (including dimension) use the extract.data() function.