sccan (1)

COMMAND:

• sccan

• A tool for sparse statistical analysis on images : scca, pscca (with options), mscca. Can also convert an imagelist/mask pair to a binary matrix image.

OPTIONS:

-h

• Print the help menu (short version). --help

• Print the help menu (long version). <VALUES>: 1 -o, --output outputImage

• Output dependent on which option is called. -p, --n_permutations 500

• Number of permutations to use in scca. -s, --smoother 0

• Smoothing function for variates -z, --row_sparseness 0

• Row sparseness - if (+) then keep values (+) otherwise allow +/- values --- always L1 -i, --iterations 20

• Max iterations for scca optimization (min 20). -n, --n_eigenvectors 2

• Number of eigenvectors to compute in scca/spca. -r, --robustify 0

• rank-based scca -c, --covering 0

• try to make the decomposition cover the whole domain, if possible -g, --uselong 0

• use longitudinal formulation ( > 0 ) or not ( <= 0 ) -l, --l1 0

• use l1 ( > 0 ) or l0 ( < 0 ) penalty, also sets gradient step size e.g. -l 0.5 ( L1 ) , -l -0.5 (L0) will set 0.5 grad descent step for either penalty --PClusterThresh 1

• cluster threshold on view P --QClusterThresh 1

• cluster threshold on view Q -e, --ridge_cca 0

• Ridge cca. --initialization NA

• Initialization file list for Eigenanatomy - must also pass mask option --initialization2 NA

• Initialization file list for SCCAN-Eigenanatomy - must also pass mask option --mask NA

• Mask file for Eigenanatomy initialization --mask2 NA

• Mask file for Eigenanatomy initialization 2 --partial-scca-option PminusRQ

• Choices for pscca: PQ, PminusRQ, PQminusR, PminusRQminusR --imageset-to-matrix [list.txt,mask.nii.gz]

• takes a list of image files names (one per line) and converts it to a 2D matrix / image in binary or csv format depending on the filetype used to define the output.

--timeseriesimage-to-matrix [four_d_image.nii.gz,three_d_mask.nii.gz, optional-spatial-smoothing-param-in-spacing-units-default-zero, optional-temporal-smoothing-param-in-time-series-units-default-zero

]

• takes a timeseries (4D) image and converts it to a 2D matrix csv format as output.If the mask has multiple labels ( more the one ) then the average time series in each label will be computed and put in the csv. --vector-to-image [vector.csv,three_d_mask.nii.gz, which-row-or-col ]

• converts the 1st column vector in a csv file back to an image --- currently needs the csv file to have > 1 columns. if the number of entries in the column does not equal the number of entries in the mask but the number of rows does equal the number of entries in the mask, then it will convert the row vector to an image. --imageset-to-projections [list_projections.txt,list_images.txt, bool do-average-not-real-projection ]

• takes a list of image and projection files names (one per line) and writes them to a csv file --- basically computing X*Y (matrices). --scca two-view[matrix-view1.mhd,matrix-view2.mhd,mask1,mask2,FracNonZero1,FracNonZero2]

• three-view[matrix-view1.mhd,matrix-view2.mhd,matrix-view3.mhd,mask1,mask2,mask3,FracNonZero1,FracNonZero2,FracNonZero3] partial[matrix-view1.mhd,matrix-view2.mhd,matrix-view3.mhd,mask1,mask2,mask3,FracNonZero1,FracNonZero2,FracNonZero3] dynsccan[matrix-view1.mhd,matrix-view2.mhd,mask1,mask2,FracNonZero1,FracNonZero2]

• Matrix-based scca operations for 2 and 3 views.For all these options, the FracNonZero terms set the fraction of variables to use in the estimate. E.g. if one sets 0.5 then half of the variables will have non-zero values. If the FracNonZero is (+) then the weight vectors must be positive. If they are negative, weights can be (+) or (-). partial does partial scca for 2 views while partialing out the 3rd view. --svd sparse[matrix-view1.mhd,mask1,FracNonZero1,nuisance-matrix] --- will only use view1 ... unless nuisance matrix is specified.

• classic[matrix-view1.mhd,mask1,FracNonZero1,nuisance-matrix] --- will only use view1 ... unless nuisance matrix is specified. cgspca[matrix-view1.mhd,mask1,FracNonZero1,nuisance-matrix] --- will only use view1 ... unless nuisance matrix is specified, -i controls the number of sparse approximations per eigenvector, -n controls the number of eigenvectors. total output will then be i*n sparse eigenvectors. prior[ matrix.mha , mask.nii.gz , PriorList.txt , PriorScale.csv , PriorWeightIn0to1 , sparseness ] ... if sparseness is set to zero, we take sparseness from the priors. network[matrix-view1.mhd,mask1,FracNonZero1,guidance-matrix] lasso[matrix-view1.mhd,mask1,Lambda,guidance-matrix] recon[matrix-view1.mhd,mask1,FracNonZero1,nuisance-matrix] recon4d[matrix-view1.mhd,mask1,FracNonZero1,nuisance-matrix]

• a sparse svd implementation --- will report correlation of eigenvector with original data columns averaged over columns with non-zero weights.