The program has a list of model identity codes, or "inversion indices" see modelfit(1) for a list; one for each possible class in the voxel classification. If the classification of a voxel is i, the program uses the inversion indexed by the i-th element of the list for the data in that voxel. If the voxel classification is greater than the length of the list, the program uses the inversion indexed by the last element of the list. By default, the list of indices is {1, 1, 1, 1, 31}, so that voxel classifications 0, 1, 2 and 3 all imply fitting a single tensor by linear least-squares to the log measurements and a classification of 4 or above implies inversion 31, which is two positive-definite tensors. The standard output format of voxelclassify(1), which is 0 for isotropic, 2 for anisotropic Gaussian and 4 for non-Gaussian, motivates the default list. With this voxel classification, elements 1 and 3 (counting from zero) of the list are redundant.
The program outputs the results, in voxel order and as big-endian eight-byte doubles, to the standard output. The output format is: [exitcode, ln(S(0)), m, a_1, D_1xx, D_1xy, D_1xz, D_1yy, D_1yz, D_1zz, a_2, D_2xx, D_2xy, D_2xz, D_2yy, D_2yz, D_2zz, ..., a_n, D_nxx, D_nxy, D_nxz, D_nyy, D_nyz, D_nzz], where n is the maximum number of tensors in each voxel (which can be set using the command line option -maxcomponents) and m is the actual number fitted and output in this voxel. If n > m, diffusion tensors above index m contain all zeros and their mixing parameters are zero. An exit code of zero indicates no problems. For a list of other exit codes, see modelfit(1). The entry S(0) is an estimate of the signal at q=0.
voxelclassify -schemefile A.scheme -ftest 1.0E-20 1.0E-6 1.0E-6 < SubjectA.Bfloat > A_VC.Bint
multitenfit -schemefile A.scheme -voxclassmap A_VC.Bint < SubjectA.Bfloat > MultiTensorA.Bdouble
Fit tensors using non-linear fitting in isotropic and anisotropic Gaussian voxels and two cylindrically symmetric positive-definite in non-Gaussian voxels:
multitenfit -schemefile A.scheme -voxclassmap A_VC.Bint -classifiedmodels 5 2 2 2 2 11 < SubjectA.Bfloat > MultiTensorA.Bdouble
Use a different voxel classification which contains zero in voxels with no clear fibre orientations, one in voxels with one fibre, two in voxels with two fibres and three in voxels with three fibres. With zero or one fibre, just fit the diffusion tensor; with two fibres fit a pair of cylindrically symmetric diffusion tensors; with three fit three cylindrically symmetric tensors:
multitenfit -schemefile A.scheme -voxclassmap A_VC2.Bint -classifiedmodels 4 2 2 11 211 < SubjectA.Bfloat > MultiTensorA.Bdouble
The new text model codes (see modelfit(1)) may also be used: {ldt, ldt, ldt, ldt, pospos} is the same as {1, 1, 1, 1, 31}. If a text code is used for a two- or three-tensor model, the single-tensor initialization is the first model in the list, so {1, pospos_eq} is the same as {1, 21} and {2, pospos_eq} is the same as {2, 22}, but {2, 21} must be specified numerically.
multitenfit also takes other standard command line options listed in modelfit(1).