int
tend_estimMain(int argc, char **argv, char *me, hestParm *hparm) {
int pret;
hestOpt *hopt = NULL;
char *perr, *err;
airArray *mop;
Nrrd **nin, *nin4d, *nbmat, *nterr, *nB0, *nout;
char *outS, *terrS, *bmatS, *eb0S;
float soft, scale, sigma;
int dwiax, EE, knownB0, oldstuff, estmeth, verbose, fixneg;
unsigned int ninLen, axmap[4], wlsi, *skip, skipNum, skipIdx;
double valueMin, thresh;
Nrrd *ngradKVP=NULL, *nbmatKVP=NULL;
double bKVP, bval;
tenEstimateContext *tec;
hestOptAdd(&hopt, "old", NULL, airTypeInt, 0, 0, &oldstuff, NULL,
"instead of the new tenEstimateContext code, use "
"the old tenEstimateLinear code");
hestOptAdd(&hopt, "sigma", "sigma", airTypeFloat, 1, 1, &sigma, "nan",
"Rician noise parameter");
hestOptAdd(&hopt, "v", "verbose", airTypeInt, 1, 1, &verbose, "0",
"verbosity level");
hestOptAdd(&hopt, "est", "estimate method", airTypeEnum, 1, 1, &estmeth,
"lls",
"estimation method to use. \"lls\": linear-least squares",
NULL, tenEstimate1Method);
hestOptAdd(&hopt, "wlsi", "WLS iters", airTypeUInt, 1, 1, &wlsi, "1",
"when using weighted-least-squares (\"-est wls\"), how "
"many iterations to do after the initial weighted fit.");
hestOptAdd(&hopt, "fixneg", NULL, airTypeInt, 0, 0, &fixneg, NULL,
"after estimating the tensor, ensure that there are no negative "
"eigenvalues by adding (to all eigenvalues) the amount by which "
"the smallest is negative (corresponding to increasing the "
"non-DWI image value).");
hestOptAdd(&hopt, "ee", "filename", airTypeString, 1, 1, &terrS, "",
"Giving a filename here allows you to save out the tensor "
"estimation error: a value which measures how much error there "
"is between the tensor model and the given diffusion weighted "
"measurements for each sample. By default, no such error "
"calculation is saved.");
hestOptAdd(&hopt, "eb", "filename", airTypeString, 1, 1, &eb0S, "",
"In those cases where there is no B=0 reference image given "
"(\"-knownB0 false\"), "
"giving a filename here allows you to save out the B=0 image "
"which is estimated from the data. By default, this image value "
"is estimated but not saved.");
hestOptAdd(&hopt, "t", "thresh", airTypeDouble, 1, 1, &thresh, "nan",
"value at which to threshold the mean DWI value per pixel "
"in order to generate the \"confidence\" mask. By default, "
"the threshold value is calculated automatically, based on "
"histogram analysis.");
hestOptAdd(&hopt, "soft", "soft", airTypeFloat, 1, 1, &soft, "0",
"how fuzzy the confidence boundary should be. By default, "
"confidence boundary is perfectly sharp");
hestOptAdd(&hopt, "scale", "scale", airTypeFloat, 1, 1, &scale, "1",
"After estimating the tensor, scale all of its elements "
"(but not the confidence value) by this amount. Can help with "
"downstream numerical precision if values are very large "
"or small.");
hestOptAdd(&hopt, "mv", "min val", airTypeDouble, 1, 1, &valueMin, "1.0",
"minimum plausible value (especially important for linear "
"least squares estimation)");
hestOptAdd(&hopt, "B", "B-list", airTypeString, 1, 1, &bmatS, NULL,
"6-by-N list of B-matrices characterizing "
"the diffusion weighting for each "
"image. \"tend bmat\" is one source for such a matrix; see "
"its usage info for specifics on how the coefficients of "
"the B-matrix are ordered. "
"An unadorned plain text file is a great way to "
"specify the B-matrix.\n **OR**\n "
"Can say just \"-B kvp\" to try to learn B matrices from "
"key/value pair information in input images.");
hestOptAdd(&hopt, "b", "b", airTypeDouble, 1, 1, &bval, "nan",
"\"b\" diffusion-weighting factor (units of sec/mm^2)");
hestOptAdd(&hopt, "knownB0", "bool", airTypeBool, 1, 1, &knownB0, NULL,
"Determines of the B=0 non-diffusion-weighted reference image "
"is known, or if it has to be estimated along with the tensor "
"elements.\n "
"\b\bo if \"true\": in the given list of diffusion gradients or "
"B-matrices, there are one or more with zero norm, which are "
"simply averaged to find the B=0 reference image value\n "
"\b\bo if \"false\": there may or may not be diffusion-weighted "
"images among the input; the B=0 image value is going to be "
"estimated along with the diffusion model");
hestOptAdd(&hopt, "i", "dwi0 dwi1", airTypeOther, 1, -1, &nin, "-",
"all the diffusion-weighted images (DWIs), as seperate 3D nrrds, "
"**OR**: One 4D nrrd of all DWIs stacked along axis 0",
&ninLen, NULL, nrrdHestNrrd);
hestOptAdd(&hopt, "o", "nout", airTypeString, 1, 1, &outS, "-",
"output tensor volume");
mop = airMopNew();
airMopAdd(mop, hopt, (airMopper)hestOptFree, airMopAlways);
USAGE(_tend_estimInfoL);
JUSTPARSE();
airMopAdd(mop, hopt, (airMopper)hestParseFree, airMopAlways);
nout = nrrdNew();
airMopAdd(mop, nout, (airMopper)nrrdNuke, airMopAlways);
nbmat = nrrdNew();
airMopAdd(mop, nbmat, (airMopper)nrrdNuke, airMopAlways);
/* figure out B-matrix */
if (strcmp("kvp", airToLower(bmatS))) {
/* its NOT coming from key/value pairs */
if (!AIR_EXISTS(bval)) {
fprintf(stderr, "%s: need to specify scalar b-value\n", me);
airMopError(mop); return 1;
}
if (nrrdLoad(nbmat, bmatS, NULL)) {
airMopAdd(mop, err=biffGetDone(NRRD), airFree, airMopAlways);
fprintf(stderr, "%s: trouble loading B-matrix:\n%s\n", me, err);
airMopError(mop); return 1;
}
nin4d = nin[0];
skip = NULL;
skipNum = 0;
} else {
/* it IS coming from key/value pairs */
if (1 != ninLen) {
fprintf(stderr, "%s: require a single 4-D DWI volume for "
"key/value pair based calculation of B-matrix\n", me);
airMopError(mop); return 1;
}
if (oldstuff) {
if (knownB0) {
fprintf(stderr, "%s: sorry, key/value-based DWI info not compatible "
"with older implementation of knownB0\n", me);
airMopError(mop); return 1;
}
}
if (tenDWMRIKeyValueParse(&ngradKVP, &nbmatKVP, &bKVP,
&skip, &skipNum, nin[0])) {
airMopAdd(mop, err=biffGetDone(TEN), airFree, airMopAlways);
fprintf(stderr, "%s: trouble parsing DWI info:\n%s\n", me, err);
airMopError(mop); return 1;
}
if (AIR_EXISTS(bval)) {
fprintf(stderr, "%s: WARNING: key/value pair derived b-value %g "
"over-riding %g from command-line", me, bKVP, bval);
}
bval = bKVP;
if (ngradKVP) {
airMopAdd(mop, ngradKVP, (airMopper)nrrdNuke, airMopAlways);
if (tenBMatrixCalc(nbmat, ngradKVP)) {
airMopAdd(mop, err=biffGetDone(TEN), airFree, airMopAlways);
fprintf(stderr, "%s: trouble finding B-matrix:\n%s\n", me, err);
airMopError(mop); return 1;
}
} else {
airMopAdd(mop, nbmatKVP, (airMopper)nrrdNuke, airMopAlways);
if (nrrdConvert(nbmat, nbmatKVP, nrrdTypeDouble)) {
airMopAdd(mop, err=biffGetDone(NRRD), airFree, airMopAlways);
fprintf(stderr, "%s: trouble converting B-matrix:\n%s\n", me, err);
airMopError(mop); return 1;
}
}
/* this will work because of the impositions of tenDWMRIKeyValueParse */
dwiax = ((nrrdKindList == nin[0]->axis[0].kind ||
nrrdKindVector == nin[0]->axis[0].kind)
? 0
: ((nrrdKindList == nin[0]->axis[1].kind ||
nrrdKindVector == nin[0]->axis[1].kind)
? 1
: ((nrrdKindList == nin[0]->axis[2].kind ||
nrrdKindVector == nin[0]->axis[2].kind)
? 2
: 3)));
if (0 == dwiax) {
nin4d = nin[0];
} else {
axmap[0] = dwiax;
axmap[1] = 1 > dwiax ? 1 : 0;
axmap[2] = 2 > dwiax ? 2 : 1;
axmap[3] = 3 > dwiax ? 3 : 2;
nin4d = nrrdNew();
airMopAdd(mop, nin4d, (airMopper)nrrdNuke, airMopAlways);
if (nrrdAxesPermute(nin4d, nin[0], axmap)) {
airMopAdd(mop, err=biffGetDone(NRRD), airFree, airMopAlways);
fprintf(stderr, "%s: trouble creating DWI volume:\n%s\n", me, err);
airMopError(mop); return 1;
}
}
}
nterr = NULL;
nB0 = NULL;
if (!oldstuff) {
if (1 != ninLen) {
fprintf(stderr, "%s: sorry, currently need single 4D volume "
"for new implementation\n", me);
airMopError(mop); return 1;
}
if (!AIR_EXISTS(thresh)) {
if (tend_estimThresholdFind(&thresh, nbmat, nin4d)) {
airMopAdd(mop, err=biffGetDone(TEN), airFree, airMopAlways);
fprintf(stderr, "%s: trouble finding threshold:\n%s\n", me, err);
airMopError(mop); return 1;
}
/* HACK to lower threshold a titch */
thresh *= 0.93;
fprintf(stderr, "%s: using mean DWI threshold %g\n", me, thresh);
}
tec = tenEstimateContextNew();
tec->progress = AIR_TRUE;
airMopAdd(mop, tec, (airMopper)tenEstimateContextNix, airMopAlways);
EE = 0;
if (!EE) tenEstimateVerboseSet(tec, verbose);
if (!EE) tenEstimateNegEvalShiftSet(tec, fixneg);
if (!EE) EE |= tenEstimateMethodSet(tec, estmeth);
if (!EE) EE |= tenEstimateBMatricesSet(tec, nbmat, bval, !knownB0);
if (!EE) EE |= tenEstimateValueMinSet(tec, valueMin);
for (skipIdx=0; skipIdx<skipNum; skipIdx++) {
/* fprintf(stderr, "%s: skipping %u\n", me, skip[skipIdx]); */
if (!EE) EE |= tenEstimateSkipSet(tec, skip[skipIdx], AIR_TRUE);
}
switch(estmeth) {
case tenEstimate1MethodLLS:
if (airStrlen(terrS)) {
tec->recordErrorLogDwi = AIR_TRUE;
/* tec->recordErrorDwi = AIR_TRUE; */
}
break;
case tenEstimate1MethodNLS:
if (airStrlen(terrS)) {
tec->recordErrorDwi = AIR_TRUE;
}
break;
case tenEstimate1MethodWLS:
if (!EE) tec->WLSIterNum = wlsi;
if (airStrlen(terrS)) {
tec->recordErrorDwi = AIR_TRUE;
}
break;
case tenEstimate1MethodMLE:
if (!(AIR_EXISTS(sigma) && sigma > 0.0)) {
fprintf(stderr, "%s: can't do %s w/out sigma > 0 (not %g)\n",
me, airEnumStr(tenEstimate1Method, tenEstimate1MethodMLE),
sigma);
airMopError(mop); return 1;
}
if (!EE) EE |= tenEstimateSigmaSet(tec, sigma);
if (airStrlen(terrS)) {
tec->recordLikelihoodDwi = AIR_TRUE;
}
break;
}
if (!EE) EE |= tenEstimateThresholdSet(tec, thresh, soft);
if (!EE) EE |= tenEstimateUpdate(tec);
if (EE) {
airMopAdd(mop, err=biffGetDone(TEN), airFree, airMopAlways);
fprintf(stderr, "%s: trouble setting up estimation:\n%s\n", me, err);
airMopError(mop); return 1;
}
if (tenEstimate1TensorVolume4D(tec, nout, &nB0,
airStrlen(terrS)
? &nterr
: NULL,
nin4d, nrrdTypeFloat)) {
airMopAdd(mop, err=biffGetDone(TEN), airFree, airMopAlways);
fprintf(stderr, "%s: trouble doing estimation:\n%s\n", me, err);
airMopError(mop); return 1;
}
if (airStrlen(terrS)) {
airMopAdd(mop, nterr, (airMopper)nrrdNuke, airMopAlways);
}
} else {
EE = 0;
if (1 == ninLen) {
EE = tenEstimateLinear4D(nout, airStrlen(terrS) ? &nterr : NULL, &nB0,
nin4d, nbmat, knownB0, thresh, soft, bval);
} else {
EE = tenEstimateLinear3D(nout, airStrlen(terrS) ? &nterr : NULL, &nB0,
(const Nrrd**)nin, ninLen, nbmat,
knownB0, thresh, soft, bval);
}
if (EE) {
airMopAdd(mop, err=biffGetDone(TEN), airFree, airMopAlways);
fprintf(stderr, "%s: trouble making tensor volume:\n%s\n", me, err);
airMopError(mop); return 1;
}
}
if (nterr) {
/* it was allocated by tenEstimate*, we have to clean it up */
airMopAdd(mop, nterr, (airMopper)nrrdNuke, airMopAlways);
}
if (nB0) {
/* it was allocated by tenEstimate*, we have to clean it up */
airMopAdd(mop, nB0, (airMopper)nrrdNuke, airMopAlways);
}
if (1 != scale) {
if (tenSizeScale(nout, nout, scale)) {
airMopAdd(mop, err=biffGetDone(TEN), airFree, airMopAlways);
fprintf(stderr, "%s: trouble doing scaling:\n%s\n", me, err);
airMopError(mop); return 1;
}
}
if (nterr) {
if (nrrdSave(terrS, nterr, NULL)) {
airMopAdd(mop, err=biffGetDone(NRRD), airFree, airMopAlways);
fprintf(stderr, "%s: trouble writing error image:\n%s\n", me, err);
airMopError(mop); return 1;
}
}
if (!knownB0 && airStrlen(eb0S)) {
if (nrrdSave(eb0S, nB0, NULL)) {
airMopAdd(mop, err=biffGetDone(NRRD), airFree, airMopAlways);
fprintf(stderr, "%s: trouble writing estimated B=0 image:\n%s\n",
me, err);
airMopError(mop); return 1;
}
}
if (nrrdSave(outS, nout, NULL)) {
airMopAdd(mop, err=biffGetDone(NRRD), airFree, airMopAlways);
fprintf(stderr, "%s: trouble writing:\n%s\n", me, err);
airMopError(mop); return 1;
}
airMopOkay(mop);
return 0;
}
int
tend_simMain(int argc, char **argv, char *me, hestParm *hparm) {
int pret;
hestOpt *hopt = NULL;
char *perr, *err;
tenEstimateContext *tec;
airArray *mop;
int E, oldstuff, seed;
Nrrd *nin, *nT2, *nbmat, *nout;
char *outS;
float b, sigma;
hestOptAdd(&hopt, "old", NULL, airTypeInt, 0, 0, &oldstuff, NULL,
"don't use the new tenEstimateContext functionality");
hestOptAdd(&hopt, "sigma", "sigma", airTypeFloat, 1, 1, &sigma, "0.0",
"Rician noise parameter");
hestOptAdd(&hopt, "seed", "seed", airTypeInt, 1, 1, &seed, "42",
"seed value for RNG which creates noise");
hestOptAdd(&hopt, "B", "B matrix", airTypeOther, 1, 1, &nbmat, NULL,
"B matrix, one row per diffusion-weighted image",
NULL, NULL, nrrdHestNrrd);
hestOptAdd(&hopt, "r", "reference field", airTypeOther, 1, 1, &nT2, "-",
"reference anatomical scan, with no diffusion weighting",
NULL, NULL, nrrdHestNrrd);
hestOptAdd(&hopt, "i", "tensor field", airTypeOther, 1, 1, &nin, "-",
"input diffusion tensor field", NULL, NULL, nrrdHestNrrd);
hestOptAdd(&hopt, "b", "b", airTypeFloat, 1, 1, &b, "1",
"b value for simulated scan");
hestOptAdd(&hopt, "o", "nout", airTypeString, 1, 1, &outS, "-",
"output image (floating point)");
mop = airMopNew();
airMopAdd(mop, hopt, (airMopper)hestOptFree, airMopAlways);
USAGE(_tend_simInfoL);
PARSE();
airMopAdd(mop, hopt, (airMopper)hestParseFree, airMopAlways);
nout = nrrdNew();
airMopAdd(mop, nout, (airMopper)nrrdNuke, airMopAlways);
if (!oldstuff) {
airSrandMT(seed);
tec = tenEstimateContextNew();
airMopAdd(mop, tec, (airMopper)tenEstimateContextNix, airMopAlways);
E = 0;
if (!E) E |= tenEstimateMethodSet(tec, tenEstimateMethodLLS);
if (!E) E |= tenEstimateValueMinSet(tec, 0.0001);
if (!E) E |= tenEstimateBMatricesSet(tec, nbmat, b, AIR_TRUE);
if (!E) E |= tenEstimateThresholdSet(tec, 0, 0);
if (!E) E |= tenEstimateUpdate(tec);
if (!E) E |= tenEstimate1TensorSimulateVolume(tec,
nout, sigma, b,
nT2, nin,
nrrdTypeFloat);
if (E) {
airMopAdd(mop, err=biffGetDone(TEN), airFree, airMopAlways);
fprintf(stderr, "%s: trouble making DWI volume (new):\n%s\n", me, err);
airMopError(mop); return 1;
}
} else {
if (tenSimulate(nout, nT2, nin, nbmat, b)) {
airMopAdd(mop, err=biffGetDone(TEN), airFree, airMopAlways);
fprintf(stderr, "%s: trouble making DWI volume:\n%s\n", me, err);
airMopError(mop); return 1;
}
}
if (nrrdSave(outS, nout, NULL)) {
airMopAdd(mop, err=biffGetDone(NRRD), airFree, airMopAlways);
fprintf(stderr, "%s: trouble writing:\n%s\n", me, err);
airMopError(mop); return 1;
}
airMopOkay(mop);
return 0;
}