dcemri.bayes-methods: Bayesian Methods for Pharmacokinetic Modeling of Dynamic...

In bjw34032/dcemriS4: A Package for Image Analysis of DCE-MRI (S4 Implementation)

Bayesian Methods for Pharmacokinetic Modeling of Dynamic Contrast-Enhanced MRI Data

Description

Bayesian analysis of contrast agent concentration time curves from DCE-MRI.

Usage

dcemri.bayes(conc, ...)

## S4 method for signature 'array'
dcemri.bayes(
  conc,
  time,
  img.mask,
  model = "extended",
  spatial = 0,
  aif = NULL,
  user = NULL,
  nriters = 3000,
  thin = 3,
  burnin = 1000,
  tune = 267,
  ab.ktrans = if (spatial == 0) c(0, 1) else {
     c(1e-04, 1e-04)
 },
  ab.kep = ab.ktrans,
  ab.vp = c(1, 19),
  ab.tauepsilon = c(1, 1/1000),
  samples = FALSE,
  parallel = FALSE,
  verbose = FALSE,
  dic = FALSE,
  ...
)

Arguments

conc	Matrix or array of concentration time series (last dimension must be time).
...	Additional parameters to the function.
time	Time in minutes.
img.mask	Mask matrix or array. Voxels with mask=0 will be excluded.
model	is a character string that identifies the type of compartmental model to be used. Acceptable models include: "weinmann" Tofts & Kermode AIF convolved with single compartment model "extended" Weinmann model extended with additional vascular compartment (default) "orton.exp" Extended model using Orton's exponential AIF "kety.orton.exp" Kety model using Orton's exponential AIF
spatial	is an integer specifying spatial smoothing of kinetic parameters 0 no spatial smoothing, fit voxelwise 2 2D smoothing, fit slices separately 3 3D smoothing; Warning: this is computationally very expensive
aif	is a character string that identifies the parameters of the type of arterial input function (AIF) used with the above model. Acceptable values are: tofts.kermode (default) or fritz.hansen for the weinmann and extended models; orton.exp (default) or user for the orton.exp and kety.orton.exp model.
user	Vector of AIF parameters. For Tofts and Kermode: a1, m1, a2, m2; for Orton et al.: Ab, mub, Ag, mug.
nriters	Total number of iterations.
thin	Thining factor.
burnin	Number of iterations for burn-in.
tune	Number for iterations for tuning. The algorithm will be tuned to an acceptance rate between 0.3 and 0.6.
ab.ktrans	Mean and variance parameter for Gaussian prior on log(Ktrans).
ab.kep	Mean and variance parameter for Gaussian prior on log(kep).
ab.vp	Hyper-prior parameters for the Beta prior on vp.
ab.tauepsilon	Hyper-prior parameters for observation error Gamma prior.
samples	If TRUE output includes samples drawn from the posterior distribution for all parameters.
parallel	If TRUE algorithm is parallelized using parallel.
verbose	Logical variable (default = FALSE) that allows text-based feedback during execution of the function.
dic	If TRUE, the deviance information criterion (DIC) and effective number of parameters (pD) will be computed. If "samples = TRUE", then samples of the DIC and pD will be given.
vp	Fractional occupancy in the plasma space.

Details

See Schmid et al. (2006) for details.

Value

Parameter estimates and their standard errors are provided for the masked region of the multidimensional array. All multi-dimensional arrays are output in nifti format. They include:

ktrans	Transfer rate from plasma to the extracellular, extravascular space (EES).
ktranserror	Error on ktrans.
kep	Rate parameter for transport from the EES to plasma.
keperror	Error on kep.
ve	Fractional occupancy by EES (the ratio between ktrans and kep).
vperror	Error on ve.
vp	Fractional occupancy by plasma.
sigma2	The residual sum-of-squares from the model fit.
time	Acquisition times (for plotting purposes).
DIC	Deviance information criterion.
DIC.map	Contribution to DIC per voxel.
pD	Effective number of parameters.
pD.map	Constribution to pD per voxel.

Note, not all parameters are available under all models choices.

Author(s)

Volker Schmid [e-mail]

References

Schmid, V., Whitcher, B., Padhani, A.R., Taylor, N.J. and Yang, G.-Z. (2006) Bayesian methods for pharmacokinetic models in dynamic contrast-enhanced magnetic resonance imaging, IEEE Transactions on Medical Imaging, 25 (12), 1627-1636.

See Also

dcemri.lm, dcemri.map, dcemri.spline

Examples

data("buckley")
xi <- seq(5, 300, by=5)
img <- array(t(breast$data)[,xi], c(13,1,1,60))
mask <- array(TRUE, dim(img)[1:3])
time <- buckley$time.min[xi]

## Bayesian estimation with Fritz-Hansen default AIF
fit.bayes <- dcemri.bayes(img, time, mask, aif="fritz.hansen",
                         nriters=1000, thin=2, burnin=200)

## Bayesian estimation with "orton.exp" function fit to Buckley's AIF
aif <- buckley$input[xi]
aifparams <- orton.exp.lm(time, aif)
aifparams$D <- 1
fit.bayes.aif <- dcemri.bayes(img, time, mask, model="orton.exp",
                              aif="user", user=aifparams,
                              nriters=1000, thin=2, burnin=200)

plot(breast$ktrans, fit.bayes$ktrans, xlim=c(0,1), ylim=c(0,1),
     xlab=expression(paste("True ", K^{trans})),
     ylab=expression(paste("Estimated ", K^{trans}, " (Bayesian)")))
points(breast$ktrans, fit.bayes.aif$ktrans, pch=2)
abline(0, 1, lwd=2, col=2)
legend("right", c("extended/fritz.hansen","orton.exp/user"), pch=1:2)

fit.lm <- dcemri.lm(img, time, mask, aif="fritz.hansen")
fit.lm.aif <- dcemri.lm(img, time, mask, model="orton.exp", aif="user",
                        user=aifparams)

plot(breast$ktrans, fit.bayes$ktrans, xlim=c(0,1), ylim=c(0,1),
     xlab=expression(paste("True ", K^{trans})),
     ylab=expression(paste("Estimated ", K^{trans})))
points(breast$ktrans, fit.bayes.aif$ktrans, pch=2)
points(breast$ktrans, fit.lm$ktrans, pch=3)
points(breast$ktrans, fit.lm.aif$ktrans, pch=4)
abline(0, 1, lwd=2, col=2)
legend("bottomright", c("Bayesian Estimation (fritz-hansen)",
                        "Bayesian Estimation (orton.exp)",
                        "Levenburg-Marquardt (weinmann/fritz.hansen)",
                        "Levenburg-Marquardt (orton.exp/user)"), pch=1:4)

bjw34032/dcemriS4 documentation built on June 2, 2022, 3:51 p.m.