robustness_analysis: Robustness of log-Euclidean and Bayes geometries to...

In astamm/bayesr: An implementation of Bayes Space for Gaussians in R

Description

Robustness of log-Euclidean and Bayes geometries to MRI-induced noise

Usage

robustness_analysis(tensor, n = 8L, B = 1000L, N = 20L, seed = NULL)

Arguments

tensor	A reference tensor.
n	Sample size (default: 8L).
B	Number of independent noisy samples (default: 1000L).
N	Number of healthy subjects used to build hypothetical template (default: 20L).
seed	Seed for random number generation (default: clock time).

Value

A tibble with simulation results.

Examples

arrow <- ggplot2::arrow(length = ggplot2::unit(0.4, "cm"), type = "closed")

refIsotropicTensor <- diag(3e-3, 3L)
data_isotropic <- robustness_analysis(refIsotropicTensor, B = 100L, seed = 1234)

data_isotropic$data %>%
 ggplot2::ggplot(ggplot2::aes(x = Sigma, y = MSE, col = Space)) +
 ggplot2::geom_point() +
 ggplot2::geom_line() +
 ggplot2::theme_minimal() +
 ggplot2::theme(legend.position = "top",
                axis.line = ggplot2::element_line(arrow = arrow)) +
 ggplot2::facet_grid(Metric ~ ., scales = "free") +
 ggplot2::scale_x_continuous(labels = scales::percent) +
 ggplot2::scale_y_log10()

refAnisotropicTensor <- diag(c(1.71e-3, 3e-4, 1e-4))
data_fascicles <- tibble::tibble()
theta <- pi * c(0, 1/6, 1/4, 1/3, 1/2)
for (a in theta) {
 R <- rbind(
   c(cos(a), sin(a), 0),
   c(-sin(a), cos(a), 0),
   c(0, 0, 1)
 )
 ref_tmp <- R %*% refAnisotropicTensor %*% t(R)
 tmp <- robustness_analysis(ref_tmp, B = 100L, seed = 1234)
 data_fascicles <- dplyr::bind_rows(
   data_fascicles,
   tmp$data %>% dplyr::mutate(Angle = round(a, 4L))
 )
}

data_fascicles %>%
  ggplot2::ggplot(ggplot2::aes(x = Sigma, y = MSE, col = Space)) +
  ggplot2::geom_point() +
  ggplot2::geom_line() +
  ggplot2::theme_minimal() +
  ggplot2::theme(legend.position = "top",
                 axis.line = ggplot2::element_line(arrow = arrow)) +
  ggplot2::facet_grid(Metric ~ Angle, scales = "free") +
  ggplot2::scale_x_continuous(labels = scales::percent) +
  ggplot2::scale_y_log10()

astamm/bayesr documentation built on May 10, 2019, 2:05 p.m.