next up previous
Next: A Simple Alternative Up: Estimation of Forman et Previous: Estimation of Forman et

Theoretical Basis

As the approximation of the derivative (equation 40) is not precise, it introduces a bias into the SPM estimation technique. This source of bias can be eliminated by using a finite difference calculation instead of the derivative-based one.

That is, the difference field is:

\begin{displaymath}
F_{\Delta}(\ensuremath{\mathbf{x}}) = F_S(\ensuremath{\mathbf{x}}+ \ensuremath{\mathbf{d}}) - F_S(\ensuremath{\mathbf{x}}).
\end{displaymath} (42)

The variance of this difference field is:

$\displaystyle E \{ F_{\Delta}(\ensuremath{\mathbf{x}}) F_{\Delta}(\ensuremath{\mathbf{x}}) \}$ $\textstyle =$ $\displaystyle E \{ F_S(\ensuremath{\mathbf{x}}+ \ensuremath{\mathbf{d}}) F_S(\e...
...nsuremath{\mathbf{x}}+ \ensuremath{\mathbf{d}}) F_S(\ensuremath{\mathbf{x}}) \}$ (43)
  $\textstyle =$ $\displaystyle 2 E\{ F_S(\ensuremath{\mathbf{x}}) F_S(\ensuremath{\mathbf{x}}) \...
...nsuremath{\mathbf{x}}+ \ensuremath{\mathbf{d}}) F_S(\ensuremath{\mathbf{x}}) \}$ (44)

Setting $\ensuremath{\mathbf{x_2}}= \ensuremath{\mathbf{x_1}}+ \ensuremath{\mathbf{d}}_x$ with $\ensuremath{\mathbf{d}}_x = (d,0,0)$, the expectations can be calculated using equation 14:

$\displaystyle E \{ F_S(\ensuremath{\mathbf{x}}) F_S(\ensuremath{\mathbf{x}}+ \ensuremath{\mathbf{d}}_x) \}$ $\textstyle =$ $\displaystyle \frac{1}{(4 \pi)^\frac{3}{2} \sigma_x \sigma_y \sigma_z} \exp\left( \frac{- d^{\, 2}}{4 {\sigma_x}^2} \right)$ (45)
$\displaystyle E \{ F_S(\ensuremath{\mathbf{x}}) F_S(\ensuremath{\mathbf{x}}) \}$ $\textstyle =$ $\displaystyle \frac{1}{(4 \pi)^\frac{3}{2} \sigma_x \sigma_y \sigma_z}.$ (46)

giving
\begin{displaymath}
E \{ F_{\Delta}(\ensuremath{\mathbf{x}}) F_{\Delta}(\ensurem...
...- \exp\left( \frac{- d^{\, 2}}{4 {\sigma_x}^2} \right) \right)
\end{displaymath} (47)

This enables the individual smoothing parameters, $\sigma_x, \sigma_y, \sigma_z$ to be found. That is:

\begin{displaymath}
{\sigma_x}^2 = \frac{- d^{\, 2}}{4 \ln\left( 1 - \frac{E \{ ...
...emath{\mathbf{x}}) F_S(\ensuremath{\mathbf{x}}) \} } \right)}.
\end{displaymath} (48)



Subsections
next up previous
Next: A Simple Alternative Up: Estimation of Forman et Previous: Estimation of Forman et
Mark Jenkinson 2001-11-07