Flow velocities can be measured using PC-MRV. Here velocity encoding gradients are applied which cause a phase evolution of moving spins proportional to the velocity while for static spins the initial phase is retained after encoding. The phase
\(\Phi\) of the Nuclear Magnetic Resonance (NMR) signal is described by
$$\Phi = \Phi _{0} + \gamma \cdot \vec{v} \cdot \vec{M}_{1} {\text{ }}$$
(1)
where
\(\gamma\) is the gyromagnetic ratio and
\(\vec{v}\) is the velocity.
\(\vec{M}_{1}\) is the first moment of the gradient applied for the duration
T defined by:
$$\vec{M}_{1} = \int\limits_{0}^{T} {{{\vec{G}}}(t) \cdot t{\text{d}}t}$$
(2)
\(\Phi _{0}\) results from background phase effects due to eddy currents, magnetic susceptibility and concomitant gradients. To filter out such effects at one-directional velocity, two measurements are done with different first moments
\(\vec{M}_{1}^{{(1)}}\) and
\(\vec{M}_{1}^{(2)}\) (e.g. using inverted gradient polarities) but under otherwise same conditions. Subtraction of the signal phases results in
\(\Delta \Phi\), which is dependent on velocity
\(\vec{v}\) and
\(\Delta \vec{M}_{1}=\vec{M}_{1}^{(1)}-\vec{M}_{1}^{(2)}\). Nonetheless, eddy current errors originating from the velocity encoding gradients are not eliminated. Therefore, an additional measurement is done with same parameters but without flow and the phase images are subtracted.
In this work the four-step balanced encoding was used (Pelc
1991).
\(Z_{1}\) to
\(Z_{4}\) are the complex images acquired for individual encoding steps. The velocity
v is determined as
$$\begin{aligned} v=\frac{\Delta \Phi }{\gamma \cdot \Delta M_{1}}. \end{aligned}$$
(3)
The phase can only be uniquely measured over an interval of
\(2\pi\). If there is flow in positive and negative direction, it is common that the VENC is thus defined as the velocity that produces a phase shift
\(\Delta \Phi =\pi\), i.e.
$$\begin{aligned} \text{VENC}=\frac{\pi }{\gamma \cdot \Delta M_{1}}. \end{aligned}$$
(4)
For too high velocities, the velocity dependent phase shift can exceed
\(\pm \pi\) and phase aliasing occurs. The phase shift can be calculated as
\(\Delta \Phi =\arg (Z_{1})-\arg (Z_{2})=\Phi _{1}-\Phi _{2}\), which is computationally costly because of two
\(\arg\) operations. Also additional aliasings may be introduced. Instead, the phase shift is calculated by
\(\Delta \Phi =\arg (Z_{1}/Z_{2})=\arg (Z_{1}\cdot {\bar{Z}}_{2})\) (Bernstein
2004). The phases in x-, y- and z-direction are calculated as follows:
$$\begin{aligned} \Delta \Phi _{\text{x}}&=\text{arg}({\bar{Z}}_{1}\cdot Z_{2}\cdot Z_{3}\cdot {\bar{Z}}_{4}) \end{aligned}$$
(5)
$$\begin{aligned} \Delta \Phi _{\text{y}}&=\text{arg}({\bar{Z}}_{1}\cdot Z_{2}\cdot {\bar{Z}}_{3}\cdot Z_{4}) \end{aligned}$$
(6)
$$\begin{aligned} \Delta \Phi _{\text{z}}&=\text{arg}({\bar{Z}}_{1}\cdot {\bar{Z}}_{2}\cdot Z_{3}\cdot Z_{4}) \end{aligned}$$
(7)
Analogously to Lee (
1995) the standard deviation of the velocity is determined as
$$\sigma _{v} = \frac{{\sqrt N }}{\pi } \cdot \frac{{{\text{VENC}}}}{{{\text{SNR}}}}$$
(8)
where
N is the number of encoding steps. For each voxel the Signal-to-Noise Ratio (SNR) is calculated from the magnitude
I of the complex signal and the standard deviation
\(\sigma\):
$$\begin{aligned} \text{SNR}=\frac{I}{\sigma } \end{aligned}$$
(9)
In magnitude images the noise distribution in regions without NMR signal is governed by a Rayleigh distribution (Gudbjartsson
1995). If the standard deviation
\(\sigma _{0}\) is determined in the signal free part of a magnitude image, the standard deviation
\(\sigma\), that is used to characterise the VNR, is given by
\(\sigma =1.5264\cdot \sigma _{0}\). In Eq. (
8), it is visible that the VENC value should be set as low as possible to achieve a low noise value
\(\sigma _{v}\) (i.e. good VNR). However, it must be noted that for
\(\text{VENC}<v_{\text{true}}\) aliasing occurs. For correction of the aliasing a reference scan with a higher VENC value is used and difference
\(D=v_{\text{high}}-v_{\text{low}}\) is determined. The number of wrappings
w is then determined as
$$\begin{aligned} w=\text{N.I.}\left(\frac{D}{2\cdot \text{VENC}_{\text{low}}}\right) \end{aligned}$$
(10)
where
\(\text{N.I.}\) is the nearest integer function. The corrected (unwrapped) velocity is calculated as:
$$\begin{aligned} v_{\text{unwr}}=v+2\cdot w\cdot \text{VENC}_{\text{low}} \end{aligned}$$
(11)
In case of three-directional velocity, an eight-step measurement can be done (four steps for low- and four steps for high-VENC measurement). This doubles the number of measurements in comparison with a single-VENC measurement but results in a VNR improvement by a factor
$$\begin{aligned} R=\frac{\text{VENC}_{\text{high}}}{\text{VENC}_{\text{low}}}=\frac{\sigma _{v-\text{high}}}{\sigma _{v-\text{low}}}. \end{aligned}$$
(12)
It should be noted that for a large
R unwrapping may fail due to large noise values in the difference
D. Neglecting pulsation effects and assuming stable conditions (temperature, electrical components) during the whole measurement operation, the condition from Lee (
1995) for a low number of wrong unwrappings can be written as
$$\begin{aligned} \text{noise}(D)<\text{VENC}_{\text{low}}. \end{aligned}$$
(13)
Regarding Eq. (
12) we formulate the noise value at each voxel by
\(\text{noise}(D)=m\cdot \sqrt{\frac{1}{R^2}+1}\cdot \sigma _{v-\text{high}}\) where
m is the number of standard deviations that can be tolerated. With Eq. (
8) it follows
$$\begin{aligned} m<\frac{\pi }{\sqrt{R^{2}+1}\cdot \sqrt{N}}\cdot \text{SNR} \end{aligned}$$
(14)
For standard normal distribution, the probability of a result outside of the interval
\([-m\cdot \sigma ;m\cdot \sigma ]\) can be calculated by
$$\begin{aligned} P(x=m\cdot \sigma )=1-\int _{-x}^{x} f(x)\, {\rm d}x=1-\text{erf}\left(\frac{m}{\sqrt{2}}\right) \end{aligned}$$
(15)
where
f(
x) is the probabilty density function and
\(\text{erf}\) the Gaussian error function. For a given ratio
R and SNR, the fraction
P of wrongly unwrapped voxels can be calculated with Eqs. (
14) and (
15).