Categoría: Espacios reflexivos

For all methods based on a subtraction between flow-sensitive and control images, it is very important to keep the flow-sensitized and control acquisitions as equal as possible, apart from the flow-sensitization. Subtraction images on a phantom (which has no flow) are a good test to see if the differences between the acquisitions are indeed only due to CBF. In practice there are always small non-zero subtraction differences on a phantom. This is called the static subtraction error (SSE).

The static subtraction error can be caused by eddy current differences between the two - labelling and control - pulse sequences. Kwong therefore suggested for FAIR to shift the slice selective gradient in time until after the inversion pulse in the global inversion sequence, to minimise the eddy current differences (Kwong et al., 1995).

Another source of static subtraction differences are pulse profile interactions between the labelling and readout pulses. To avoid any such interactions between the inversion and readout pulse profiles, an inversion to readout slice thickness ratio of 3:1 is commonly used in FAIR (Kim, 1995), and a saturation pulse twice the slice thickness is used for EPISTAR (Edelman, 1994). However, this does lengthen the transit time and decreases the sensitivity to slow flow. By improving the rf pulse profiles the gaps between the labelling and readout slices can be reduced (Frank et al., 1997). FOCI pulses have been advocated for this (Ordidge et al., 1996; Yongbi et al., 1999). Recently, Yongbi and

colleagues (Yongbi et. al., 2000) suggested an in vivo method for assessing the static

subtraction error and the minimum labelling-readout gap needed: by using Gd-DTPA in a calibration session on humans the blood Ti will be dramatically reduced and this will give subtraction images that consist only of static subtraction signal.

Zhou and colleagues suggested another cause of the static subtraction error: radiation damping. Radiation damping is caused by the reaction of the induced current in the coil to the transverse magnetization inducing this current (Zhou et al., 1998). It leads to an apparent increase in Ti relaxation. The extent of this increased longitudinal relaxation

depends on the total equilibrium magnetization Mq. For global inversion the total

difference in Ti relaxation which is not due to perfusion. Zhou et al. recommend a sequence (‘FAIRER’) with small gradients (~ 0.06 G/cm) that leave no large transverse

magnetization at any time before the acquisition. Radiation damping depends on field strength, shimming and rf coil characteristics. It is less relevant at lower, clinical, field strengths.

More recently another approach to minimize the static subtraction error was reported: the SEEPAGE sequence (Blamire and Styles, 2000) avoids the need for image subtraction and thus the subtraction problems. All spins in the slice of interest are pre­ saturated, followed by a train of non-selective inversion pulses interspersed with crusher gradients. The pulses, gradients and acquisition are timed in such a way that the slice of interest signal is ‘trapped’ at zero, while the unsaturated inflowing spins approach

saturation with a time constant Ti. Thus an image can be obtained that only contains spins that have entered the slice after the initial saturation. This approach seems very promising; more experimental evaluation (e.g. sensitivity to pulse imperfections, etc.) will be required to establish it further.

Even though the transit times are reduced for pulsed techniques compared to continuous methods, they can still be substantial. As discussed before, transit time can be mimimized by optimal rf pulse design, which allows for a small gap between the readout and labelling slices. However, for multi-slice sequences where one wide inversion slab encompasses (FAIR, (Kim and Tsekos, 1997)) or lies below (EPISTAR, (Edelman and Chen, 1998)) all slices that are imaged, longer transit times for at least some of the slices are inevitable. To deal with the residual transit time after pulse sequence optimizations, it is possible to incorporate the transit time delay 5t into the model: equation (3.10) can be expanded into (Thomas, D.L., 1999):

(3.11)

AM = 0 t < A

A M - 2ûfoM o —

exp(-(TI - a ) / T ,J - e x p ( - ( T I -

This uses a mean transit time for the whole slice. This approach has been successfully implemented by Yang and colleagues (Yang et al, 1998).

Another problem highlighted by Calamante et al. (Calamante et al, 1996) is the physical width of a tag: in FAIR the global inversion only pertains to spins within the rf

coil. After some time t, fresh spins will move in. This will lead to an underestimation of

CBF. Similarly in EPISTAR, the inversion slab is of a limited width determined by the inversion pulse and the slice selection gradient. Again, fresh spins will move in after some time T, leading to an underestimation of calculated CBF.

One way around this is to ensure that the inversion times used are shorter than the fresh spin inflow time x. On the other hand, the modelled Ti decay of the label will also limit the range over which a fresh spin that started out fresh can be distinguished from a fresh spin that was inverted before. Yang et al. (Yang et al., 1998) studied this

phenomenon: they increased the slice selective inversion width from 80-1000 mm using a body rf coil. They showed that spins further away than 300-400 mm did not contribute to the FAIR signal of the slice of interest at a TI of 1.2s due to the transit time and label decay. Hypothesizing this would not change dramatically for slightly longer inversion times, they proposed that using a standard rf coil of 300 mm diameter should not give any significant errors due to fresh inflow.

Buxton et al. (Buxton et al., 1998) describe the effects of transit times and the fresh inflow by considering the labelled spins as a bolus of finite width coming into the slice of interest. Their results reduce back to the formula above for standard ASL assumptions (plug flow of bolus, single compartment kinetics leading to a constant blood-brain partition

coefficient X and immediate extraction of water from vascular bed into tissue upon arrival

in voxel).

Wong and colleagues (Wong et al., 1998) have specifically designed sequences with minimised sensitivity to transit time delays and fresh inflow. These are called QUIPSS and QUIPSS II. They solve the problem of the transit time variation and limited length of the tag by setting these parameters themselves. In QUIPSS I, after the labelling below the slice, a saturation pulse is applied to the slice of interest at a time TIi to remove all inflow history. The signal is then acquired at a time TL. If TIi is chosen larger than the longest transit time and TI] shorter than that transit time + the bolus length, then transit time and bolus length

disappear from the equations. QUIPSS II applies the saturation pulse to the labelling region after time TIi (< bolus width). This cuts off the tag at this point. The bolus of length TIi

thus created can be followed into the slice and imaged at time TI2 (> max transit time).

Again the equations become independent of transit time and tag width. QUIPSS II can easily be extended into multi-slice, whereas QUIPSS cannot. Obviously the timing of the saturation pulse and readout is crucial, and difficult to get right when a range o f transit times is present. If one would want to characterise the bi-exponential curve completely, this would be a problem. Wong and colleagues approximate the effect of the bi-exponential with a correction factor q. In a later report (Luh et al., 1999) these researchers minimised pulse related quantification errors by optimising the saturation profile. The QUIPSS

techniques have a lower signal difference compared to standard pulsed ASL methods due to

label decay during TI2.