Research article Received: 5 November 2013,
Revised: 16 January 2014,
Accepted: 7 March 2014,
Published online in Wiley Online Library: 2 April 2014
(wileyonlinelibrary.com) DOI: 10.1002/nbm.3104
Effects of diffusion on high-resolution quantitative T2 MRI Wendy Oakdena* and Greg J. Stanisza,b Carr–Purcell–Meiboom–Gill-based sequences are often assumed to be insensitive to diffusion. However, imaging gradients always contribute some degree of diffusion weighting which increases with resolution. This may cause an apparent decrease in T2 when using a multi-echo sequence, such as quantitative T2 (qT2) at high resolution. This study investigated the impact of diffusion on the qT2 sequence. An equation was developed relating the diffusion factor associated with each echo (bqT2) to the underestimation of T2, which was strongly dependent on both the actual T2 and the apparent diffusion coefficient of the tissue. The diffusion dependence of the measured T2 was demonstrated in rat spinal cord. The measured T2 was independent of the imaging plane in gray matter, where diffusion was isotropic, and orientation dependent in white matter, where diffusion was strongly anisotropic. The dependence of the measured T2 on the actual T2 value was also demonstrated in MnCl2 phantoms. The relationship between the resolution and underestimation of T2 was investigated both theoretically and experimentally for the original readout and a fully refocused readout. The fully refocused readout increased the resolution at which diffusion effects could be neglected whilst measuring T2. To avoid the misidentification of cerebrospinal fluid when applying qT2 in the brain or spinal cord, a minimum in-plane voxel dimension of 0.2 mm was suggested for the standard qT2 sequence and 0.1 mm for the refocused readout. Simulations of myelin water fraction measurement indicated that signal-to-noise ratio requirements were increased in the presence of diffusion. Finally, the use of decreasing spoiler gradients to attenuate stimulated echoes should be avoided, as it was found to distort the T2 distribution when the slice thickness was less than 1 mm. Copyright © 2014 John Wiley & Sons, Ltd. Keywords: Quantitative T2; Relaxometry < Endogenous Contrast Methods < Methods and Engineering; Diffusion; Myelin; CPMG; White Matter
INTRODUCTION
672
The highest achievable resolution in MRI is on the order of 10–100 μm, depending on the magnetic field strength, strength of the imaging gradients and the physical properties of the tissue being imaged (1). Although it is possible to image a single cell, it is extremely time consuming (2). Quantitative imaging methods can be used to gain information about water molecules in different compartments within a cell, or group of cells, even though the precise physical locations of these compartments cannot be determined. Quantitative T2 (qT2) is one such technique, and reveals the presence of multiple water components relaxing with different T2 relaxation times (3). This is especially informative in white matter (WM), where water trapped in the myelin sheath surrounding the axons has a significantly shorter T2 than either intra- or extracellular water. The T2 decay curve of healthy WM can be decomposed into three separate pools based on the T2 relaxation time. The shortest of these relaxation times (10–50 ms) is correlated with water trapped within the myelin sheath, the intermediate relaxation time (70–90 ms) with intra-/extracellular (I/E) water, and the longest (~2 s) with cerebrospinal fluid (CSF) (3). These values have been established in healthy human brain at 1.5 and 3 T, and are generally accepted, as T2 is considered to be weakly dependent on the magnetic field (4,5). Alterations in the relaxation time of these pools have been shown to correlate with particular pathology, such as demyelination (6,7) or inflammation (8–10). As a result of the sensitivity of T2 to microstructural changes, histology is required to determine the pathological significance of a particular T2 distribution. Once the underlying
NMR Biomed. 2014; 27: 672–680
pathology is well understood, qT2 can be a powerful tool for the investigation of disease, as is the case with multiple sclerosis (6). In order to fully explore the effects of various pathologies on qT2, animal models of brain disease are often required. A higher resolution is necessary for the examination of these smaller brains and to reduce within-voxel heterogeneity. Unfortunately, as the resolution increases, the gradients required begin to have a noticeable effect on the T2 decay curves. High-field, high-resolution qT2 and multi-slice multi-echo (MSME) imaging of ex vivo human brain (7), in vivo mouse brain (11,12) and ex vivo rat spinal cord (13,14) all report significantly shorter T2 values than qT2 images acquired at 1.5 and 3 T.
* Correspondence to: W. Oakden, Sunnybrook Health Sciences Centre, 2075 Bayview Ave, S605, Toronto, ON, Canada, M4N 3 M5. E-mail: [email protected] a W. Oakden, G. J. Stanisz Department of Medical Biophysics, University of Toronto, Toronto, ON, Canada b G. J. Stanisz Physical Sciences Platform, Sunnybrook Research Institute, Toronto, ON, Canada Abbreviations used: BW, bandwidth; CPMG, Carr–Purcell–Meiboom–Gill; CSF, cerebrospinal fluid; ETL, echo train length; FOV, field of view; GM, gray matter; I/E, intra-/extracellular; MSME, multi-slice multi-echo; MWF, myelin water fraction; NNLS, non-negative least squares; qT2, quantitative T2; SNR, signal-to-noise ratio; WM, white matter.
Copyright © 2014 John Wiley & Sons, Ltd.
EFFECTS OF DIFFUSION ON HIGH-RESOLUTION QUANTITATIVE T2 MRI In homogeneous systems, T2 is well defined and only very weakly dependent on the magnetic field (4). In tissue, microscopic field variations arise from differing tissue susceptibilities, as well as paramagnetic or superparamagnetic particles, which are not completely compensated by the spin echo and can cause a decrease in the measured T2 which will depend on both the field strength and TE (5). Increasing field strength also leads to increased signal-to-noise ratio (SNR), allowing for higher resolution. The imaging gradients used to achieve these higher resolutions increase the diffusion weighting and can also decrease the measured T2, leading to a decrease in the measured T2 which only appears to be dependent on the field strength (15,16). Sequences designed to measure diffusion deal with imaging gradients by calculating diffusion based on the entire sequence rather than only those gradients deliberately being used to add diffusion weighting (17,18). Imaging sequences, especially those with long echo trains, can suffer from a diffusion-related decrease in signal or change in contrast depending on the resolution and tissue (19–22). An understanding of these changes is vitally important when interpreting images and when comparing values across studies. Finally, there are sequences which aim to quantify T2 as consistently as possible, where it is necessary to minimize the effects of diffusion and be aware of the magnitude of possible T2 miscalculation. NMR microscopy acquires a single echo per TR, with modifications to the conventional spin echo readout prewinder in order to avoid the underestimation of T2 as a result of imaging gradients (23), but is too long to be used in vivo. The qT2 sequence is based on the Carr–Purcell–Meiboom–Gill (CPMG) sequence (24,25), and uses a multi-echo readout from which T2 decay curves can be obtained and analyzed. The CPMG sequence substantially minimizes the effects of diffusion as a result of a constant background gradient (24); however, the additional gradients required for imaging also act as diffusionsensitizing gradients and can contribute substantially to the signal decay (20,26). The effect on the qT2 sequence is more significant than a simple loss of signal with time; the diffusion sensitization increases with each echo, altering the decay curve and decreasing the apparent T2 values. This is similar to the effect seen in conventional MSME or turbo spin echo sequences (20,21), but simpler to understand, as the diffusion weighting in each image is constant across that image. This article describes the incorporation of a fully refocused readout gradient, typically used for flow compensation (27,28), into the original qT2 sequence first proposed by Poon and Henkelman (29), and investigates the relationship between T2 underestimation and resolution for both the original and refocused readout sequence. It also explains how to calculate the underestimation of T2 resulting from both imaging and spoiler gradients, demonstrates the issues arising from anisotropic diffusion, and examines the effects on myelin water fraction (MWF) estimation. In addition, we discuss the distortions in the T2 distribution resulting from decreasing spoiler gradient schemes, similar to the complications from non-linear TE spacing (30).
the b value parameter describing the diffusion weighting is given by: b ¼ ðγGδÞ2 Δ 3δ , where γ is the Larmor frequency, G is the gradient strength, δ is the gradient duration and Δ is the time between gradients. The qT2 sequence consists of a 90° excitation, followed by a train of 180° refocusing pulses (Fig. 1). Each echo contributes to a different image, meaning that each image has its own echo time, with TE typically referring to the first echo time. Each 180° pulse is bracketed by a pair of spoiler gradients (Fig. 1a, c), which are similar to the diffusion-sensitizing gradients used in a pulsed gradient spin echo experiment (15). Imaging gradients (Fig. 1b, d) can also be considered as pairs of diffusion-sensitizing gradients. The b values arising from the gradients bracketing a 180° pulse can be calculated using the sequence parameters, or estimated based on the linear voxel size (Δx) and slice thickness (Δz). Assuming that the echoes are evenly spaced, let b1 refer to the b value resulting from the gradients between the initial excitation and the first echo at time TE, and bi refer to the b value resulting from the gradients between times TE i –1 and TE i; then the signal for echo n is equal to: n X TEn T2
SðnÞ ¼ M0 e
D
e
bi i¼1
[1]
If the spoiler and imaging gradients do not change with echo number (Fig. 1b, c), then bi is constant and equal to bqT2, and the signal for echo n is equal to:
SðnÞ ¼ M0 e
TEn T2
ebqT2 Dn
[2]
Neglecting the effects of diffusion, this signal can be fitted to the following equation:
SðnÞ ¼ M0 e
TEn T 2app
[3]
resulting in an apparent T2 value (T2app) which is dependent on b, D and TE: 1 1 bqT2 D [4] ¼ þ T 2app T 2 TE
THEORY Effect of diffusion on the measured T2 value
NMR Biomed. 2014; 27: 672–680
Copyright © 2014 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/nbm
673
In the pulsed gradient spin echo sequence developed by Stejskal and Tanner (15), the measured signal S is attenuated relative to S0, the signal measured in the absence of diffusion weighting, by S = S0e Db, where D is the apparent diffusion coefficient and
Figure 1. Radiofrequency (RF), spoiler and readout gradients used in a quantitative T2 (qT2) sequence. The standard qT2 sequence described by Poon and Henkelman (29) uses decreasing spoilers (a) and places the readout prewinder at the beginning of the sequence (b). The diffusion-minimized qT2 sequence uses equal spoilers (c) and refocuses the readout gradients after each readout (d).
W. OAKDEN AND G. J. STANISZ This situation becomes slightly more complicated in the case of refocused readout gradients (Fig. 1d) where 2b1 = bi for i > 1. Then, setting bi = bqT2, b1 = ½bqT2, giving:
SðnÞ ¼ M0 e
TEn T2
ebqT2 Dðn2Þ 1
which results in a modification to Equation [4]: 1 1 bqT2 D 1 1 ¼ þ T 2app ðnÞ T 2 TE 2n
[5]
[6]
Although the apparent T2 value is dependent on the number of echoes n, the dependence is minimal and can either be neglected for small values of bqT2, or a small gradient with a b value of ½bqT2 may be added to the beginning of the sequence, reducing this to Equation [4]. If the spoiler gradient amplitude decreases with echo number (Fig. 1a), then T2app is a function of the echo number n: 1 1 D ¼ T 2app ðnÞ T 2 TEn
n X
bi
[7]
i¼1
The resulting signal decay curve can no longer be fitted with a mono-exponential model. Readout gradients The readout gradient Gx depends on the bandwidth (BW) and the field of view (FOV): Gx δ ¼
2πBW 2πN Ts π δ ¼ ¼ γFOV γFOVT s 2 γΔx
echoes. The main disadvantage of the decreasing spoiler gradient scheme is that the final spoiler should completely dephase the magnetization across the slice, and the first spoiler is multiplied by the echo train length (ETL). For thin slices and long echo trains, this initial spoiler may be impossible to achieve. Prasloski et al. (31) have developed a new technique which corrects for stimulated echoes. Non-selective refocusing pulses improve performance across the excited slice, reducing stimulated echoes, but exciting magnetization outside the slice of interest. Equally sized spoilers (Fig. 1c) are sufficient to spoil the out-of-slice signal as it is not excited by the initial slice-selective excitation pulse. The size of these spoilers depends on the geometry of the sample being scanned and the coil sensitivity. Setting these spoiler gradients appropriately is a matter of trial and error, and there may be a temptation to make them as large as possible whilst avoiding eddy current effects. Phase-encoding gradients Phase-encoding gradients which bracket the readout, as in most MSME-style sequences, contribute diffusion weighting unevenly across the image, as each phase-encoding step acquires its own diffusion weighting. However, they are smallest near the centre of k space in which the majority of diffusion contrast is encoded (17,18). A study of the diffusion sensitivity of MSME-style sequences, also referred to as turbo spin echo sequences, has demonstrated that the phase-encoding gradients account for
Revised: 16 January 2014,
Accepted: 7 March 2014,
Published online in Wiley Online Library: 2 April 2014
(wileyonlinelibrary.com) DOI: 10.1002/nbm.3104
Effects of diffusion on high-resolution quantitative T2 MRI Wendy Oakdena* and Greg J. Stanisza,b Carr–Purcell–Meiboom–Gill-based sequences are often assumed to be insensitive to diffusion. However, imaging gradients always contribute some degree of diffusion weighting which increases with resolution. This may cause an apparent decrease in T2 when using a multi-echo sequence, such as quantitative T2 (qT2) at high resolution. This study investigated the impact of diffusion on the qT2 sequence. An equation was developed relating the diffusion factor associated with each echo (bqT2) to the underestimation of T2, which was strongly dependent on both the actual T2 and the apparent diffusion coefficient of the tissue. The diffusion dependence of the measured T2 was demonstrated in rat spinal cord. The measured T2 was independent of the imaging plane in gray matter, where diffusion was isotropic, and orientation dependent in white matter, where diffusion was strongly anisotropic. The dependence of the measured T2 on the actual T2 value was also demonstrated in MnCl2 phantoms. The relationship between the resolution and underestimation of T2 was investigated both theoretically and experimentally for the original readout and a fully refocused readout. The fully refocused readout increased the resolution at which diffusion effects could be neglected whilst measuring T2. To avoid the misidentification of cerebrospinal fluid when applying qT2 in the brain or spinal cord, a minimum in-plane voxel dimension of 0.2 mm was suggested for the standard qT2 sequence and 0.1 mm for the refocused readout. Simulations of myelin water fraction measurement indicated that signal-to-noise ratio requirements were increased in the presence of diffusion. Finally, the use of decreasing spoiler gradients to attenuate stimulated echoes should be avoided, as it was found to distort the T2 distribution when the slice thickness was less than 1 mm. Copyright © 2014 John Wiley & Sons, Ltd. Keywords: Quantitative T2; Relaxometry < Endogenous Contrast Methods < Methods and Engineering; Diffusion; Myelin; CPMG; White Matter
INTRODUCTION
672
The highest achievable resolution in MRI is on the order of 10–100 μm, depending on the magnetic field strength, strength of the imaging gradients and the physical properties of the tissue being imaged (1). Although it is possible to image a single cell, it is extremely time consuming (2). Quantitative imaging methods can be used to gain information about water molecules in different compartments within a cell, or group of cells, even though the precise physical locations of these compartments cannot be determined. Quantitative T2 (qT2) is one such technique, and reveals the presence of multiple water components relaxing with different T2 relaxation times (3). This is especially informative in white matter (WM), where water trapped in the myelin sheath surrounding the axons has a significantly shorter T2 than either intra- or extracellular water. The T2 decay curve of healthy WM can be decomposed into three separate pools based on the T2 relaxation time. The shortest of these relaxation times (10–50 ms) is correlated with water trapped within the myelin sheath, the intermediate relaxation time (70–90 ms) with intra-/extracellular (I/E) water, and the longest (~2 s) with cerebrospinal fluid (CSF) (3). These values have been established in healthy human brain at 1.5 and 3 T, and are generally accepted, as T2 is considered to be weakly dependent on the magnetic field (4,5). Alterations in the relaxation time of these pools have been shown to correlate with particular pathology, such as demyelination (6,7) or inflammation (8–10). As a result of the sensitivity of T2 to microstructural changes, histology is required to determine the pathological significance of a particular T2 distribution. Once the underlying
NMR Biomed. 2014; 27: 672–680
pathology is well understood, qT2 can be a powerful tool for the investigation of disease, as is the case with multiple sclerosis (6). In order to fully explore the effects of various pathologies on qT2, animal models of brain disease are often required. A higher resolution is necessary for the examination of these smaller brains and to reduce within-voxel heterogeneity. Unfortunately, as the resolution increases, the gradients required begin to have a noticeable effect on the T2 decay curves. High-field, high-resolution qT2 and multi-slice multi-echo (MSME) imaging of ex vivo human brain (7), in vivo mouse brain (11,12) and ex vivo rat spinal cord (13,14) all report significantly shorter T2 values than qT2 images acquired at 1.5 and 3 T.
* Correspondence to: W. Oakden, Sunnybrook Health Sciences Centre, 2075 Bayview Ave, S605, Toronto, ON, Canada, M4N 3 M5. E-mail: [email protected] a W. Oakden, G. J. Stanisz Department of Medical Biophysics, University of Toronto, Toronto, ON, Canada b G. J. Stanisz Physical Sciences Platform, Sunnybrook Research Institute, Toronto, ON, Canada Abbreviations used: BW, bandwidth; CPMG, Carr–Purcell–Meiboom–Gill; CSF, cerebrospinal fluid; ETL, echo train length; FOV, field of view; GM, gray matter; I/E, intra-/extracellular; MSME, multi-slice multi-echo; MWF, myelin water fraction; NNLS, non-negative least squares; qT2, quantitative T2; SNR, signal-to-noise ratio; WM, white matter.
Copyright © 2014 John Wiley & Sons, Ltd.
EFFECTS OF DIFFUSION ON HIGH-RESOLUTION QUANTITATIVE T2 MRI In homogeneous systems, T2 is well defined and only very weakly dependent on the magnetic field (4). In tissue, microscopic field variations arise from differing tissue susceptibilities, as well as paramagnetic or superparamagnetic particles, which are not completely compensated by the spin echo and can cause a decrease in the measured T2 which will depend on both the field strength and TE (5). Increasing field strength also leads to increased signal-to-noise ratio (SNR), allowing for higher resolution. The imaging gradients used to achieve these higher resolutions increase the diffusion weighting and can also decrease the measured T2, leading to a decrease in the measured T2 which only appears to be dependent on the field strength (15,16). Sequences designed to measure diffusion deal with imaging gradients by calculating diffusion based on the entire sequence rather than only those gradients deliberately being used to add diffusion weighting (17,18). Imaging sequences, especially those with long echo trains, can suffer from a diffusion-related decrease in signal or change in contrast depending on the resolution and tissue (19–22). An understanding of these changes is vitally important when interpreting images and when comparing values across studies. Finally, there are sequences which aim to quantify T2 as consistently as possible, where it is necessary to minimize the effects of diffusion and be aware of the magnitude of possible T2 miscalculation. NMR microscopy acquires a single echo per TR, with modifications to the conventional spin echo readout prewinder in order to avoid the underestimation of T2 as a result of imaging gradients (23), but is too long to be used in vivo. The qT2 sequence is based on the Carr–Purcell–Meiboom–Gill (CPMG) sequence (24,25), and uses a multi-echo readout from which T2 decay curves can be obtained and analyzed. The CPMG sequence substantially minimizes the effects of diffusion as a result of a constant background gradient (24); however, the additional gradients required for imaging also act as diffusionsensitizing gradients and can contribute substantially to the signal decay (20,26). The effect on the qT2 sequence is more significant than a simple loss of signal with time; the diffusion sensitization increases with each echo, altering the decay curve and decreasing the apparent T2 values. This is similar to the effect seen in conventional MSME or turbo spin echo sequences (20,21), but simpler to understand, as the diffusion weighting in each image is constant across that image. This article describes the incorporation of a fully refocused readout gradient, typically used for flow compensation (27,28), into the original qT2 sequence first proposed by Poon and Henkelman (29), and investigates the relationship between T2 underestimation and resolution for both the original and refocused readout sequence. It also explains how to calculate the underestimation of T2 resulting from both imaging and spoiler gradients, demonstrates the issues arising from anisotropic diffusion, and examines the effects on myelin water fraction (MWF) estimation. In addition, we discuss the distortions in the T2 distribution resulting from decreasing spoiler gradient schemes, similar to the complications from non-linear TE spacing (30).
the b value parameter describing the diffusion weighting is given by: b ¼ ðγGδÞ2 Δ 3δ , where γ is the Larmor frequency, G is the gradient strength, δ is the gradient duration and Δ is the time between gradients. The qT2 sequence consists of a 90° excitation, followed by a train of 180° refocusing pulses (Fig. 1). Each echo contributes to a different image, meaning that each image has its own echo time, with TE typically referring to the first echo time. Each 180° pulse is bracketed by a pair of spoiler gradients (Fig. 1a, c), which are similar to the diffusion-sensitizing gradients used in a pulsed gradient spin echo experiment (15). Imaging gradients (Fig. 1b, d) can also be considered as pairs of diffusion-sensitizing gradients. The b values arising from the gradients bracketing a 180° pulse can be calculated using the sequence parameters, or estimated based on the linear voxel size (Δx) and slice thickness (Δz). Assuming that the echoes are evenly spaced, let b1 refer to the b value resulting from the gradients between the initial excitation and the first echo at time TE, and bi refer to the b value resulting from the gradients between times TE i –1 and TE i; then the signal for echo n is equal to: n X TEn T2
SðnÞ ¼ M0 e
D
e
bi i¼1
[1]
If the spoiler and imaging gradients do not change with echo number (Fig. 1b, c), then bi is constant and equal to bqT2, and the signal for echo n is equal to:
SðnÞ ¼ M0 e
TEn T2
ebqT2 Dn
[2]
Neglecting the effects of diffusion, this signal can be fitted to the following equation:
SðnÞ ¼ M0 e
TEn T 2app
[3]
resulting in an apparent T2 value (T2app) which is dependent on b, D and TE: 1 1 bqT2 D [4] ¼ þ T 2app T 2 TE
THEORY Effect of diffusion on the measured T2 value
NMR Biomed. 2014; 27: 672–680
Copyright © 2014 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/nbm
673
In the pulsed gradient spin echo sequence developed by Stejskal and Tanner (15), the measured signal S is attenuated relative to S0, the signal measured in the absence of diffusion weighting, by S = S0e Db, where D is the apparent diffusion coefficient and
Figure 1. Radiofrequency (RF), spoiler and readout gradients used in a quantitative T2 (qT2) sequence. The standard qT2 sequence described by Poon and Henkelman (29) uses decreasing spoilers (a) and places the readout prewinder at the beginning of the sequence (b). The diffusion-minimized qT2 sequence uses equal spoilers (c) and refocuses the readout gradients after each readout (d).
W. OAKDEN AND G. J. STANISZ This situation becomes slightly more complicated in the case of refocused readout gradients (Fig. 1d) where 2b1 = bi for i > 1. Then, setting bi = bqT2, b1 = ½bqT2, giving:
SðnÞ ¼ M0 e
TEn T2
ebqT2 Dðn2Þ 1
which results in a modification to Equation [4]: 1 1 bqT2 D 1 1 ¼ þ T 2app ðnÞ T 2 TE 2n
[5]
[6]
Although the apparent T2 value is dependent on the number of echoes n, the dependence is minimal and can either be neglected for small values of bqT2, or a small gradient with a b value of ½bqT2 may be added to the beginning of the sequence, reducing this to Equation [4]. If the spoiler gradient amplitude decreases with echo number (Fig. 1a), then T2app is a function of the echo number n: 1 1 D ¼ T 2app ðnÞ T 2 TEn
n X
bi
[7]
i¼1
The resulting signal decay curve can no longer be fitted with a mono-exponential model. Readout gradients The readout gradient Gx depends on the bandwidth (BW) and the field of view (FOV): Gx δ ¼
2πBW 2πN Ts π δ ¼ ¼ γFOV γFOVT s 2 γΔx
echoes. The main disadvantage of the decreasing spoiler gradient scheme is that the final spoiler should completely dephase the magnetization across the slice, and the first spoiler is multiplied by the echo train length (ETL). For thin slices and long echo trains, this initial spoiler may be impossible to achieve. Prasloski et al. (31) have developed a new technique which corrects for stimulated echoes. Non-selective refocusing pulses improve performance across the excited slice, reducing stimulated echoes, but exciting magnetization outside the slice of interest. Equally sized spoilers (Fig. 1c) are sufficient to spoil the out-of-slice signal as it is not excited by the initial slice-selective excitation pulse. The size of these spoilers depends on the geometry of the sample being scanned and the coil sensitivity. Setting these spoiler gradients appropriately is a matter of trial and error, and there may be a temptation to make them as large as possible whilst avoiding eddy current effects. Phase-encoding gradients Phase-encoding gradients which bracket the readout, as in most MSME-style sequences, contribute diffusion weighting unevenly across the image, as each phase-encoding step acquires its own diffusion weighting. However, they are smallest near the centre of k space in which the majority of diffusion contrast is encoded (17,18). A study of the diffusion sensitivity of MSME-style sequences, also referred to as turbo spin echo sequences, has demonstrated that the phase-encoding gradients account for
