Ultra-High Contrast (UHC) MRI of the Brain, Spinal Cord and Optic Nerves in Multiple Sclerosis Using Directly Acquired and Synthetic Bipolar Filter (BLAIR) Images

Abstract

In this educational review, the basic physics underlying the use of ultra-high contrast (UHC) bipolar filter (BLAIR) sequences, including divided subtracted inversion recovery (dSIR), is explained. These sequences can increase the contrast produced by small changes in T₁ by a factor of ten or more compared with conventional IR sequences. In illustrative cases, the sequences were used in multiple sclerosis (MS) patients during relapse and remission and were compared with positionally matched conventional (T₂-weighted spin echo, T₂-FLAIR) images. Well-defined focal lesions were seen with dSIR sequences in areas where little or no change was seen with conventional sequences. In addition, widespread abnormalities affecting almost all of the white matter of the brain were seen during relapses when there were no corresponding abnormalities seen on conventional sequences (the whiteout sign). Grayout signs, in which there is a loss of contrast in gray matter or between gray matter and CSF, were also seen, as well as high signal boundaries around lesions. Disruption of the usual high signal boundary between white and gray matter was seen in leucocortical lesions. Lesions in the spinal cord were better seen or only seen with dSIR sequences. Generalized change was observed in the optic nerve with the dSIR sequence in a case of optic neuritis. UHC BLAIR sequences may be of considerable value for recognition of abnormalities in clinical practice and in research studies on MS.

1. Introduction

Ultra-high contrast (UHC) MRI is a term used to describe MRI which shows high-contrast demonstration of abnormalities in situations where little or no change from normal is seen on conventional state-of-the-art MR images. It is achieved without increase in static or gradient field strength.

With conventional MR sequences, change in a tissue property such as T₁ is usually used only once to create contrast. With one form of UHC MRI, the same change in a tissue property is used three or four times in the same sequence to create much greater contrast [1]. An example of this is the divided subtracted inversion recovery (dSIR) sequence, where changes in T₁ are used repeatedly to increase contrast. The contrast from small changes in T₁ with dSIR sequences may be ten or more times greater than that obtained with conventional IR sequences such as MP-RAGE (magnetization prepared-rapid acquisition gradient echo). Using this approach, widespread changes have been seen in mild traumatic brain injury (mTBI) and hypoxic injury to the brain in regions which show no abnormality with state-of-the-art T₂-weighted spin echo (T₂-wSE) and/or T₂-weighted fluid attenuated inversion recovery (T₂-FLAIR) sequences [2,3,4]. This is likely to be because the changes in tissue properties such as T₁ and T₂ are too small to show abnormalities with conventional sequences, but produce obvious changes with the greater contrast amplification provided by dSIR sequences.

There is little point in demonstrating lesions that are already seen with high contrast using conventional sequences with even greater contrast, so the focus in applications of UHC MRI has been on areas of the brain that show little or no lesion contrast and appear normal with conventional sequences [5].

In an earlier paper, a single case of multiple sclerosis (MS) was described in order to demonstrate the technical capabilities of a dSIR sequence [1]. In this paper, we demonstrate more cases of MS imaged with dSIR sequences, as well as a new bipolar filter sequence, namely, log then subtracted inversion recovery (lSIR). The paper also illustrates how the T₁-dependent dSIR and lSIR sequences can be supplemented by other tissue properties such as T₂ in the form of composite bipolar filter sequences. MS patients were examined in remission and during relapse.

In the next section of this educational review, the basic physics underlying the use of bipolar sequences is described. This is followed by a comparison of conventional and bipolar sequences to determine where the new sequences are likely to be of clinical or research value.

2. Basic Physics

The first part of this section is a condensed and updated version of work published previously [1,5]. The second part describes new techniques.

2.1. Tissue Property Filters (TP-Filters) and the Inversion Recovery (IR) Sequence

Instead of illustrating image contrast in the usual way, by plotting longitudinal and then transverse magnetization (M_Z and then M_XY) against time using the Bloch equations, it is possible to use plots of signals against T₁ and T₂, respectively. These plots are tissue property filters (TP-filters); the TPs may be mobile proton density, T₁, T₂, T₂*, D*, etc.

For TR >> T₁, the T₁ dependent part of the IR sequence has the following signal equation:

S_T1 = (1 − 2e^−TI/T₁)

(1)

where S_T1 is the signal from the T₁-filter, TI in the inversion time, and T₁ is the longitudinal magnetization relaxation time. Two T₁-filters of the IR sequence with different values of TI (TI_s = TI_short and TI_i = TI_intermediate) are shown in magnitude form in Figure 1A. The curves are negative unipolar filters.

The subtraction is as follows: first, TI_s T₁-unipolar filter minus the second TI_i T₁-unipolar filter produces the subtracted IR (SIR) T₁-bipolar filter seen in Figure 1B. The vertical dashed lines at the null points of the two IR T₁-filters seen in Figure 1 divide the X axes in Figure 1B,C into the lowest Domain (lD), the middle Domain (mD), and the highest Domain (hD). In the mD in Figure 1B (between the two dashed lines), the size of the slope of the SIR T₁-bipolar filter is about double that of the IR T₁-unipolar filters in the mD shown in Figure 1A. This is because the slope of the SIR filter in its mD in Figure 1B is the positive slope of the TI_s filter in the mD shown in Figure 1A minus the negative slope of the TI_i filter in the mD also shown in Figure 1A.

The two IR T₁-filters shown in Figure 1A can also be added to give the Added IR (AIR) T₁-filter shown in Figure 1C. In its mD, which is bounded by the vertical dashed lines, the signal is reduced to about 0.20 compared with its value of 2 at T₁ = 0 (i.e., about one tenth of the value).

Figure 2A shows a divided SIR (dSIR) T₁-bipolar filter in which the SIR T₁-bipolar filter in Figure 1B is divided by the AIR T₁-filter in Figure 1C, i.e., ${\displaystyle \frac{\mathrm{difference}}{\mathrm{sum}}}$. The signal S_dSIR of the dSIR filter is given by the following:

$${\mathrm{S}}_{\mathrm{dSIR}}={\displaystyle \frac{{\mathrm{S}}_{\mathrm{TIs}}-{\mathrm{S}}_{\mathrm{TIi}}}{{\mathrm{S}}_{\mathrm{TIs}}+{\mathrm{S}}_{\mathrm{TIi}}}}$$

(2)

where S_TIs is the signal from the shorter TI filter and S_TIi is the signal from the intermediate TI_i filter.

The resulting dSIR T₁-bipolar filter (Figure 2A) shows a negative pole and a positive pole. Its mD shows a highly positive nearly linear slope whose magnitude is about ten times greater than the slopes of the IR T₁-filters shown in Figure 1A. The dSIR filter has maximum and minimum signal values of ±1. (The contributions from mobile proton density and T₂ to conventional IR sequences cancel out with dSIR sequences, which are, therefore, T₁ maps. As a result, the whole sequence can be accurately represented by its T₁-bipolar filter.)

Figure 2B compares the contrast generated by a conventional S_TIs IR T₁-unipolar filter (pink) to that from the SIR T₁-bipolar filter (blue) shown in Figure 1B, from the same increase in T₁ (ΔT₁) (horizontal positive green arrow). Using the small change approximation of differential calculus, ΔT₁ is multiplied by the slopes of the respective S_TI IR and SIR filters (red lines) to produce the differences in signal ΔS shown, i.e., the contrast resulting from each of them. This is given by the vertical pink and blue arrows on the right. The SIR T₁-bipolar filter generates about twice the contrast (blue arrow) of the S_TIs IR T₁-unipolar filter (pink arrow) from the same increase in T₁, ΔT₁. The S_TIs IR T₁-filter is that of a conventional IR sequence such as MP-RAGE.

Figure 2C compares the contrast produced by the S_TIs IR T₁-unipolar filter (pink) to that produced by the dSIR T₁-bipolar filter shown in Figure 2A (blue). The increase in T₁ ΔT₁ (horizontal positive green arrow) is multiplied by the slopes of the respective S_TIs IR and dSIR filters (red lines) to produce the differences in signal ΔS shown, i.e., the contrast generated by the two filters. This is given by the vertical pink and blue arrows on the right. From the same increase in T₁ (ΔT₁), the dSIR T₁-bipolar filter produces contrast about ten times greater than the contrast produced by the S_TIs IR T₁-unipolar filter.

To produce the large increase in contrast shown in Figure 2C, the dSIR sequence is typically targeted at small increases in the T₁ of normal tissue produced by disease (shown by the horizontal green arrow). These increases are positioned within the steeply sloping mD of the dSIR T₁-bipolar filter. This is achieved by choosing appropriate values of TI_s and TI_i. The target is often small increases in the T₁ of white matter.

High contrast can also be produced by difference or change in T₁ within the lower part of the hD and the upper part of the lD, where the dSIR T₁-bipolar filter has relatively steep slopes. (The term ‘tissue’ is used to include fluids in the rest of this paper unless otherwise specified).

2.2. Contrast at Tissue Boundaries

MRI tissue boundaries take different forms, such as a gradual change in signal from one tissue to another, as well as sharply defined high and low signal boundaries between tissues.

In the boundary region between two pure tissues (such as between white and gray matter), the T₁s of voxels which contain mixtures of the two tissues typically span the range of T₁ values between those of the two pure tissues. If a narrow mD dSIR T₁-bipolar filter (e.g., with TI_s nulling normal white matter and TI_i longer than TI_s but less than that needed to null gray matter) is used, there are mixtures of the two tissues in voxels in the boundary region, with intermediate T₁ values (T_1W,G) which correspond with the peak signal of the T₁-bipolar filter (S_W,G), as shown in Figure 3 (vertical blue arrow). This produces a high signal boundary between white matter and gray matter. The dSIR T₁-bipolar filter shows much higher contrast than that produced between white matter and gray matter by a conventional white matter nulled IR sequence such as MP-RAGE (Figure 3) (vertical pink arrow). High contrast boundaries are also seen at the junction between white matter and CSF at ventricular boundaries through the same mechanism.

2.3. T₁ Maps and Qualitative—Quantitative MRI

To better understand the dSIR T₁-bipolar filter, a linear equation of the form y = mx + c can be used to approximate the filter in its mD. The equation is produced by fitting a straight line between the first and last points of the mD (i.e., first point x = ${\displaystyle \frac{{\mathrm{TI}}_{\mathrm{s}}}{\ln 2}}$ and y = −1, and last point x = ${\displaystyle \frac{{\mathrm{TI}}_{\mathrm{i}}}{\ln 2}}$ and y = +1). In the mD, the signal of the dSIR sequence S_dSIR, is given by the following:

$${\mathrm{S}}_{\mathrm{dSIR}} \approx {\displaystyle \frac{\ln 4}{∆\mathrm{T}\mathrm{I}}} {\mathrm{T}}_1-{\displaystyle \frac{\Sigma \mathrm{T}\mathrm{I}}{∆\mathrm{T}\mathrm{I}}}$$

(3)

where ΔTI = TI_i − TI_s (i.e., second TI minus first TI), which is positive, and ΣTI = TI_s + TI_i, which is also positive. Note that, because ΔTI is positive, the slope ${\displaystyle \frac{\ln 4}{∆\mathrm{T}\mathrm{I}}}$ is positive. The offset is negative. T₁ is the longitudinal magnetization relaxation time.

The expression in Equation (3) illustrates four key features of the dSIR T₁-bipolar filter, firstly, the near linear change in signal (i.e., S_dSIR) with T₁ in the mD, secondly, the filter has a slope in the mD equal to ${\displaystyle \frac{\ln 4}{\Delta \mathrm{TI}}}$, thirdly, the filter shows a high sensitivity to small changes in T₁ when the size of ΔTI is small. When ΔT₁ is small, the size of ΔTI can be decreased to match it, and so the sensitivity of the filter is scaled up. This is because the reduction in ΔTI increases the steepness of the T₁-filter in the mD and, thus, the amplification of the contrast produced from ΔT₁. It compensates for the small value of ΔT₁ until the sequence becomes SNR and/or artefact limited. This is not the case with conventional IR sequences, where, if ΔT₁ is decreased, contrast is decreased.

Fourthly, Equation (3) can be used to calculate T₁ values in the mD, as follows:

$${\mathrm{T}}_1\approx {\displaystyle \frac{∆\mathrm{T}\mathrm{I}}{\mathrm{l}\mathrm{n}4}} {\mathrm{S}}_{\mathrm{dSIR}}+{\displaystyle \frac{\Sigma \mathrm{T}\mathrm{I}}{\mathrm{l}\mathrm{n}4}}$$

(4)

The terminology is the same as for Equation (3).

The linear approximation is only valid in the mD.

T₁s in the mD occupy the full range of the display gray scale from black to white. Outside of the mD, in the lD and hD, the magnitude of the T₁-bipolar filter signal decreases towards zero at minimum and maximum values of T₁. T₁s, from the shortest and longest T₁s in the lD and hD, respectively, only occupy parts of the display gray scale.

2.4. Log then Subtracted Inversion Recovery (lSIR) Sequences

Another bipolar filter is the natural logarithm (ln) then subtracted inversion recovery (lSIR) filter. This is half of the following subtraction: ln short S_TIs minus ln intermediate S_TIi. Figure 4 shows a dSIR T₁-bipolar filter in blue and the corresponding lSIR T₁-bipolar filter with the same TI_s and TI_i in orange. The two filters appear similar in the lowest region of the lD and in the highest region of the hD, as well as in the central region of the mD. The slope of the dSIR filter (blue) in the mD is essentially constant and equal to ${\displaystyle \frac{\ln 4}{\Delta \mathrm{TI}}}$. The slope of the lSIR filter (orange) is the same as that of the dSIR filter at the center of the mD, but greater around the lower and higher nulling TI values. The signal of the lSIR filter asymptotically approaches negative and positive infinity at the two corresponding nulling T₁s. The lSIR filter increases its slope, and, thus, its contrast amplification, as the change in T₁ from the nulling value of T₁ becomes smaller. This compensates for the decrease in the size of the change in T₁. The low and high signal peaks of the lSIR filter are narrower than those of the corresponding dSIR filter, which results in narrower boundaries between white and gray matter.

Like the dSIR filter, the lSIR filter essentially eliminates the mobile proton density and T₂ dependence of the full IR sequence and so can be used to model the contrast behavior of the full IR sequence.

From a practical point of view, the lSIR T₁-filter has 2–3 times the slope of the dSIR T₁-filter for very small differences or changes in T₁ around the null points, and, thus, has a 20–30 times greater contrast than conventional IR sequences for the same very small differences or changes in T₁. The maximum and minimum values of the lSIR filter are usually shown as ±2–3 (rather than ±infinity). This compares with values of ±1 for the dSIR filter.

The lSIR filter signal, S_lSIR, is the inverse hyperbolic tangent of the dSIR filter signal, S_dSIR, in the mD (S_lSIR = atanh S_dSIR) and the inverse hyperbolic cotangent of the dSIR filter signal in the lD and hD.

2.5. Composite (c) Bipolar Filters (T₁ as well as T₂, T₂*, and/or D*)

It is possible to introduce an attenuation filter S_OF (S_{other filter}) in the form of an additional segment of the IR pulse sequence and apply it to one of the two IR sequences used for dSIR T₁-bipolar filter imaging in Figure 1 and Figure 2. S_OF may equal e^−ΔTE/T2, e^−ΔTE/T2*, or e^−ΔbD* (where ΔTE and Δb are the differences in TE or b [the diffusion sensitivity parameter] between the two IR sequences). This, respectively, introduces T₂, T₂*, and D* dependence. It provides additional contrast to the T₁ contrast of the dSIR T₁-bipolar filter. When ΔTE equals zero, or Δb equals zero, S_OF = 1 and there is no attenuation. If S_OF is set to 1.0, 0.5, and 0.25, the curves shown in Figure 5A,B result when using a narrow mD dSIR sequence (TI_s = 350 ms and TI_i = 500 ms). The nulling times are the same for all values of S_OF.

In Figure 5A, attenuation of the first IR filter with TI_s is shown with S_OF set at 1 (blue), 0.5 (orange), and 0.25 (yellow). The yellow and orange curves are wider than the unattenuated dSIR curve (blue) and are seen around the outside of the first (negative) pole of the blue curve at the first TI_s. At the second positive pole, the yellow and orange curves are narrower than the unattenuated blue curve and are seen inside it. This results in a broadening, loss of contrast, and lower signal white matter around the first (negative) pole, as well as increased contrast and narrower boundaries between white and gray matter around the second (positive) pole.

In (A), as S_OF is decreased from 1.0 to 0.5 and 0.25, the two latter filters become wider around the first (negative) pole and are positioned outside the blue filter. At the second (positive) pole, the attenuated filters are inside the blue filter. They show higher slopes.

(B) shows the conventional T₁-bipolar filter (blue) with S_OF = 1, an attenuated TI_s filter with S_OF = 0.5 (red), and a further attenuated TI_s filter with S_OF = 0.25 (yellow). As S_OF is attenuated, the S_OF = 0.5 and S_OF = 0.25 filters around the first (negative) pole are situated inside the blue curve, become narrower, and have higher slopes. The attenuated filters around the second (positive) pole are situated outside the blue curve. They are wider and show lower slopes. For T₂, a short T₂ increases attenuation relative to a longer T₂, and this corresponds to the narrower and sharper peak, with greater negative contrast and spatial resolution of the contrast in (A).

In Figure 5B, attenuation of the second IR filter with TI_i is shown. As S_OF is decreased to 0.5 and 0.25, the filter narrows around the inside of the blue filter at the first (negative) pole and widens around the outside of the blue filter at the second (positive) pole. The negative contrast produced for the same difference in T₁ is increased around the first pole and decreased around the second pole.

The same principles can be applied to the lSIR filter. Composite (c) forms of the bipolar filters can be designated as cSIR, cdSIR, clSIR, etc.

Susceptibility and magnetization transfer (MT) can also be used to attenuate signals and produce supplementary contrast. The combination of T₁ and T₂* contrast may be of particular value in demonstrating the paramagnetic rim sign and the central vein sign in MS.

3. Methods

With approval from the New Zealand Health and Disability Ethics Committee (EXP 11360, 2022), and informed consent from each subject, MR scans were performed on four adult normal controls and nine adult patients with MS during relapse and remission or with suspected MS (one case). A 3T scanner (General Electric Healthcare Premier) was used. Two-dimensional IR fast spin echo (FSE) sequences were performed with a TIs nulling normal white matter, and a longer TIi was chosen to produce narrow mD dSIR images targeted at small increases in the T₁ of white matter (Figure 2C). Two-dimensional and three-dimensional wide mD dSIR images with a short TI_s and a long TI_l were also obtained. Synthetic narrow mD dSIR images were created from wide mD dSIR images [1], and dSIR images were used to synthetically produce lSIR images. Positionally matched 2D and 3D T₂-wSE and/or T₂-FLAIR images were acquired, as well as filtered susceptibility weighted images (

Table 1. Pulse sequences and pulse sequence parameters.

#	Sequence	TR (ms)	TI (ms)	TE (ms)	Matrix Size Voxel Sizes (mm)	Number of Slices	Slice Thickness (mm)
1	2D FSE IR (for white matter nulling)	9192	350	7	256 × 224	26	4
					0.9 × 0.1
					Z512
					0.4 × 0.4
2	2D FSE IR (used with #1 for narrow mD dSIR)	5796	500	7	256 × 224	26	4
					0.9 × 0.1
					Z512
					0.4 × 0.4
3	2D FSE IR (used with T1 for wide mD dSIR)	5796	800	7	256 × 224	26	4
					0.9 × 0.1
					Z512
					0.4 × 0.4
4	3D BRAVO (for white matter nulling)	2000	400		256 × 256	220	0.8
					0.8 × 0.8
					Z512
5	3D BRAVO (used with #4 for wide mD dSIR)	2000	800		256 × 256	220	0.8
					0.8 × 0.8
					Z512
6	2D T2-FLAIR	6300	1851	102	320 × 240	26	4
					0.7 × 0.7
					Z512
					0.4 × 0.4
7	3D T2-FLAIR without/with fat saturation	6300	1850	102	256 × 256	252	0.8
					0.8 × 0.8
					Z512
					0.6 × 0.6
8	3D susceptibility weighted	40	-	32	300 × 300	110	2
					0.8 × 0.8
					Z512

Z = zipped.

).

4. Illustrative Cases

Figure 6 shows mid-ventricular images of a 41-year-old female patient with MS in remission. No abnormality is seen on the T₂-wSE image (Figure 6A). On the narrow mD dSIR image (Figure 6B), normal peripheral white matter is dark and has a low signal (thick arrows). The more central superior longitudinal fasciculi show a higher signal with a gradient of signal decreasing from posterior to anterior. Within them are the corticospinal tracts (short thin arrows). Three focal lesions are seen in the peripheral white matter and at the leucocortical junctions (long thin arrows).

At a supraventricular level in the same patient (Figure 7), the T₂-wSE image is again normal (Figure 7A). The corresponding narrow mD dSIR image shows only a small region of low signal normal white matter (dark arrow). The corticospinal tracts are seen (short thin white arrows), as well as a lesion (long thin white arrow). There are widespread asymmetric and patchy abnormalities in the white matter.

Figure 8 shows positionally matched supraventricular T₂-FLAIR (Figure 8A) and narrow mD dSIR (Figure 8B) images in a patient with MS during a relapse. A single lesion is seen on the T₂-FLAIR image, and this is also seen on the dSIR image (long white arrows). An additional six lesions are only seen on the dSIR image (short white arrows). Many of the lesions on the dSIR image have high signal boundaries (rims). In addition, there are bilateral symmetrical widespread relatively uniform increases in signal in white matter, with some sparing of more peripheral white matter. These are features of a high grade (4–5) (maximum grade 5) whiteout sign [6]. This differs from the normal dark appearance of white matter seen on the narrow mD dSIR image in Figure 6B. No evidence of a whiteout sign is seen on the T₂-FLAIR image.

Figure 9 and Figure 10 show a 38-year-old patient with MS in remission (Figure 9A and Figure 10A, left columns) and during a relapse two years and five months later (Figure 9B and Figure 10B, right columns) imaged with the same narrow mD dSIR sequence. Figure 9 shows three positionally matched lower levels in the brain and Figure 10 shows two positionally matched higher levels in the brain. White matter shows a generally low signal during remission (whiteout sign grade 1–2) (Figure 9A and Figure 10A), and a high signal (whiteout sign grade 4–5) during the relapse (Figure 9B and Figure 10B). The multiple levels within the brain show the wide distribution of the high grade whiteout sign in the cerebellar and cerebral hemispheres, as well as in the brainstem. No evidence of a whiteout sign was seen on the corresponding T₂-FLAIR images.

Figure 11 shows narrow mD dSIR images in a normal adult control (Figure 11A) and in a 77-year-old patient with MS (Figure 11B) during a relapse. In addition to the whiteout sign, there is a loss of contrast between gray and white matter in the thalamus of the patient (Figure 11B). Also, the heads of the caudate nuclei, as well as the insular and peripheral cortices, appear isointense with CSF on the dSIR image. These are grayout signs. No evidence of a whiteout sign or grayout signs was seen on the corresponding T₂-FLAIR images.

Figure 12 shows a 41-year-old female patient with MS during a relapse. Wide mD dSIR (Figure 12A) and phase filtered susceptibility weighted gradient echo (Figure 12B) images are shown. The dSIR image shows well defined high signal boundaries around two lesions (Figure 12A, white arrows). These lesions show less obvious paramagnetic rim signs on the filtered susceptibility weighted images (Figure 12B, white arrows). The high signal in Figure 11A could be due to increased T₁ in the lesions beyond the second T₁ nulling value and/or reduction in T₁ in long T₁ lesions due to free iron.

Figure 13 shows a leucocortical lesion in a 41-year-old female patient with MS in remission. The lesion appears blurred and relatively featureless on the narrow mD dSIR image (Figure 13A), but shows a disrupted boundary within the lesion on the lSIR image (Figure 13B). Figure 13C shows dSIR (blue) and lSIR (orange) signal profiles across the lesion. These are plots of signal vs. distance in mms on the image. The lSIR profile has higher signal and generally steeper slopes than the dSIR profile, corresponding to the graphs in Figure 4.

Figure 14 shows another leucocortical lesion in a 24-year-old female patient with MS in remission. dSIR (Figure 14A) and cdSIR (Figure 14B) images are compared. The cdSIR image shows higher contrast. In addition, the frontal white matter has a more uniform low signal intensity on the cdSIR image. These are features consistent with the graphs in Figure 5A.

Figure 15 shows T₂-FLAIR (Figure 15A) and wide mD dSIR (Figure 15B) sagittal images of the upper cervical spinal cord in a 43-year-old female patient with MS in remission. The T₂-FLAIR image shows a poorly defined smudge of increased signal (white arrow). In the corresponding position, the dSIR image (Figure 15B) shows a high-contrast lesion with sharply defined boundaries, which is much more extensive than in (Figure 15A) (lower three arrows). An additional lesion is seen in the medulla on the dSIR image (highest arrow), but is not seen on the T₂-FLAIR image.

Figure 16 shows the upper cervical spinal cord in a 42-year-old female patient with MS in remission. T₂-FLAIR (Figure 16A), T₂-wSE (Figure 16B), and wide mD dSIR (Figure 16C) images are compared. A large lesion is seen on all three images, but with greatest contrast on the dSIR image (long white arrows). In addition, three small lesions are only seen on the dSIR image (Figure 16C, short white arrows).

In a patient with optic neuritis and suspected MS, Figure 17A,B (upper row) show axial fat saturated T₂-FLAIR (Figure 17A) and narrow mD dSIR (Figure 17B) images of their optic nerves. Figure 17C,D (lower row) show parasagittal oblique fat saturated T₂-FLAIR (Figure 17C) and narrow mD dSIR (Figure 17D) images of the left optic nerve. The right optic nerve appears normal in Figure 17A,B. The left optic nerve shows a distal abnormality on the T₂-FLAIR axial and parasagittal images (white arrows), but shows proximal and distal abnormalities on the corresponding dSIR images (white arrows).

5. Discussion

Use of bipolar filter sequences such as dSIR, lSIR, and cdSIR demonstrated abnormal appearances in the brain, spinal cord, and optic nerve in MS patients which were not shown, or were only poorly shown, with T₂-wSE and T₂-FLAIR sequences. These included focal lesions, the whiteout sign, grayout signs, high signal boundaries/rims around lesions, disruption of boundaries between white and gray matter, and generalized changes in the optic nerve.

Unipolar filter IR sequences such as intermediate TI IR, STIR, MP-RAGE, and MP2RAGE have been employed in clinical MRI for many years, but bipolar filter sequences (

Table 2. Bipolar filters.

Bipolar Filter	Reverse Bipolar Filter	Tissue Property
SIR	rSIR	T1
dSIR	drSIR	T1
cdSIR	cdrSIR	T1, T2, T2*, D*
lSIR	lrSIR	T1
clSIR	clrSIR	T1, T2, T2*, D*
Synthetic SIR, dSIR, lSIR	rSIR, drSIR, lrSIR	T1
Synthetic cdSIR, clSIR	cdrSIR, clrSIR	T1, T2, T2*, D*

) are a new development [1,4]. Detail of the mathematics describing them is included in

Table 3. Bipolar filters signal equations and other functions.

#	Filter, Other Functions	Signal Equation	Figure #
1	IR, TIs	STIs = 1 − 2e−TIs/T1	Figure 1, Figure 2 and Figure 3
2	IR, TIi	STIi = 1 − 2e−TIi/T1	Figure 1, Figure 2 and Figure 3
3	SIR	SSIR = STIs − STIi	Figure 1
4	dSIR	${\mathrm{S}}_{\mathrm{dSIR}}={\displaystyle \frac{{\mathrm{S}}_{\mathrm{TIs}}-{ \mathrm{S}}_{\mathrm{TIi}}}{{\mathrm{S}}_{\mathrm{TIs}}+{+\mathrm{S}}_{\mathrm{TIi}}}}\left({\displaystyle \frac{\mathrm{difference}}{\mathrm{sum}}}\right)$	Figure 2 and Figure 3
5	cdSIR	${\mathrm{S}}_{\mathrm{dSIR}}={\displaystyle \frac{{\mathrm{S}}_{\mathrm{TIs}}·{\mathrm{S}}_{\mathrm{OF}}-{ \mathrm{S}}_{\mathrm{TIi}}}{{\mathrm{S}}_{\mathrm{TIs}}·{\mathrm{S}}_{\mathrm{OF}}+{\mathrm{S}}_{\mathrm{TIi}}}}$  ${\mathrm{S}}_{\mathrm{dSIR}}={\displaystyle \frac{{\mathrm{S}}_{\mathrm{TIs}}-{ \mathrm{S}}_{\mathrm{TIi}}·{\mathrm{S}}_{\mathrm{OF}}}{{\mathrm{S}}_{\mathrm{TIs}}+{\mathrm{S}}_{\mathrm{TIi}}·{\mathrm{S}}_{\mathrm{OF}}}}$	Figure 5
6	cdSIR, SOF	SOF = ±e−ΔTE/T2, ±e−ΔTE/T2*, ±e−ΔbD*, etc.	Figure 5
7	dSIR, SdSIR	${\mathrm{S}}_{\mathrm{dSIR}} \approx {\displaystyle \frac{\ln 4}{∆\mathrm{T}\mathrm{I}}} {\mathrm{T}}_1-{\displaystyle \frac{\Sigma \mathrm{T}\mathrm{I}}{∆\mathrm{T}\mathrm{I}}}$(in mD)	Figure 2
8	dSIR, T1	${\mathrm{T}}_1\approx {\displaystyle \frac{∆\mathrm{T}\mathrm{I}}{\mathrm{l}\mathrm{n}4}} {\mathrm{S}}_{\mathrm{dSIR}}+{\displaystyle \frac{\Sigma \mathrm{T}\mathrm{I}}{\mathrm{l}\mathrm{n}4}}$(in mD)	Figure 2
9	lSIR	SlSIR = ½(ln STIs − ln STIi)	Figure 4
10	clSIR	${\mathrm{S}}_{1\mathrm{SIR}}{=\mathrm{lnS}}_{\mathrm{Tis}}·{\mathrm{S}}_{\mathrm{OF}}-{\mathrm{lnS}}_{\mathrm{TIi}}$${\mathrm{S}}_{1\mathrm{SIR}}{=\mathrm{lnS}}_{\mathrm{Tis}}-{\mathrm{lnS}}_{\mathrm{TIi}}·{\mathrm{S}}_{\mathrm{OF}}$	Figure 4
11	clSIR, lSIR	${\mathrm{S}}_{\mathrm{c}1\mathrm{SIR}}{=\mathrm{S}}_{\mathrm{c}1\mathrm{SIR}}\pm {\displaystyle \frac{\Delta \mathrm{TE}}{{\mathrm{T}}_2}}$$, \pm {\displaystyle \frac{\Delta \mathrm{TE}}{{{\mathrm{T}}_2}^{*}}}$, ±ΔbD*, etc	Figure 4
12	lSIR, dSIR †	SlSIR = atanh SdSIR	Figure 4

^† The lSIR filter is the inverse hyperbolic tangent of the dSIR filter.

. The equations also provide a basis for implementing synthetic bipolar sequences which can be produced from existing T₁ sequences or a combination of TP maps and synthetic bipolar filters.

UHC MRI with bipolar filter sequences uses IR sequences which already exist on most clinical MR systems and require minimal image processing. They can be implemented in a short time at very low cost and can be used at any clinical static field.

5.1. T₁ Measurements and Magnetization Transfer (MT)

The dSIR sequence is a T₁ map with a linear relationship between S_dSIR and T₁ in the mD (Equation (4)). T₁ measurements obtained with dSIR sequences have been validated in phantom and human studies [7,8]. The observed values of T₁ (T_1obs) are generally shortened by incidental MT effects arising from off-resonance 180° pulses when FSE data collections are used. MT results in a decrease in observed mobile proton density and a proportionate decrease in T_1obs. As a general rule, in diseased tissue compared with normal tissue, MT effects are decreased relative to normal. This results in a relative increase in T₁ in diseased tissue, which is synergistic with increases in T₁ arising in disease from other causes.

5.2. Signs

5.2.1. Whiteout Sign

The whiteout sign consists of a bilateral, symmetrical, widespread, generally uniform increase in signal in the white matter of the cerebral and cerebellar hemispheres, as well as the brainstem. The anterior and posterior central corpus callosum, as well as peripheral white matter of the cerebral hemispheres are usually relatively spared. Whiteout signs can be graded from one (normal) to five (abnormal) [6]. They are reversible and have been seen in mild traumatic brain injury (mTBI), hypoxic injury to the brain, cerebral tumors, long COVID, and MS. Whiteout signs are generally associated with neuroinflammation, although edema, demyelination, and degenerative change may also be implicated. When seen with the dSIR sequence nulled for normal white matter, they are due to widespread small increases in the T₁ of white matter which are insufficient to generate visible contrast with conventional clinical sequences.

5.2.2. Grayout Signs

Grayout signs can be manifest as a reduction or loss of contrast within a gray matter structure such as the thalamus, as well as reduction or loss of contrast between gray matter and white matter or between gray matter and CSF. On narrow mD dSIR sequences, a grayout sign can be caused by an increase in T₁ in gray matter within the hD. Not infrequently, grayout signs are seen with higher grade whiteout signs, and, thus are features of an encephalopathy rather than a leucoencephalopathy (when only white matter is involved).

High signal boundaries and lesions with narrow mD dSIR sequences can arise with an increase in T₁ in the lesion sufficient to cross the peak filter value and enter the hD. This corresponds with the dark rim around MS lesions seen with gray matter nulled double IR sequences, which occurs when there are relatively large increases in T₁. This finding was seen more frequently in MS than in other diseases [9].

The high signal boundary/rim can also arise from a decrease in T₁ in an MS lesion induced by free iron from myelin at the edge of an MS lesion. The decrease in T₁ is from the hD back to the mD and passes through the peak of the bipolar filter.

High signal boundaries around lesions may be less obvious in the presence of a whiteout sign because of the widespread high signal in white matter, although many are apparent in Figure 8B.

5.3. Clinical Issues

5.3.1. Activity

The usual signs of activity in MS are an increase in number and/or size with T₂-weighted imaging in repeat studies, as well as gadolinium based contrast agent (GBCA) enhancement. To this, it may be possible to add the presence of a whiteout sign or increase in size or extent of a whiteout sign.

By using drSIR sequences sensitized to decreases in T₁, it may be possible to increase the sensitivity of conventional IR sequences to GBCA enhancement by an order of magnitude, and, thus, increase the sensitivity of MRI to disease activity [1].

5.3.2. Acute Clinical Episodes with Stable MRI (ACES)

Clinical episodes without changes in MRI appearances were recorded in 26.1% of episodes in a real-world study of MS [10]. Clinical radiological mismatch in MS during relapse has also been described in other papers [11,12]. There is a need for greater MRI sensitivity to correlate better with clinical features. This could be provided by bipolar filter sequences.

Smoldering lesions are recognized by progressive increase in size of abnormalities on T₂-weighted images and paramagnetic rim lesions without GBCA enhancement [13,14,15,16,17], and could benefit from more sensitive sequences. The same applies to the use of T₂-FLAIR imaging and GBCA enhancement to assess therapeutic response. Paramagnetic lesion rims [18,19,20,21] and the central vein sign [22,23,24,25,26] could also benefit from increased sensitivity.

5.3.3. Diagnostic Criteria for MS

The core sequences for recognition of MS lesions are T₂-wSE and/or T₂-FLAIR sequences [27,28,29,30], and thus are based on changes in T₂, not the changes in T₁ seen with dSIR sequences. The sensitivity of dSIR sequences to the presence of MS lesions is supported by the belief that increases in T₂ are accompanied by increases in T₁, and so detection of either may be of similar clinical significance. The initial study of MS with MRI was performed with a T₁-weighted IR sequence.

Conventional diagnostic criteria for MS treat patchy changes due to increased white matter signal seen with T₂-weighted sequences as of no diagnostic significance. These features may be much more obvious with dSIR sequences (Figure 7). In addition, whiteout signs may be seen in a characteristic pattern throughout the white matter of the brain.

5.3.4. Clinical Use

In clinical practice, it is possible to positionally match T₂-FLAIR sequences with dSIR T₁-bipolar filter (BLAIR) sequences. Larger changes in T₂ are shown by T₂-FLAIR sequences and smaller changes in T₁ by T₁-BLAIR sequences, which are, thus, complementary, as seen in Figure 8. cdSIR, T₁, T₂-BLAIR sequences can be used to add T₂-weighting to the T₁-weighting of dSIR sequences.

5.4. Other Sequences Using Two IR Sequences

Two other sequences employing two IR sequences are the nearest comparison to the BLAIR sequences described in this paper.

The first is the MP2RAGE sequence which uses two IR sequences with different TIs [31]. These signals are multiplied together and divided by the sum of their squares. This results in a negative unipolar filter with slope for white matter greater than that of the original IR sequences [32,33].

The second is the FLAWS (fluid and white matter suppression) group of sequences, which also uses two IR sequences with different TIs [32,33] that are subtracted and divided by their sum. Compared with narrow mD dSIR sequences, the ΔTIs of FLAWS high contrast (FLAWShc) and FLAWS high contrast opposite (FLAWShco) are wide and the sequences are monotonic, not bipolar, because they use signed or phase-sensitive signal values, not magnitude values.

The FLAWSmin sequence takes the minimum value of the two IR sequences and produces a positive unipolar filter with a wide ΔTI. It has shown halo signs of increased signal at the margins of MS lesions in gray matter or at the leucocortical boundary [34,35]. It has similarities to the dSIR sequence.

5.5. Further Developments

Further developments include 2D and 3D registration of images with different TIs as part of the basic sequence, but also registration of images before and after GBCA administration and for serial studies at different examinations.

The T₁, T₂*-BLAIR sequence is being developed to produce synergistic contrast from paramagnetic effects due to free iron.

Longer TI values are being used to null blood, and so visualize synergistic T₁ and T₂* effects due to central veins within MS lesions.

Particular variants of sequences designated to specifically see white matter changes in the spinal cord are being assessed together with narrow mD sequences with longer TIs for the thalamus.

It is possible to visualize lesions in the cortex using a longer second TI of 900 ms or 950 ms so that the high signal boundary is between the cortex and CSF not between white and gray matter. Boundaries can also be narrowed by the use of lSIR sequences rather than dSIR sequences.

6. Conclusions

Bipolar filter sequences promise a variety of options which appear well suited to imaging changes in the brain in MS. A program of technical development is in place to expand the range of available sequences and improve their effectiveness in diagnosis. Detailed clinical evaluation will be required. UHC MRI with bipolar filters could provide a substantial advance in the imaging of MS in clinical and research studies with MRI.