**4. Results**

132 Advances in the Biology, Imaging and Therapies for Glioblastoma

Fig. 5. Modeling of the signal-to-noise ratio using the above equations for: ab) the TrueFISP sequence, c) the gradient-echo sequence, and d) the spin-echo sequence. The signal was estimated as a function of the flip angle and TR for a), b), and c). For the spin-echo sequence, the signal was estimated as a function of the TEs and TRs of the sequence. The T1 and T2 values for the brain (1,295 ms and 53 ms, respectively) were used in a), c), and d). The T1 and T2 values for the tumor (1,525 ms and 72 ms, respectively) were used in b). The same color scale was used in the 4 simulations. To allow for the 2-dimensional nature of the spin-

The spin-echo sequence should provide a high S/N, provided the TR is long and TE very short. However, these acquisition conditions do not provide sufficient contrast, as shown in

We can thus clearly see the advantage of the TrueFISP sequence, as far as the S/N is concerned, for an equivalent total acquisition time, compared with both a standard 3D

The contrast was then assessed theoretically, based on the signal difference between the tumor and the healthy brain. For the TrueFISP sequence (Fig. 6a), optimal contrast was

The gradient-echo sequence provided almost no contrast compared with the TrueFISP sequence, while the spin-echo sequence provided improved contrast for TRs over 3,500 ms

echo sequence, a correction factor was applied.

gradient-echo sequence and a 2D RARE spin-echo sequence.

achieved with a short TE and a flip angle between 25° and 50°.

the following figure.

and TEs greater than 50 ms.

**3.4 Contrast** 

#### **4.1 Correction of TrueFISP banding artifacts**

TrueFISP images were acquired at 4.7 T using a cross-polarized antenna. A U87 glioma was implanted in a mouse, and imaging was performed 4 days later.

Fig. 7.a shows the image acquired using the TrueFISP sequence with a TE of 2.5 ms, a flip angle of 35°, and a Δφ value of 180°. As expected, black-signal bands are clearly visible in the image. These hinder visualization of parts of the animal's brain. Modifying the Δφ value (to 0°, 90°, and 270°) shifts the positions of the null-signal bands. For some of these values (Δφ = 270°), the tumor is clearly visible, while for others (Δφ = 90°), the null-signal zone covers the area containing the tumor, making all subsequent analyses of the tumor impossible.

3D TrueFISP MRI Provides Accurate Longitudinal Measurements of Glioma Volumes in Mice 135

As shown in the simulations, for a TR of 5 ms, the S/N should be optimal with a flip angle between 15° and 35°. On the other hand, contrast between tumor and brain should be optimal with a flip angle between 25° and 55°. For a fixed TE, TrueFISP images were acquired with a flip angle between 8° and 35°. Above this range, the S/N is reduced and, more importantly, signals from sub-cutaneous fat and fluids (such as cerebrospinal fluid) become very intense. This leads to a strong reduction in signal dynamics for the brain and

Fig. 8. Images obtained using the 3D TrueFISP sequence after correction of banding artifacts

The best contrast between brain and tumor was achieved at 35°. For α = 25 and 30°, good contrast was maintained, providing easy tumor detection, and the S/N was increased. Below this, the signal for healthy brain was increased and the tumor became difficult to distinguish. An 8° angle gave rise to more TrueFISP artifacts in different positions. At this value, reconstruction could not generate an image without null-signal bands, and the tumor

The table below gives the S/N and contrast between the brain and tumor for the various flip angles. To summarize, flip angles between 25 and 35° allow tumors to be segmented semi-

according to the α flip angle. TE was set to 2.5 ms.

became indistinguishable from the surrounding brain.

automatically based on tissue contrast.

**4.2 Influence of flip angle on S/N and contrast on TrueFISP imaging** 

tumor, and generates images which are more difficult to interpret.

Fig. 7. Brain images from a mouse with an implanted tumor on day 4 after implantation. Images were obtained using the 3D TrueFISP sequence at 4.7 T and Δφ = 180° (a), 0° (b), 90° (c), 270° (d). Null-signal bands are clearly visible and shift in position. Image e was obtained by SOS reconstruction from a) and b). The arrow indicates a non-compensated low-signalintensity band. Image f was obtained by reconstruction from a), b), and c). The arrow indicates a high-signal-intensity band. Finally, Image g was obtained from the 4 images a), b), c), and d). The arrow shows the tumor, with a volume of less than 2 µL.

To eliminate these null-signal bands, the images acquired in a), b), c), and d) can be combined. To do so, various methods have been described in the literature, such as the sum of k-spaces, the sum of magnitude images, or the square root of the sum of squares of magnitude images. The latter method will be used here as it has been shown to be the most effective [14].

As shown in Images a-b, the black-signal bands do not appear to overlap. These two images were therefore combined to produce Image e. This eliminated the very pronounced nullsignal bands, and the tumor became clearly visible. Nevertheless, the signal was not homogeneous throughout the brain, and a band (indicated by the white arrow in Image e) shows a reduced signal intensity compared with other areas of the brain. This type of artifact appears in various positions for all types of combinations of 2 images (ac, ad, bc, etc.).

This low-signal-intensity band could be eliminated by adding a third image to the combination (a, b, and c). However, a band of slightly higher signal intensity (white arrow in Image f) appeared. The position of this band corresponds to the black-signal band in Image d.

Finally, by adding Image d to the final combination, Image g was obtained (g = (a2 + b2 + c2 + d2)1/2. In this case, the brain appears perfectly homogeneous and the tumor is clearly visible.

### **4.2 Influence of flip angle on S/N and contrast on TrueFISP imaging**

As shown in the simulations, for a TR of 5 ms, the S/N should be optimal with a flip angle between 15° and 35°. On the other hand, contrast between tumor and brain should be optimal with a flip angle between 25° and 55°. For a fixed TE, TrueFISP images were acquired with a flip angle between 8° and 35°. Above this range, the S/N is reduced and, more importantly, signals from sub-cutaneous fat and fluids (such as cerebrospinal fluid) become very intense. This leads to a strong reduction in signal dynamics for the brain and tumor, and generates images which are more difficult to interpret.

134 Advances in the Biology, Imaging and Therapies for Glioblastoma

Fig. 7. Brain images from a mouse with an implanted tumor on day 4 after implantation. Images were obtained using the 3D TrueFISP sequence at 4.7 T and Δφ = 180° (a), 0° (b), 90° (c), 270° (d). Null-signal bands are clearly visible and shift in position. Image e was obtained by SOS reconstruction from a) and b). The arrow indicates a non-compensated low-signalintensity band. Image f was obtained by reconstruction from a), b), and c). The arrow indicates a high-signal-intensity band. Finally, Image g was obtained from the 4 images a),

To eliminate these null-signal bands, the images acquired in a), b), c), and d) can be combined. To do so, various methods have been described in the literature, such as the sum of k-spaces, the sum of magnitude images, or the square root of the sum of squares of magnitude images. The latter method will be used here as it has been shown to be the most

As shown in Images a-b, the black-signal bands do not appear to overlap. These two images were therefore combined to produce Image e. This eliminated the very pronounced nullsignal bands, and the tumor became clearly visible. Nevertheless, the signal was not homogeneous throughout the brain, and a band (indicated by the white arrow in Image e) shows a reduced signal intensity compared with other areas of the brain. This type of artifact appears in various positions for all types of combinations of 2 images (ac, ad, bc,

This low-signal-intensity band could be eliminated by adding a third image to the combination (a, b, and c). However, a band of slightly higher signal intensity (white arrow in Image f) appeared. The position of this band corresponds to the black-signal band in

Finally, by adding Image d to the final combination, Image g was obtained (g = (a2 + b2 + c2 + d2)1/2. In this case, the brain appears perfectly homogeneous and the tumor is clearly

b), c), and d). The arrow shows the tumor, with a volume of less than 2 µL.

effective [14].

etc.).

Image d.

visible.

Fig. 8. Images obtained using the 3D TrueFISP sequence after correction of banding artifacts according to the α flip angle. TE was set to 2.5 ms.

The best contrast between brain and tumor was achieved at 35°. For α = 25 and 30°, good contrast was maintained, providing easy tumor detection, and the S/N was increased. Below this, the signal for healthy brain was increased and the tumor became difficult to distinguish. An 8° angle gave rise to more TrueFISP artifacts in different positions. At this value, reconstruction could not generate an image without null-signal bands, and the tumor became indistinguishable from the surrounding brain.

The table below gives the S/N and contrast between the brain and tumor for the various flip angles. To summarize, flip angles between 25 and 35° allow tumors to be segmented semiautomatically based on tissue contrast.

3D TrueFISP MRI Provides Accurate Longitudinal Measurements of Glioma Volumes in Mice 137

The reference method for tumor detection by anatomic imaging is T2-weighted spin-echo imaging. Images were therefore acquired with a T2-weighted 2D RARE sequence. This sequence detects the tumor despite its small size (see Fig. 10d). In addition, as expected from simulations, contrast between the tumor and healthy brain is higher than with the TrueFISP sequence (Fig. 10a). The total acquisition time for the 2D RARE sequence was comparable with that for the 3D TrueFISP sequence. However, as the images were acquired in 2D by multi-slice imaging, reconstruction in other dimensions generates images with poor resolution and low

The two imaging methods were next used to measure tumor volume. With the TrueFISP sequence, the volume was estimated at 2.4 µL, while with the RARE sequence it was estimated at 4.3 µL. This overestimation of the volume with the RARE sequence is due to the low spatial resolution of this sequence in the third dimension. Actually, significant partialvolume effects are generated by this sequence and lead to volume-measurement errors. A 3D RARE sequence validated the 2.4 µL volume determined using the TrueFISP sequence.

Fig. 10. 3D TrueFISP (a, b, and c) and multi-slice 2D RARE imaging (d, e, and f) of a mouse with implanted glioma: a) & d) sagittal sections; b) & e) reconstruction of transverse sections; c) & f) reconstruction of coronal sections. The white arrows indicate the tumor.

Volume overestimation based on 2D imaging was confirmed by using the two sequences to image a phantom consisting of a known volume of water. With the 2D RARE sequence, the error in volume measurement was around 20%, while with the TrueFISP sequence the error

**4.4 Comparison between RARE and TrueFISP imaging** 

The 3D RARE image was acquired over 45 minutes.

was only 4%.

quality (Figs. 10ef) compared with the 3D TrueFISP sequence (Figs. 10bc).


Table 1. S/N and contrast for brain and tumor, as a function of the flip angle, α. TE was set to 2.5 ms.

#### **4.3 Influence of TE on S/N and contrast in TrueFISP imaging**

Using a low bandwidth generates images with a high S/N. However, a decrease in bandwidth causes an increase in the sequence's TE. The influence of this was measured over the range between 1.5 and 10 ms (with the flip angle set at 35°). With a low TE (1.5 ms), the contrast appeared better than in the reference image acquired at 2.5 ms, but the S/N was significantly reduced, and the signal from fatty tissue was increased due to its relatively low T1. This resulted in a lower-quality image. Interestingly, at higher TEs (5 and 10 ms), the TrueFISP sequence generated few susceptibility artifacts compared with a FLASH sequence with the same TE (data not shown). However, at TE = 10 ms, the tumor was no longer visible due to a strongly reduced contrast between brain and tumor.

As described in the table below, only a TE value of 2.5 ms provides a contrast greater than 10, allowing the tumor to be segmented semi-automatically.

Fig. 9. Images acquired using the 3D TrueFISP sequence after correction of banding artifacts based on TE. The flip angle was set to 35°.


Table 2. S/N and contrast for brain and tumor, as a function of the echo time. The flip angle α was set at 35°.
