No significant difference found in PET/MRI CBF values reconstructed with CT-atlas-based and ZTE MR attenuation correction

A new hybrid scanner combining positron emission tomography (PET) and MRI, the PET/MRI system, should be very beneficial in the neuropsychiatric field, especially for the precise fusion of high-grade anatomical MR images and PET molecular imaging [1]. The quantitative PET images are also anticipated, although several issues remain to be confirmed for global as well as regional values. MR-based attenuation correction (MRAC) is one of the major concerns in quantitative brain PET images because the brain is covered by the skull. MR segmentation images using the Dixon method are usually applied for the reconstruction of body PET images. However, for brain PET images, MRAC using the Dixon-based method showed greater bias and underestimation of PET values compared with CT-based attenuation correction methods because of inaccurate estimation of the bone attenuation [2, 3]. The single- and multi-CT atlas-based (AB) attenuation correction methods were also compared, and multi-AB-MRAC showed PET quantitative values very close to those from individual CT attenuation correction (CTAC) [4]. The ultrashort echo time (UTE) and zero echo time (ZTE) pulse sequences have been applied to brain PET image reconstruction for bone identification in the MRAC method, and biases of quantitative values were reported to be reduced [5‐7]. A multicenter study showed that these newer MRAC methods produced better and more reliable quantitative images than those from the simple Dixon-based method [8]. These recent studies suggested that the MRAC method has been improved in terms of quantitative PET evaluation using PET/MRI scanners.

PET measurement of cerebral blood flow (CBF) using H₂¹⁵O PET usually requires arterial blood sampling from individual patients to estimate the input function of the tracer. Recently, several methods to obtain the image-derived arterial input function (IDIF) have been proposed by using time-of-flight (TOF) MR angiography images to detect the location of the internal carotid arteries (ICA) at the skull base [9‐12]. The method is ideal for the PET/MRI scanner because both image data can be acquired simultaneously. The IDIF method in H₂¹⁵O PET/MRI is useful and practical to estimate the arterial input function directly from the cerebral arteries [13], and the CBF values using IDIF were consistent with previous ¹⁵O-PET studies [14, 15]. However, the brain counts in the skullbase were reported to be underestimated in previous studies [4, 6], while a recent study suggested that both AB- and ZTE-MRAC tended to overaetimate PET counts slightly in the whole brain even in the skullbase [16]. The PET counts at the skullbase may influence on the quantitative values in the IDIF method, because it usually uses arterial counts at the ICA in the skullbase area.

The present study investigated the differences of IDIF curves at the ICA area using the two different MRAC methods as well as their influence on CBF values in the whole brain. Since the ZTE approach is reported to show closer PET counts to ordinary CTAC, AB-MRAC may provide greater bias in brain PET images and IDIF counts [16].

Figure 2 shows MRAC maps and ZTE images for the PET attenuation correction. Although the ZTE-MRAC (Fig. 2c) shows similar head AC maps to AB-MRAC (Fig. 2a), slight differences can be observed in the frontal sinus region and thickness of skull bone, especially in the frontal part.

Figure 3 shows the average differences in arterial radioactivity at each point of the dynamic PET frame. The differences were less than 5%, although the variances (SD) were as large as ± 10% (Fig. 3a). Figure 3b shows a representative IDIF curve from a patient with error bars calculated from the average differences of each time point. Estimation errors were small, especially in the phase of lower radioactivity, indicating that only small differences may be expected in the area under the curve of IDIF from the AB- and ZTE-MRAC PET images. As the result of this small difference in IDIF curves, the CBF images calculated from AB-MRAC reconstruction and those from ZTE-MRAC were very close, and the regional values were almost the same (Fig. 4). Table 1 shows the regional average values at baseline and after the acetazolamide challenge test. The regional values averaged for the bilateral hemispheres showed no significant differences between the two MRAC methods. A scatter plot of the regional CBF values from the 310 ROIs used for CBF measurement showed a good correlation. Figure 5 shows two representative cases of regression analysis; one of them had excellent correlations for both at baseline and after acetazolamide administration (Fig. 5a, R² = 1.0). Even in the second case, which had greater variance, the correlation coefficient was excellent (Fig. 5b, R² = 0.96). The means of slopes and correlation coefficients for all PET scans were 1.02 ± 0.08 and 0.97 ± 0.03, respectively. The Bland-Altman plot for the latter case showed no significant bias, although the CBF image with AB-MRAC tended to show greater values in the higher mean CBF range (mean difference 3.0 ± 3.8; range − 4.4–10.4 mL/min/100 g) (Fig. 5c).

Figure 6 shows the mean percentage differences in CBF images calculated from individual CBF by 2 × [AB-MRAC − ZTE-MRAC] / [AB-MRAC + ZTE-MRAC]. Red color indicates higher values and blue indicates lower values for the AB-MRAC method. The parietal to occipital region shows higher CBF values for the AB-MRAC method compared with those of the ZTE-MRAC method, although the difference was less than 5%. Other brain regions showed no differences and the quantitative CBF values were very close.

The present study evaluated the effects of a new MRAC method using ZTE images on the quantitative PET and CBF values calculated from brain H₂¹⁵O PET/MRI. We compared arterial input curves from the IDIF method and regional CBF values calculated separately from two different dynamic PET images of AB- and ZTE-MRAC reconstruction, and found no significant differences between the two methods. The TACs from the IDIF showed mean differences less than 5% at each time point, and did not show significant differences in the areas under the curves of the TACs. The 3-min H₂¹⁵O-PET average images as well as quantitative CBF images from the AB-MRAC reconstruction method tended to show greater regional values compared with the ZTE-MRAC reconstruction method (Fig. 6). However, the values at the skull base, where the IDIF-TAC was obtained, were similar. Previous FDG-PET studies also showed a tendency of overestimation of AB-MRAC images in the parietal region compared with images from CTAC [6, 21]. Thus, the ZTE-MRAC method seems to be appropriate for the quantitative evaluation of brain PET. Our previous study showed that overall CBF values from H₂¹⁵O PET/MRI using the IDIF method were reasonable in healthy volunteers [13], and were consistent with a previous multicenter study in Japan using ¹⁵O-PET for healthy volunteers [14].

There were no significant differences in the regional CBF values from each brain area obtained from patients with cerebrovascular disease in the present study. Although recent FDG-PET/MRI studies showed little regional differences in PET counts when using several new MRAC methods [5, 21], about 5% regional difference at maximum may affect results of statistical mapping analysis if we use normal database obtained from different calculation methods of PET image. To apply brain PET/MRI images for statistical mapping analysis, the same method of PET/MRI reconstruction should be used for both control as well as patient images.

Several previous studies reported that the AB-MRAC method provided reasonable PET values, although AB-MRAC tended to overestimate in the parietal regions, and the ZTE-MRAC method showed a tendency of slight underestimation for the whole brain [4, 6, 21]. These results, comparing FDG accumulation in the brain, are consistent with the present study. If we had compared our CBF data with images from other AC methods such as CTAC, similar results may have been observed because the ZTE-MRAC method would provide quantitative images very close to the CTAC PET images [16]. One of the reasons for the higher CBF values in the parietal region with the AB-MRAC method may have been caused by the thickness of the skull CT atlas prepared by the vendor. The frontotemporal part of the upper skull showed a thicker bone map on AB-MRAC images compared with ZTE-MRAC (Fig. 2, right column in axial view). This small difference may have induced the overestimation in AB-MRAC images.

The TOF technique for PET reconstruction using PET/MRI seems to be important, especially for quantitative images. The frontal part of the skull sometimes shows substantial differences between AB-MRAC and CTAC due to the frontal sinus. If the difference between MRAC and CTAC strictly affects the PET image calculations, brain cortices close to the sinus would show significant differences in regional radioactivity concentration, as well as on quantitative images. However, even in the cases with a relatively large frontal sinus as well as temporal bone defects after neurosurgical operation, the CBF values close to the bone defect were not significantly different (Fig. 7). This result indicates that the TOF acquisition technique would substantially compensate for the errors in MRAC [6, 24]. TOF reconstruction is reported to be less sensitive to errors in attenuation correction, and is beneficial in multimodal systems such as PET/MRI [25].

A limitation of the present study was that we evaluated only a small number of patients with cerebrovascular disease who had unilateral occlusive changes in the major cerebral arteries. Since each IDIF was obtained from an unoccluded part of the internal carotid artery, and all patients had excellent input curves very close to those of healthy volunteers [13], the IDIF curves were considered to be appropriate and did not affect the CBF calculation. Furthermore, the purpose of the study was to compare the difference in quantitative results obtained from two different MRAC methods. Although the patient number was limited, 25 PET scans were compared in total. Thus, we consider that the results of the present study are reliable.

The CBF images calculated from AB- and ZTE-MRAC reconstruction showed no significant differences in quantitative regional values, although the parietal region tended to show greater values on AB-MRAC images compared with ZTE-MRAC. Arterial radioactivity at the skull base was very similar between them, and almost the same IDIFs were obtained, with small errors, which provided similar CBF values for the brain. Since the commercially available single-CT-atlas-based method transforms the CT-atlas into AB-MRAC using individual Dixon-MRI images, it may induce slight bias on MRAC maps as well as regional PET counts [4, 6, 16]. The ZTE-MRAC reconstruction method, estimating bone attenuation from individual ZTE-MRI images, is expected to be used as a standard method for quantitative brain PET studies.