The experimental framework was presented in Fig.1. This study was composed of three steps: A) Image process pipeline, which introduced the two quantitative approach, modified IDIF and SUVR process, B) IDIF validation, which introduced the CMRGlc’s validity in healthy controls, C) IDIF application, which introduced the diagnostic assessment and postoperative measurement in ischemic cerebrovascular disease patients.
Participants
The data were collected during a hybrid TOF PET/MR system (Signa, GE Healthcare) imaging study conducted from June 2017 through January 2021 in the Department of Radiology and Nuclear Medicine at Xuanwu Hospital Capital Medical University, which included 26 ischemic cerebrovascular disease patients (22 male and 4 female, 50.2±8.6 y) with preoperative and postoperative imaging examination. The detailed diagnosis-specific exclusion and inclusion criteria included: 1) patients were confirmed diagnosis of ICA or MCA occlusive disease based on digital subtraction angiography; 2) evidence of cerebral hypoperfusion on CT or MR perfusion imaging consistent with affected side; 3) patients had a history of transient ischemic attacks or complete stroke involving the relevant ICA or MCA territory and treatment with ineffective medication[18]; 4) all patients completed PET-MR imaging within 1 month before STA-MCA bypass surgery and had confirmed vascular connection success based on digital subtraction angiography after surgery. Patients were excluded if 1) had an acute stroke (less than one month); 2) ICA or MCA occlusion presented bilaterally; 3) with any contraindication for MRI and artefacts on MRI [19]. The severity of cerebrovascular diseases was measured using the National Institutes of Health Stroke Scale (NIHSS) and the Modified Rankin Scale (mRS). An imaging control group included 16 healthy controls (HCs, 5 male and 11 female, 46.8±10.6 y) were recruited from the community. HCs were chosen based on negative screening for neurological disorders, structural MRI images and MR angiography (MRA) image [20].
This study has been approved by the institutional review board of Xuanwu Hospital and conducted in accordance with the Declaration of Helsinki, and written informed consent was obtained from all subjects prior to the examinations.
PET/MR acquisition protocol
PET and MR sequence images were simultaneously acquired as shown in Supplementary Fig.1. Dynamic PET and MR images of all subjects were acquired on a hybrid PET/MR system (Signa, GE Healthcare). Each subject was instructed to fast for at least 6 h to reach a serum glucose level lower than 8 mmol/L. Before scanning, each subject was measured the glucose concentration (mmol/L) in blood. All imaging sessions were acquired in the resting state, without performing any task. A 19-channel head and neck union coil were used so that undertake a high signal-to-noise ratio of the PET/MR imaging. All subjects were placed in a supine position to ensure in the center of the field of view, and instructed to remain calm with their eyes closed.
Multiple MRI sequences included: a T1-weigthed MRI sequence (voxel size, 1×1×1 mm; echo time (TE), 3.2 ms; repetition time (TR), 8.5 ms; matrix, 256×256; 178 slices) for the anatomic localization, MR angiography (MRA) sequence (voxel size, 0.43×0.43×0.7 mm; TE, 3.7 ms; TR, 25 ms; matrix, 512×512; 272 slices) for the segmentation of the internal carotid vasculature, and a T2 fluid-attenuated inversion recovery (FLAIR) sequence (voxel size, 0.47×0.47×4 mm; TE, 144 ms; TR, 11000 ms; matrix, 512×512; 32 slices) for the definition of infarction area. At the same time, PET list-model acquisition was initiated with manual intravenous injection of 18F-FDG (3.7 MBq/kg) and a 70-min dynamic PET image was acquired.
After scanning, the PET list mode data were re-binned into a dynamic frame sequence with 31 frames (10s×9, 30s×3, 60s×4, 180s×6, 300s×9; voxel size, 1.17×1.17×2.78 mm; matrix, 256×256; 89 slices) [12, 17, 21]. Corrected PET data were reconstructed using a time-of-flight, point spread function, ordered subset expectation maximization (TOF+PSF+OSEM) with 8 iterations and 28 subsets, and a 3-mm cut-off filter. Meanwhile, a single static PET frame was also reconstructed (50-60 min, voxel size: 1.82×1.82×2.78 mm; matrix, 192×192; 89 slices). All PET emission data were corrected for attenuation, scatter, random, decay, and deadtime. Attenuation correction was performed based on MR images, and the default attenuation correction sequence (Dixon MR sequences) was automatically prescribed and acquired as follows: LAVA-Flex (GE Healthcare) axial acquisition, voxel size = 1.95×2.93×5.2 mm, TE, 1.7 ms, TR, 4 ms, and 120 slices. Corrected PET data were reconstructed using TOF+PSF+OSEM algorithm with 8 iterations and 32 subsets, and a 3-mm cut-off filter.
IDIF method
Dynamic FDG-PET scans were analyzed using a fully automated processing pipeline to obtain an IDIF and further support the noninvasive absolute quantification of CMRGlc as described previously [16, 17]. The processing pipeline was denoted in Fig.1A. Firstly, individual TOF-MRA image was used to segment ICA by the automated carotid arteries segment method [16]. The petrous segment of ICA was regarded as VOI to extract IDIF. Of note, the unilateral ICA mask was acquired for ischemic cerebrovascular disease patients because of their progressive ICA steno-occlusive changes. The whole carotid vasculature (CV) was identified using histogram-based quantile thresholding (0.995) and automatic seeded region growing with a connectedness constraint. According to ICA’s specific morphology, we employed morphological feature vector (Gz) to characterize the shape of the vascular tree, according to the mathematical equation:
Gz = Nz × Mz × Rz (1)
where normalized mean intensity (Nz), major axis length (Mz) and ratio of major to minor axis length (Rz) were obtained for CV slices with each cerebral hemisphere. The feature curve with elliptical structures presented as prominent peaks. The structure with the global max peak was regarded as the petrous segment. For patients with steno-occlusive ICA, the corresponding Gz peak of infraction side would lower than other structures. Along the slices from caudal to cranial, the ICA with lower Gz peak compared with last peak (circle of Willis) was regarded as steno-occlusive ICA and further removed from ICA mask. The above steps resulted in unilateral ICA mask for ischemic cerebrovascular disease patients, but bilateral ICA mask for HCs.
The individual T1-MRI sequence was regarded as the reference volume and all subsequent MRI scans were rigidly co-registered to native space (SPM 12, Wellcome Trust Center for Neuroimaging, UCL). All dynamic PET images were corrected for motion by rigid co-registration between each frame with the individual T1-MRI image. Subsequently, a modified Mueller-Gaertner method with spill-out and spill-in corrections was employed for partial volume effect (PVE) correction [16, 22]. The IDIF was derived and interpolated with a step length of 1. The voxel-wise Patlak graphical analysis was employed to generate the absolute CMRGlc map by the time-activity curve (IDIF) derived from corrected PET frames [23, 24]. For the assessment of the potential of noninvasive quantification modeling, the SUVR map was obtained by count normalizing each voxel’s intensity to the mean activity concentration in whole cerebellum (Supplementary methods).
The supratentorial cerebral arterial territories are of key clinical importance for evaluation of cerebrovascular diseases and may help in the assessment of actual territorial contribution of individual collateral arteries in ischemic cerebrovascular disease patients [25, 4]. We employed this vascular territory to compare the performance of two quantitative methods. Vascular territory mapping including left and right anterior, middle, and posterior cerebral artery (ACA, MCA, and PCA), basilar artery (BA), and cerebellar artery (CA) territory in the standard space (Supplementary Fig.2) were manually drawn by two radiologists according to the maps described previously [26, 27]. All territories were warped to individual native space using deformation parameters. The average ipsilateral (surgery side, and excluding the infraction region) metabolic values of VOIs (gray matter, white matter, and 5 territories) were extracted from CMRGlc and SUVR maps for further analysis. The infraction area of each patient was available from T2-FLAIR sequence by two experienced radiologists. Each radiologist independently reviewed the infarction areas on FLAIR images in blind model. The resulting infarction volume was available by taking the average of two individual reader infarction areas.
IDIF validation and application
Consider the steno-occlusive ICA for ischemic cerebrovascular disease patients, the IDIF method used only unilateral ICA mask as blood-pool region to extract IDIF curve. To ensure that the quantification results were not influenced by unilateral ICA, we hypothesized that the quantification results derived from bilateral ICA should be similar to those of unilateral ICA. To test it, we compared quantitative CMRGlc results of bilateral ICA with those of unilateral ICA by characterizing the within-subject concordance in healthy controls. We used the intraclass correlation coefficient (ICC) and absolute percentage error to measure consistency between unilateral and bilateral CMRGlc values. We quantified and compared the CMRGlc values within 7 VOIs from healthy controls subjects, respectively.
To assess whether the IDIF quantification could accurately track the pathophysiological changes and the postoperative recovery effect, we calculated and compared the quantitative values derived from CMRGlc and SUVR maps for ischemic cerebrovascular disease pre- and post- operative visits using univariate ANOVAs with post-hoc tests. We evaluated which quantitative value was sensitive to metabolic abnormality and pathophysiological changes in ischemic cerebrovascular disease progression. We applied metabolic values of vascular territories into diagnostic assessment to classify the healthy controls and pre- and post- operative ischemic cerebrovascular disease patients via receiver operating characteristic curve (ROC). The relative change index, (postoperative value / preoperative value-1)*100%, was employed to measure the metabolic changes after surgery. Furthermore, to illustrate which quantitative value could track the degree of ischemic cerebrovascular severity, we used the general linear model to evaluate the association between clinical assessments (NIHSS and mRs scores) and quantitative values of surgery side cerebral hemispheres.
Statistical analyses
All average data were reported as mean ± standard deviation. Before statistical test, all variances were statistically analyses using normality tests. Metabolic difference across subjects were assessed separately using pared t-test, 1-way ANOVA and post hoc Dunnett tests. ROC and area under the curve (AUC) were employed to assess the diagnostic ability of quantitative values, and DeLong test was sued to measure the diagnostic differences [28]. All statistical analyses were conducted using SPSS software, version 24 (IBM). All measurements were considered significant at the P<0.05 level, two-tailed.