Study population
The study population was identical to that investigated in a previous study undertaken by our group [10]. Briefly, it originated from a prospective, cross-sectional study and comprised 121 patients (February 2018 to September 2020) suffering from WHO CNS grade 3 or 4 gliomas, according to the 2016 WHO classification [15]. Patients were referred for hybrid MR/PET imaging for treatment monitoring or suspicion of recurrent tumor. The inclusion criteria were: an ECOG performance score of 0–1, no major depression, no seizures, and fluency in the German language. The local ethics committee approved the protocol, and all patients provided informed written consent according to the Declaration of Helsinki. Patient characteristics are shown in Table 1. The median interval between therapy initiation and imaging was 14 months (mean, 30 months; range, 1–214 months), and 105 patients had
Table 1 Patient Characteristics
Median (Range) unless otherwise stated; ECOG-PS: Eastern Cooperative Oncology Group Performance Score; ISCED: International Standard Classification of Education (1997); GBM: glioblastoma; NOS: not otherwise specified; AA: anaplastic astrocytoma; AOD: anaplastic oligodendroglioma; IDH: Isocitrate-Dehydrogenase; 1p19q codel: 1p19q codeleted; FET: O-(2-[18F]fluoroethyl)-L-tyrosine
completed tri-modality primary treatment, including biopsy/resection, definitive/postoperative radiotherapy, and concomitant/adjuvant chemotherapy with alkylating agents. Most patients had already received one series (100 patients) or two series (14 patients) of local radiotherapy (dose range 59–62 Gy in 92% of patients), with a median interval of 13 months (mean, 32 months; range, 2–213 months) between the start of irradiation and imaging. All patients except one were right-handed.
Assessment of quality of life
Quality of life was assessed at the date of imaging using the validated German version of the EORTC Quality of Life Questionnaire QLQ-C30 and the EORTC QLQ-BN20 (https://qol.eortc.org) [16, 17]. The EORTC QLQ-C30 comprises 30 questions covering five functional domains: physical functioning (Physical), role functioning (Role), cognitive deficits (Cognitive), emotional functioning (Emotional), social functioning (Social), and a global measure of health status (QL). The EORTC QLQ-BN20 is more explicitly adapted to brain cancer and has 20 questions from four domains: future uncertainty (Future), visual disorder (Visual), motor dysfunction (Motor), and communication deficit (Communication). The scores for each domain were computed according to the EORTC QLQ scoring manuals, where the raw scores from 2-5 questions are averaged and rescaled to a range of 0-100 points. Additionally, sociodemographic variables comprising age, gender, and education (International Standard Classification of Education, ISCED, (http://uis.unesco.org/sites/default/files/documents/international-standard-classification-of-education-1997-en_0.pdf) were recorded.
Multimodal imaging, image segmentation, and determination of lesion-specific damage to functional cortical regions and white matter tracts
These procedures have been described in detail before [10]. Briefly, FET positron emission tomography (PET) and MR data were acquired simultaneously using a high-resolution 3T hybrid PET/MR scanner (Siemens Tim-TRIO/BrainPET, Siemens Medical Systems, Erlangen). Structural MR imaging included a 3D T1-weighted magnetization-prepared rapid acquisition gradient-echo (MPRAGE) anatomical scan, a contrast-enhanced T1-weighted image (T1-CE) obtained from a second MPRAGE scan following the injection of gadolinium, or high-resolution T1-weighted, contrast-enhanced MR scans available from the referring institution. Moreover, T2-weighted and T2-weighted fluid-attenuated inversion recovery (FLAIR) structural images were acquired. In addition to these imaging procedures, all patients underwent resting-state fMRI (rs-fMRI), where 300 functional volumes were acquired within 11 minutes using a gradient-echo echo planar imaging (GE-EPI) pulse sequence (36 axial slices, slice thickness 3.1 mm, repetition time TR = 2200 ms, echo time TE = 30 ms, flip angle = 90°, FoV = 200 ×200 mm2, in-plane voxel-size 3.1 ×3.1 mm2). The patients were instructed to relax and to let their minds wander but not to fall asleep [10].
Lesion masks were generated for resection cavities, T1-CE enhancing lesions, T2/FLAIR hyperintensities, and lesions with pathologically increased FET uptake. Resection cavities were manually contoured by a radiation oncologist (M.K.), whereas the T1-CE-enhancing lesions and T2/FLAIR hyperintensities were automatically segmented (deep-learning-based software HD-GLIO-AUTO, https://github.com/NeuroAI-HD/HDGLIO-AUTO). FET PET segmentation was implemented by an FSL (https://fsl.fmrib.ox.ac.uk) custom script using a tumor-to-brain ratio (TBR) of > 1.6 as the threshold for pathological FET uptake [18]. All segmentations were visually inspected and manually corrected if needed. For lesion-function mapping, the rs-fMRI-based cortical parcellation atlas of Schaefer et al. [19], comprising 2x50 functional nodes belonging to 7 resting-state networks (visual, somato-motor, dorsal attention, ventral attention, limbic, fronto-parietal control, and default mode), and an atlas of white matter tracts (2 × 24 tracts) from the Stereotaxic White Matter Atlas of the Johns Hopkins University (JHU) [20] were applied. All images were spatially normalized by elastic registration to the MNI-152 brain template by means of the SPM12 toolbox (www.fil.ion.ucl.ac.uk/spm/software/spm12) and, finally, the partial overlapping volume of the nodes or tracts with each of the lesion segments was computed, see Fig. 1.
Determination of resting-state functional connectivity
The determination of resting-state functional connectivity (RSFC) was performed similarly to Kocher et al. [21]. Briefly, functional images were subjected to the standard pre-processing steps of the SPM12/CONN toolbox comprising motion correction with removal of outliers and cerebro-spinal fluid/ white matter signals, slice timing correction, smoothing with 5 mm FWHM, bandpass-filtering to 0.008–0.09 Hz and denoising. All structural and functional images were non-rigidly co-registered to the MNI-152 standard brain template using the SPM12/CONN unified segmentation/registration algorithm. For the determination of RSFC, the cortical parcellation of Schaefer et al. [19] was imported into the CONN toolbox and used to compute full connectivity matrices from the z-transformed Pearson correlation coefficients between the respective nodes. From these z-values, the within-network connectivity for each node of the respective network [22] was computed.
Statistical Analysis
SPSS (IBM SPSS Statistics version 27, IBM Corporation, Armonk, NY, USA) and the Python software package scipy.stats (version 1.5.4, https://scipy.org) were used for all analyses. Correlation coefficients between EORTC QLQ domain scores and the volumetric overlap between lesions and functional nodes or white matter tracts were computed using a 2-sided Kendall tau-b rank correlation that corrects for ties and is less sensitive to outliers. Moreover, the Pearson correlation coefficients between within-network functional connectivity of individual nodes and the EORTEC QLQ domain scores were calculated. In general, p-values <0.05 were regarded significant. No formal correction was applied for massive univariate testing [23]. However, only correlations with p<0.001 for lesion-QoL and p<0.01 for RSFC-QoL analysis were considered significant.