Impact of attenuation correction on image-derived input functions and cerebral blood flow quantification with simultaneous [15O]-water PET/MRI

173 Objectives: Considerable technical advances have improved attenuation correction (AC) for simultaneous PET/MRI in the brain1, but few studies have evaluated the impact of AC on PET kinetic modeling and parameter quantification. In particular, image-derived input function (IDIF) methods can reduce the invasiveness of quantitative PET studies by avoiding arterial blood sampling2. However, brain IDIFs typically derive from large arteries near the skull base that are highly susceptible to attenuation effects of bone. This simultaneous PET/MRI study evaluates the effect of AC on IDIF, kinetic modeling of cerebral blood flow (CBF) values, and measurement reproducibility of [15O]-water. Methods: Simultaneous time-of-flight 3 Tesla PET/MRI (GE Healthcare Signa) was acquired in 8 healthy volunteers (22-66 yrs, 6 female). PET imaging of CBF was performed using [15O]-water (833±148 MBq) before and after injection of Diamox, a vasodilator that augments CBF. Four subjects received successive baseline scans to evaluate scan-rescan reproducibility. Dynamic PET frames were reconstructed with three AC methods: (1) Atlas-based AC3 using a two-point Dixon MRI; (2) Zero echo time (ZTE)-AC4 which segments tissue on a ZTE MRI (nominal echo time of 0ms) and assigns continuous attenuation values to bone; (3) Deep learning (DL) network that takes a single input (ZTE MRI) and outputs a pseudo-CT image that was trained on real CT scans in a separate group of 50 patients5. The DL-AC method has shown quantitative accuracy to CT at multiple sites5 and served as the reference standard in this study. IDIFs were created for each AC method from the cervical arteries, correcting for spill-over effects with the true arterial volume (segmented on MR angiogram)2. Dynamic PET data were spatially normalized in PMOD3.5 and time activity curves were extracted from pre-defined volumes of interest6. Kinetic modeling with a one-tissue compartment model was used to quantify absolute baseline CBF (ml/100g/min), post-Diamox CBF, and cerebrovascular reactivity (% increase in CBF). Mixed effect statistical models were used to compare CBF across the AC types in R3.4 software. Results: The deep learning network produced high quality pseudo-CT images; ZTE and DL attenuation maps preserved bone structures in the skull base and cervical spine that were not well defined on atlas-based maps (Figure 1). As a result, atlas-based AC underestimated the IDIF peak by 4.3% (P=0.003), and by as much as 20% in an individual, compared to ZTE- and DL-AC (Figure 2). Mixed effects regression (Figure 3) revealed that atlas-based AC overestimated baseline CBF by 5.4 ml/100g/min (8.0%, P<0.0001) and overestimated post-Diamox CBF by 15.5 ml/100g/min (18.4%, P=0.0001), compared to the DL reference. Atlas-based AC also led to overestimation of cerebrovascular reactivity by 19.8% (P=0.02). No statistical differences were observed in CBF or cerebrovascular reactivity between ZTE- and DL-AC. In subjects with repeated acquisitions, scan-rescan coefficient of variation (CoV) of baseline CBF measurements was low (~10%) and not different between the AC types (Figure 4). However, inter-subject CoV of CBF was greater for atlas-based AC by 14.0% (P Conclusions: Atlas-based AC underestimated IDIF peak values compared to ZTE or DL-AC, likely because the cervical arteries used for IDIF were close to the skull base and not well seen on atlas-based AC maps. Underestimation of the IDIF peak directly led to CBF overestimation from kinetic modeling of dynamic images with atlas-based AC, in various brain perfusion states. Furthermore, the high inter-subject CoV of CBF values suggests that atlas-based AC contributes additional subject-dependent variance unrelated to perfusion (a confounder). In contrast, ZTE-based AC gave reproducible quantification of absolute CBF, comparable to the deep learning AC reference that was trained on real CT scans.