A novel partial volume correction method for accurate quantification of [¹⁸F] flortaucipir in the hippocampus

The PET tracer [¹⁸F]flortaucipir([¹⁸F]AV-1451) is a promising biomarker for in vivo assessment of tau pathology in AD [1]. However, cautious interpretation of [¹⁸F] flortaucipir data is warranted, especially in the hippocampus. Indeed, although tau pathology affects the hippocampus relatively early in the course of the disease [2], [¹⁸F] flortaucipir uptake in the hippocampus does not distinguish AD patients from controls [3, 4]. A possible explanation may be that the relatively low spatial resolution of PET leads to an underestimation of the signal in a small volume of interest (VOI) such as that of the hippocampus [4], especially in case of atrophy [3]. Resulting partial volume effects (PVE) may cause spill-in or spill-out to adjacent regions with higher or lower activity, respectively [5]. In the hippocampal region, this could also be the case as high off-target binding in the adjacent choroid plexus (CP) [6] may cause spill-in of [¹⁸F] flortaucipir [3, 7].

Partial volume correction (PVC) methods are used to correct for PVE. Previous studies used several different MR-based PVC methods [8‐11] to process [¹⁸F] flortaucipir scans [7, 11‐17]. Although some studies reported no significant effects of PVC on the correlation between tau uptake and cerebrospinal fluid (CSF) measures, cognition, and diagnostic accuracy [13, 15, 17], the specific impact of the CP signal on hippocampal [¹⁸F] flortaucipir uptake has yet to be evaluated.

In this study, PVC was performed using a method that combines Van Cittert (VC) iterative deconvolution (IDM) with highly constrained back-projection (HYPR) denoising. The combination of HYPR and VC IDM (Hypr-IDM-Hypr, HDH) was recently developed and validated [18] and allows for more accurate quantification of PET images. The purpose of the present study was to assess the impact of CP activity on quantification of hippocampal [¹⁸F] flortaucipir binding and to correct for spill-in using HDH PVC.

Ten controls (average age 67.7 ± 6.8 years, MMSE score 29.2 ± 0.6) and 10 probable AD patients (average age 63.9 ± 7.8 years, MMSE score 23.9 ± 3.1) were included. There was no significant difference in age between AD patients and controls (p < 0.05). Additional file 2: Figure S2 illustrates the time activity curves for choroid plexus and hippocampus VOIs (with and without PVC).

Before PVC, the relationship between the hippocampus and CP in controls decreased for an eroded VOI (r² = 0.45, slope = 0.53) compared with the complete VOI (r² = 0.59, slope = 0.80). After PVC, a further decrease was observed for both complete VOI (r² = 0.29, slope = 0.27) and eroded VOI (r² = 0.15, slope = 0.15). The relationship between hippocampus and CP uptake was significantly reduced when using an optimized VOI in combination with PVC (p for interaction < 0.05) (Fig. 2 and Additional file 3: Figure S3). An even stronger effect was observed in case of AD patients when using PVC in combination with the optimized hippocampal VOI (r² = 0.54, slope = 0.98 without PVC versus r² = 0.01, slope = − 0.05 with PVC; p for interaction < 0.05) (Figs. 1 and 2).

In addition, there was a significant decrease in hippocampal V_T after using both PVC and eroded VOI compared with the uncorrected data (Fig. 3 and Additional file 4: Figure S4, p < 0.05). V_Ts obtained for the choroid plexus and hippocampus (complete and eroded) both before and after PVC for all the subjects included in the study (mean ± SD) are presented in the Additional file 5: Figure S5.

In order to improve the quantification of [¹⁸F] flortaucipir uptake in the hippocampus, a novel HDH PVC method was used together with an eroded hippocampal VOI to correct for possible spill-in effects from the CP. After PVC and optimizing hippocampal VOI, only a weak correlation remained between hippocampal V_T and CP V_T, suggesting that spill-in from the CP was successfully reduced thus allowing for more accurate quantification of hippocampal [¹⁸F] flortaucipir uptake.

Wang et al. [16] showed that PVC [10] reduced the correlation between [¹⁸F] flortaucipir hippocampal and CP uptake (r = 0.39, p = 0.003 before PVC vs r = 0.14, p = 0.31 after PVC). Wang et al. seem to have mitigated the CP’s spill over without the use of hippocampal VOI erosion. Although the PVC implementation used by Wang et al. is not similar to the method used in this study, it can be argued that the relationship between choroid plexus and hippocampus is mitigated when applying PVC. With regard to the use of eroded hippocampal VOI, the possible reason for Wang et al. to not require erosion of hippocampal VOI could be that relationship present in their data between choroid plexus and hippocampus was not as strong (r = 0.39) as in our data (r = 0.76, combining both AD and controls). In case of a weak relation, possibly just PVC is sufficient; however, in case of a stronger relation, erosion might be necessary. Another point to note is that before PVC, no significant difference in the V_Ts was observed between the complete hippocampal VOI and eroded hippocampal VOI (Additional file 5: Figure S5). This suggests that the benefit of erosion was only observed along with the PVC. In addition, Schöll et al. [7] showed that PVC [8] caused a large increase in CP signal, indicating that spill-out from the CP could be reduced, potentially leading to a more accurate estimation of tau uptake in the adjacent hippocampus. Both studies suggest that PVC can be used to reduce spill-over effects from the CP. Nevertheless, both studies used MR-based PVC methods, and segmentation problems could result in a potential bias [10].

One of the strengths of the present study is that the VC IDM method leads to more accurate quantification of PET images and does not use MR scans for PVC, thereby avoiding potential segmentation and co-registration error. The addition of HYPR reduces the poor SNR of VC IDM, thus preserving image quality [18].

A potential weakness of this study is the fact that CP VOI was defined on PVC PET images rather than on MR images. However, CP activity could clearly be distinguished from that of surrounding tissue, as it was both visually and quantitatively much higher. Even then, a manual VOI definition is prone to error and inter-subject variability. In this study, the same researcher worked on the VOI definition to avoid inter-subject variability and to mitigate the error in VOI definition, and the manual VOIs of CP on PET PVC images were validated using MR FLAIR images. PET PVC-based VOI definitions showed good correspondence to the MR FLAIR images, suggesting that PVC PET images can be used for reliable definition of the CP VOI; however, an automatic or a semiautomatic VOI erosion method is warranted. Another limitation would be that there is no ground truth (no autopsy data available) to check the accuracy of the implementation. However, based on the presented analysis, it can be stated that the proposed methodology mitigates the spill-over effects of choroid plexus on hippocampus uptake.

The use of our new PVC method in combination with an optimized hippocampus VOI significantly reduces spill-in of CP activity into the hippocampus and improves accuracy of hippocampal V_T.