Andre Thielcke1, Mario Amend1, Suril Gohel2, Bharat Biswal2, Bernd J. Pichler1, and Hans F. Wehrl1
1Department of Preclinical Imaging and Radiopharmacy, Werner Siemens Imaging Center, Eberhard Karls University of Tuebingen, Tuebingen, Germany, 2Department of Biomedical Engineering, New Jersey Institute of Technology, Newark, NJ, United States
Synopsis
Recent
advancement in hardware and software has enabled researchers to study systems
level neuroscience using simultaneous PET/MRI. In this study, simultaneous PET/MR
was used to investigate resting state networks (RSN) in rats, comparing [18F]FDG
PET vs. BOLD-fMRI. RSNs such as default mode network (DMN) have been shown to
be disrupted in clinical populations. ICA and ROI-analysis was used to elucidate
the complementary nature between PET/MR and visualize brain connectivity. ICA PET
and MR data showed prominent RSNs. However, ROI-analysis illustrated different
connectivity between network-involved areas. This work suggests the
complimentary nature of metabolic connectivity mapping (Cometomics). Purpose
Basic research in resting state
networks (RSN) has demonstrated altered connectivity pattern in regards to
neuronal diseases and is important for clinical translation. It is known that
structural and functional connectivity is changed in clinical populations e.g.
Azheimer´s disease, or Epilepsia. Such changes can be investigated by using
imaging technologies like MRI or positron emission tomography (PET). In this
study we investigated prominent resting state networks in rats assessed by
simultaneous PET/MR. The complementary nature between PET and MR could deliver
more detailed information about neuronal network organization and communication,
not only on a functional but also on a metabolic scale.
Methods
Lewis rats (n=4, male, ca 350 g ± 50 g) were measured using a simultaneous PET/MR system, designed for
small animals (7 T MR; isolfurane anesthesia). fMRI-BOLD was measured over 20
min to reveal functional connectivity during resting state (EPI, TR=3000 ms, TE
18 ms, Matrix Size 64×64×16, Voxel Size: 0.5×0.5×1
mm
3, 400 volumes). Metabolic information was measured by using [
18F]FDG-PET
with an acquisition time of 60 min (29.6 MBq; OSEM reconstruction into 60
frames). PET and MR data were preprocessed, using SPM 12. Statistic analyses
were performed by using Matlab (2012b) to evaluate Pearson´s correlation
coefficient between ROI timeseries; GIFT for independent component analysis
(ICA) (40 components, ICASSO algorithm). After ICA, prominent RSNs e.g. DMN
were assessed. PET and MR network images were overlaid and compared, using dice
coefficients (dc). ROI and the correlation strengths between them were
identified by a ROI-analysis using PMOD and Matlab (2012b). Furthermore, distances of areas within
networks were compared with their connectivity strength. Additionally, the
cerebellum was included to the DMN to investigate complementary information
between PET and MR.
Results
Prominent RSNs were identified in both modalities, for instance the cingulate
cortex (Cg1, Cg2 dc: 0.47, overlap: 68%), motor cortex (M1, dc 0.20, overlap:
22%), caudate putamen (CPu, dc: 40, overlap: 48%). Furthermore, DMN-like
structure could be identified with MR (8 ICA components) as well as with PET (7
ICA components). In general, Z-scores from MR data were higher compared to PET.
However, also disparities could be observed between both modalities. Therefore,
connection strengths within specific networks were investigated (Figure 1, 2). ROI-analysis
based on PET data presented additional weak and negative correlation between
areas involved in the DMN (Figure 3). In contrast, MR showed only strong positive
correlation within the DMN. No influence of the regional distance on
connectivity strength could be observed in both modalities (Figure 4).
Furthermore, PET suggested a participation of the cerebellum to the DMN in
contrast to MR.
Discussion
This study though preliminary in nature, demonstrates
that RSN could be imaged not only with MR but also with PET. This novel
technique of metabolic connectivity mapping (Cometomics) allows an accurate
comparison between PET/MR. The study revealed similarities but also
discrepancies between PET/MR images in motor and ganglia networks during
resting state. Discrepancies can be explained by the complementary nature (BOLD-fMRI:
hemodynamic, [
18F]FDG-PET: metabolic) between modalities that offers a new perspective
to combine functional and metabolic connectivity in order to receive a deeper
insight into neuronal processing. Although, ICA showed high similarities
between PET and MR, ROI-analysis revealed significant differences within RSNs. Therefore, ROI-analysis is suited for a more detailed look at global neuronal processing. Correlation
coefficients between network-involved areas were significant higher in MR data than
in PET data, probably due to larger amounts of images involved in MR analysis
compared to PET. In contrast to PET, MR data indicated only positive connectivity
between all areas that are involved in the DMN, whereas PET revealed predominately
negative connectivity. In that case PET could indicate that metabolic signals show
other neuronal processes than MR e.g. inhibition (possibly negative
correlation) vs. excitation (possibly positive correlation) processes. Furthermore,
the distance factor between areas within networks didn´t indicates an influence
on connectivity strength, which illustrates the stability of such networks. In
addition, we included the cerebellum to the DMN. Interestingly, MR data
postulated no connection between DMN and cerebellum. However, PET data
illustrated negative correlation between DMN and cerebellum. Here, further
investigations of PET/MR data are needed.
Conclusion
In conclusion, based on simultaneous PET/MR we
showed nearly optimal matching between certain PET and MR derived brain
networks, using ICA. However, via ROI-analysis, differences within networks became more prominent. This work serves as
a catalyzer for further research on the interpretation and evaluation of hemodynamic
and metabolic signals ultimately allowing a deeper insight into brain processing.
Acknowledgements
This work was funded by:
BMBF Grant No: 01GQ1415,
NIH R01
DA038895,
University of Tuebingen, fortuene:
2209-0-0.
References
Wehrl
HF et al.,
Nature
Med.,
2013
Biswal
B et al., Magn Reson
Med.,
1995