Hierarchical Multivariate Bayesian Reference Tissue Modelling of PET Data
Poster No:
1988
Submission Type:
Abstract Submission
Authors:
Granville Matheson1,2, Johan Lundberg2, Martin Gärde2, Emma Veldman2, Amane Tateno3, Yoshiro Okubo3, Mikael Tiger2, R Todd Ogden1
Institutions:
1Columbia University, New York, NY, 2Karolinska Institutet, Solna, Stockholm, 3Nippon Medical School, Tokyo, Tokyo
First Author:
Co-Author(s):
Introduction:
Positron emission tomography (PET) is an in vivo imaging methodology essential for studying the molecular pathophysiology of psychiatric and neurological disease. PET analysis is conventionally performed as a two-stage process of quantification followed by analysis. Quantification typically involves fitting pharmacokinetic (PK) models to each time activity curve (TAC) from each region of each individual independently to estimate several PK parameters. For analysis, the parameter representing target binding is entered into a statistical model. We recently introduced SiMBA (Simultaneous Multifactor Bayesian Analysis) (1), which is a hierarchical model which performs quantification for all regions of all individuals at once. In this way accuracy of parameter estimates is improved by borrowing strength across the sample. Moreover, SiMBA performs both quantification and analysis simultaneously, improving inferential efficiency through both effective error propagation, as well as by exploiting multivariate relationships between all of the estimated PK parameters. Until now, SiMBA has been restricted to invasive models, i.e. which require the collection of arterial blood data. We have now extended this approach to a non-invasive reference tissue implementation which applies the simplified reference tissue model (SRTM) (2).
Methods:
We applied the model to PET data with the radioligand [11C]AZ10419369, which binds selectively to the serotonin 1B receptor, using the cerebellar grey matter as reference [3]. We created simulated datasets based on the mean and variance of the estimated parameters. In simulated data, we assessed accuracy of the estimated PK parameters, as well as the power, precision and false positive rate of simulated treatment effects relative to placebo in a two-group design. Next, using data collected at three different PET centres (n=139, n=47 and n=39), we examined the consistency of estimates of regional rate of change of receptor availability with age, which has previously been reported (4). We also compared inferences made using a combined model with those derived from each individual centre.
Results:
In simulated data SiMBA improves quantitative accuracy, reducing error by 60% for BPND, and 74% for R1 compared to conventional approaches. There was no increase in the false-positive rate throughout. Inferential efficiency was greatly improved, with precision of estimated differences and power equivalent to inferences made using conventional means with approximately double the sample size for 20 or more participants per group. However these improvements were modest for sample sizes of 10 per group. Examining empirical age associations between centres, we observe previously shown decreases in BPND across centres which are replicated across all three samples. Furthermore, across centres we also replicate the regional differences in the rate of these changes, with the dorsal brain stem exhibiting the most rapid age-related decreases in BPND, and the ventral striatum and thalamus showing the least rapid decreases in BPND. Finally, the estimates from the combined model are not only consistent with individual estimates from each of the centres, but also exhibit greater precision.
·Figure 1
·Figure 2
Conclusions:
We present a novel approach for non-invasive quantification and analysis of PET time activity curve data which not only improves quantification and inferences, but also yields inferences which are highly consistent across data collected at different PET centres, and allows combining data from different centres into a single model. The primary disadvantage of this approach is its high computational burden, taking up to a week to run for typical sample sizes. Another obstacle to its adoption is the necessity for defining priors over all parameters, however this can also be seen as an advantage as it allows for the incorporation of outside knowledge into the model definition.
Lifespan Development:
Aging
Modeling and Analysis Methods:
Bayesian Modeling 2
Methods Development
PET Modeling and Analysis 1
Novel Imaging Acquisition Methods:
PET
Keywords:
Aging
Data analysis
Modeling
Multivariate
Neurotransmitter
Positron Emission Tomography (PET)
RECEPTORS
Seretonin
Statistical Methods
1|2Indicates the priority used for review
Provide references using author date format
2. Lammertsma, A. A., & Hume, S. P. (1996). Simplified reference tissue model for PET receptor studies. Neuroimage, 4(3), 153-158.
3. Varnäs, K., Nyberg, S., Halldin, C., Varrone, A., Takano, A., Karlsson, P., ... & Farde, L. (2011). Quantitative analysis of [11C] AZ10419369 binding to 5-HT1B receptors in human brain. Journal of Cerebral Blood Flow & Metabolism, 31(1), 113-123.
4. Nord, M., Cselenyi, Z., Forsberg, A., Rosenqvist, G., Tiger, M., Lundberg, J., ... & Farde, L. (2014). Distinct regional age effects on [11C] AZ10419369 binding to 5-HT1B receptors in the human brain. Neuroimage, 103, 303-308.