Bloodstream: a BIDS App for Processing of Blood Data for Analysis of PET Data
Poster No:
1989
Submission Type:
Abstract Submission
Authors:
Granville Matheson1,2, Paul Wighton3, Anthony Galassi4, Martin Nørgaard5
Institutions:
1Columbia University, New York, NY, 2Karolinska Institutet, Solna, Stockholm, Sweden, 3Martinos Center for Biomedical Imaging at MGH, Boston, MA, 4National Institute of Mental Health, Bethesda, MD, 5Copenhagen University, Copenhagen, NA
First Author:
Co-Author(s):
Introduction:
Positron emission tomography (PET) is an in vivo imaging method for measurement of the concentration of specific proteins, peptides and biochemical functions. With its much higher biochemical sensitivity and specificity, PET serves a complementary role to MRI. The gold standard for quantification of PET data involves modelling together with an arterial input function (AIF). This requires sampling the arterial blood throughout the PET examination, from which the blood plasma must be extracted, and from which the relative concentrations of the parent compound and its radiometabolites must be determined. This procedure is labour-intensive, can be uncomfortable for participants, and requires experienced staff for arterial cannulation, blood measurement, as well as analysis. Although non-invasive reference tissue methods or simplified semi-quantitative approaches can serve as reasonably good substitutes for blood sampling for some targets, there exist many targets for which blood sampling is required for valid quantification. With the recent incorporation of PET into the BIDS specification, PET and its associated blood data can now be shared in a standardised and machine-readable format (1). With the increase of data-sharing practices, there are an increasing number of PET datasets which include blood data being shared. Research groups who regularly work with blood data have mostly developed their own in-house approaches for storing and processing this data for later modelling. However even in many PET-focused research groups, the skills and experience are lacking to work with blood data.
Methods:
To this end, we have developed bloodstream, which is a BIDS application for automatically processing blood data so that it can be used for invasive PET quantification. When applied directly to the data without any configuration file, bloodstream will combine the different measured curves into an interpolated arterial input function which can be used for modelling using simple linear interpolation. For the application of more complex approaches, in which models are fit to one or more of the constituent series of measurements, a configuration file can be created which specifies the models which should be applied and allows the customisation of various attributes prior to fitting. To ease the creation of these configuration files, we have created a web app with drop-down menus to select the models and customise their settings, and the output can be downloaded. The bloodstream BIDS app returns not only the interpolated curves, but also a full analysis notebook with diagnostic plots. bloodstream is developed in R (2) and based on functions in the kinfitr R package (3, 4). For datasets requiring more tailored modelling and analysis, the code used to perform all the modelling can be retrieved from the output notebooks to allow users to run the analysis interactively at each step with greater customisation. bloodstream can be installed as an R package, or invoked from within a docker container, which further simplifies its use.
·Figure 1
Results:
bloodstream is shared on GitHub at the following link: https://github.com/mathesong/bloodstream. It has been successfully applied to one open dataset [5] using the [11C]PS13 radiotracer, as well as several soon-to-be-open datasets. We have also recorded a demonstration video covering the theory of PET blood processing, as well as a hands-on tutorial of the application of bloodstream to an open dataset linked on the GitHub README.
·Figure 2
Conclusions:
In summary, we present a new tool for performing reproducible blood processing and modelling for PET analysis, which can be used for later invasive quantification of PET data. In this way, access is improved to complex and costly datasets for groups lacking experience working with blood data. Moreover, this tool incentivises the curation of PET data to the BIDS structure both for internal use of this tool, but also hopefully leads to more externally shared datasets.
Modeling and Analysis Methods:
Methods Development
PET Modeling and Analysis 1
Neuroinformatics and Data Sharing:
Workflows 2
Novel Imaging Acquisition Methods:
PET
Keywords:
Acquisition
Blood
Data analysis
Modeling
Positron Emission Tomography (PET)
Statistical Methods
Workflows
1|2Indicates the priority used for review
Provide references using author date format
2. R Core Team, R. (2023). R: A language and environment for statistical computing.
3. Matheson, G. J. (2019). kinfitr: reproducible PET pharmacokinetic modelling in R. bioRxiv, 755751.
4. Tjerkaski, J., Cervenka, S., Farde, L., & Matheson, G. J. (2020). Kinfitr—an open-source tool for reproducible PET modelling: validation and evaluation of test-retest reliability. EJNMMI research, 10, 1-11.
5. Kim, M, Lee, J, Anaya F. J., Hong, J. and Miller, W., Telu, S., Singh, P., Cortes, M. Y., Henry, K., Tye, G. L., Frankland, M. P., Santamaria, J. A. M., Liow, J., Zoghbi, S. S., Fujita, M., Pike, V. W. & Innis, R. B. (2022). First-in-human evaluation of [11C]PS13, a novel PET radioligand, to quantify cyclooxygenase-1 in the brain. OpenNeuro. [Dataset] doi: doi:10.18112/openneuro.ds004230.v2.3.1