A unified approach for identifying PET-based neuronal activation and molecular connectivity with the functional PET toolbox

Abstract

Functional PET (fPET) identifies stimulation-specific changes of physiological processes, individual molecular connectivity and group-level molecular covariance. Since there is currently no consistent analysis approach available for these techniques, we present a toolbox for unified fPET assessment. The toolbox supports analysis of data obtained with a variety of radiotracers, scanners, experimental protocols, cognitive tasks and species. It includes general linear model (GLM)-based assessment of task-specific effects, percent signal change and absolute quantification, and data-driven independent component analysis (ICA). It allows computation of molecular connectivity via temporal correlations of PET signals and molecular covariance as between-subject covariance using static images. Toolbox performance was evaluated by comparison to previous results obtained using established protocols, demonstrating strong agreement (r = 0.91–0.99). Stimulation-induced changes in metabolism ([¹⁸F]FDG) and neurotransmitter dynamics (6-[¹⁸F]FDOPA, [¹¹C]AMT) were detected across different cognitive tasks. Molecular connectivity demonstrated metabolic interactions between networks, whereas group-level covariance highlighted interhemispheric relationships. These results underscore the toolbox’s flexibility in capturing dynamic molecular processes. The toolbox offers a comprehensive, reproducible, user-friendly approach for analyzing fPET data across various experimental settings. This facilitates sharing of analyses pipelines and comparison across centres to advance the study of brain metabolism and neurotransmitter dynamics in health and disease.

Keywords

Glucose metabolism dopamine synthesis serotonin synthesis molecular connectivity molecular covariance

Introduction

Identifying neuronal activation in response to specific stimulation is a cornerstone of neuroscience. Blood-oxygen level dependent (BOLD) fMRI is one of the most commonly used techniques and since its discovery in the 1990s¹ numerous software tools have been made available, such as SPM (https://www.fil.ion.ucl.ac.uk/spm/), FSL (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/) and AFNI (https://afni.nimh.nih.gov/). While fMRI has greatly increased our understanding of brain function in health and disease, the BOLD signal represents an indirect marker of neuronal activation, comprising relative changes in cerebral blood flow, volume and oxygenation.^2,3

The PET-based alternative to assess neuronal activation is the use of radioligands such as [¹⁵O]H₂O or [¹⁸F]FDG.^4,5 While the former represents the gold standard to obtain cerebral blood flow, the experimental setup is considerably challenging due to the ¹⁵O half-life of only 2 min. On the other hand, the latter is a validated proxy of glucose metabolism, however, the assessment of task effects requires multiple scans and/or radiotracer applications when using a conventional bolus protocol. A more recent approach to identify neuronal activation is functional positron emission tomography (fPET) with [¹⁸F]FDG.^6
–8 Since the radioligand is taken up by both neurons⁹ and astrocytes,¹⁰ the individual contribution in response to stimulation still needs to be determined. Similar to fMRI, fPET employs repeated stimulation and analysis techniques such as the general linear model (GLM) or independent component analysis (ICA) to extract task-specific effects. Since [¹⁸F]FDG binds almost irreversibly during the scan, a (bolus+) constant infusion protocol is required to maintain free radioligand throughout the experiment, enabling it to bind according to the actual metabolic demands.⁷ A major advantage of this technique is the gain in experimental flexibility, allowing several conditions to be imaged in a single scan, whereas conventional bolus application would require separate scans for each condition. The temporal resolution has been gradually improved from minutes to just a few seconds, allowing direct comparison between different signals in the temporal domain.¹¹

This technique has demonstrated robust increases in glucose metabolism across various cognitive tasks,^12
–15 with high test-retest reliability.¹⁶ Furthermore, the approach has been successfully extended to image dynamics of neurotransmitter synthesis during task performance, such as the dopamine¹⁷ and serotonin system¹⁸ with 6-[¹⁸F]FDOPA and [¹¹C]AMT, respectively. As fPET provides specific information about physiological processes (such as glucose metabolism, neurotransmitter action, etc.), it offers critical complementary insights into neuronal activation,^13,14,19 making it particularly valuable to explain effects observed with less specific but commonly used imaging approaches such as BOLD fMRI or EEG. However, until now researchers have relied on their own in-house software implementations for fPET data analysis, which pose challenges such as the need for local adaptation and the risk of variability in outcome metrics, leading to reduced reproducibility across studies and imaging centres.

We aim to solve the above issues by providing a versatile toolbox, filling a gap in neuroimaging data analysis. This allows for standardized analyses of fPET data, which facilitates comparison across different tasks, studies and imaging centres as well as sharing of analysis procedures. Thus, the overarching aim of the toolbox is to provide a flexible and robust platform for a variety of analysis techniques. The performance of the software included in the toolbox is validated by demonstrating replication of numerous previous results obtained within a single software package. The current work consolidates a decade of experience with this technique^7,11
–25 into a freely available open-source software package. The toolbox includes Matlab code that has been successfully applied to various data sets involving different stimulation paradigms (visual stimulation, finger tapping, Tetris, working memory, monetary incentive delay, optogenetic stimulation, resting-state) at a range of temporal resolutions (from 60 to 3 s). It supports diverse imaging and infusion protocols (constant infusion and bolus + constant infusion), radiotracers ([¹⁸F]FDG, 6-[¹⁸F]FDOPA and [¹¹C]AMT), scanner systems (GE Advance, Siemens Biograph Vision mMR, 600 Edge and Quadra, Brain Biosciences CerePET, Bruker small-animal PET insert for 7T MRI ClinScan) and species (humans, non-human primates, rodents). The toolbox provides scripting for efficient batch processing and includes a user-friendly graphical user interface (GUI). It will be regularly updated based on new developments and community feedback. This equips basic science and clinical researchers with a generalizable, reproducible analysis approach, which facilitates the widespread use of fPET.

Methods

Implementation

The workflow of the fPET toolbox is shown in Figure 1. The process begins by creating an input variable that contains all relevant information for subsequent calculations. This can be done within a Matlab script or automatically via the GUI provided with the toolbox (see Supplementary Material for a GUI example). The input variable ‘fpetbatch’ is a Matlab structure array (struct), which includes details about the process to be executed (e.g. GLM, ICA, molecular connectivity or covariance), input files (e.g. 4D fPET data, masks, blood data, etc.) and further optional parameter settings (e.g. design matrix, filter specifications, etc.). Once the master function is called with the input parameters, all required subfunctions are executed internally without requiring further user input. All settings and intermediate calculations are stored in a file within the designated results directory, facilitating documentation, analyses sharing and troubleshooting.

Figure 1.

Schematic overview of fPET toolbox functionalities. The input is defined by a script or the GUI and contains all relevant parameters such as the desired calculation, input data, masks, atlases, etc. The toolbox is started by calling a single master function, all other computations are carried out internally, without any further user input. After evaluating the correctness and completeness of the input, the selected calculations are done (GLM, ICA, molecular connectivity or covariance). The corresponding results are then saved in the results folder and can be used for further analyses.

Functionality

When the toolbox is initiated, it first checks whether data already exists in the specified results directory and provides an option to overwrite the data. It then verifies the completeness and validity of all input variables, alerting the user to any necessary input modifications. Depending on the chosen functionality, calculations are performed on a voxel-wise level for GLM and ICA or region of interest basis for molecular connectivity and covariance.

General linear model

The GLM separates stimulation specific effects from baseline by estimating the fit of a predefined model to the data.^7,12 The design matrix setup is highly flexible with options to define task regressors as ramp functions (slope = 1 kBq/min), custom regressors (e.g. nuisance regressors, non-standard stimulation regressors, etc.) and realignment parameters from motion correction. Baseline radiotracer uptake is defined by an average time activity curve (TAC) as extracted from a user-defined mask (typically grey matter). This baseline can be refined by excluding certain brain areas (e.g. fMRI activation maps, meta-analysis masks, anatomical regions, etc.). Additionally, a third order polynomial function can be used to model the baseline,⁷ yielding similar stimulation-induced effects,^12,20 and test–retest reliability,¹⁶ but less optimal fitting.¹²

Before GLM calculations, data can be low-pass filtered and regressors orthogonalized. Solving the GLM (equation (1)) yields 3D maps of beta values and t-statistics for each regressor (including a constant term) and residuals:

Y = β \times X + \in

(1)

\begin{array}{l} TAC = β_{base} {regressor}_{base} + β_{task n} {regressor}_{task n} + \\ β_{add m} {regressor}_{add m} + β_{motion k} {regressor}_{motion k} + ε \end{array}

Percent signal change

For each stimulation regressor defined in the GLM, 3D maps of percent signal change can be calculated as recently validated.²¹ In short, GLM beta estimates of task effects are contrasted with the baseline. This is feasible as under certain assumptions the radiotracer concentration in blood plasma cancels out and percent signal change can be calculated as in equation (2):

%SC \propto \frac{K_{i, task}}{K_{i, BL}} \propto \frac{C_{T, task} (t)}{C_{T, BL} (t)} \propto \frac{β_{task} {regressor}_{task} (t)}{β_{BL} {regressor}_{BL} (t)}

(2)

This non-invasive alternative has been shown to yield strong agreement with percent signal change obtained from the absolutely quantified values.²¹

Absolute quantification

If blood sampling data is available, absolute quantification of the net influx constant K_i can be done for both the baseline and stimulation regressors defined in the GLM using equation (3):

K_{i} = \frac{K_{1} k_{3}}{k_{2} + k_{3}}

(3)

The calculation requires a plasma input function characterizing the concentration of radiotracer in blood or to compute this via a whole-blood input function and plasma-to-whole blood ratio. Blood data are linearly interpolated to match the fPET data frame times. Plasma-to-whole blood values can be either averaged or fitted with a linear function. Blood data are provided as text files with information of time and activity/ratio. It is assumed that combination of automatic and manual blood samples as well as calibration to fPET data has been done beforehand. The quantification is then carried out with the Patlak plot²⁶ (equation (4)), where the start of the linear fit is set to t* = 1/3 of the scan duration:

\frac{C_{T} (t)}{C_{P} (t)} = K_{i} \frac{\int_{0}^{t} C_{P} (τ) d τ}{C_{P} (t)} + intercept

(4)

For [¹⁸F]FDG the cerebral metabolic rate of glucose (CMRGlu) can be calculated (equation (5)) if blood glucose levels (Glu_P) are provided, using a default lumped constant of LC = 0.89^27,28:

CMRGlu = K_{i} \frac{{Glu}_{P}}{LC} \times 100

(5)

Visualizing task effects

In addition to voxel-wise maps as outcome parameters, the temporal characteristics of stimulation effects can be visualized. This function isolates the task-specific TAC for a particular brain region (as defined by a mask image) by subtracting all non-stimulation related parameters of the GLM (see equation (1)). Furthermore, the TACs of raw data and baseline radiotracer uptake can be plotted, which is recommended for quality control (Table 1). Visualization can be performed at both the individual and group level.

Table 1.

Recommended experimental settings for fPET studies. These settings have been successfully applied across various analysis techniques (GLM, ICA, molecular connectivity and covariance), condition (rest or task), radiotracer ([¹⁸F]FDG, 6-[¹⁸F]FDOPA, [¹¹C]AMT), species (humans, non-human primates, rodents) and PET scanners. Nevertheless, we acknowledge that special applications and further advancements may re-define the optimal settings.

Aspect	Recommendation	Reason
Radiotracer administration	Bolus + constant infusion	Enhances overall SNR and produces nearly linear TACs, enabling more accurate motion correction and model robustness
Radiotracer administration	Bolus activity = 20%	Ensures that the majority (80%) of radiotracer activity is constantly infused throughout the measurement and is thus readily available for stimulation-induced effects
Task design	Initial rest period ~6–8 min	Allows for stabilization of radiotracer uptake, establishing a reliable baseline for subsequent task-induced signal changes
	Rest periods between tasks	Aids in isolating task-related changes by differentiating them from baseline activity, improving signal attribution
	Final rest period	Helps distinguishing task-related signal changes at the end of the scan, reducing ambiguity in identifying post-task activity
	Similar length for task and rest	Balances task and rest durations, which can lead to more consistent task responses and more reliable signal detection
Blood sampling	During rest periods	Minimizes interference with task engagement, reducing movement-related artifacts and ensuring clearer signal attribution to task versus baseline
GLM analysis	Calculate percent signal change	Provides a standard, non-invasive metric for comparing activation across studies, facilitates reproducibility, eliminates need for blood sampling without extra user input
GLM analysis	Visualize TACs	Enables to inspect data quality, detect artifacts and issues with radiotracer infusion, assess model fits
ICA analysis	Remove initial rest period	Avoids contamination of ICA by initial frames with low SNR, ensuring more accurate task-related component extraction
Molecular connectivity analysis	Remove initial rest period	Prevents initial uptake fluctuations from skewing connectivity estimates, leading to more reliable connectivity patterns
Molecular covariance analysis	Appropriate normalization mask	Reduces intersubject variability by accounting for individual differences in uptake, allowing for more robust cross-subject comparisons

SNR: signal-to-noise ratio; TACs: time activity curves.

Independent component analysis

fPET data can also be analyzed using ICA, a data-driven alternative to the GLM that identifies stimulation-specific effects without prior knowledge of their timing or spatial extent. The ICA approach is model-free, reducing the initial setup complexity but requiring careful interpretation of the results. Both, task-specific and resting-state fPET data can be processed.

The ICA implementation uses FastICA,²⁹ which is widely used for analysis of, for example, BOLD fMRI data. However, since radiotracers currently used for fPET exhibit irreversible kinetics, it is required to remove the baseline uptake beforehand.^19,30 This is achieved by intensity normalization and z-scoring before and after dimensionality reduction via principal component analysis. Depending on the number of input files, time courses of each individual are concatenated to enable group ICA analysis. Since FastICA uses random starting points, repeated use may lead to different results. To provide stable component estimates, FastICA is run ten times internally. For each run, the kurtosis of ICs is computed¹⁹ and the solution with the highest kurtosis (average of the top 50% of ICs) is selected.

Molecular connectivity

Molecular connectivity assesses the similarity between PET signals across brain regions based on their molecular characteristics. We would like to emphasize that the terminology is employed according to a recent consensus within the field,²⁴ which has also been employed by other groups.³¹ In essence, the term ‘connectivity’ describes the relationship between brain regions based on the similarity of their signal’s temporal characteristics, like in fMRI and EEG research. As such, this is calculated within a single individual. In contrast, the term ‘covariance’ refers to the calculation of associations between groups of subjects.

Similar to ICA, this requires removal of the baseline radiotracer uptake, which can be achieved by fitting each ROI’s TAC with a third-order polynomial, regression against a representative baseline TAC or spatio-temporal filtering.^23,32 An ROI atlas is used to compute pairwise correlations between regions, resulting in a molecular connectivity matrix for each individual.

Molecular covariance

Molecular covariance refers to the calculation of associations between subjects using static 3D maps (SUV, K_i, volume of distribution, binding potential, etc.) and an ROI atlas to calculate pairwise regional correlations. This type of analysis already began over 40 years ago.^33,34 Unlike molecular connectivity (which uses temporal information), covariance is computed across subjects rather than within an individual, enabling to draw inferences at the group level only. For more detailed discussions about this and related issues see other recent work.^24,32,35,36 For the computation of molecular covariance, an intensity normalization may be required to remove spurious intersubject variability.^37,38

Environment

The free and open-source toolbox is written in Matlab and tested on versions R2023a, R2021b update 3 and R2018a. Matlab was selected as underlying framework to provide a widely available solution to accommodate users with different backgrounds (simple scripting and GUI). Tests on Windows (10) and Linux (Ubuntu) enable independence of the operating system, which is a relevant consideration for collaborative and multi-site research. Moreover, the few required dependencies make it easy to set up and maintain while also minimizing the potential for version conflicts. The toolbox relies on SPM12 for data handling (https://www.fil.ion.ucl.ac.uk/spm/) and the FastICA toolbox v2.5 (http://research.ics.aalto.fi/ica/fastica/index.shtml), both of which are distributed under the GNU General Public License.

License and disclaimer

The toolbox is and will be actively maintained by the authors through official channels, but also accepts contributions from the research community.

The fPET toolbox is open-source software released under the terms of the GNU General Public License version 2 as published by the Free Software Foundation (https://www.gnu.org/licenses/old-licenses/gpl-2.0.html). In short, it may be used, modified and distributed for non-commercial purposes, provided the original source is cited. The software is provided ‘as is’, without any warranty, express or implied, including but not limited to warranties of fitness for a particular purpose. The developers shall not be held liable for any damages or issues arising from its use.

Results

The toolbox’s functionality is demonstrated using previously published human data. This includes visual stimulation and finger tapping (n = 5)¹² and the video game Tetris (n = 24),¹⁵ both scanned with [¹⁸F]FDG. Data from 6-[¹⁸F]FDOPA (n = 10)²² and [¹¹C]AMT (n = 16)¹⁸ were obtained while subjects performed the monetary incentive delay reward task. Molecular connectivity and covariance were computed using resting-state [¹⁸F]FDG bolus + infusion data from n = 8 subjects.²³ An extensive description of the experimental design, imaging protocols and demographic information can be found within the respective manuscripts. Direct comparison with previously reported results were obtained by Pearson correlation of group-level statistical maps for activation studies and molecular connectivity/covariance matrices, respectively. This resulted in strong agreement for all of the previous analyses, with r = 0.91–0.99 (see details below). Thus, application of the toolbox effectively reproduced the previously reported findings across various settings, including different radiotracers, scanner systems and cognitive tasks.

The GLM analysis separated task effects from baseline radiotracer uptake and absolute quantification with arterial input functions yielded CMRGlu and K_i. Data obtained with [¹⁸F]FDG while subjects completed the Tetris task yielded significant activation in the intraparietal sulcus, frontal eye field and visual cortex (Figure 2(a)),^14,15,20 with percent signal changes of ~20% from baseline. Likewise, the monetary incentive delay task showed increases in 6-[¹⁸F]FDOPA dopamine synthesis rates in the ventral striatum (Figure 2(b))²² as well as [¹¹C]AMT serotonin synthesis rates in the ventral striatum and anterior insula (Figure 2(c)).¹⁸ Here, task-induced changes reached up to ~150% and 40% for the two neurotransmitter systems, respectively. Evaluation of task-specific time activity curves confirmed the model fits in the temporal domain for each of the different radiotracers. Correlation with previously reported activation maps were r = 0.95 for the [¹⁸F]FDG study, r = 0.99 for 6-[¹⁸F]FDOPA and r = 0.91 for [¹¹C]AMT.

Figure 2.

Example results obtained with the GLM. The approach separates baseline radiotracer uptake (column 1) from task-specific effects (columns 2–4) as defined by the design matrix. Quantification was done in an absolute manner with the Patlak plot (CMRGlu, K_i) and relative to baseline as percent signal change (%). Task-specific effects were subject to a one sample t-test (p < 0.05 FWE corrected cluster level, following p < 0.001 uncorrected voxel level). Task-specific time activity curves were extracted to visualize model fits. The cerebral metabolic rate of glucose (CMRGlu, a) was obtained with [¹⁸F]FDG, while subjects completed the Video game Tetris^®.^14,15,20 The net influx constant K_i of dopamine (b) and serotonin synthesis rates (c) was obtained with 6-[¹⁸F]FDOPA²² and [¹¹C]AMT,¹⁸ respectively, while subjects carried out the monetary incentive delay reward task.

ICA of [¹⁸F]FDG data during visual stimulation and right finger tapping¹² identified the corresponding components of task-specific activation. This included the primary visual cortex as well as the primary motor cortex and supplementary motor area (Figure 3(a)) within the first two independent components (i.e. those with highest kurtosis values). These effects also matched those identified in the GLM analysis (Figure 3(b)). However, for ICA activation in the visual cortex was split into two components (IC 1 and IC 7). The lower kurtosis values of IC 7 also seem to reflect the less pronounced effects in the GLM analysis.

Figure 3.

Example results obtained with ICA. (a) After removal of baseline radiotracer uptake, the approach isolates independent components based on similar spatio-temporal characteristics. Task effects were obtained with [¹⁸F]FDG, while subjects underwent visual stimulation and simultaneously performed a right finger tapping task.¹² Two independent components captured activation in the visual cortex (IC 1 and IC 7), whereas another one comprised effects in the motor cortex (IC 2). (b) For comparison, results from the GLM analysis as obtained with the toolbox from the same data are shown on the right.

Computation of molecular connectivity and covariance from resting-state [¹⁸F]FDG scans was done with dynamic fPET data (50 × 1 min frames) and static images thereof (average of last 10 min), respectively.²³ As demonstrated previously,^23,32 within-subject molecular connectivity and between-subject covariance yielded different network patterns. Molecular connectivity was characterized by strong links between different modules, while covariance showed strongest links between homologous brain regions (Figure 4(a) and (b)). Correlation with previously reported matrices were r = 0.98 for molecular connectivity and r = 0.99 for molecular covariance.

Figure 4.

Example results of molecular connectivity and covariance. (a) For molecular connectivity, the baseline was removed by fitting a third-order polynomial to each region’s TAC. Pairwise regional correlations were calculated for each individual as temporal correlations. (b) Molecular covariance was computed from static images as correlation between subjects. Resting-state fPET data for both analysis were acquired with [¹⁸F]FDG.²³ For (a) the full dynamic dataset was included, while for (b) a static image was created as mean of the last 10 min. Cortical and subcortical regions were taken from the Harvard–Oxford atlas (left and right hemisphere separated). Thresholded matrices were obtained by retaining connections, which are stronger than mean + 3 standard deviations.

Discussion

We introduce a versatile toolbox for the analysis of fPET data, enabling the identification of stimulation-specific changes across various physiological processes, including glucose metabolism as well as dopamine and serotonin synthesis. In addition to task-based analyses, the software also supports computation of individual molecular connectivity and group-level covariance matrices with virtually any radioligand. The centralization of all parameter settings in a single file provides an effective approach to document, share and troubleshoot analysis workflows.

Methodological considerations

Although this work does not aim to provide a comprehensive comparison between all available methodologies, a few advantages and limitations are noteworthy. The GLM approach is particularly suitable for situations where a clear hypothesis exists regarding the timing of activation, often the case in task-based studies.^39,40 This approach enables robust detection of task-specific effects with intuitive interpretability due to the pre-defined model, with clear linkage between experimental design and neural response. Furthermore, GLM estimates can be extended to derive percent signal change or perform absolute quantification of the net influx constant.^7,21 However, the GLM effectiveness depends on how well the model matches the actual data. Underestimation of effects may occur if the signal contains unexpected or complex temporal variations. In contrast, ICA is data-driven and identifies spatially remote effects based on shared temporal structure, without reference to the task design. As such, it does not require a priori knowledge on the timing and shape of the effects, making it suitable for detecting spontaneous or unknown activity sources that exhibit more complex signals.^8,30 ICA is also advantageous for noise removal,⁴¹ particularly for high-temporal resolution data where physiological artefacts such as respiratory or cardiac signals, can become significant.⁴² However, ICA has its limitations. The resulting components can be biologically challenging to interpret, particularly when the data contains mixed sources or artefacts and it may show lower sensitivity and specificity to task-relevant activation, due to its model-free nature. This flexibility in capturing individual variation may come at the cost of reduced consistency across participants. Furthermore, the selection of the number of components can be subjective, requiring validation.⁴³ Practically, the two approaches also require different preprocessing steps, which will result in different outcomes. For instance removal of baseline radiotracer uptake and dimensionality reduction is required before ICA, whereas a design matrix with baseline and task effects needs to be defined for GLM. However, a comprehensive comparison of the two approaches (e.g. optimal preprocessing, sensitivity analyses with respect to noise) exceeds the scope of the current work and thus needs to be carried out in future studies.

The fPET toolbox supports both individual molecular connectivity and group-level molecular covariance analyses, each offering distinct insights. Molecular connectivity captures the temporal similarity of radiotracer uptake between brain regions within an individual, akin to functional connectivity in fMRI. This is particularly useful for studying network interactions based on metabolic demands and neurotransmitter systems like dopamine and serotonin, revealing synchronized activity patterns in response to tasks or pharmacological challenges. However, interpreting such connections still requires further work to as this is not yet established beyond [¹⁸F]FDG. On the other hand, group-level covariance aggregates data across individuals to explore consistent patterns of static molecular characteristics. This approach is more suitable for population-level studies, such as comparing patient groups or examining broad network disruptions. Covariance analysis highlights how molecular processes differ across subjects, providing a macroscopic view of brain function. Thus, the two methods offer complementary insights. Molecular connectivity is ideal for individualized, real-time analysis, while covariance excels in population studies or situations where no dynamic data is available. The toolbox allows for flexible application of both approaches, enabling tailored analyses depending on the research focus.

Recommendations for future studies

To stimulate future research with fPET, we provide several key recommendations for its optimization (Table 1). While early fPET studies utilized a constant radioligand infusion,^6,7 the bolus + constant infusion protocol offers clear advantages. This infusion protocol increases the signal-to-noise ratio, particularly at early time points,¹² thereby improving motion correction accuracy. Moreover, it generates TACs with near-linear characteristics, enabling more robust modelling and reduced bias task effect estimates.^7,12,44 Although the optimal ratio between bolus and subsequent infusion has yet to be identified,^12,44 higher infusion rates lead to more radioactivity for the estimation of task effects and thus more robust task estimation, due to a steeper slope of the TAC. This aligns with the earliest work on task-specific changes in metabolic demands, which applied the stimulus during the initial bolus,⁴⁵ coinciding with the TAC’s steepest slope.

We further recommend delaying the onset of stimulation until a stable baseline with near-linear uptake is established, ~6 min after start of the radiotracer administration. Incorporating rest periods between task blocks, similar to fMRI protocols, aids in distinguishing task-related changes from baseline activity. The inclusion of a rest period following the final task block also helps to avoid ambiguity in task-related effects at the end of the scan. While consensus on such resting periods in functional imaging is limited, a similar length as the stimulation block has been shown to yield robust results, for example, Hahn et al.²⁰

For simultaneous PET/MR imaging, the choice between a hierarchical⁸ or continuous task design²⁰ depends on the actual experiment and aim of the study. As a hierarchical design contains short task (and rest) blocks for fMRI within long task blocks for fPET, this might be well suited for simultaneous assessment of neuronal activation proxies obtained with fMRI and fPET. However, one needs to be aware that changes in the [¹⁸F]FDG fPET signal do indeed occur within seconds, and thus also short task and rest blocks will influence the signal.¹¹ On the other hand, a continuous design employs continuous task performance for longer time periods of, for example, 5 min, without resting periods during that time. This is particularly suited for the simultaneous assessment of functional connectivity^15,20 or other MRI techniques like arterial spin labelling or spectroscopy.

When employing the GLM for the identification of task effects, we recommend always calculating percent signal change. This non-invasive measure offers a comparable and scalable outcome across different experiments, radiotracers and research centres, without the need for invasive blood sampling or extra user input. Additionally, visualizing TACs remains a simple and effective means to inspect data quality, evaluate model fits and detect potential artefacts in the data or potential issues with radioligand infusion.

Finally, the authors offer widespread support for fPET studies, including experimental design, measurement protocol, practical implementation, data analysis and multimodal combinations, based on various international collaborations. Examples of these include previous work on the combination of metabolic demands and functional connectivity,²⁰ the divergence between glucose metabolism and the BOLD signal,^13,14,19 optimization of the preprocessing pipeline⁴⁶ and measurement protocols.²²

Limitations, outlook and conclusions

Currently, the toolbox focuses on deriving task estimates or connectivity for individual subjects as well as covariance across groups. However, it does not implement statistical testing. The decision is deliberate, as the scope of this work is to provide individual-level outputs that can then be further processed. These can be realized using well-established external tools for group-level statistics, such as SPM, FSL and AFNI for task effects as well as network based statistics⁴⁷ and the brain connectivity toolbox⁴⁸ for molecular connectivity. Nevertheless, we aim to expand the toolbox with further potential features such as intersubject correlations, a currently lacking statistical framework for molecular covariance analysis, distinct analysis options for high-temporal resolution fPET data and a simulation section. A dedicated support platform is inherently available via GitHub, where users can report issues.

With a decade of international experience, the unified approach offered by this toolbox aims to stimulate further advancements in the field of fPET. This generalization may enhance the comparability of results across sites or studies, which is particularly important when pooling data in multi-centre research, as it ensures consistent interpretation among researchers, reduces ambiguity and supports reproducibility. Furthermore, imaging the dynamics of neurotransmitter synthesis offers a promising application with unprecedented stimulation-induced signal changes around 100% from baseline for 6-[¹⁸F]FDOPA¹⁷ and 40% for [¹¹C]AMT,¹⁸ even for cognitive tasks. Looking forward, fPET holds considerable promise for the identification of pharmacological effects. Particularly, targeting and imaging the same neurotransmitter system may provide insights into the mode of action of neuropsychopharmacological treatment agents and drugs of abuse. These approaches could be further extended to patient populations, where the exploitation of functional dynamics of molecular processes offers a novel view on brain disorders as compared to conventional techniques.⁴⁹ Here, the availability of percent signal change as a quantitative metric makes this technique more accessible by eliminating the need for invasive blood sampling.

Supplemental Material

sj-pdf-1-jcb-10.1177_0271678X251370831 – Supplemental material for A unified approach for identifying PET-based neuronal activation and molecular connectivity with the functional PET toolbox

Supplemental material, sj-pdf-1-jcb-10.1177_0271678X251370831 for A unified approach for identifying PET-based neuronal activation and molecular connectivity with the functional PET toolbox by Andreas Hahn, Murray B Reed, Christian Milz, Pia Falb, Matej Murgaš and Rupert Lanzenberger in Journal of Cerebral Blood Flow & Metabolism

Footnotes

Acknowledgements

We thank the graduated team members and the diploma students of the Neuroimaging Lab (NIL, head: R Lanzenberger) as well as the clinical colleagues from the Department of Psychiatry and Psychotherapy for clinical and/or administrative support. In detail, we would like to thank S Kasper, K Papageorgiou, GM Godbersen, S Klug, P Michenthaler, T Vanicek, A Basaran, M Hienert, L Silberbauer, J Unterholzner and G Gryglewski for their medical support. L Rischka and M Klöbl for acquisition and analysis support. V Ritter, K Einenkel, E Sittenberger and D Gomola for subject recruitment. We are further grateful to L Nics, C Vraka, C Philippe, V Pichler, J Raitanen, J Völkle, A Pomberger and L Aichinger for radioligand synthesis. The scientific project was performed with the support of the Medical Imaging Cluster of the Medical University of Vienna.

Ethical considerations

Not applicable. All studies were approved by an ethics committee as reported in the original publications.

Code availability statement

All code and documentation for the fPET toolbox is available on GitHub and is distributed under the GNU general public license version 2.0 ().

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship and/or publication of this article: This research was funded in whole or in part by the Austrian Science Fund (FWF; grant DOI 10.55776/KLI610 and 10.55776/KLI1151, PI: A Hahn; grant DOI 10.55776/KLI1006, PI: R Lanzenberger) and the Vienna Science and Technology Fund (WWTF, DOI 10.47379/CS18039), Co-PI: R Lanzenberger. For open access purposes, the author has applied a CC BY public copyright license to any author accepted manuscript version arising from this submission. C. Milz is a recipient of a DOC Fellowship (27221) from the Austrian Academy of Sciences at the Department of Psychiatry and Psychotherapy, Medical University of Vienna.

Declaration of conflicting interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship and/or publication of this article: R. Lanzenberger received investigator-initiated research funding from Siemens Healthcare regarding clinical research using PET/MR. He is a shareholder of the start-up company BM Health GmbH since 2019. All other authors report no conflict of interest in relation to this study.

Author contributions

Study design: A Hahn, MB Reed, R Lanzenberger. Methods: A Hahn, MB Reed. Data analysis: A Hahn, MB Reed. Manuscript preparation: A Hahn, MB Reed, C Milz, P Falb, M Murgaš. All authors discussed the implications of the findings and approved the final version of the manuscript.

Consent to participate

Not applicable. All subjects provided written informed consent as reported in the original publications.

Consent for publication

Not applicable.

ORCID iDs

Andreas Hahn

Murray B Reed

Christian Milz

Pia Falb

Rupert Lanzenberger

References

Ogawa

Lee

Kay

, et al. Brain magnetic resonance imaging with contrast dependent on blood oxygenation. Proc Natl Acad Sci U S A 1990; 87(24): 9868–9872.

Heeger

Ress

What does fMRI tell us about neuronal activity?

Nat Rev Neurosci 2002; 3(2): 142–151.

Norris

DG.

Principles of magnetic resonance assessment of brain function. J Magn Reson Imaging 2006; 23(6): 794–807.

Fang

Ohba

Hashimoto

, et al. Imaging mitochondrial complex I activation during a vibrotactile stimulation: a PET study using [(18)F]BCPP-EF in the conscious monkey brain. J Cereb Blood Flow Metab 2020; 40(12): 2521–2532.

Mintun

Lundstrom

Snyder

, et al. Blood flow and oxygen delivery to human brain during functional activity: theoretical modeling and experimental data. Proc Natl Acad Sci U S A 2001; 98(12): 6859–6864.

Villien

Wey

Mandeville

, et al. Dynamic functional imaging of brain glucose utilization using fPET-FDG. Neuroimage 2014; 100: 192–199.

Hahn

Gryglewski

Nics

, et al. Quantification of task-specific glucose metabolism with constant infusion of ¹⁸F-FDG. J Nucl Med 2016; 57(12): 1933–1940.

Jamadar

Ward

, et al. Simultaneous task-based BOLD-fMRI and [18-F] FDG functional PET for measurement of neuronal metabolism in the human visual cortex. Neuroimage 2019; 189: 258–266.

Lundgaard

Xie

, et al. Direct neuronal glucose uptake heralds activity-dependent increases in cerebral metabolism. Nat Commun 2015; 6: 6807.

10.

Zimmer

Parent

Souza

, et al. [(18)F]FDG PET signal is driven by astroglial glutamate transport. Nat Neurosci 2017; 20(3): 393–395.

11.

Hahn

Reed

Vraka

, et al. High-temporal resolution functional PET/MRI reveals coupling between human metabolic and hemodynamic brain response. Eur J Nucl Med Mol Imaging 2023; 51(5): 1310–1322.

12.

Rischka

Gryglewski

Pfaff

, et al. Reduced task durations in functional PET imaging with [(18)F]FDG approaching that of functional MRI. Neuroimage 2018; 181: 323–330.

13.

Stiernman

Grill

Hahn

, et al. Dissociations between glucose metabolism and blood oxygenation in the human default mode network revealed by simultaneous PET-fMRI. Proc Natl Acad Sci U S A 2021; 118(27): e2021913118.

14.

Godbersen

Klug

Wadsak

, et al. Task-evoked metabolic demands of the posteromedial default mode network are shaped by dorsal attention and frontoparietal control networks. eLife 2023; 12: e84683.

15.

Klug

Godbersen

Rischka

, et al. Learning induces coordinated neuronal plasticity of metabolic demands and functional brain networks. Commun Biol 2022; 5(1): 428.

16.

Rischka

Godbersen

Pichler

, et al. Reliability of task-specific neuronal activation assessed with functional PET, ASL and BOLD imaging. J Cereb Blood Flow Metab 2021; 41(11): 2986–2999.

17.

Hahn

Reed

Pichler

, et al. Functional dynamics of dopamine synthesis during monetary reward and punishment processing. J Cereb Blood Flow Metab 2021; 41(11): 2973–2985.

18.

Hahn

Reed

Murgas

, et al. Dynamics of human serotonin synthesis differentially link to reward anticipation and feedback. Mol Psychiatry 2024; 2: 600–607.

19.

Haas

Bravo

Ionescu

, et al. Functional PET/MRI reveals active inhibition of neuronal activity during optogenetic activation of the nigrostriatal pathway. Sci Adv 2024; 10(43): eadn2776.

20.

Hahn

Breakspear

Rischka

, et al. Reconfiguration of functional brain networks and metabolic cost converge during task performance. Elife 2020; 9: e52443.

21.

Godbersen

Falb

Klug

, et al. Non-invasive assessment of stimulation-specific changes in cerebral glucose metabolism with functional PET. Eur J Nucl Med Mol Imaging 2024; 51(8): 2283–2292.

22.

Reed

Handschuh

Schmidt

, et al. Validation of cardiac image derived input functions for functional PET quantification. Eur J Nucl Med Mol Imaging 2024; 51: 2625–2637.

23.

Reed

Ponce de León

Vraka

, et al. Whole-body metabolic connectivity framework with functional PET. Neuroimage 2023; 271: 120030.

24.

Reed

Cocchi

Sander

, et al. Connecting the dots: approaching a standardized nomenclature for molecular connectivity in positron emission tomography. Eur J Nucl Med Mol Imaging. Epub ahead of print June 02, 2025.https://doi.org/10.1007/s00259-025-07357-1.

25.

Klug

Murgas

Godbersen

, et al. Synaptic signaling modeled by functional connectivity predicts metabolic demands of the human brain. Neuroimage 2024; 295: 120658.

26.

Patlak

Blasberg

RG.

Graphical evaluation of blood-to-brain transfer constants from multiple-time uptake data. Generalizations. J Cereb Blood Flow Metab 1985; 5(4): 584–590.

27.

Graham

Muzi

Spence

, et al. The FDG lumped constant in normal human brain. J Nucl Med 2002; 43(9): 1157–1166.

28.

Wienhard

Measurement of glucose consumption using [(18)F]fluorodeoxyglucose. Methods 2002; 27(3): 218–225.

29.

Hyvarinen

Oja

Independent component analysis: algorithms and applications. Neural Netw 2000; 13(4–5): 411–430.

30.

Jamadar

Ward

PGD

, et al. Analysis of continuous infusion functional PET (fPET) in the human brain. Neuroimage 2020; 213: 116720.

31.

Deery

Liang

Siddiqui

, et al. Reconfiguration of metabolic connectivity in ageing. Commun Biol 2024; 7(1): 1600.

32.

Jamadar

Ward

PGD

Liang

, et al. Metabolic and hemodynamic resting-state connectivity of the human brain: a high-temporal resolution simultaneous BOLD–fMRI and FDG–fPET multimodality study. Cereb Cortex 2021; 31(6): 2855–2867.

33.

Horwitz

Duara

Rapoport

SI.

Intercorrelations of glucose metabolic rates between brain regions: application to healthy males in a state of reduced sensory input. J Cereb Blood Flow Metab 1984; 4(4): 484–499.

34.

McIntosh

Grady

Ungerleider

, et al. Network analysis of cortical visual pathways mapped with PET. J Neurosci 1994; 14(2): 655–666.

35.

Sala

Lizarraga

Ripp

, et al. Static versus functional PET: making sense of metabolic connectivity. Cereb Cortex 2021; 32(5): 1125–1129.

36.

Jamadar

Egan

. Resting-state FDG–PET connectivity: covariance, ergodicity, and biomarkers. Response to Commentary by Sala et al.; static versus functional PET: making sense of metabolic connectivity. Cereb Cortex 2022; 32(9): 2054–2055.

37.

Vanicek

Hahn

Traub-Weidinger

, et al. Insights into intrinsic brain networks based on graph theory and PET in right- compared to left-sided temporal lobe epilepsy. Sci Rep 2016; 6: 28513.

38.

Biswal

BB.

Metabolic brain covariant networks as revealed by FDG–PET with reference to resting-state fMRI networks. Brain Connect 2012; 2(5): 275–283.

39.

Friston

Holmes

Worsley

, et al. Statistical parametric maps in functional imaging: a general linear approach. Hum Brain Mapp 1994; 2(4): 189–210.

40.

Poline

J-B

Brett

The general linear model and fMRI: does love last forever?

Neuroimage 2012; 62(2): 871–880.

41.

Glasser

Coalson

Bijsterbosch

, et al. Using temporal ICA to selectively remove global noise while preserving global signal in functional MRI data. Neuroimage 2018; 181: 692–717.

42.

Bancelin

Bachrata

Bollmann

, et al. Unsupervised physiological noise correction of functional magnetic resonance imaging data using phase and magnitude information (PREPAIR). Hum Brain Mapp 2023; 44(3): 1209–1226.

43.

Jamadar

Ward

PGD

, et al. Estimation of simultaneous BOLD and dynamic FDG metabolic brain activations using a multimodality concatenated ICA (mcICA) method. Neuroimage 2021; 226: 117603.

44.

Jamadar

Liang

Zhong

, et al. Monash DaCRA fPET–fMRI: a dataset for comparison of radiotracer administration for high temporal resolution functional FDG–PET. Gigascience 2022; 11: giac031.

45.

Phelps

Kuhl

Mazziota

JC.

Metabolic mapping of the brain’s response to visual stimulation: studies in humans. Science 1981; 211(4489): 1445–1448.

46.

Reed

Ponce de León

Klug

, et al. Optimal filtering strategies for task-specific functional PET imaging. J Cereb Blood Flow Metab. Epub ahead of print April 2, 2025. https://doi.org/10.1177/0271678X251325668

47.

Zalesky

Fornito

Bullmore

ET.

Network-based statistic: identifying differences in brain networks. Neuroimage 2010; 53(4): 1197–1207.

48.

Rubinov

Sporns

Complex network measures of brain connectivity: uses and interpretations. Neuroimage 2010; 52(3): 1059–1069.

49.

Stiernman

Grill

McNulty

, et al. Widespread fMRI BOLD signal overactivations during cognitive control in older adults are not matched by corresponding increases in fPET glucose metabolism. J Neurosci 2023; 43(14): 2527–2536.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.73 MB