Connecting the dots: approaching a standardized nomenclature for molecular connectivity in positron emission tomography

Abstract

Positron emission tomography (PET)-based connectivity analysis provides a molecular perspective that complements fMRI-derived functional connectivity. However, lack of standardized terminology and diverse methodologies in PET connectivity studies has resulted in inconsistencies, complicating the interpretation and comparison of results across studies. A standardized nomenclature is thus needed to reduce ambiguity, enhance reproducibility, and facilitate interpretability across radiotracers, imaging modalities and studies. Here, we define and differentiate the terms "molecular connectivity" and "molecular covariance". Drawing parallels from other imaging modalities, we propose "molecular connectivity" as an umbrella term to characterize statistical dependencies between the measured PET signal across brain regions at a within-subject level. Like fMRI resting-state functional connectivity, "molecular connectivity" leverages spatio-temporal associations in the PET signal to derive brain network associations. Conversely, "molecular covariance" denotes group-level computations of covariance matrices between-subjects. Further specification of the terminology can be achieved by including the target of the employed radioligand, such as "metabolic connectivity/covariance" for [¹⁸F]FDG or "amyloid covariance" for [¹⁸F]flutemetamol and "tau covariance" for [¹⁸F]flortaucipir. While this approach to standardization aims to clarify terminology, open questions remain about the neurobiological underpinnings of these connectivity metrics. Future research should focus on elucidating these mechanisms and developing advanced computational methodologies that evaluate diverse feature relationships and improve the robustness of PET-based connectivity metrics.

Keywords: Consensus; Functional PET (fPET); Metabolic connectivity; Molecular covariance; Terminology.

Fig. 1

Graphical overview of common neurophysiological techniques used to assess brain connectivity. (a) Acquisition: Each box represents a common imaging method where the position on the x-axis indicates the temporal resolution of each technique as an input for different connectivity measures, whereas position on the y-axis represents the spatial resolution of the technique. (b) Radiotracer administration: In the field of PET imaging, application of the radiotracer as bolus enables to obtain static images and tracer kinetics, while a bolus + constant infusion protocol additionally allows to capture moment-to-moment signal fluctuations. In theory, moment-to-moment fluctuations may also be obtained after a bolus application, however, the rationale for bolus + infusion protocol is to provide free radiotracer throughout the scan, which then binds according to such rapid fluctuations (dashed green line) (c) Input for connectivity estimation: dynamic approaches enable the computation of within-subject connectivity, which is typically calculated by correlating the time courses among brain regions. For PET this is derived from either bolus plus constant infusion or simple bolus. In contrast, between-subject covariance is estimated over a group of participants as it lacks a temporal component. As such, these techniques use different input signals to estimate connectivity, i.e., moment-to-moment fluctuations in the signal or radiotracer kinetics for within-subject metrics, and covariance of PET outcome measures for between-subjects PET signal covariance (i.e., static images). Furthermore, static images can also be obtained from dynamic PET data, e.g., through kinetic modeling, and subsequently enable estimation of covariance metrics. EEG: electroencephalography, fMRI: functional magnetic resonance imaging, fPET: functional positron emission tomography, SUV(R): standardized uptake value (ratio)

Fig. 2

Summary of techniques and proposed nomenclature of PET-based connectivity. Within-subject connectivity is broadly separated into the assessment of moment-to-moment signal fluctuations and raw time activity curves (TACs) or compartment time courses. The former requires the elimination of the low-frequency baseline radiotracer uptake and reducing high-frequency noise, which can be achieved using different detrending methods. Similar to fMRI-based functional connectivity, the correlation of moment-to-moment signal fluctuations are used to estimate correlationed-based “molecular connectivity” (cMC). The Euclidean distance can be used to estimate similarities using either raw TACs or radiotracer compartment time courses between brain regions, thus termed Eucledean-based “molecular connectivity” (eMC). The between-subject metrics use any type of static PET image as input and after computation using methods like covariance/correlation or SICE, are termed as “molecular covariance” (mCov). Moreover, within the mCov term, similarity-based methods that utilize static images to define network relationships using metrics such as cosine similarity, inverse Euclidean distance, or graph-based measures are also included