Human metabolic homeostasis relies on complex neuronal, vascular, and humoral mechanisms at the level of the whole body. Simultaneous non-linear interactions among organs form distinct physiological networks. Many systemic diseases have an underlying disturbance in the inter-organ physiological interaction networks [
1]. The existing methods for detecting such a disturbance work mostly at the organ level and developing generalized methodologies capable of adequately quantifying the abnormalities at the system level remain a challenge so far. Most of the studies conducted to date on such topics have utilized non-imaging tools. Thiele et al. [
2] developed a metabolic network reconstruction approach in which organ-specific information from the literature and omics data were used. The data sources in that study included 20 organs, 6 sex organs, 6 blood cell types, and 13 biofluid compartments. Barajas-Martínez et al. [
3] developed a physiological network based on anthropometric measurements, fasting blood tests, and other vital sign measurements. The authors concluded that the specific structural properties of the network change across the human lifespan and could, therefore, serve as indicators of the health status. Cui et al. [
4] reconstructed the global mammal metabolic network for different tissues and cell types, through which they attempted to connect organs with the inter-organ metabolite transport. In separate studies, Bashan et al. [
5] and Bartsch et al. [
6] developed a framework to probe the interactions among diverse body systems and identified a physiological network that represented the interplay between network topology and function.
Imaging approaches have been used mostly to investigate the functional interactions related to brain dysfunction. Brain disorders, such as dementia, have their origin and the associated functional impairment not in distinct regions but rather in a network of connected regions [
7]. Alterations in these inter-regional brain connectivity networks, if quantified, could reflect the status of various neurological diseases. For instance, structural connectivity has been investigated using diffusion tensor imaging (DTI) [
8,
9], and functional connectivity has been investigated using functional magnetic resonance imaging (fMRI) [
10‐
12]. DTI or fMRI connectivity patterns could, in principle, be determined on the subject level by correlating fiber connections or time-series signals. Metabolic connectivity deciphered using PET measurements, in contrast, is often derived using a population-based approach [
13,
14]. The reason is that a routine PET scan usually performs static acquisition 60 min post injection, which measures the summed activity concentration in a certain period (typically 5–30 min). However, conventional methods cannot compute the connectivity (correlation) between the region without access to the real-time dynamic activity signal. Certain recent studies have suggested deriving the metabolic connectivity at the individual brain level based on relationships in the regional activities [
15,
16]. On the other hand, few studies have investigated whole-body metabolic connectivity using PET imaging. Horsager et al. [
17] used three different radiotracers to investigate the alpha-synuclein interaction pathology between organs, and the findings supported their hypothesis regarding the existence of two subtypes of Parkinson’s disease (brain-first top-down and body-first bottom-up types). Heiskanen et al. [
21] utilized
18F-FDG PET to gain a system-level understanding of how exercise training affects the crosstalk in the central metabolism. Dias et al. [
18] compared whole-body FDG uptakes and glucose metabolic rates between the diabetes group and non-diabetes group. They successfully derived difference of organ crosstalk in these two groups hence conclude the impact of diabetes on glucose homeostasis. Sundar et al. [
20] compared the metabolic rates at multiple organs between healthy male group and female group, based on group-averaged normative correlation analysis of the measured time-activity curves. Suchacki et al. [
19] reported an approach to understanding murine bone metabolic interactions in vivo by analyzing the correlations of the
18F-FDG time-activity curves in bones.
In this context, the present study proposes a framework capable of constructing a network that would reveal the individual metabolic abnormality using a subject’s whole-body SUV image and a normal control database. The analysis does not require access to dynamic acquisition, and is not limited to scanners with long axial field-of-view such as uEXPLORER but can also be applied on conventional scanners. The key concept underlying the proposed framework is to model the individual differences based on the knowledge of normative modeling using a control database. The implementation of the proposed framework was demonstrated and validated in different diseases.