Multi-organ metabolic connectivity mapping in advanced pulmonary arterial hypertension: a group-level 18 F-FDG PET-controlled study Authors: Marc-Antoine Rey, Suzanne Varet, Anne Roche, Jérôme Ghidaglia, Remy Mas, Yannick Jentreau, David Montani, Stéphane Onephandara, Marc Humbert, Aude Amoin, Laurent Savale, Sylvain Faure, Florent L Besson Published: 2026 ABSTRACT AbstractBackgroundTo explore potential systemic alterations in carbohydrate metabolism associated with advanced pulmonary arterial hypertension (PAH) using group-level PET-based metabolic connectivity mapping.MethodsThis retrospective, controlled study analysed 18F-FDG PET-CT scans from 75 individuals (29 with PAH, 46 controls). Eleven major organs were segmented using AI-based tools and voxel-level PET data were extracted. Inter-organ metabolic profiles in PAH and control groups were evaluated using a Maximum mean discrepancy (MMD) framework with extensive permutation testing (1,000,000 permutations) to assess intra-group homogeneity and detect between-group distributional differences. Upon confirmation of within-group homogeneity, organ-level metabolic connectomes were derived from Spearman correlation matrices, with Holm-adjusted multiple testing correction and significance filtering (t-test, p < 0.05). Between-group comparisons were similarly performed using MMD, flowed by post-hoc organ-pair analyses and construction of differential connectomes to localize statistical metabolic networks divergence.ResultsStandard SUV analysis revealed no significant intergroup differences across most organs except for increased uptake in the right heart in PAH patients. MMD testing confirmed intra-group homogeneity in both controls (0,0048) and PAH (0,0465), with no rejection of H0 at α = 0.05, while demonstrating significant between-group differences (H0 rejected). Spearman-based PET connectomes, retaining only significant correlations (𝜌 ≠ 0; p < 0.05) revealed perturbated metabolic network in PAH. This altered network involved the heart, adipose tissue, liver, spleen, muscle, and bone marrow.ConclusionGroup-level whole-body 18F-FDG PET connectivity analysis may provide additional insights into systemic metabolic alterations in advanced PAH that are not readily captured by conventional regional SUV assessments. These findings suggest that PET-based connectomics could complement existing methods for assessing metabolic involvement in PAH.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12880-026-02510-0.Keywords: 18F-FDG PET, Automated segmentation, Connectivity mapping, Pulmonary arterial hypertension Background Pulmonary hypertension (PH) is defined by a mean pulmonary artery pressure (mPAP) > 20 mmHg at rest, confirmed by right heart catheterization. Among the five PH classifications, Group 1 pulmonary arterial hypertension (PAH) is a progressive disease affecting the pulmonary arterioles [1]. It leads to increased pulmonary vascular resistance and right heart failure. PAH may be idiopathic (IPAH), heritable (HPAH), or associated with conditions like connective tissue disease, congenital heart defects, HIV, and drug exposure. PAH is rare but severe, with a prevalence of 48–55 per million adults. Clinical symptoms are nonspecific—dyspnea, fatigue, and syncope—and diagnosis is often delayed. Diagnosis involves clinical evaluation, echocardiography, pulmonary function tests, and definitive RHC [1]. Risk stratification tools like REVEAL and ESC/ERS guides help assess prognosis and direct therapy, focusing on lung and heart dysfunctions [1]. Although most patients are managed with combination therapy, management strategy underestimate its systemic nature, in which multi-organ dysregulation and systemic inflammation appear key factors in disease progression and clinical deterioration [2, 3]. In the era of precision medicine, such consideration implies to develop holistic tools of the disease process to meet the personalized treatment approach. Over the past three decades, positron emission tomography (PET) has revolutionized the diagnostic landscape of numerous oncologic and chronic inflammatory diseases. Although early seminal studies demonstrated the utility of FDG-PET in identifying right ventricular dysfunction in pulmonary arterial hypertension (PAH) [4–7], its application has largely remained limited to regional assessments. However, recent advancements in data science and the development of artificial intelligence (AI)-driven tools now enable large-scale, automated extraction of quantitative, multi-tissue PET metrics at the whole-body level [8]. This technological shift represents a major paradigm change, transitioning PET imaging from an organ-centric approach to a more holistic, systems-level analysis. Functional metabolic network modeling—initially developed in the field of computational neuroimaging—is increasingly being applied to extracerebral domains. These approaches allow for the characterization of inter-organ metabolic interactions across the entire body, forming the basis of a new analytical framework known as PET connectomics [9–12]. By capturing systemic metabolic alterations linked to various pathophysiological conditions, PET connectomics provides emerging biomarkers of global homeostatic regulation and dysregulation. These systemic biomarkers have shown promise across a range of chronic diseases, underpinned by a robust theoretical rationale. Despite growing recognition of PAH as a systemic disease with multi-organ involvement, the application of whole-body PET connectivity analysis in this context remains unexplored. This study aims to evaluate, at the group level, the utility of whole-body PET connectomics for characterizing inter-organ metabolic dysregulation in PAH. Methods The present study was conducted in compliance with the declaration of Helsinki, and was approved by our Institutional Review Board - the Research Ethics Committee of Université Paris-Saclay (cer.polethis@universite-paris-saclay.fr; ref: CER-PARIS-SACLAY 2024-77). According to the rules of our institution, all patients are systematically informed for data collection and research purposes. According to our institutional review board, the included patients were specifically informed. In addition, all patients were enrolled in the registry of the French PH Network. This registry was set up in agreement with French Bioethics laws (Commission Nationale de l’Informatique et des Libertés; 842063). All patients provided written informed consent/assent prior to genetic analysis. Population study The 18F-FDG PET data of twenty-nine patients with advanced PAH (heritable, idiopathic or congenital heart disease exclusively) who underwent PET-CT in our department to rule-out any contraindication before transplantation was retrospectively collected. PET data from forty-six control patients—matched for age, with no history of PAH, no abnormalities on PET scans, and no active medical conditions within 1.5 months prior to the PET—were identified using Cohort360, a large-scale medical data querying tool developed by our institution, and served as controls. All examinations were performed using the same integrated PET-CT scanner (Biograph mCT FlowMotion, Siemens Healthineers, Erlangen, Germany). All the PET data were reconstructed using the same 3D iterative algorithm (3D time-of-flight–ordered subset expectation maximization method, two iterations and 21 subsets with time-of-flight and point spread function modeling, and with random, dead time, scatter, decay, and attenuation corrections; matrix size = 400 × 400), with postfiltering (Gaussian filter: 3.0 mm). All patients fulfilled the international procedure guidelines for FDG PET imaging, fasting for 6 h and ensuring blood glucose levels were below 11 mmol/L at the time of the imaging procedure. The whole-body PET-CT scan were performed from vertex to mid-thigh, starting 60 min after the intravenous injection of 3.5 MBq/kg of 18F-FDG. Image processing and statistical analysis An overview of image processing is provided in Fig. 1. all the PET-CT series were spatially resampled within the PET native spaces and segmented using TotalSegmentator version 2, an open-source tool for robust and automated segmentation of anatomical structures (https://github.com/wasserth/TotalSegmentator [13]. The metabolic SUV data were extracted voxel-wise within 11 predefined organs: brain, left heart, right heart, lungs, liver, spleen, axial skeleton, appendicular skeleton, subcutaneous adipose tissue (SAT), visceral adipose tissue (VAT), and skeletal muscles. For each patient, voxel-wise SUV data were filtered to exclude outliers below the 25th percentile minus 1.5 times the interquartile range and above the 75th percentile plus 1.5 times the interquartile range and were averaged at the organ level. Image processing workflow Evaluation of inter-organ metabolic profiles in the PAH and control groups, and their differences, was performed as follows: Within-group analyses Assessment of intra-group homogeneity: the assumption of homogeneity in the distribution of metabolic activities within each group was evaluated using the Maximum Mean Discrepancy (MMD) framework [14] (see Supplementary MMD). Each group was randomly partitioned into two subgroups of equal size, and the metabolic distributions across the 11 organs were compared using the MMD test. To approximate the probability distribution of MMD under H₀ (homogeneity), 1,000,000 permutations were performed. An observed MMD value exceeding the corresponding quantile of this distribution led to rejection of H₀. This procedure was repeated 10,000 times. With a type I error rate (α) set at 0.05, intra-group homogeneity was rejected when the ratio (number of rejected MMD tests / N = 10,000 procedures) exceeded α.Inter-organ PET-derived metabolic connectomes.When intra-group homogeneity was not rejected, pairwise Spearman correlation matrices were computed across organs within each group. The p-values for the correlations were estimated using the asymptotic distribution of the correlation coefficient. Resulting p-values were adjusted for multiple comparisons using the Holm method. Finally, correlations were then filtered to retain only statistically significant, non-zero associations (ρ ≠ 0; p < 0.05). Connectomes were subsequently constructed for each group, in which nodes represent organs and edges represent the strength of inter-organ correlations. Within-group analyses Assessment of intra-group homogeneity: the assumption of homogeneity in the distribution of metabolic activities within each group was evaluated using the Maximum Mean Discrepancy (MMD) framework [14] (see Supplementary MMD). Each group was randomly partitioned into two subgroups of equal size, and the metabolic distributions across the 11 organs were compared using the MMD test. To approximate the probability distribution of MMD under H₀ (homogeneity), 1,000,000 permutations were performed. An observed MMD value exceeding the corresponding quantile of this distribution led to rejection of H₀. This procedure was repeated 10,000 times. With a type I error rate (α) set at 0.05, intra-group homogeneity was rejected when the ratio (number of rejected MMD tests / N = 10,000 procedures) exceeded α.Inter-organ PET-derived metabolic connectomes.When intra-group homogeneity was not rejected, pairwise Spearman correlation matrices were computed across organs within each group. The p-values for the correlations were estimated using the asymptotic distribution of the correlation coefficient. Resulting p-values were adjusted for multiple comparisons using the Holm method. Finally, correlations were then filtered to retain only statistically significant, non-zero associations (ρ ≠ 0; p < 0.05). Connectomes were subsequently constructed for each group, in which nodes represent organs and edges represent the strength of inter-organ correlations. Assessment of intra-group homogeneity: the assumption of homogeneity in the distribution of metabolic activities within each group was evaluated using the Maximum Mean Discrepancy (MMD) framework [14] (see Supplementary MMD). Each group was randomly partitioned into two subgroups of equal size, and the metabolic distributions across the 11 organs were compared using the MMD test. To approximate the probability distribution of MMD under H₀ (homogeneity), 1,000,000 permutations were performed. An observed MMD value exceeding the corresponding quantile of this distribution led to rejection of H₀. This procedure was repeated 10,000 times. With a type I error rate (α) set at 0.05, intra-group homogeneity was rejected when the ratio (number of rejected MMD tests / N = 10,000 procedures) exceeded α. Assessment of intra-group homogeneity: the assumption of homogeneity in the distribution of metabolic activities within each group was evaluated using the Maximum Mean Discrepancy (MMD) framework [14] (see Supplementary MMD). Each group was randomly partitioned into two subgroups of equal size, and the metabolic distributions across the 11 organs were compared using the MMD test. To approximate the probability distribution of MMD under H₀ (homogeneity), 1,000,000 permutations were performed. An observed MMD value exceeding the corresponding quantile of this distribution led to rejection of H₀. This procedure was repeated 10,000 times. With a type I error rate (α) set at 0.05, intra-group homogeneity was rejected when the ratio (number of rejected MMD tests / N = 10,000 procedures) exceeded α. Inter-organ PET-derived metabolic connectomes.When intra-group homogeneity was not rejected, pairwise Spearman correlation matrices were computed across organs within each group. The p-values for the correlations were estimated using the asymptotic distribution of the correlation coefficient. Resulting p-values were adjusted for multiple comparisons using the Holm method. Finally, correlations were then filtered to retain only statistically significant, non-zero associations (ρ ≠ 0; p < 0.05). Connectomes were subsequently constructed for each group, in which nodes represent organs and edges represent the strength of inter-organ correlations. Inter-organ PET-derived metabolic connectomes. When intra-group homogeneity was not rejected, pairwise Spearman correlation matrices were computed across organs within each group. The p-values for the correlations were estimated using the asymptotic distribution of the correlation coefficient. Resulting p-values were adjusted for multiple comparisons using the Holm method. Finally, correlations were then filtered to retain only statistically significant, non-zero associations (ρ ≠ 0; p < 0.05). Connectomes were subsequently constructed for each group, in which nodes represent organs and edges represent the strength of inter-organ correlations. Between-group analysesBetween-group differences in the distribution of metabolic activities were assessed using the MMD framework (1,000,000 permutations; single procedure). Rejection of the null hypothesis H₀ indicated between-groups non homogeneity. In such cases, post hoc MMD tests were conducted for each organ pair to identify the specific contributors to these differences. An MMD-based connectome was then generated, along with connectomes representing between-group differences in inter-organ correlations for organ pairs in which H₀ was rejected. Between-group analyses Between-group differences in the distribution of metabolic activities were assessed using the MMD framework (1,000,000 permutations; single procedure). Rejection of the null hypothesis H₀ indicated between-groups non homogeneity. In such cases, post hoc MMD tests were conducted for each organ pair to identify the specific contributors to these differences. An MMD-based connectome was then generated, along with connectomes representing between-group differences in inter-organ correlations for organ pairs in which H₀ was rejected. All the statistical analyses were performed with Python (version 3.6; Python Software Foundation, www.python.org). Results Patients’ characteristics and data extraction The general characteristics of the PAH and control patients are summarized in Table 1. Briefly, the two groups were comparable in terms of age, weight, body mass index and creatinin (p ≥ 0.05). A higher prevalence of females was observed in the PAH group (59% vs. 35%), while diabetes mellitus was more frequent in the control group (13% vs. 3%). In both groups, millions of voxels were automatically segmented at the patient level, ranging from a few tens of thousands for organs such as the heart and spleen to several million for muscle tissue. Table 1Patients’ characteristics at the time of PET-CTPAH (n = 29)Controls (n = 46) Age - years 47.6 ± 1152 ± 26.2 Female sex – n (%) 17 (59%)16 (35%) Weight - kg Body mass index – kg/m 2 67.8 ± 13.924 ± 3.974.9 ± 17.525.6 ± 5.4 Comorbidities – n (%) - Hypertension- Diabetes mellitus- Coronary artery disease- Obesity1 (3)1 (3)1 (3)2 (7)11 (24)6 (13)0 (0)10 (22) PAH aetiology – n (%) Not applicable - Idiopathic- Heritable- Congenital Heart disease6 (21)15 (52)8 (27) NYHA FC II/III/IV – n (%) 6-min walk distance - m 11(38) / 17(59) / 1(3)441 (355–504)* Not applicable Hemodynamics Not applicable - RAP – mmHg- PAWP – mmHg- mPAP – mmHg- Cardiac index – l/min/m2- PVR - WU8 ± 411 ± 3*60 ± 142.5 ± 0.7*12.3 ± 6.2 Biological variables - NTproBNP- Creatinin - µmol/L- eGFR – mL/min/1.73m2- Total bilirubin – mmol/L- Glycaemia (mmol/l)339 (112 − 34)*80 ± 2088 ± 208 ± 45.3 ± 0.9 Not applicable 77 ± 35**103 ± 29** Not applicable 5.2 ± 0.8 Volume segmented per patient (cm 3 ) - Brain- Right Heart- Left Heart- Lungs- Axial skeleton- Appendicular skeleton- Liver- Spleen- SAT- VAT- Muscles1743.6 ± 228.3538.7 ± 173.6185.9 ± 57.02343.9 ± 849.03021.0 ± 430.52078.8 ± 448.21965.1 ± 862.0337.3 ± 196.621498.9 ± 10673.52189.4 ± 1442.819061.8 ± 5082.81828.3 ± 232.1283.8 ± 78.1134.6 ± 37.52060.9 ± 813.93234.1 ± 581.62309.5 ± 544.92038.8 ± 411.8279.3 ± 133.523930.6 ± 13554.73594.6 ± 2478.022967.9 ± 7524.8Data are expressed as mean ± SD or median with interquartile range (IQR, 25–75%) depending on their distribution and availability*PAWP: 5 missing values; CI: 4 missing values; 6-min walk distance: 6 missing values; NT-proBNP: 1 m** Creatinin: 11 missing values; eGFR: 12 missing values Patients’ characteristics at the time of PET-CT Weight - kg Body mass index – kg/m 2 67.8 ± 13.9 24 ± 3.9 74.9 ± 17.5 25.6 ± 5.4 - Hypertension - Diabetes mellitus - Coronary artery disease - Obesity 1 (3) 1 (3) 1 (3) 2 (7) 11 (24) 6 (13) 0 (0) 10 (22) - Idiopathic - Heritable - Congenital Heart disease 6 (21) 15 (52) 8 (27) NYHA FC II/III/IV – n (%) 6-min walk distance - m 11(38) / 17(59) / 1(3) 441 (355–504)* - RAP – mmHg - PAWP – mmHg - mPAP – mmHg - Cardiac index – l/min/m2 - PVR - WU 8 ± 4 11 ± 3* 60 ± 14 2.5 ± 0.7* 12.3 ± 6.2 - NTproBNP - Creatinin - µmol/L - eGFR – mL/min/1.73m2 - Total bilirubin – mmol/L - Glycaemia (mmol/l) 339 (112 − 34)* 80 ± 20 88 ± 20 8 ± 4 5.3 ± 0.9 Not applicable 77 ± 35** 103 ± 29** Not applicable 5.2 ± 0.8 - Brain - Right Heart - Left Heart - Lungs - Axial skeleton - Appendicular skeleton - Liver - Spleen - SAT - VAT - Muscles 1743.6 ± 228.3 538.7 ± 173.6 185.9 ± 57.0 2343.9 ± 849.0 3021.0 ± 430.5 2078.8 ± 448.2 1965.1 ± 862.0 337.3 ± 196.6 21498.9 ± 10673.5 2189.4 ± 1442.8 19061.8 ± 5082.8 1828.3 ± 232.1 283.8 ± 78.1 134.6 ± 37.5 2060.9 ± 813.9 3234.1 ± 581.6 2309.5 ± 544.9 2038.8 ± 411.8 279.3 ± 133.5 23930.6 ± 13554.7 3594.6 ± 2478.0 22967.9 ± 7524.8 Data are expressed as mean ± SD or median with interquartile range (IQR, 25–75%) depending on their distribution and availability *PAWP: 5 missing values; CI: 4 missing values; 6-min walk distance: 6 missing values; NT-proBNP: 1 m ** Creatinin: 11 missing values; eGFR: 12 missing values Two illustrative cases are presented in Fig. 2, highlighting the whole-body physiological biodistribution of the FDG radiotracer in both groups. Visual qualitative analysis revealed comparable FDG uptake across the body, except in the cardiac region, where uptake in the PAH group was markedly increased. The voxel-wise distribution of standardized uptake values (SUVs) per organ is summarized in Table 2; Fig. 3. In the control group, SUV values ranged from 0.4 ± 0.1 (subcutaneous adipose tissue, SAT) to 6.1 ± 2.0 (brain), while in the PAH group, values ranged from 0.35 ± 0.07 (SAT) to 5.9 ± 1.0 (brain). In both groups, the brain, left and right heart, axial skeleton, liver, and spleen exhibited SUV values above 1.0, whereas the lungs, SAT, visceral adipose tissue (VAT), muscles, and skeleton showed SUV values at or below 1.0. The only statistically significant differences between groups were observed in the right heart, where SUVs were significantly higher in the PAH group (p < 0.05). Maximum intensity projections of illustrative cases of control and PAH patients. Left: control patient without pulmonary arterial hypertension (PAH), a 29-year-old male referred to our PET unit to investigate possible large vessel vasculitis in the context of cerebral vasculitis. No abnormal uptake was detected. Right: patient with advanced PAH, a 45-year-old male referred to our PET unit to rule out any active disease prior to transplantation. The cardiac region shows intense and diffuse 18F-FDG uptake. No other abnormal uptake is visually observed Table 2Values of SUV per organ and between group difference (t-test)ControlsPAHt valuep valueMean SUVSDmean SUVSDBrain6.12.05.91.00.720.47Lungs0.50.20.60.1-0.830.40Skeletonaxial1.20.21.20.20.650.52Skeletonappendicular1.00.21.00.20.001.00Spleen1.70.41.70.20.270.79Liver2.20.42.10.31.300.20SAT0.40.10.40.11.920.06VAT0.60.20.70.2-1.750.08Muscle0.70.10.70.10.370.71 Right heart* 1.7 0.3 2.6 1.3 -3.83 0.0006 Left heart2.11.32.71.4-1.780.08*Between group difference are statistically significant (p value < 0.05) Values of SUV per organ and between group difference (t-test) *Between group difference are statistically significant (p value < 0.05) SUV boxplots per organ. Statistically significant differences between groups were observed only in the right heart. Axial SK = axial skeleton; App SK = appendicular skeleton; SAT: subcutaneous fat; VAT: visceral fat Within-group analyses The assessment of intra-group homogeneity yielded the following results: in the control group, the proportion of rejected MMD tests (number of rejections out of N = 10,000 procedures) was 0.0448, compared with 0.0465 in the PAH group. In both groups, the null hypothesis (H0) was not rejected at a type I error rate of α = 0.05, supporting the assumption of distributional homogeneity of metabolic activities within each group. Under this assumption, all parameters characterizing the distributions are identical within each group, particularly the correlations. Accordingly, pairwise Spearman correlation matrices were computed across organs at the group level. Figure 4 presents the organ-wise connectivity strength profiles—i.e., PET-derived metabolic connectomes—for both groups, corrected for multiple comparisons and filtered to retain only statistically significant, non-zero interactions (𝜌 ≠ 0; p < 0.05). In the control group (Fig. 4A), significant pairwise correlations were observed for the majority of organ pairs, forming a diffuse, star-like pattern of metabolic connectivity with Spearman 𝜌 values generally exceeding 0.5, except for subcutaneous adipose tissue (SAT) and visceral adipose tissue (VAT). In contrast, the PAH group (Fig. 4B) exhibited a distinct connectivity pattern, characterized by strong associations between the right and left heart, as well as between the lungs and several organs, including the liver, bone marrow (both axial and appendicular skeleton), and spleen. Group-level PET connectome profiles. These connectomes show the organ pairwise correlations which are statistically significant within each group (ρ ≠ 0; p-value < 0.05). The color scale on the right represents the strength of interaction (ρ). A. Controls. B. PAH Between-groups analyses The MMD test assessing between-group homogeneity yielded a test statistic of 0.052 (p = 0.01), leading to rejection of the null hypothesis at a significance level of α = 0.05. These results indicate that the two groups differ significantly in overall metabolic activity. Post-hoc MMD-based connectome analysis, together with connectomes representing between-group differences in organ-wise correlations (Δ connectomes), is presented in Fig. 5. MMD analyses revealed significant between-group discrepancies in organ network structure, primarily involving the heart, liver, spleen, SAT, VAT and bone marrow (axial and appendicular skeleton), for which the null hypothesis was rejected (Fig. 5A). Corresponding differences in organ-wise correlations showed generally higher connectivity in controls across most organ pairs, with a few exceptions in which correlations were higher in the PAH group, notably lungs – VAT, lung – left heart, appendicular skeleton – VAT and skeletal muscle – VAT interactions (Fig. 5B). After filtering non-significant correlation differences using Fisher z-transformation, persistent between-group differences remained in metabolic network interactions involving the heart, liver, spleen, adipose tissue, bone marrow (Fig. 5C). Δ connectomes between the two groups. Between-group differences in organ pairwise connectivity strengths. A. Connectome with MMD’s values. The color scale on the right represents the MMD values – MMD > 0.03 (95-percentile estimation) reject H0, meaning the organ pair in the two groups is non-similar. B. Connectome with correlations differences among organ pairs rejecting H0 C. Connectome with Fisher z-transformed correlations differences among organ pairs rejecting H0. In both cases, the color scale on the right represents the between-group difference in correlations values. Red indicates organ pairs with lower correlation in the PAH group compared to the control group, while blue indicates higher correlation in the PAH group relative to controls Discussion PET metabolic connectivity mapping is emerging as a promising approach at the whole-body level. By leveraging voxel-wise data extraction with advanced AI-based segmentation tools, we captured whole-body SUV distributions and evaluated, at the group level, the potential of PET connectomics to characterize alterations in organ metabolic networks in advanced PAH. To this end, (i) a total of 10,000 within-group statistical analyses were conducted using MMD tests with one million permutations to assess intra-group organ-wise metabolic homogeneity; (ii) between-group statistical analyses were performed using the same permutation-based MMD framework to evaluate inter-group differences in organ-wise metabolism; and (iii) differences in organ pairwise connectivity strengths between groups were modelized using PET-derived connectomes. These statistically significant between-group differences (𝜌 ≠ 0; p < 0.05) were primarily observed in the metabolic networks of the heart, subcutaneous adipose tissue, visceral adipose tissue, liver, spleen, muscle, and bone marrow. Notably, connectivity strengths within these organ-specific networks were higher in the control group compared with the PAH group (see Supplementary Correlations), suggesting disrupted or attenuated metabolic interactions in PAH. While the connectivity alterations between groups were statistically significant, organ-level radiotracer uptake did not differ — except in the right cardiac region. A diffuse increase in FDG uptake within the right heart is a well-established indicator of cardiac remodeling in advanced PAH. It may predict long-term prognosis [4, 15–17] and has been shown in some studies to correlate with exercise capacity, NT-pro BNP levels, and right ventricular function [6, 7, 18]. Beyond right ventricular dysfunction, chronic systemic multi-organ dysregulation and systemic inflammation are increasingly recognized as key factors in disease progression and clinical deterioration [2, 3]. While targeted therapies remain central to management, the disease trajectory may also be modulated by pharmacological, interventional, or supportive strategies. Early recognition of these often-overlooked systemic alterations offers promising opportunities for adjunctive therapeutic interventions within the framework of precision medicine [3]. In our study, the metabolic connectivity strengths of the heart, adipose tissue, bone marrow (skeleton), muscle, liver, and spleen networks showed clear differences between the PAH and control groups. Notably, the network-level differences observed in the PAH group strongly echo several systemic hallmarks of the disease, including cardiac deconditioning and dysfunction [19–22], liver impairment [23], disturbances in iron homeostasis [24, 25], skeletal muscle dysfunction [26], and metabolic syndrome [27–29]. Given that the precise mechanisms underlying these systemic alterations in PAH remain poorly understood, further research is essential to clarify the prognostic value of these PET-derived network signatures. Recently, Sun and coworkers proposed a perturbation-based covariance network to discriminate, at both the group and individual levels, multi-organ cross-talk alterations in few mixed miscellaneous conditions (10 lung cancers, 1 covid-19 discharge and 1 gut bleeding [10]. Such covariance network approach was further applied on 26 overweight patients, exhibiting altered metabolic interactions at the body level compared to normal weight patients, in particular pancreas, spleen, VAT and bone [11]. In both cases, several regions of interest were semi automatically drawn within key organ structures. In both cases, multiple regions of interest were semi-automatically delineated within key organ structures. When applying these approaches alongside whole-body, large-scale data extraction to our dataset, we were unable to identify robust PAH perturbation networks, highlighting the need for tailored-made methodologies. In our study, we employed TotalSegmentator, a robust and validated AI-based morphological segmentation tool [13, 30, 31]. To minimize the risk of urinary tracer contamination, which was frequently observed, the kidneys were excluded from the analysis, thereby precluding the evaluation of kidney–heart crosstalk. As the PET-CT series were spatially resampled within the native PET space, quality control was performed across the entire dataset by three independent readers under blinded conditions, using a predefined 5-point Likert scale (1 = very poor, 2 = poor, 3 = acceptable, 4 = good, 5 = very good). Inter-reader agreement was optimal (see Supplementary Tables 1, 2 and 3), supporting the robustness of the segmentation procedure. The bootstrap framework within each group further reinforced the consistency and robustness of our results in this particular clinical context. Our study has several limitations. The dataset was limited and retrospective, conducted at a single center. However, the specialized expertise of our tertiary care center in PAH ensured a well-defined patient selection, all belonging to the same pathophysiological group (Group I). The control group included patients with various clinical indications. Due to ethical concerns related to radiation exposure, inclusion of healthy volunteers was not feasible. To address this, we selected clinically normal PET scans, confirmed their normality through careful review, and verified that no medical events occurred within 1.5 months post-scan. This multi-step screening process supports that the control group was in a physiological state. We did not adjust our analyses for potential confounding factors such as age, BMI, sex, or diabetes. However, the patient and control groups were matched for age, and there were no significant differences in weight or BMI, suggesting these factors are unlikely to have influenced the group-level analyses. Sex distribution differed slightly between groups, which may represent a study limitation. Notably, Geist et al. [32] compared networks from 55 male and 55 female healthy subjects matched for BMI, age, glucose, and CRP levels and found no significant differences between the networks, suggesting minimal sex-related effects. Additionally, 91% of our dataset was unaffected by diabetes, further minimizing its potential impact on group-level analyses. Finally, individual-level assessment of connectivity networks was not feasible; we are currently developing methods to translate group-level analyses into robust, individual multi-organ connectivity metrics. Conclusions To conclude, this study demonstrates the effectiveness of FDG PET connectomics in characterizing group-level fingerprints of altered whole-body metabolic networks in advanced PAH. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements During the preparation of this manuscript, the authors used ChatGPT 3.5 for language editing. After utilizing this tool, the authors carefully reviewed and revised the content as needed and take full responsibility for the final publication. Abbreviations Artificial intelligence Cardiac index Cardiac output Estimated glomerular filtration rate Fluorodeoxyglucose Maximum Mean Discrepancy Mean pulmonary arterial pressure Pulmonary arterial hypertension Pulmonary arterial wedge pressure Pulmonary hypertension Right atrial pressure Pulmonary vascular resistance Subcutaneous fat Standardized uptake value Visceral adipose tissue World Health Organization (WHO) functional class 6 minutes’ walk test Author contributions All authors contributed to the study conception and design. J.G, Y.J, S.O, A.A have made substantial contributions to the acquisition of the data. Material preparation, data collection and analysis were performed by M.A.R, F.L.B, S.V, S.F, A.R, R.M, D.M, M.H and L.S. The first draft of the manuscript was written by F.L.B and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript. Funding The authors declare that no funds, grants, or other support were received during the preparation of this manuscript. Data availability The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Declarations Ethical approval and consent to participate The present study was conducted in compliance with the declaration of Helsinki, and was approved by our Institutional Review Board - the Research Ethics Committee of Université Paris-Saclay (cer.polethis@universite-paris-saclay.fr; ref: CER-PARIS-SACLAY 2024-77). According to the rules of our institution, all patients are systematically informed for data collection and research purposes. According to our institutional review board, the included patients were specifically informed. In addition, all patients were enrolled in the registry of the French PH Network. This registry was set up in agreement with French Bioethics laws (Commission Nationale de l’Informatique et des Libertés; 842063). All patients provided written informed consent/assent prior to genetic analysis. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Marc-Antoine Rey and Suzanne Varet contributed equally to this work. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.