CN114533102B - A method to investigate systemic metabolic abnormalities at the individual level using whole-body SUV images - Google Patents

Abstract

The present invention discloses a method for investigating individual-level systemic metabolic abnormalities using whole-body SUV images. , through which sampling areas were selected; the subjects included healthy controls, lung cancer groups, subjects 30 days after Covid‑19 discharge, and subjects with unexplained gastrointestinal bleeding; construct The individual connection network of different subjects, and the degree of metabolic abnormality in each sampling area of different subjects is obtained through the individual connection network of different subjects; the control group is determined by the degree of metabolic abnormality in each sampling area of different subjects Homogeneity analysis, lung cancer group heterogeneity analysis, group- and individual-level network analysis, and individual network and single-organ analysis. The methods of the present invention can potentially identify systemic metabolic abnormalities.

Description

A method to investigate whole-body metabolic abnormalities at the individual level using whole-body SUV images

technical field

The invention belongs to the technical field of medical image processing, and in particular relates to a method for investigating an individual-level systemic metabolic abnormality using a whole-body SUV image.

Background technique

Human metabolic homeostasis relies on complex neuronal, vascular and humoral mechanisms at the systemic level. Simultaneous nonlinear interactions between organs form distinct physiological networks. Many systemic diseases are due to or related to the interference of physiological interactions between organs. Although there are established methods for studying this interference at the organ level, further development of a general method sufficient to quantify at the system level presents challenges.

To date, most studies on these topics have used non-imaging tools. Thiele et al. developed a metabolic network reconstruction method using organ-specific information from literature and omics data. Includes data sources for 20 organs, 6 sex organs, 6 blood cell types, and 13 biofluidic compartments. Barajas-Martínez et al. introduced physiological networks based on anthropometric measurements, fasting blood tests and other vital signs. They concluded that specific structural properties of the network will change throughout the human lifespan and provide indicators of health status. Cui et al. reconstructed global mammalian metabolic networks in tissues and cell types and attempted to link organ-to-organ metabolite transport. Bashan et al. and Bartsch et al. developed a framework to probe interactions between different systems and identify a physiological network that exhibits interactions between network topology and function. But none of the above methods can be quantified at the system level.

SUMMARY OF THE INVENTION

Aiming at the problem that the existing metabolic abnormality analysis methods cannot be quantified at the system level, the present invention proposes a method for investigating the systemic metabolic abnormality at the individual level by using the whole body SUV image.

In order to achieve the above object, the present invention adopts the following technical solutions:

A method of investigating whole-body metabolic abnormalities at the individual level using whole-body SUV images, including:

Step 1, data collection and processing: collect 18F-FDG PET/CT scan data of different subjects in different sampling areas, obtain images of different subjects with standard uptake value SUV, and select sampling areas through the images; The subjects included a healthy control group consisting of subjects without any disease records, a lung cancer group consisting of subjects diagnosed with lung cancer, with different lesion sites, subjects 30 days after discharge from Covid-19, reasons Subjects with unidentified gastrointestinal bleeding;

Step 2, construct the individual connection network of different subjects, and obtain the metabolic abnormality degree of each sampling area of different subjects through the individual connection network of different subjects;

In step 3, the homogeneity analysis of the control group, the heterogeneity analysis of the lung cancer group, the network analysis at the group and individual levels, and the individual network and single-organ analysis were carried out through the metabolic abnormality degree of each sampling area of different subjects.

Further, the step 1 includes:

Step 1.1: CT scan is performed first for attenuation correction, followed by 18F-FDG PET/CT acquisition; scan data is reconstructed into a voxel-sized fixed matrix using a 3D ordered subset expectation maximization algorithm; CT-based attenuation correction is used Attenuation and scatter correction were performed on the map; the reconstructed active images were then converted to images with standard uptake values SUV by normalizing injected dose and weight;

Step 1.2: For each scan, delineate the sampling area of interest for all organs of interest on the SUV image; the sampling area includes whole brain, blood, left ventricle, lung, liver, pancreas, spleen, left/right kidney, muscle and spine; brain cell division using statistical parametric mapping, extraction of planes containing brains in reconstructed images as new volumes and spatial normalization in Montreal Neurological Institute space; normalized images were subjected to Gaussian filters Smoothed and then divided into regions defined by the AAL2 atlas, of which brainstem, whole cerebellum, cerebrospinal fluid, whole white matter, caudate nucleus, putamen and frontal cortex were selected as new sampling areas, along with the delineated organs, each time A total of 18 sampling areas were scanned for analysis.

Further, the step 2 includes:

Step 2.1, construct a reference metabolic network refNET from the healthy control group, which is obtained by calculating the Pearson correlation coefficient between the SUVs of each region pair, using a covariance network structure;

Step 2.2: Add a patient to the healthy control group to form a new group to construct a new structural covariance network, which is labeled as the perturbation network ptbNET;

In step 2.3, the difference between the perturbed network ptbNET and the reference network refNET is calculated as the residual network resNET; a threshold of 0.3 is set to eliminate weak correlations; the Z-score map of the residual network resNET is obtained:

where N is the total number of subjects in the new group, the remaining networks essentially represent abnormal levels of connectivity, and each network consists of 153 edges connecting 18 regions, each edge exhibiting varying degrees of metabolic changes; to quantify the degree of abnormality , which defines the intensity STR of each regional anomaly:

是集合，

M＝18是区域数，ZCC_mi表示区域m与其相邻区域i之间的Z分数图的相关系数，相邻节点的总数等于m-1。where m is the region index number,

is the set,

M=18 is the number of regions, ZCC _mi represents the correlation coefficient of the Z-score graph between region m and its adjacent region i, and the total number of adjacent nodes is equal to m-1.

Further, in the step 3, the homogeneity analysis of the control group is carried out as follows:

Individual network analyses were performed on subjects in each healthy control group to measure within-group similarity by averaging the Pearson correlation coefficient of Z-scores between any pair of networks, followed by resampling to test reproducibility of subject selection .

Further, in the step 3, the lung cancer group heterogeneity analysis is performed as follows:

The strength of each lung cancer patient's personal network was compared to the strength in the reference network, and the difference between the lung cancer patient networks was calculated by measuring the mean of the between-subject Pearson correlation coefficients between paired Z-scores on all 153 edges. similarity between.

Further, in the step 3, network analysis at the group and individual levels is carried out as follows:

Group-level metabolic networks were constructed for the patient group and healthy control group, respectively, and the normalization between the two group-level metabolic networks was constructed by calculating the Pearson correlations of pairs of brain regions of all subjects in the two groups. The normalized difference is treated as a group-level difference network Diff _{group, i} :

where i represents the edge, patNET represents the group-level metabolic network of the patient group, and refNET represents the group-level metabolic network of the healthy control group;

Construct the mean residual network Diff _individual,i by computing the mean of the Z-scores for each edge in all patient networks:

where j is the patient index number and _Np is the total number of patients;

Calculate the Pearson correlation coefficient between the mean residual network Diff _individual,i and the group-level difference network Diff _group,i .

Further, in the step 3, personal network and single organ analysis are carried out as follows:

Separate connectivity networks were constructed for a subject 30 days post-discharge with Covid-19 and a subject with unexplained gastrointestinal bleeding to quantify changes in SUV based on organ network strength.

Compared with the prior art, the present invention has the following beneficial effects:

The present invention can construct an individual metabolic abnormality network using the whole body PET/CT SUV image of the subject and the healthy control group. The method proposed by the present invention can investigate the systemic metabolic abnormalities at the individual level and characterize the molecular connectivity of glucose metabolism at the individual level, which is not possible with current analytical methods. The methods of the present invention can potentially identify systemic metabolic abnormalities, which cannot be derived by traditional population-level approaches due to the large heterogeneity among lung cancer groups. In addition, the present invention can study how the brain and other organs interact in health and disease conditions from a systems perspective.

Description of drawings

1 is a basic flow chart of a method for investigating an individual-level systemic metabolic abnormality using a whole-body SUV image according to an embodiment of the present invention;

FIG. 2 is a target area of 2D sampling according to an embodiment of the present invention;

3 is an overall framework diagram of an individual metabolic network obtained in an embodiment of the present invention;

4 is a box plot of the connection strength of individual metabolic networks in the lungs of the healthy control group and the lung cancer group according to the embodiment of the present invention and the corresponding SUV values in the two groups of lungs;

Figure 5 shows the individual metabolic connectivity of a subject 30 days after discharge from Covid-19 and a subject with unexplained gastrointestinal bleeding in the embodiment of the present invention;

FIG. 6 is a correlation diagram between the organ |ΔSUV| in the embodiment of the present invention and the network strength calculated by the network in FIG. 4 .

Detailed ways

The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments:

In the present invention, we propose a framework to construct individual metabolic abnormality networks using whole-body PET/CT SUV images of subjects and healthy controls. The implementation details are demonstrated first, followed by verification. Finally, example applications are demonstrated and their performance compared to traditional group-level connectivity and single-organ uptake analysis.

Specifically, as shown in Figure 1, a method for investigating whole-body metabolic abnormalities at the individual level using whole-body SUV images includes:

Step S101, data collection and processing: collect 18F-FDG PET/CT scan data of different subjects in different sampling areas, obtain images of different subjects with standard uptake value SUV, and select the sampling areas through the images; The subjects included a healthy control group consisting of subjects without any disease records, a lung cancer group consisting of subjects diagnosed with lung cancer, with different lesion sites, subjects 30 days after discharge from Covid-19, reasons Subjects with unidentified gastrointestinal bleeding;

Step S102, constructing individual connection networks of different subjects, and obtaining the metabolic abnormality degree of each sampling area of different subjects through the individual connection networks of different subjects;

In step S103, the homogeneity analysis of the control group, the heterogeneity analysis of the lung cancer group, the network analysis at the group and individual level, and the individual network and single organ analysis are performed according to the metabolic abnormality degree of each sampling area of different subjects.

Specifically, this embodiment includes a total of 36 18F-FDG PET/CT scans. Of the subjects, twenty-four were young, healthy subjects with no documented disease; ten were diagnosed with lung cancer with different lesions; one was performed 30 days after discharge from Covid-19; one was unexplained gastrointestinal bleeding. The demographics of these studies are listed in Table 1.

Table 1 Demographics of Subjects

During the data acquisition and processing of step S101, all scans were acquired on a uExplorer PET/CT scanner at the People's Hospital of Henan Province, China. These studies were approved by the local ethics committee. Written consent was obtained from each subject prior to scanning. The scanning process and data format are as follows. A CT scan was first performed for attenuation correction. A 60-minute list-mode PET acquisition was then initiated with a bolus of 18F-FDG injected intravenously from the lower extremity (see Table 1 for the dose of injection). To obtain the SUV images, the scan data over 50-60 minutes was reconstructed into a 192×192×80 matrix of voxel size 3.125×3.125× ^2.866mm3 using the 3D ordered subset expectation maximization (OSEM) algorithm on the workstation . Reconstruction applied 3 iterations, 28 subsets and 2mm Gaussian post-smoothing. Attenuation and scatter correction were performed using a CT-based attenuation correction map. The reconstructed active images (in Bq/cc) were then converted to images with standard uptake values (SUV) by normalizing injected dose and weight.

Regions of interest (ROI) for all organs of interest were manually delineated on the SUV images for each scan. The target area for sampling is shown in Figure 2, with a total of 11 sampling areas, including whole brain, blood, left ventricle, lung, liver, pancreas, spleen, left/right kidney, muscle and spine, where brain refers to the whole brain, The lung refers to the left or right lung where the lesion is located, excluding the lesion itself. Brain cell division was further performed using statistical parametric mapping (SPM12). The plane containing the brain in the reconstructed image was extracted as a new volume and the FDG-PET template was spatially normalized in the Montreal Neurological Institute (MNI) space. Normalized images were smoothed with a Gaussian filter of 8 mm FWHM and then divided into 94 regions defined by the AAL2 atlas. Sub-regions were further extracted from the brain according to the AAL2 atlas defined in the context. In this example, only the brainstem, whole cerebellum (CER), cerebrospinal fluid (CSF), whole white matter (WM), caudate nucleus, putamen and frontal cortex (SF) were selected for subsequent analysis. Along with the organs depicted, a total of 18 sampling areas were analyzed per scan. Network analysis was performed using the Brain Connectivity Toolbox. All statistical analyses were performed using the Statistics and Machine Learning Toolbox in Matlab R2018b.

In step S102, in order to construct an individual connection network, we employ a method for brain anatomy MRI. The overall framework is shown in Figure 3. The basic idea is to obtain the abnormal information by analyzing the interference of the individual sample with the overall control sample set. First, a reference metabolic network, refNET, was constructed from a healthy control group (24 healthy subjects), which was obtained by calculating the partial Pearson correlation coefficient (age, gender as covariates) between the SUVs for each region pair , using a covariance network structure. The node of the network is the target area, and the connection edge is the correlation coefficient between the nodes. The larger the correlation coefficient, the stronger the connection. Generally speaking, the constructed reference network has common characteristics of all control samples. We then added a patient to the healthy control group, thus forming a new group of 25 subjects, to construct a new structural covariance network, labeled the perturbation network ptbNET. Next, the difference between the perturbed network ptbNET and the reference network refNET is calculated as the residual network resNET. A threshold of 0.3 was set to remove weak correlations that might come from noise. Obtained the Z-score graph for resNET:

where N is the total number of subjects in the new group, i.e. the total number of samples. The residual networks essentially represent abnormal levels of connectivity, each consisting of 153 edges connecting 18 regions. Each edge exhibited varying degrees of metabolic changes and resulted in deviations from normal values in the control group. Under normal conditions, the body network structure is stable and connected. When metabolic indices changed in subjects due to disease, network links changed accordingly. To quantify the degree of anomaly, we define the intensity STR of anomalies in each region:

是集合，

M＝18是区域数。ZCC_mi表示区域m与其相邻区域i之间的Z分数图的相关系数。相邻节点的总数等于m-1。where m is the region index number,

is the set,

M=18 is the number of regions. ZCC _mi represents the correlation coefficient of the Z-score map between region m and its neighboring region i. The total number of adjacent nodes is equal to m-1.

In the data and statistical analysis of step S103, we first investigate the overall consistency of the healthy control group by measuring the within-group similarity. Second, the heterogeneity of the individual network of lung cancer patients was confirmed. Third, compare measures of networking at the individual level with measures of networking at the group level. Finally, the ability of individual networks to reveal abnormalities in individual organs was tested by measuring the correlation between SUV and network strength. Details on how to conduct these investigations are listed below.

(1) Homogeneity analysis of the control group

A separate network analysis was performed on subjects in each healthy control group. Within-group similarity was measured by averaging the correlation coefficient of Z-scores between any pair of networks. We then perform a resampling procedure to test the repeatability of topic selection. The idea is that two random groups from the same population should not be different from each other. There were a total of 30 normal samples from which we randomly selected 24 samples (4/5 of 30) as the control group. The sampling process was repeated 20 times. We constructed refNETs from each randomly sampled group using the proposed method. The reproducibility between these refNETs was quantified by averaging the correlation coefficient of Z-scores between any pair of networks. We further performed a resampling test for sensitivity to the sample size of the control group. The idea is to investigate the minimum normal number of subjects needed to construct a control group. We randomly selected 10, 15, 18, 20 and 24 samples (20 times each) from the above 30 normal subjects as the control group. New resNETs were then calculated for each patient in the lung cancer group using the proposed method and compared with resNETs in context.

(2) Heterogeneity analysis of lung cancer group

The strength of each patient's personal network was compared to the strength in the reference network. Similarities between patient networks were calculated by measuring the mean of the between-subject Pearson correlation coefficients between paired Z-scores on all 153 edges.

(3) Network analysis at group and individual levels

We constructed group-level metabolic networks for patients and healthy controls, respectively. Two group-level metabolic networks (both structural covariance networks) were constructed by calculating Pearson correlations for pairs of brain regions for all subjects in the two groups. The normalized differences between the two group-level metabolic networks were regarded as the group-level difference network Diff _{group, i} .

where i denotes edges, patNET denotes the group-level metabolic network of the patient group, and refNET denotes the group-level metabolic network of the healthy control group.

We also construct the mean residual network Diff _{individual, i} by computing the mean of the Z-scores for each edge in the network of all patients:

where j is the patient index number and _Np is the total number of patients.

Calculate the Pearson correlation coefficient between the mean residual network Diff _individual,i and the group-level difference network Diff _group,i .

(4) Personal network and single organ analysis

We constructed a separate network of connections for a subject discharged from Covid-19 and a patient with gastrointestinal bleeding. Changes in SUV were quantified according to the network strength of the organ to reveal the ability of the proposed method to reveal abnormalities at the organ level.

Specifically, the analysis results are as follows:

(1) Homogeneity analysis of the control group

Within-group similarity was measured by averaging the correlation coefficient of Z-scores between any pair of networks. The similarity coefficient was 0.921 ± 0.133, indicating low inter-subject variability among the control group. A resampling process was performed to quantify the reproducibility between these refNETs by averaging the correlation coefficient (0.872 ± 0.152) of Z-scores between any pair of networks. This shows that the method is robust to the selection of control objects. We further performed a resampling test for sensitivity to the sample size of the control group. For sample sizes 10, 15, 18, 20, 22, and 24, the overall mean similarity relative to sample size 30 was 0.58±0.18, 0.70±0.157, 0.79±0.142, 0.89±0.126, 0.92±0.124, and 0.97±0.102, respectively . This shows that as the number of subjects increases, the average similarity increases, while the corresponding variance decreases. The more samples in the control group, the more stable the proposed method. The data of 24 healthy subjects used in this example are sufficient to construct a control group.

(2) Heterogeneity analysis of lung cancer group

The strength of each patient's individual network was significantly different from that in the reference network (P < 0.01, Bonferroni corrected for 153 edges). However, the similarity between patient networks was low, with an average between-subject Pearson correlation coefficient between paired Z-scores for all 153 edges of 0.196 ± 0.182. Despite significant organ-wide heterogeneity in individual patients, the connections that make up the lung are much more abnormal than others (Fig. 4, where A is the individual network connections in the lungs of the control (i.e., healthy) and diseased (i.e., lung cancer) groups) Boxplots of intensities. B are the corresponding SUV values in the two groups of lungs (50-60 min; connection intensities from a single network appear to be more able to separate control and disease groups). The overall burden of structural deviation, or the number of markedly altered margins per patient, reflects the severity of the abnormality.

(3) Network analysis at the individual and group levels

The Pearson correlation coefficient between the average residual network Diff_individual and the group-level difference network Diff_group obtained by the proposed method is 0.78. This suggests that each subject collectively contributes to group differences in glucose metabolism levels, although lung cancer group heterogeneity demonstrates a high degree of heterogeneity between subjects.

(4) Personal network and single organ analysis

We constructed a separate connectivity network for a subject 30 days after discharge from Covid-19 and a patient with unexplained gastrointestinal bleeding, as shown in Figure 5. On the left is the connectivity matrix of all organs of interest and on the right is the network connectivity, where darker lines represent stronger connections between nodes and the degree of black represents the strength of a given node. Figure 5 reveals metabolic connectivity information between organs. For subjects discharged from Covid-19, the lung as an abnormal hub had the strongest connectivity strength, especially for the left ventricle and brain (Fig. 5B). As shown in Figure 6, intensity was significantly correlated with organ ΔSUV (R=0.973 and R=0.893, P<0.05), indicating how much SUV deviates from the control mean. These suggest that a single network can reveal metabolic abnormalities at the network and organ levels.

In conclusion, we propose a framework that applies network principles to analyze whole-body PET data, providing a platform to identify metabolic dysfunction at the systemic level, which is currently impossible to analyze. Subtle biases in metabolic connectivity can be revealed and are highly correlated with organ-level SUV measurements. The present invention can potentially identify systemic metabolic abnormalities that cannot be derived by traditional population-level approaches due to the large heterogeneity among disease groups. This heterogeneity may be due to differences in disease expression or altered system function. In other words, from the network point of view, the proposed method is complementary to traditional methods. It should be noted that the proposed method does not provide a real metabolic connectivity network for scanning, but a perturbed network for the reference network, which reflects the variation between normal and disease samples at the system level. Whole-body SUV PET is generally available for routine PET imaging, although a reference data set is required. Another approach to deriving metabolic networks at the individual level is to utilize dynamic PET, in which the time-activity curves of organs are correlated at the subject level. However, regiodynamics carry information on nonspecific tracer binding and delivery, which may hide the specific interaction of the tracer with its target.

We constructed individual metabolic networks based on PET/CT SUV images. However, our method can be adapted to use other functional parameters such as net metabolic rate, blood flow, phosphorylation rate when dynamic scans are available. Likewise, it can easily be applied to non-FDG tracers, such as those that visualize neurotransmitters, to reveal abnormalities in brain-organ interaction at the subject level.

The present invention can construct an individual metabolic abnormality network using the whole body PET/CT SUV image of the subject and the healthy control group. The method proposed by the present invention can investigate the systemic metabolic abnormalities at the individual level and characterize the molecular connectivity of glucose metabolism at the individual level, which is not possible with current analytical methods. The methods of the present invention can potentially identify systemic metabolic abnormalities, which cannot be derived by traditional population-level approaches due to the large heterogeneity among lung cancer groups. In addition, the present invention can study how the brain and other organs interact in health and disease conditions from a systems perspective.

The above are only the preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, without departing from the principles of the present invention, several improvements and modifications can be made. It should be regarded as the protection scope of the present invention.

Claims (6)

1. A method for investigating systemic metabolic abnormalities at an individual level using a systemic SUV image, comprising:

step 1, data acquisition and processing: acquiring 18F-FDG PET/CT scanning data of different sampling areas of different subjects to obtain images with standard uptake values SUV of the different subjects, and selecting the sampling areas through the images; the subjects included a healthy control group consisting of subjects without any disease record, a lung cancer group consisting of subjects diagnosed with lung cancer with different lesions, subjects 30 days after discharge of Covid-19, subjects with unexplained gastrointestinal bleeding;

the step 1 comprises the following steps:

step 1.1: firstly, carrying out CT scanning to carry out attenuation correction, and then carrying out 18F-FDG PET/CT acquisition; reconstructing the scan data into a voxel size-fixed matrix using a 3D ordered subset expectation-maximization algorithm; performing attenuation and scatter correction using the CT-based attenuation correction map; then, converting the reconstructed moving images into images with standard uptake values SUV by normalizing the injection dosage and the weight;

step 1.2: for each scan, a sampled region of interest of all organs of interest is delineated on the SUV image; the sampling region includes the whole brain, blood, left ventricle, lung, liver, pancreas, spleen, left/right kidney, muscle and spine; performing brain cell division by using statistical parameter mapping, extracting a plane containing the brain in a reconstructed image into a new volume, and performing spatial normalization in a Montreal neurological institute space; smoothing the normalized image by a Gaussian filter, dividing the normalized image into areas defined by an AAL2 map, selecting a brainstem, a whole cerebellum, cerebrospinal fluid, a whole white matter, a caudate nucleus, a putamen and a frontal cortex as new sampling areas, and analyzing 18 sampling areas in total in each scanning together with a depicted organ;

step 2, constructing individual connection networks of different subjects, and obtaining the metabolic abnormal degree of each sampling area of the different subjects through the individual connection networks of the different subjects;

and 3, performing control group homogeneity analysis, lung cancer group heterogeneity analysis, group-to-individual network analysis and individual network-to-single organ analysis according to the metabolic abnormality degree of each sampling region of different subjects.

2. The method for investigating individual-level general metabolic abnormalities using a general SUV image according to claim 1, wherein said step 2 comprises:

step 2.1, constructing a reference metabolic network refNET from the health control group, wherein the reference metabolic network is obtained by calculating Pearson correlation coefficients between SUVs of each region pair and adopts a covariance network structure;

step 2.2: adding a patient to the healthy control group to form a new group so as to construct a new structural covariance network, wherein the network is marked as a perturbation network ptbNET;

step 2.3, calculating the difference between the disturbance network ptbNET and the reference network refNET as a residual network resNET; setting a threshold of 0.3 to eliminate weak correlation; a Z-score map of the remaining network resNET is obtained:

where N is the total number of topics in the new group, the remaining networks essentially represent an abnormal level of connectivity, each network consisting of 153 edges connecting 18 regions, each edge exhibiting a different degree of metabolic variation; to quantify the degree of anomaly, the intensity STR of each region anomaly is defined:

where m is the index number of the region,

is a set of the data to be transmitted,

m =18 is the number of zones, ZCC _mi Representing the correlation coefficient of the Z-fraction graph between the region m and its neighboring regions i, the total number of neighboring nodes being equal to m-1.

3. The method for investigating individual-level general metabolic abnormalities using whole-body SUV images as claimed in claim 2, wherein in said step 3, the control homogeneity analysis is performed as follows:

subjects in each healthy control group were subjected to individual network analysis, intra-group similarity was measured by averaging the Z-fractional Pearson correlation coefficient between any pair of networks, and then re-sampling was performed to test subject selection repeatability.

4. The method for investigating individual-level general metabolic abnormalities using whole-body SUV images as claimed in claim 2, wherein in said step 3, lung cancer group heterogeneity analysis is performed as follows:

the intensity of the personal network of each lung cancer patient was compared to the intensity in the reference network and the similarity between the lung cancer patient networks was calculated by measuring the average of the Pearson correlation coefficients between the subjects between the paired Z-scores on all 153 edges.

5. The method for investigating metabolic abnormalities of the whole body at an individual level using whole body SUV images as set forth in claim 2, wherein in step 3, the group-to-individual level network analysis is performed as follows:

establishing group-level metabolic networks for the patient group and the healthy control group, respectively, establishing two group-level metabolic networks by calculating Pearson correlations of pairs of brain regions of all subjects in the two groups, normalized differences between the two group-level metabolic networks being considered as group-level difference networks Diff _group,i ：

Where i represents an edge, patNET represents a group-level metabolic network of a patient group, refNET represents a group-level metabolic network of a healthy control group;

constructing an average residual network Diff by averaging the Z-scores for each edge across all patient networks _individual,i ：

Where j is the patient index number, N _p Is the total number of patients;

calculating average residual error network Diff _individual,i And group-level differentiated network Diff _group,i Pears in betweenThe on correlation coefficient.

6. The method for investigating metabolic disorders of the whole body at the individual level using whole body SUV images as claimed in claim 2, wherein in step 3, the personal network and single organ analysis is performed as follows:

separate connection networks were constructed for one subject 30 days after Covid-19 discharge and one subject with unexplained gastrointestinal bleeding, and the change in SUV was quantified based on the network strength of the organ.

Images