A feature marker combination based on PET / CT molecular imaging and a human body aging clock evaluation method

Abstract

The application discloses a feature marker combination based on PET / CT molecular imaging and a human body aging clock evaluation method, and belongs to the technical field of medical data. The technical problem to be solved is to solve the defects of the existing human body aging evaluation method, such as insufficient comprehensiveness and limited accuracy. The technical solution points are to provide a feature marker combination composed of functional feature markers and volume feature markers; a support vector machine regression (SVR) algorithm is used to construct a model, the mean absolute error (MAE) in the independent verification set is 5.68 years, the determination coefficient R 2 is 0.85. The application can be used for human body aging clock evaluation, and provides a new perspective of aging evaluation based on multiple organ perspectives and verifiable marker support.

Description

Technical Field

[0001] This invention belongs to the field of medical data analysis technology, specifically relating to a combination of characteristic biomarkers based on PET / CT molecular imaging and a method for assessing the human aging clock. Background Technology

[0002] With the accelerating pace of global population aging, the accurate assessment of the degree of human aging has become a research hotspot in the fields of life sciences and clinical medicine. Aging is a complex systemic life process involving multiple aspects such as tissue homeostasis imbalance, organismal self-renewal, replication, regulation and repair, and reproduction. Essentially, it is a comprehensive manifestation of the disordered coordinated operation of multiple systems throughout the body.

[0003] Currently, methods for estimating biological age mainly include DNA methylation detection and local organ imaging characterization analysis. Among these, DNA methylation has the advantages of high stability and strong correlation with age, and peripheral leukocytes are commonly used as the detection sample. However, the blood DNA methylation clock cannot fully reflect the true aging state of the brain and organs throughout the body. On the other hand, local organ imaging assessment methods based on chest X-rays, abdominal CT, or MRI can only capture the structural composition changes of a single organ during the aging process, and cannot achieve a comprehensive aging assessment of multiple organs throughout the body. Therefore, the accuracy and comprehensiveness of the assessment are both insufficient.

[0004] PET / CT (Positron Emission Tomography-Computed Tomography) is an advanced imaging technique that combines functional metabolic imaging (PET) with anatomical imaging (CT). It is administered via intravenous injection. 18 F-FDG imaging agent can enter cells via the glucose transporter (GLUT) on the cell membrane. After being phosphorylated by hexokinase, it cannot be further metabolized and remains in the cell, thus enabling imaging of cellular metabolic activity. PET / CT can obtain tomographic images of multiple parts of the whole body (including the brain and trunk) in a single imaging session, providing a direct and comprehensive reflection of the functional metabolic state of all organs.

[0005] Studies have shown that aging leads to a decline in the efficiency of systemic glucose metabolism and insufficient cellular energy supply, while long-term metabolic disorders further accelerate the aging process, creating a vicious cycle. Therefore, there is an urgent need to provide a method for accurately assessing the human aging clock by constructing a combination of functional and structural biomarkers for multiple organs and tissues in the brain and trunk based on PET / CT, thereby overcoming the limitations of existing assessment methods.

[0006] Relevant patent documents retrieved: The document, published in China (CN114450750A) on May 6, 2022, discloses a method for creating a biological aging clock for a subject. The method may include: (a) receiving proteomic features from a subject's tissue or organ; (b) creating an input vector based on the proteomic features; (c) inputting the input vector into a machine learning platform; (d) having the machine learning platform generate a predicted biological aging clock for the tissue or organ based on the input vector, wherein the biological aging clock is tissue- or organ-specific; and (e) compiling a report including a biological aging clock that identifies the predicted biological age of the tissue or organ.

[0007] The prior art represented by the aforementioned documents has at least the following unresolved technical problems or defects: The aging clock constructed based on proteomics features has a high degree of invasiveness and poor clinical accessibility, making it difficult to achieve non-invasive, in vivo, and simultaneous assessment of multiple organs throughout the body. Furthermore, specific, stable, and reproducible biomarker combinations that are highly correlated with aging have not been disclosed, the criteria for feature selection are unclear, and the reliability and reproducibility of the model are insufficient.

[0008] In solving the above problems or overcoming the above defects, the present invention encountered the following difficulties and obstacles: PET / CT images can extract a large number of multi-organ features, covering functional and structural features of multiple systems throughout the body. However, some features have a weak correlation with aging and poor stability, making it difficult to accurately screen out core biomarkers that are highly correlated with the human aging process, have good repeatability, and strong anti-interference ability. They are easily affected by irrelevant features, individual differences, and detection errors, which can affect the accuracy of subsequent evaluation models.

[0009] Different machine learning algorithms show significant differences in their performance in age prediction. Traditional algorithms are prone to overfitting, which results in models performing well on the training set but lacking generalization ability on the independent validation set. This makes it impossible to achieve accurate age prediction for different age groups and individuals. It is necessary to overcome the challenge of balancing model generalization and prediction accuracy through extensive parameter tuning, algorithm comparison, and validation optimization. Summary of the Invention

[0010] The purpose of this invention is to provide: A combination of characteristic biomarkers based on PET / CT molecular imaging and a method for assessing the human aging clock, and related technologies, to address the shortcomings of existing human aging assessment methods, such as insufficient comprehensiveness and limited accuracy, or a combination thereof.

[0011] Terminology Explanation: Unless otherwise defined, all technical terms in this document have the same meanings as commonly understood by one of ordinary skill in the art to which the subject matter of the claims pertains. Unless otherwise stated, all patents, patent inventions, and publications cited in this document are incorporated herein by reference in their entirety. If multiple definitions exist for terms in this document, the definitions in this chapter shall prevail.

[0012] It should be understood that the above brief description and the following detailed description are exemplary and for illustrative purposes only, and do not limit the subject matter of the invention in any way. In this invention, the singular is used in conjunction with the plural unless otherwise specifically stated. It should also be noted that, unless otherwise stated, the use of “or” or “or” means “and / or”. Furthermore, the use of the term “comprising” and other forms such as “including,” “containing,” and “contains” are not limiting.

[0013] The definition of the standard chemical terminology can be found in the reference "PET / CT Diagnostics" (Yu Lijuan and Pan Zhongyun, People's Medical Publishing House, 2009).

[0014] Unless otherwise stated, conventional methods within the scope of the art shall be used, such as conventional acquisition and preprocessing methods for PET / CT images, conventional extraction methods for medical image features, conventional parameter tuning methods for machine learning models, and conventional verification methods for data statistical analysis.

[0015] Unless specifically defined herein, the use of all commercially available products herein employs standard techniques. For example, it may be carried out using the manufacturer's instructions for use with the kit, or in accordance with methods known in the art or the description of this invention. The techniques and methods described herein can generally be implemented according to conventional methods well known in the art, based on the descriptions in the various summary and more specific documents cited and discussed in this specification.

[0016] The terms "optional / arbitrary" or "optionally / arbitrarily" mean that an event or situation described below may or may not occur, including both the occurrence and non-occurrence of the event or situation. For example, according to the definition below: "The age assessment product includes any one or more of reagents, kits, and chips" means that the age assessment product can be a reagent, or the age assessment product can be a kit, or the age assessment product can be a chip, or the age assessment product can be a reagent, a kit, and a chip / The term "brain region markers" used in this article refers to characteristic markers derived from relevant regions of human brain tissue, specifically including the left lateral ventricle (excluding the temporal horn), the right lateral ventricle (excluding the temporal horn), the left caudate nucleus, the right caudate nucleus, and the whole brain (functional characteristic markers, from which SUVR values ​​are extracted), which are important components of the functional characteristic markers of this invention.

[0017] The term "trunk region markers" used in this article refers to characteristic markers derived from organs and tissues in the human trunk, specifically including the left and right hip bones (functional markers, from which SUVR values ​​are extracted) and the aorta, pulmonary artery, left atrium, left and right iliac arteries, left and right iliopsoas muscles (volume markers, from which volume values ​​are extracted). These markers are used to reflect the metabolic and structural aging status of the trunk and, together with brain region markers, constitute a multi-organ aging assessment system.

[0018] The term “left lateral ventricle (excluding the temporal horn)” used in this article refers to the cavity within the left cerebral hemisphere, that is, the branching structure on the left side of the lateral ventricle. It is an important component of the ventricular system, does not include the temporal horn structure extending into the temporal lobe, and is an important cerebrospinal fluid circulation cavity in the brain.

[0019] The term “right lateral ventricle (excluding the temporal horn)” used in this article refers to the cavity within the right cerebral hemisphere, namely the right branch of the lateral ventricle, which is symmetrically distributed with the left lateral ventricle (excluding the temporal horn) and does not include the temporal horn structure.

[0020] The term "left hip bone" as used in this article refers to the irregular flat bone that forms the left side of the pelvis, which is formed by the fusion of the ilium, ischium, and pubis at the acetabulum.

[0021] The term "right hip bone" as used in this article refers to the irregular flat bone that forms the right side of the pelvis. It is formed by the fusion of the ilium, ischium, and pubis at the acetabulum and is symmetrically distributed with the left hip bone.

[0022] The term "left caudate nucleus" used in this article refers to the gray matter nucleus located in the left basal ganglia region of the brain.

[0023] The term "right caudate nucleus" used in this article refers to the gray matter nucleus located in the right basal ganglia region of the brain, which is symmetrically distributed with the left caudate nucleus.

[0024] The term "whole brain" as used in this article refers to the entire human brain tissue, encompassing all brain structures such as the cerebral cortex, basal ganglia, and ventricular system.

[0025] The term "aorta" used in this article refers to the largest artery in the human body, which originates from the left ventricle and is divided into the ascending aorta, the aortic arch, and the descending aorta.

[0026] The term "pulmonary artery" as used in this article refers to the artery that connects the right ventricle to the lungs, including the left and right pulmonary arteries.

[0027] The term "left atrium" as used in this article refers to one of the four chambers of the heart, located in the upper left of the heart.

[0028] The term "left iliac artery" as used in this article refers to the left branch of the descending aorta.

[0029] The term "right iliac artery" used in this article refers to the right branch originating from the descending aorta, which is symmetrically distributed with the left iliac artery.

[0030] The term "left iliopsoas muscle" as used in this article refers to the muscle located on both sides of the left iliac fossa and lumbar vertebrae, and is composed of the left iliacus muscle and the left psoas major muscle.

[0031] The term "right iliopsoas muscle" as used in this article refers to the muscle located on both sides of the right iliac fossa and lumbar vertebrae, composed of the right iliacus and right psoas major muscles, and symmetrically distributed with the left iliopsoas muscle.

[0032] The term "in this article" 18 "F-FDG" refers to: 18 F-fluorodeoxyglucose, fluorine-18 labeled deoxyglucose, is a radioactive tracer used in PET imaging. It can enter cells via the glucose transporter (GLUT) on the cell membrane, and after being phosphorylated by hexokinase, it remains in the cell to reflect cellular metabolic activity.

[0033] The term "PET image" as used in this article refers to a positron emission tomography (PET) image, which is created by detecting... 18 The distribution of radioactive tracers such as F-FDG in the body directly reflects the functional metabolic state of various organs and tissues in the human body, but does not show anatomical structural details.

[0034] The term "CT image" used in this article refers to a computed tomography (CT) image, which is generated by X-rays penetrating the human body and being processed by a computer to produce cross-sectional anatomical images of various organs and tissues in the human body, clearly showing the shape, size and location information of the organs.

[0035] The term "healthy person" as used in this article refers to a subject who, after clinical examination (including physical examination, laboratory examination and imaging examination), is confirmed to have no major organic diseases, no chronic metabolic diseases, no malignant tumors, no neurological diseases, and whose liver and kidney functions, cardiopulmonary functions are normal and whose mental state is good.

[0036] The term "DICOM file" used in this article refers to the international standard file format for digital imaging and communications in medicine, which is mainly used to store and transmit medical imaging data such as X-ray, CT, MRI, and ultrasound, as well as related patient information.

[0037] The term "MMIS post-processing platform" used in this article refers to the Medical Image Post-processing System, a professional software platform for preprocessing, organ segmentation, feature extraction, and image analysis of medical images (including PET, CT, MRI, etc.). It can realize functions such as image registration, resampling, and automatic segmentation, and support the subsequent analysis and modeling of image data.

[0038] The term "isotropic voxel" used in this article refers to a voxel with equal side lengths along the x, y, and z coordinate axes in three-dimensional space. In PET / CT image resampling, it can ensure that the spatial resolution of the image is consistent in all directions, avoid organ segmentation deviation and feature extraction error caused by voxel anisotropy, and ensure the accuracy of subsequent multi-organ feature analysis.

[0039] The term "SUV" used in this article refers to Standardized Uptake Value, which is a core functional indicator for quantifying the glucose metabolic activity of organs and tissues in PET / CT examinations. It is calculated by standardizing the radioactive uptake concentration of organs and tissues with the injection dose and the subject's weight. The higher the value, the stronger the metabolic activity of the corresponding organ or tissue, and vice versa.

[0040] The term "volume feature" used in this article refers to the three-dimensional volume quantification index of organ tissues calculated based on CT images after automatic segmentation of organ tissues using the MMIS post-processing platform, with the unit being cm. 3 It can clearly reflect the morphological size and structural changes of organs and tissues, and is an important structural indicator for assessing the aging state of organs.

[0041] The term "SUVR" used in this article refers to the Standardized Uptake Value Ratio, which is calculated by dividing the average SUV value of the target organ or brain region by the average SUV value of the reference region (the aortic blood pool for organs and the cerebellum for brain regions). It can reduce the impact of individual differences and the accuracy of detection equipment on the assessment of metabolic activity and improve the stability and reliability of the detection data.

[0042] The term "scale function" used in this article refers to the built-in function in R software used for data standardization. Its basic syntax is scale(x, center = TRUE, scale = TRUE). By default, it first subtracts the mean from the data (center=TRUE) and then divides it by the standard deviation (scale=TRUE), which can quickly achieve Z-score standardization of data and simplify the data preprocessing process.

[0043] The term "Z-score standardization" used in this article, also known as standard score standardization, refers to a commonly used data preprocessing method. Its calculation formula is Z=(x-μ) / σ (where x is the original data, μ is the data mean, and σ is the data standard deviation). It can convert original data of different dimensions and ranges into standardized data with a mean of 0 and a standard deviation of 1, which is used to eliminate the influence of dimensional differences on subsequent statistical analysis and model construction.

[0044] The term "MAE" used in this paper refers to Mean Absolute Error, a quantitative indicator used to evaluate the predictive accuracy of regression models. It is calculated as the average of the absolute differences between all predicted values ​​and their corresponding true values. The smaller the value, the smaller the deviation between the model's prediction and the true value, and the higher the prediction accuracy. This paper uses it to evaluate the predictive performance of prediction models for different age groups.

[0045] The term "coefficient of determination R" used in this article 2 "R" refers to the core indicator for measuring the goodness of fit of a regression model, with a value range of 0-1. It is calculated as the ratio of the regression sum of squares to the total sum of squares. 2 The closer the value is to 1, the better the model fits the data and the higher the degree of variable variation it can explain. This paper uses it to evaluate the fitting performance of prediction models for different ages.

[0046] The "quartile division method" used in this paper employs quartiles (Q1, Q3) as the division threshold. This is a standardized grouping method commonly used in the fields of biomedicine and aging assessment. Its core advantage is that it can effectively avoid the influence of outliers on the grouping results. Simultaneously, it divides the study population into three categories: "below the lower quartile, middle quartile, and above the upper quartile," which highly aligns with the natural stratification of human aging rates (a minority of people age slowly, most people age normally, and a minority age rapidly). Specifically, Q1 = -2.94 represents 25% of individuals in the healthy population whose aging rate is below this value (i.e., slow aging), Q3 = 3.59 represents 25% of individuals whose aging rate is above this value (i.e., rapid aging), and the middle 50% of individuals fall within the normal aging range. This division method not only meets the requirements of statistical rigor but also accurately distinguishes between populations with different aging rates, possessing clear operability and reproducibility.

[0047] In a first aspect, the present invention provides: a combination of characteristic biomarkers based on PET / CT molecular imaging.

[0048] This includes technical features: combinations of characteristic markers.

[0049] The combination of feature markers consists of the SUVR value of each functional feature marker and the volume value of each volume feature marker.

[0050] Preferably, the functional markers are: the left lateral ventricle (excluding the temporal horn), the right lateral ventricle (excluding the temporal horn), the left hip bone, the right hip bone, the left caudate nucleus, the right caudate nucleus, and the whole brain.

[0051] Preferably, the volumetric feature markers are: aorta, pulmonary artery, left atrium, left iliac artery, right iliac artery, left iliopsoas muscle, and right iliopsoas muscle.

[0052] Preferably, the combination of feature markers is used as input to a support vector machine regression algorithm to construct an age assessment model.

[0053] Specifically, the SUVR value is calculated as follows: SUVR value of brain region markers = Average SUV value of the corresponding brain region / Average SUV value of the cerebellum; The SUVR value of a trunk region marker = the average SUV value of the corresponding trunk region / the average SUV value of the aortic blood pool.

[0054] Preferably, the brain region markers include the left lateral ventricle (excluding the temporal horn), the right lateral ventricle (excluding the temporal horn), the left caudate nucleus, the right caudate nucleus, and the whole brain.

[0055] Preferably, the trunk area landmarks include the left hip bone and the right hip bone.

[0056] Specifically, the volume value is a three-dimensional volume quantification index of volume characteristic markers.

[0057] Preferably, the volume value is calculated by performing three-dimensional segmentation of the corresponding organ or tissue region in the PET / CT image.

[0058] Based on further solutions to the technical problems of the present invention, or simultaneous solutions to multiple technical problems, the preferred solution in the technical solution provided in the first aspect of the present invention includes: The first preferred approach: The functional biomarkers are: the left lateral ventricle (excluding the temporal horn), the right lateral ventricle (excluding the temporal horn), the left hip bone, the right hip bone, the left caudate nucleus, the right caudate nucleus, and the whole brain. This approach addresses the technical problem of "existing aging assessment biomarkers being singular and lacking representativeness," and further solves the technical problem of "relying solely on local organs to comprehensively reflect the overall aging state." By screening multi-organ functional metabolic characteristics highly correlated with aging, the stability and predictive reliability of biomarker combinations are significantly improved.

[0059] The second preferred option: The volumetric biomarkers are: aorta, pulmonary artery, left atrium, left iliac artery, right iliac artery, left iliopsoas muscle, and right iliopsoas muscle. This technical solution, while addressing the technical problem that "aging assessment focuses only on functional metabolism and ignores structural changes," further addresses the technical problem that "single-dimensional characteristics are easily affected by individual differences."

[0060] The third preferred solution: The aforementioned feature biomarker combination is used as input to a support vector machine regression algorithm to construct an age assessment model. This technical solution, in addition to solving the technical problem that "existing feature biomarker combinations lack clear application scenarios and cannot be directly used for aging assessment modeling," further addresses the technical problem that "the mismatch between feature biomarkers and modeling algorithms leads to low model prediction accuracy and weak generalization ability."

[0061] Secondly, this invention provides an age assessment model based on PET / CT molecular imaging.

[0062] This includes a technical feature: an age assessment model.

[0063] The age assessment model incorporates the characteristic marker combination of this invention.

[0064] Specifically, the age assessment model is constructed using a support vector machine regression algorithm.

[0065] Preferably, the parameters of the support vector machine regression algorithm are: C=100.0, gamma=0.01, and kernel function is rbf.

[0066] Based on further solutions to the technical problems of the present invention, or simultaneous solutions to multiple technical problems, the preferred solution in the technical solution provided in the second aspect of the present invention includes: The first preferred solution: The parameters of the support vector machine regression algorithm are: C=100.0, gamma=0.01, and kernel function is rbf. This technical solution, while solving the technical problem of "mismatch between feature markers and modeling algorithms", further solves the technical problem of "unoptimized support vector machine regression algorithm parameters leading to easy overfitting and insufficient prediction accuracy".

[0067] Thirdly, the present invention provides: an age assessment product based on PET / CT molecular imaging.

[0068] This includes a technical feature: an age assessment product.

[0069] The age assessment product is used to detect the feature values ​​of the combination of feature markers used in this invention, wherein the feature values ​​are the SUVR values ​​of each functional feature marker and the volume values ​​of each volume feature marker.

[0070] Preferably, the age assessment product includes any one or more of reagents, kits, and chips.

[0071] Based on further solutions to the technical problems of the present invention, or simultaneous solutions to multiple technical problems, the preferred solution in the technical solution provided in the third aspect of the present invention includes: The first preferred option: The age assessment product includes any one or more of reagents, kits, and chips. This technical solution, while addressing the technical problem of "difficulty in applying combined detection of characteristic biomarkers," further solves the technical problem of "providing standardized, modular, and combinable detection product formats."

[0072] Fourthly, the present invention provides an evaluation method based on PET / CT molecular imaging.

[0073] This includes technical features: evaluation methods.

[0074] The assessment methods include age assessment methods and human aging clock assessment methods.

[0075] Specifically, the age assessment method includes the following steps: (1) Obtain PET / CT molecular imaging data of the individual to be evaluated; (2) Detect the feature values ​​of the combination of feature markers; (3) Input the feature values ​​into the age assessment model and output the predicted age of the individual to be assessed to complete the age assessment.

[0076] Specifically, the human aging clock assessment method includes the following steps: S1. Obtain PET / CT molecular imaging data of the individual to be evaluated; S2. Detect the feature values ​​of the combination of feature markers; S3. Input the feature values ​​into the age assessment model and output the predicted age of the individual to be assessed; S4. Based on the difference between the predicted age and the actual age, the individuals to be evaluated are classified into slow aging, normal aging, and fast aging.

[0077] Preferably, the method for acquiring PET / CT molecular imaging data in step (1) or step S1 includes: intravenous injection of the individual to be evaluated. 18 After F-FDG imaging agent, a whole-body scan was performed using PET / CT equipment to acquire PET images, CT images, and corresponding DICOM files, thus completing the acquisition of PET / CT molecular imaging data.

[0078] Preferably, the age assessment model described in step (3) or step S3 is a support vector machine regression (SVR) model. The construction method is to use the feature marker combination described in the first aspect of the present invention as input, and construct it based on the support vector machine regression algorithm called by Python software. The parameters are C=100.0, gamma=0.01, and kernel function is rbf.

[0079] Preferably, the dividing criterion in step S4 is: using the first and third quartiles of the difference distribution as the dividing thresholds. A difference less than the first quartile indicates slow aging. The range between the first and third quartiles indicates normal aging; A value above the third or fourth quartile indicates rapid aging.

[0080] Based on further solutions to the technical problems of the present invention, or simultaneous solutions to multiple technical problems, the preferred solution in the technical solution provided in the fourth aspect of the present invention includes: The first priority solution provides specific steps for age assessment and human aging clock assessment methods. This solution addresses the shortcomings of existing human aging assessment methods, such as insufficient comprehensiveness and limited accuracy. Furthermore, it standardizes the entire process from image acquisition to result output by clearly defining the specific steps for age assessment and human aging clock assessment. This ensures that different operators can perform assessments according to a unified process, improving the repeatability and stability of the assessment methods and providing practical methodological support for accurate assessment of the human aging clock.

[0081] The second priority solution provides a standard for classifying aging rates in the human aging clock assessment method. This solution addresses the shortcomings of existing human aging assessment methods, such as insufficient comprehensiveness and limited accuracy. Furthermore, it utilizes quartiles as threshold values ​​to standardize and objectively stratify aging rates, clearly defining the criteria for slow, normal, and rapid aging. This allows for precise differentiation of individuals in different aging states, providing quantitative evidence for anti-aging intervention program development and efficacy evaluation, and significantly enhancing the clinical application value of the assessment method.

[0082] In this invention, embodiments 1-4 at least support the protection scope of the technical feature "combination of feature markers".

[0083] The technical feature "combination of feature markers" is derived from the aforementioned explanation and / or the corresponding technical features in Examples 1-4, including the SUVR values ​​of each functional feature marker and the volume values ​​of each volumetric feature marker, through the common feature "based on PET / CT molecular image extraction, used for human aging clock assessment, simultaneously including the SUVR values ​​(metabolic function dimension) of functional feature markers and the volume values ​​(anatomical structure dimension) of volumetric feature markers, and this combination has been experimentally verified to be highly correlated with the human aging process and can be stably used as input for age assessment models." Therefore, those skilled in the art can reasonably infer that the technical feature "combination of feature markers," its subordinate concepts, the technical means essentially equivalent to "combination of feature markers," and technical means that can replace "combination of feature markers" within the scope of conventional technical means and common knowledge based on the existing technical level should all fall within the protection scope of this invention. For example, replacing "combination of feature markers" with marker combinations or feature combinations while keeping other technical features unchanged still falls within the protection scope of this invention.

[0084] In this invention, embodiments 1-4 at least support the protection scope of the technical feature "age assessment model".

[0085] The technical feature "age assessment model" is derived from the SVR model and other technical features explained above and / or in Examples 1-4, summarized by the common feature "using the combination of the feature markers of this invention as input, for human aging clock and age assessment, constructed using machine learning regression algorithms, and after parameter optimization, can stably achieve high-accuracy age prediction." Therefore, those skilled in the art can reasonably infer that the technical feature "age assessment model," its subordinate concepts, the technical means essentially equivalent to "age assessment model," and technical means that can replace "age assessment model" based on existing technology and conventional technical means and common knowledge should all fall within the protection scope of this invention. For example, replacing "age assessment model" with assessment model, model, etc., while keeping other technical features unchanged, still falls within the protection scope of this invention.

[0086] In this invention, embodiments 1-4 at least support the protection scope of the technical feature "age assessment product".

[0087] The technical feature "age assessment product" is derived from the reagents, kits, chips, etc., described above and / or in Examples 1-4, based on the common feature "used to detect the feature values ​​(SUVR values ​​of each functional feature marker and volume values ​​of each volume feature marker) of the combination of feature markers of the present invention, adapted to age assessment related to PET / CT molecular imaging and human aging clock assessment, and can achieve standardized and accurate detection of feature values, providing reliable input for age assessment models." Therefore, those skilled in the art can reasonably infer that the technical feature "age assessment product," its subordinate concepts, the technical means essentially equivalent to "age assessment product," and the technical means that can replace "age assessment product" within the scope of conventional technical means and common knowledge based on the existing technical level should all fall within the protection scope of the present invention. For example, replacing "age assessment product" with assessment products, age assessment kits, etc., while keeping other technical features unchanged, still falls within the protection scope of the present invention.

[0088] In this invention, embodiments 1-4 at least support the protection scope of the technical feature "evaluation method".

[0089] The technical feature "assessment method" is summarized by the common feature "based on PET / CT molecular imaging data, using the feature marker combination of the first aspect of this invention and the age assessment model of the second aspect as the core support, for human age prediction and aging state stratification, including a complete standardized process of image acquisition, feature detection, model input, and result output / stratification, which can stably achieve accurate and objective assessment of human aging state." Therefore, those skilled in the art can reasonably infer that the technical feature "assessment method," its subordinate concepts, the basically equivalent technical means of "assessment method," and the technical means that can replace "assessment method" based on the existing technical level within the scope of conventional technical means and common knowledge should all fall within the protection scope of this invention. For example, replacing "assessment method" with age determination method, human aging rate assessment method, etc., while keeping other technical features unchanged, still falls within the protection scope of this invention.

[0090] The present invention has at least the following beneficial effects: Compared with existing technologies, this invention provides a complete PET / CT molecular imaging-mediated human aging clock assessment system by offering a combination of characteristic biomarkers, an age assessment model, age assessment products, and assessment methods. This comprehensively addresses the shortcomings of existing human aging assessment methods, such as insufficient comprehensiveness, limited accuracy, poor operability, and weak clinical applicability. Specific beneficial effects are as follows: 1. Advantages of the combination of feature biomarkers: 14 core feature biomarkers (7 functional features + 7 volumetric features) that are highly correlated with the human aging process, have strong stability and outstanding anti-interference ability were selected. These biomarkers cover the brain and trunk regions, achieving dual-dimensional characterization of metabolic function and anatomical structure. This overcomes the problems of feature redundancy, weak correlation and poor repeatability of existing technologies, and provides accurate and reliable core targets for aging assessment. Furthermore, the standardized SUVR value calculation method effectively reduces the interference of individual differences and detection errors, and improves the consistency and repeatability of feature data.

[0091] 2. Advantages of the age assessment model: Utilizing a parameter-optimized Support Vector Machine Regression (SVR) model, the optimal parameter combination (C=100.0, gamma=0.01, kernel function rbf) is clearly defined, adapting to the feature marker combination of this invention. This effectively balances model fit and generalization ability, avoiding overfitting or underfitting. On the independent validation set, it achieves an MAE of 5.68 years and an R... 2 With an excellent prediction performance of 0.85, it significantly outperforms existing traditional machine learning algorithms, solving the technical problems of low prediction accuracy and weak generalization ability of existing models, and enabling accurate age prediction for different age groups and individuals. Attached Figure Description

[0092] Figure 1 This is to show the prediction performance of the support vector regression model.

[0093] Figure 2 A flowchart for predicting human biological age and assessing the aging clock. Detailed Implementation

[0094] The following non-limiting embodiments are intended to enable those skilled in the art to gain a more comprehensive understanding of the present invention, but do not limit the invention in any way. The following content is merely an exemplary description of the scope of protection claimed by the present invention, and those skilled in the art can make various changes and modifications to the present invention based on the disclosed content, and such changes should also fall within the scope of protection claimed by the present invention.

[0095] The present invention will be further described below by way of specific embodiments. Unless otherwise specified, all instruments, devices, equipment, reagents, products, etc., used in the embodiments of the present invention are obtained through conventional commercial means.

[0096] Example 1: Data Preprocessing Flow Collected from 177 healthy subjects 18F-FDG PET / CT image DICOM files were collected from 91 males and 86 females, aged 14-94 years. The MMIS post-processing platform was used to rigidly register the PET and CT images and resample the images to isotropic voxels to complete image preprocessing. The preprocessed images were then input into the organ segmentation module for automatic organ segmentation, obtaining the average normalized uptake value (SUV) and volumetric characteristics of 104 organs and tissues, including the brain, liver, spleen, pancreas, heart, lungs, kidneys, bones, major blood vessels, and trunk muscles.

[0097] SUV is a functional indicator used in PET / CT scans to quantify the glucose metabolic activity of organs and tissues; a higher value indicates stronger metabolic activity, and vice versa. Volumetric characteristics are structural indicators obtained from automatically segmented organs and tissues based on CT images. The organ-standardized uptake ratio (SUVR) was calculated by dividing the average SUV value of each organ by the average SUV value of the aortic blood pool; the SUVR value of each brain region was calculated by dividing the average SUV value of each brain region by the average SUV value of the cerebellum. The obtained organ SUVR values ​​(functional characteristics) and volumetric values ​​(volumetric characteristics) were Z-score standardized using the scale function in R software (version 4.1.1) to eliminate the influence of different dimensions on subsequent analyses.

[0098] Example 2: Extraction of Organ Tissue Features from PET / CT The 371 standardized features (including 184 functional features and 187 volumetric features) and subject gender information were imported into Python software (version 3.9.13). The f_regression function was used to perform an F-test to evaluate the p-value of feature relevance and the importance score of each feature. The top 20 functional features (SUVR metabolic features) and 20 structural features (volume features) are shown in Table 1 and Table 2, respectively.

[0099] Table 1. SUVR metabolic characteristics (functional characteristics) of organs and tissues detected by PET / CT

[0100] Table 2 Structural characteristics (volume characteristics) of organ tissues from PET / CT scans

[0101] Example 3: Evaluation Model Construction 1. Feature Filtering The top 7 features in terms of importance scores from the aforementioned SUVR metabolic (functional) and structural (volume) features were selected to construct an age prediction regression model. Specifically: Functional features include: left lateral ventricle (excluding the temporal horn), right lateral ventricle (excluding the temporal horn), left hip bone, right hip bone, left caudate nucleus, right caudate nucleus, and SUVR value of the whole brain; The volume characteristics are: the volume values ​​of the aorta, pulmonary artery, left atrium, left iliac artery, right iliac artery, left iliopsoas muscle, and right iliopsoas muscle.

[0102] 2. Model Import Four machine learning prediction models were built using Python software (version 3.9.13): Lasso (Minimum Absolute Shrinkage and Selection Operator), Support Vector Regression (SVR), Gradient Boosting, and Random Forest. The models were imported as follows: to import a linear regression model, execute the code "from sklearn.linear_model import Lasso"; to import an ensemble regression model, execute the code "from sklearn.ensemble import RandomForestRegressor,GradientBoostingRegressor"; and to import a support vector machine regression model, execute the code "from sklearn.svmimport SVR".

[0103] 3. Model parameter values In the Lasso regression model, the regularization parameter alpha can take values ​​of 0.001, 0.01, 0.1, 1.0, and 10.0.

[0104] In the support vector regression model, the regularization parameter C has values ​​of 0.1, 1.0, 10.0, and 100.0; the gamma parameter has values ​​of 'scale', 'auto', 0.1, and 0.01; and the kernel function type is defined as 'linear' and 'rbf'.

[0105] In the random forest model, n_estimators can take values ​​of 50, 100, and 200; the maximum depth can take values ​​of 0, 5, 10, and 15; and the minimum number of sample splits can take values ​​of 2, 5, and 10.

[0106] The gradient boosting model has learning_rate values ​​of 0.01, 0.1, and 0.2; maximum depth values ​​of 3, 5, and 7; and n_estimators values ​​of 50, 100, and 200.

[0107] 4. Model parameter optimization process The GridSearchCV module of the sklearn library in Python software (version 3.9.13) is used to perform grid search, optimize model parameters, and determine the optimal parameter combination for each model.

[0108] Taking the support vector regression model as an example: The regularization parameter C can be selected from 0.1, 1.0, 10.0, and 100.0; the gamma parameter can be selected from 'scale', 'auto', 0.1, and 0.01; and the kernel function type can be selected from 'linear' and 'rbf'. One parameter combination is regularization parameter C = 0.1, gamma = 0.1, and kernel function type is linear; another parameter combination is regularization parameter C = 1.0, gamma = 0.01, and kernel function type is rbf. Each combination of parameter values ​​forms a parameter network, and so on, for a total of 32 different parameter combinations.

[0109] Using the feature data of 177 healthy subjects as the training set, the training set was randomly divided into 5 mutually exclusive subsets: 4 for training and 1 for testing. This data was randomly split 5 times, and the model performance was evaluated for each of the 32 different parameter combinations in the support vector regression model. The average of the 5 results was taken as the model performance. Based on the model performance evaluation metrics, the optimal parameter combination for model performance was selected.

[0110] Performance evaluation metrics for regression models include: Mean Absolute Error (MAE) and Coefficient of Determination (R²). 2 The smaller the MAE, the closer R² is to 1, indicating better model performance. In this case, the optimal parameter combination of the model is matched.

[0111] Similarly, the optimal parameters and model performance of the Lasso regression model (5 different parameter combinations), the random forest model (36 different parameter combinations), and the gradient boosting model (27 different parameter combinations) were obtained through grid search and five-fold cross-validation, respectively.

[0112] After parameter optimization, the optimal parameters and performance of each model are as follows: The MAE of the Lasso regression model was 6.1, and the R-squared value was [missing value]. 2 The value is 0.80, and the optimal regularization parameter alpha is 1.0.

[0113] The SVR model has a MAE of 4.20 and R... 2 The value is 0.88, the optimal regularization parameter C is 100.0; gamma is 0.01, and the kernel function type is rbf.

[0114] The random forest model has a MAE of 2.33 and R...2 The value is 0.96, the optimal parameter n_estimators is 200, and the minimum number of sample splits is 2.

[0115] The gradient boosting model has a MAE of 2.37 and R0. 2 The value is 0.97, the optimal parameters are learning_rate 0.1, maximum depth 3, and n_estimators 50.

[0116] Example 4 Evaluation Model Validation A validation set of 71 healthy participants was collected, including 36 males and 35 females, aged 12-83 years. Following the data preprocessing workflow described above, the data of the validation set participants were... 18 F-FDG PET / CT images were registered and resampled, organs were automatically segmented, and the SUVR values ​​(functional features) and volume features of each organ were obtained. The features were Z-score standardized using the scale function of R software (version 4.1.1).

[0117] The 14 standardized features (7 functional features and 7 volumetric features) of the validation set were imported into Python software (version 3.9.13). The four regression models with the determined optimal parameters were then called, and the code "from sklearn.metricsimport mean_squared_error, mean_absolute_error, r" was executed. 2 The performance evaluation metric for each model was calculated using the "_score", and the results are shown in Table 3. Among them, the SVR model exhibited the best predictive performance, with an MAE of 5.68 years and R0.05 on the independent validation set. 2 With a value of 0.85, its predictive performance is as follows: Figure 1 As shown in Table 4, the specific prediction results of the SVR model validation set are as follows: The entire age prediction process is as follows: Figure 2 As shown.

[0118] Table 3. Predictive performance of four regression models on human biological age on the independent validation set.

[0119] Table 4 Age prediction results on the validation set

[0120] Example 5: A method for assessing the human aging clock based on multi-organ PET / CT molecular imaging The individual to be evaluated 18The raw F-FDG PET / CT data were processed according to the above data preprocessing and feature extraction procedures. The resulting 14 standardized features were then imported into the optimal SVR model (optimal parameters: C=100.0, gamma=0.01, kernel function rbf) to obtain the predicted age of the individual.

[0121] The individual aging rate is defined as predicted age minus actual age. Based on the difference between the predicted age and actual age of the SVR model in the training set population, the median of the difference is calculated to be -2.94 for the quarter and 3.59 for the three-quarters. According to the individual aging rate, the human aging level is divided into three categories: Slow-aging population: Individual aging rate < -2.94; For people aging normally: the individual aging rate ranges between -2.94 and 3.59; People who are aging rapidly: Individual aging rate > 3.59.

[0122] Comparative Example 1: Combination of Feature Markers The top three features in terms of importance scores from the aforementioned SUVR metabolic (functional) and structural (volume) characteristics were selected to construct an age prediction regression model. Specifically: Functional characteristics include: SUVR values ​​of the left lateral ventricle (excluding the temporal horn), the right lateral ventricle (excluding the temporal horn), and the right hip bone; The volume characteristics are: the volume values ​​of the aorta, pulmonary artery, and left atrium.

[0123] Comparative Example 2: Combination of Feature Markers The top 5 features in terms of importance scores from the aforementioned SUVR metabolic (functional) and structural (volume) features were selected to construct an age prediction regression model. Specifically: The functional characteristics are: SUVR values ​​of the left lateral ventricle (excluding the temporal horn), the right lateral ventricle (excluding the temporal horn), the right hip bone, the left hip bone, and the right caudate nucleus; The volume characteristics are: the volume values ​​of the aorta, pulmonary artery, left atrium, left iliac artery, and right iliac artery.

[0124] Comparative Example 3: Combination of Characteristic Markers The top eight features in terms of importance scores from the aforementioned SUVR metabolic (functional) and structural (volume) features were selected to construct an age prediction regression model. Specifically: Functional characteristics include: SUVR values ​​of the left lateral ventricle (excluding the temporal horn), right lateral ventricle (excluding the temporal horn), left hip bone, right hip bone, left caudate nucleus, right caudate nucleus, whole brain, and sacrum; The volume characteristics are: the volume values ​​of the aorta, pulmonary artery, left atrium, left iliac artery, right iliac artery, left iliopsoas muscle, right iliopsoas muscle, and the remaining area of ​​the lower lateral region of the right parietal lobe.

[0125] Comparative Example 4: Combination of Characteristic Markers Nine features with the highest importance scores from the aforementioned SUVR metabolic (functional) and structural (volume) features were selected to construct an age prediction regression model. Among them: Functional characteristics include: SUVR values ​​of the left lateral ventricle (excluding the temporal horn), right lateral ventricle (excluding the temporal horn), left hip bone, right hip bone, left caudate nucleus, right caudate nucleus, whole brain, sacrum, and right anterior cingulate cortex; The volume characteristics are: the volume values ​​of the aorta, pulmonary artery, left atrium, left iliac artery, right iliac artery, left iliopsoas muscle, right iliopsoas muscle, the remaining area of ​​the right parietal lobe, and the left first rib.

[0126] Experiment Example 1: Verification of the effect of feature marker combination A validation set of 71 healthy participants was collected, including 36 males and 35 females, aged 12-83 years. Following the data preprocessing workflow described above, the data of the validation set participants were... 18 F-FDG PET / CT images were registered and resampled, organs were automatically segmented, and the SUVR values ​​(functional features) and volume features of each organ were obtained. The features were Z-score standardized using the scale function of R software (version 4.1.1).

[0127] The standardized feature marker combinations of the validation set (Comparative Examples 1-4) were imported into Python software (version 3.9.13), and the optimal parameters were obtained as described above: The optimal regularization parameter alpha of the Lasso regression model is 0.1 (Comparative Examples 1-4).

[0128] The optimal parameters for SVR are: regularization parameter C = 100.0; gamma = 0.01; and kernel function type rbf (Comparative Examples 1-4).

[0129] The optimal parameters for the random forest model are: n_estimators = 200 and minimum number of splits = 10 (Comparative Example 1 and Comparative Example 2); n_estimators = 100 and minimum number of splits = 2 (Comparative Example 3); n_estimators = 100 and minimum number of splits = 5 (Comparative Example 4).

[0130] The optimal parameters for the gradient boosting model are learning_rate 0.1, maximum depth 3, and n_estimators 50 (Comparative Examples 1-3); and learning_rate 0.2, maximum depth 3, and n_estimators 50 (Comparative Example 4).

[0131] Then execute the code "from sklearn.metrics import mean_squared_error, mean_absolute_error, r 2 The performance evaluation metric for each model is calculated using the "_score" function, and the results are shown in Table 5.

[0132] Table 5 Predictive performance of Comparative Examples 1-4

[0133] The above results show that the combination of 14 feature markers in this invention is the optimal feature combination. This specific combination, in synergy with the optimal parameter SVR model, can achieve the highest aging age prediction accuracy and the strongest generalization performance in the independent validation set, and can be stably and accurately applied to the non-invasive assessment of the human aging clock.

[0134] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, and is not intended to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions made by those skilled in the art to the technical solution of the present invention do not depart from the essence and scope of the technical solution of the present invention.

Claims

1. A combination of biomarkers based on PET / CT molecular imaging, characterized in that, The feature marker combination consists of the SUVR value of each functional feature marker and the volume value of each volume feature marker; the feature marker combination is used as input to the support vector machine regression algorithm to construct an age assessment model; The aforementioned functional markers are: left lateral ventricle (excluding the temporal horn), right lateral ventricle (excluding the temporal horn), left hip bone, right hip bone, left caudate nucleus, right caudate nucleus, and the whole brain; The volumetric markers are: aorta, pulmonary artery, left atrium, left iliac artery, right iliac artery, left iliopsoas muscle, and right iliopsoas muscle.

2. The signature combination of claim 1, wherein, The SUVR value is calculated as follows: SUVR value of brain region marker = average SUV value of the corresponding brain region / average SUV value of the cerebellum; SUVR value of trunk region marker = average SUV value of the corresponding trunk region / average SUV value of the aortic blood pool; the volume value is a three-dimensional volume quantification index of the volume characteristic marker.

3. A PET / CT molecular imaging-based age assessment model, characterized by, The age assessment model comprises the combination of feature markers as described in any one of claims 1-2, and the age assessment model is constructed using a support vector machine regression algorithm.

4. The age assessment model of claim 3, wherein, The parameters of the support vector machine regression algorithm are: C=100.0, gamma=0.01, and kernel function is rbf.

5. A product for age assessment based on PET / CT molecular imaging, characterized in that, The age assessment product is used to detect the feature values ​​of the combination of feature markers according to any one of claims 1-2, wherein the feature values ​​are the SUVR values ​​of each functional feature marker and the volume values ​​of each volume feature marker.

6. The age assessment product of claim 5, wherein, The age assessment products mentioned include any one or more of reagents, kits, and chips.

7. An evaluation method based on PET / CT molecular imaging, characterized in that, The assessment methods include age assessment methods and human aging clock assessment methods, and the assessment methods include using the combination of characteristic markers as described in any one of claims 1-2, the age assessment model as described in any one of claims 3-4, or the age assessment product as described in any one of claims 6-7.

8. The evaluation method according to claim 7, characterized in that The age assessment method includes the following steps: (1) Obtain PET / CT molecular imaging data of the individual to be evaluated; (2) Detect the feature values ​​of the combination of feature markers; (3) Input the feature values ​​into the age assessment model and output the predicted age of the individual to be assessed to complete the age assessment.

9. The evaluation method according to claim 7, characterized in that The aforementioned method for assessing the human aging clock includes the following steps: S1. Obtain PET / CT molecular imaging data of the individual to be evaluated; S2. Detect the feature values ​​of the combination of feature markers; S3. Input the feature values ​​into the age assessment model and output the predicted age of the individual to be assessed; S4. Based on the difference between the predicted age and the actual age, the individuals to be evaluated are classified into slow aging, normal aging, and fast aging.

10. The evaluation method according to claim 9, characterized in that The dividing criterion is as follows: the first and third quartiles of the difference distribution are used as the dividing thresholds. A difference less than the first quartile indicates slow aging. The range between the first and third quartiles indicates normal aging; A value above the third or fourth quartile indicates rapid aging.

Images