Organ system phenotypic aging assessment method, device and equipment and storage medium

Abstract

The invention relates to an organ system phenotypic aging assessment method, device and equipment and a storage medium. According to the main technical scheme, the method comprises the steps of obtaining phenotypic aging marker information of all organ systems of a to-be-evaluated object, preprocessing the phenotypic aging marker information, determining a model modal type used for processing the phenotypic aging marker information according to the information type of the phenotypic aging marker information and the gender of the to-be-evaluated object, and determining the sex of the to-be-evaluated object according to the model modal type. Predicting the biological age of the organ system based on a pre-trained organ system aging model according to the model modal type and the pre-processed phenotypic aging marker information, and determining the age of the organ system according to the actual age of the to-be-evaluated object, the biological age of the organ system and pre-stored population sample information. And evaluating the phenotypic senescence state of the organ system of the to-be-evaluated object, so that the phenotypic senescence state of each organ system of the human body can be comprehensively and accurately evaluated.

Description

Technical Field

[0001] The present application relates to the technical field of aging health and anti-aging intervention health management, and in particular to a method, apparatus, device and storage medium for evaluating phenotypic aging of organ systems. Background Art

[0002] Human aging is a complex, multifactorial process influenced by both genetics and the environment, affecting various organs and systems throughout the body. It is considered the root cause of many chronic diseases, such as heart disease, neurodegenerative diseases, respiratory diseases, osteoporosis, and even cancer. Aging is a heterogeneous process, with different individuals aging at different rates. Aging involves nearly every organ system and involves numerous interconnected biological processes. Even within the same individual, the aging state of different organ systems varies. Furthermore, the aging of one organ selectively affects the aging of other organ systems, resulting in a multi-organ aging network.

[0003] Studies have found that the phenotypic markers of various organs in the human body change with age. By monitoring the changes in these phenotypic markers, adverse health changes caused by the aging of various organ systems can be discovered in a timely manner. At present, human aging assessments are mostly overall aging assessments, that is, a holistic aging assessment model is constructed using several or dozens of phenotypic markers. For individuals, this assessment is relatively rough. In practical applications, only the overall aging status of the individual is understood, but the aging status of the specific organ system is unknown. This is insufficient guidance for personalized precision health care and intervention. Although there are aging assessments that are targeted at a certain organ or system, such as kidney aging assessments and liver aging assessments, the aging of organ systems is correlated. Knowing only the aging status of a certain organ or system is still insufficient for aging intervention, and it is difficult to comprehensively and accurately assess the phenotypic aging status of various organ systems in the human body. Summary of the Invention

[0004] Based on this, the present application provides an organ system phenotypic aging assessment method, device, equipment and storage medium, which can comprehensively and accurately assess the phenotypic aging status of various organ systems of the human body.

[0005] In a first aspect, a method for assessing phenotypic aging of an organ system is provided, the method comprising: obtaining phenotypic aging marker information of each organ system of a subject to be assessed, and preprocessing the phenotypic aging marker information; Determining a model modality type for processing the phenotypic aging marker information based on the information type of the phenotypic aging marker information and the gender of the subject to be evaluated; According to the model modality type and pre-processed phenotypic aging marker information, the biological age of the organ system is predicted based on the pre-trained organ system aging model; The phenotypic aging status of the organ system of the subject to be evaluated is assessed based on the actual age of the subject to be evaluated, the biological age of the organ system and pre-existing population sample information.

[0006] According to one achievable method in an embodiment of the present application, the model modality types include a unimodal model and a multimodal model; determining the model modality type for processing the phenotypic aging marker information based on the information type of the phenotypic aging marker information and the gender of the subject to be evaluated includes: If the information type of the phenotypic aging marker information is one, determining that the model modality type used to process the phenotypic aging marker information is a single modality model corresponding to the gender of the subject to be evaluated; If the information type of the phenotypic aging marker information is at least two, the model modality type used to process the phenotypic aging marker information is determined to be a multimodal model corresponding to the gender of the subject to be evaluated.

[0007] According to one achievable method in an embodiment of the present application, the pre-stored population sample information includes sample phenotypic aging marker information, sample actual age, and sample gender of multiple sample individuals; based on the actual age of the subject to be evaluated, the biological age of the organ system, and the pre-stored population sample information, the aging status of the organ system of the subject to be evaluated is evaluated, including: Based on the sample genders of multiple sample individuals, the actual age of the sample is stratified according to a preset age gradient to obtain multiple sample age layers; Classify the sample individuals corresponding to each sample age layer in multiple sample age layers according to the organ system type, and obtain the classification group of each organ system in each sample age layer; Based on the sample phenotypic aging marker information and classification groups, the first quartile and third quartile of biological age for each classification group were determined based on the pre-trained organ system aging model; The phenotypic aging status of the organ system of the subject to be evaluated was assessed based on the chronological age of the subject to be evaluated, the biological age of the organ system, the first quartile of biological age for each classification group, and the third quartile of biological age.

[0008] According to one achievable method in an embodiment of the present application, the phenotypic aging state of the organ system of the subject to be evaluated is evaluated based on the actual age of the subject to be evaluated, the biological age of the organ system, the first quartile of biological age of each classification group, and the third quartile of biological age, including: Determine the age group to which the subject to be evaluated belongs based on the subject's actual age; Determine the first quartile and third quartile of biological age corresponding to the organ system of the subject to be evaluated based on the age group to which the subject to be evaluated belongs and the first quartile and third quartile of biological age of each classification group; The phenotypic aging status of the organ system of the subject to be evaluated is assessed based on the biological age of the organ system, the first quartile of biological age corresponding to the organ system, and the third quartile of biological age.

[0009] According to one achievable method in an embodiment of the present application, the single-modality model includes a first aging model, a second aging model, a third aging model, and a fourth aging model; before obtaining phenotypic aging marker information of each organ system of the subject to be evaluated, the method further includes: Acquiring sample information of various organ systems, wherein the sample information includes three-dimensional original images, two-dimensional original images, original measurement signals, quantitative phenotypic aging marker information, and actual age of the various organ systems; Preprocessing three-dimensional original images, two-dimensional original images, original measurement signals, and quantitative phenotypic aging marker information; According to the pre-processed three-dimensional original image and actual age, a pre-built three-dimensional convolutional neural network is trained to obtain a first aging model; According to the pre-processed two-dimensional original image and actual age, a pre-built two-dimensional convolutional neural network is trained to obtain a second aging model; According to the pre-processed raw measurement signals and actual age, a pre-built one-dimensional convolutional neural network is trained to obtain the third aging model; According to the pre-processed quantitative trait phenotypic aging marker information and actual age, the fourth aging model is constructed based on the preset machine learning algorithm.

[0010] According to one achievable method in an embodiment of the present application, before obtaining the phenotypic aging marker information of each organ system of the subject to be evaluated, the method further includes: Select at least two unimodal models; Randomly select a set of initial weight combinations, and determine the next set of optimal weight combinations based on the preset proxy model and the preset acquisition function; Determine the model weight combination based on the optimal weight combination and a preset validation set, wherein the preset validation set includes sample phenotypic aging marker information of different information types, sample actual age, and sample gender; A multimodal aging model is determined based on at least two unimodal models and a combination of model weights.

[0011] According to an achievable method in an embodiment of the present application, the method further includes: Based on the phenotypic aging status of the organ system of the subject to be evaluated, generating a phenotypic aging status evaluation result of each organ system; The phenotypic aging status assessment results of each organ system are displayed on the human organ diagram.

[0012] In a second aspect, a device for assessing aging of organ system phenotypes is provided, the device comprising: An acquisition module is used to obtain phenotypic aging marker information of each organ system of the subject to be evaluated and preprocess the phenotypic aging marker information; a determination module, configured to determine a model modality type for processing the phenotypic aging marker information based on the information type of the phenotypic aging marker information and the gender of the subject to be evaluated; A prediction module is used to predict the biological age of an organ system based on a pre-trained organ system aging model according to the model modality type and pre-processed phenotypic aging marker information; The evaluation module is used to evaluate the phenotypic aging status of the organ system of the subject to be evaluated based on the actual age of the subject to be evaluated, the biological age of the organ system and pre-stored population sample information.

[0013] According to a third aspect, a computer device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores computer instructions that can be executed by the at least one processor, and the computer instructions are executed by the at least one processor to enable the at least one processor to perform the method involved in the first aspect above.

[0014] In a fourth aspect, a computer-readable storage medium is provided, on which computer instructions are stored, characterized in that the computer instructions are used to enable a computer to execute the method involved in the above-mentioned first aspect.

[0015] According to the technical content provided in the embodiments of the present application, phenotypic aging marker information of each organ system of the subject to be evaluated is obtained and preprocessed. Based on the information type of the phenotypic aging marker information and the gender of the subject to be evaluated, the model modality type for processing the phenotypic aging marker information is determined. Based on the model modality type and the preprocessed phenotypic aging marker information, the biological age of the organ system is predicted based on a pre-trained organ system aging model. Based on the actual age of the subject to be evaluated, the biological age of the organ system, and pre-stored population sample information, the phenotypic aging status of the organ system of the subject to be evaluated is evaluated. This method can comprehensively and accurately evaluate the phenotypic aging status of various organ systems in the human body. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A diagram of an application environment for an organ system phenotypic aging method according to an embodiment; Figure 2 Schematic diagram of a process for organ system phenotypic aging in one embodiment; Figure 3 A structural block diagram of an organ system phenotypic aging device in one embodiment; Figure 4 Schematic diagram of the structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0017] The present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0018] For ease of understanding, the system to which this application is applicable is first described. The organ system phenotypic aging assessment method provided in this application can be applied to Figure 1 In the system architecture shown. The terminal 110 communicates with the server 120 via a network. The terminal 110 obtains the phenotypic aging marker information of each organ system of the subject to be evaluated, and pre-processes the phenotypic aging marker information. According to the information type of the phenotypic aging marker information and the gender of the subject to be evaluated, the model modality type used to process the phenotypic aging marker information is determined. According to the model modality type and the pre-processed phenotypic aging marker information, the biological age of the organ system is predicted based on the pre-trained organ system aging model. According to the actual age of the subject to be evaluated and the biological age of the organ system, the phenotypic aging status of the organ system of the subject to be evaluated is evaluated. The terminal 110 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers and portable wearable devices, and the server 120 can be implemented as an independent server or a server cluster consisting of multiple servers.

[0019] Figure 2 This is a flow chart of a method for evaluating organ system phenotypic aging provided in an embodiment of the present application. The method can be performed as follows: Figure 1 The terminal 110 in the system shown executes. Figure 2 As shown, the method may include the following steps: S210 , obtaining phenotypic aging marker information of each organ system of the subject to be evaluated, and preprocessing the phenotypic aging marker information.

[0020] Among them, organ system is a general term for organs or systems, including but not limited to brain organs, heart organs, lung organs, liver organs, kidney organs, vascular system, musculoskeletal system, metabolic system, and immune system.

[0021] Phenotypic aging marker information refers to biomarkers that can intuitively reflect the aging phenotype of cells or tissues. The screening process for phenotypic aging marker information involves searching several existing literature databases, such as PubMed (English), CNKI (Chinese), and the Aging Atlas, using the keywords "aging" and "senescence." For example, aging biomarkers associated with specific organ systems are obtained. These aging markers are then linked to healthcare big data resources, such as primary healthcare or commonly used clinical healthcare or diagnosis resources, to identify phenotypic aging markers and examination items for specific organ systems. After review by relevant professionals, a reference database of phenotypic aging biomarker characteristics for important human organ systems is constructed. Phenotypic aging marker information for the organ system being evaluated can be obtained from the database. By monitoring changes in these phenotypic markers, adverse health changes associated with aging in each organ system can be promptly detected.

[0022] The methods of obtaining phenotypic aging marker information for different organ systems are different. For example, the phenotypic aging marker information for brain organs is mainly imaging examination markers, including brain MRI original images, brain CT original images, etc.; the phenotypic aging marker information for heart organs is mainly clinical measurement markers, including electrocardiogram original measurement signals, etc.; the phenotypic aging marker information for lung organs is mainly imaging examination markers, including: chest CT original images, chest X-ray original images, etc., as well as clinical measurement markers: respiratory rate, forced vital capacity, forced expiratory volume in the first second, FEV1 / FVC, peak expiratory flow rate, etc.

[0023] Phenotypic aging markers for the liver are primarily laboratory markers, including alanine aminotransferase, aspartate aminotransferase, total bilirubin, direct bilirubin, indirect bilirubin, bile acid, γ-glutamyl transferase, alkaline phosphatase, lactate dehydrogenase, total protein, albumin, prealbumin, globulin, albumin-to-globulin ratio, cholinesterase, type III procollagen, type IV collagen, hyaluronic acid, laminin, and prothrombin time. Phenotypic aging markers for the kidney are primarily laboratory markers, including urea, creatinine, uric acid, cystatin C, β2-microglobulin, total protein, albumin, phosphate, sodium, potassium, chloride, and phosphorus.

[0024] The phenotypic aging markers of the vascular system mainly include clinical measurement markers, laboratory detection markers and imaging examination markers. Among them, clinical measurement markers include: systolic blood pressure, diastolic blood pressure, heart rate, left / right brachial-ankle pulse wave velocity, left / right ankle-brachial index, etc.; laboratory detection markers include: total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, etc.; imaging examination markers include: bilateral carotid artery ultrasound examination, including the inner diameter of the left common carotid artery, the thickness of the left carotid intima-media, The number of left carotid artery plaques, the degree of left carotid artery stenosis, the left pulse wave transmission velocity, the inner diameter of the right common carotid artery, the thickness of the right carotid intima-media, the number of right carotid artery plaques, the degree of right carotid artery stenosis, the right pulse wave transmission velocity, the carotid-femoral pulse wave conduction velocity, etc.; transcranial Doppler ultrasound examination, including the systolic peak velocity, end-diastolic velocity, average velocity, pulsatility index, resistance index, etc. of the middle cerebral artery, anterior cerebral artery, posterior cerebral artery, ophthalmic artery, internal carotid artery siphon segment, bilateral vertebral arteries, basilar artery, etc.

[0025] The phenotypic aging markers of the musculoskeletal system mainly include clinical measurement markers, laboratory detection markers and imaging examination markers. Among them, clinical measurement markers include grip strength, height, weight, BMI, waist and hip circumference, waist-to-hip ratio, ankle joint width, etc.; laboratory detection markers include blood calcium, phosphate, vitamin D3, parathyroid hormone, calcitonin, bone-specific alkaline phosphatase, osteocalcin, type I procollagen C-terminal propeptide / N-terminal propeptide, osteoprotegerin, tartrate-resistant acid phosphatase, etc.; imaging examination markers include dual-energy X-ray absorptiometry to measure bone density original images, dual-energy X-ray absorptiometry to measure bone density values, including cervical spine, thoracic spine, lumbar spine and other parts of each vertebra and the total BMD (g / cm 2 )、BMC(g)、Area(cm 2 ), T-score, Z-score; human body composition indicators measured by electrical impedance method or dual-energy x-ray absorptiometry, including body fat percentage, fat mass, non-fat mass, body water content, and body water percentage.

[0026] The phenotypic aging marker information of the metabolic system mainly includes clinical measurement markers, including: basal metabolic rate, BMI, waist circumference, hip circumference, waist-to-hip ratio, systolic blood pressure, diastolic blood pressure, etc.; and laboratory detection markers, including: total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, lipoprotein A, apolipoprotein A, apolipoprotein B, fasting blood glucose, glycated hemoglobin, fasting insulin, thyroid-stimulating hormone, alanine aminotransferase, aspartate aminotransferase, blood creatinine, blood uric acid, blood urea, etc.

[0027] The information on phenotypic aging markers of the immune system is mainly laboratory test markers, including: C-reactive protein, white blood cell count, neutrophil count, neutrophil percentage, lymphocyte count, lymphocyte percentage, monocyte count, monocyte percentage, eosinophil count, eosinophil percentage, basophil count, basophil percentage, red blood cell count, hematocrit, mean corpuscular volume, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, red blood cell distribution width, hemoglobin, mean platelet volume, platelet count, hematocrit, platelet distribution width, large platelet ratio, total T lymphocyte count, helper T lymphocyte count, suppressor T lymphocyte count, suppressor T lymphocyte percentage, CD3 cell count, CD4 cell count, CD8 cell count, etc.

[0028] Preprocessing operations are performed on three-dimensional original images (such as MRI original images, CT original images) and two-dimensional original images (such as X-ray original images), including format conversion, truncation normalization, and resampling; the original signal data are preprocessed by format conversion, denoising, unified sampling rate, signal normalization, signal segmentation, etc.; clinical measurement data and laboratory test data are matched according to the organ system phenotypic aging marker system to prepare for the subsequent call of the organ system-specific aging model for calculation.

[0029] S220 , determining a model modality type for processing the phenotypic aging marker information according to the information type of the phenotypic aging marker information and the gender of the subject to be evaluated.

[0030] Phenotypic aging marker information is primarily categorized by data format into three-dimensional raw imagery, two-dimensional raw imagery, raw measurement signals, and quantitative phenotypic aging markers. Model modality types include unimodal models and multimodal models. Unimodal models include a first aging model, a second aging model, a third aging model, and a fourth aging model, which are organ system aging models trained based on three-dimensional raw imagery, two-dimensional raw imagery, raw measurement signals, and quantitative phenotypic aging markers, respectively. Multimodal models are trained based on at least two of the unimodal models. Both unimodal and multimodal models are constructed separately based on gender, meaning each type of organ system aging model is divided into male and female categories.

[0031] First, determine the types of information available for phenotypic aging markers for each organ system. If only one type of information is available, select a single-modal model; if multiple types are available, select a multimodal model. Then, based on the gender of the subject being evaluated, select an organ system aging model from the selected modality models.

[0032] S230, predicting the biological age of the organ system based on a pre-trained organ system aging model according to the model modality type and the pre-processed phenotypic aging marker information.

[0033] Pre-trained organ system aging models include pre-trained single-modality and multi-modality models, including the first aging model, the second aging model, the third aging model, the fourth aging model, and the multi-modality aging model. Based on the model modality type, the corresponding organ system aging model for each organ system is selected from the pre-trained organ system aging models. The pre-processed phenotypic aging marker information serves as the input for each organ system's corresponding organ system aging model, and the biological age of the organ system serves as the output of the organ system aging model. The training process for each organ system aging model will be detailed later.

[0034] S240, assessing the phenotypic aging status of the organ system of the subject to be assessed based on the actual age of the subject to be assessed, the biological age of the organ system, and pre-stored population sample information.

[0035] The pre-stored population sample information comes from pre-stored medical and health big data, and may include the sample phenotypic aging marker information, sample actual age, and sample gender of multiple sample individuals. Multiple sample individuals are stratified according to the sample actual age, and sample individuals in the same age group are classified according to organ system. The biological age of the organ system of the sample individual is obtained based on the sample phenotypic aging marker information and the pre-trained organ system aging model. The interquartile range of each organ system in each age group is calculated according to the biological age of the organ system. The phenotypic aging state of the organ system of the subject to be evaluated is evaluated based on the range of the organ system's biological age within the interquartile range of the organ system in the age group of the subject's actual age. Phenotypic aging states include decelerated aging state, accelerated aging state, and normal aging state.

[0036] It can be seen that the embodiments of the present application obtain phenotypic aging marker information of each organ system of the subject to be evaluated, pre-process the phenotypic aging marker information, determine the model modality type for processing the phenotypic aging marker information according to the information type of the phenotypic aging marker information and the gender of the subject to be evaluated, predict the biological age of the organ system based on the model modality type and the pre-processed phenotypic aging marker information based on the pre-trained organ system aging model, and evaluate the phenotypic aging status of the organ system of the subject to be evaluated according to the actual age of the subject to be evaluated, the biological age of the organ system, and pre-stored population sample information, thereby comprehensively and accurately evaluating the phenotypic aging status of each organ system of the human body.

[0037] As an achievable approach, the model modality type for processing the phenotypic aging marker information is determined based on the information type of the phenotypic aging marker information and the gender of the subject to be evaluated, including: If the information type of the phenotypic aging marker information is one, determining that the model modality type used to process the phenotypic aging marker information is a single modality model corresponding to the gender of the subject to be evaluated; If the information type of the phenotypic aging marker information is at least two, the model modality type used to process the phenotypic aging marker information is determined to be a multimodal model corresponding to the gender of the subject to be evaluated.

[0038] If the information type of the phenotypic aging marker information is one, a single-modal model is directly selected to process the phenotypic aging marker information based on the gender of the subject to be evaluated. If the information type of the phenotypic aging marker information is at least two, and the single-modal model cannot fully predict the biological age of the organ system, a multimodal model is selected to process the phenotypic aging marker information.

[0039] As a feasible approach, the aging status of the subject's organ system can be assessed based on the subject's actual age, the biological age of the organ system, and pre-existing population sample information, including: Based on the sample genders of multiple sample individuals, the actual age of the sample is stratified according to a preset age gradient to obtain multiple sample age layers; Classify the sample individuals corresponding to each sample age layer in multiple sample age layers according to the organ system type, and obtain the classification group of each organ system in each sample age layer; Based on the sample phenotypic aging marker information and classification groups, the first quartile and third quartile of biological age for each classification group were determined based on the pre-trained organ system aging model; The phenotypic aging status of the organ system of the subject to be evaluated was assessed based on the chronological age of the subject to be evaluated, the biological age of the organ system, the first quartile of biological age for each classification group, and the third quartile of biological age.

[0040] The preset age gradient can be set according to research needs. For example, it can be set to 1, that is, 1 year old is 1 age layer. Based on the sample genders of multiple sample individuals, the actual age of the samples is stratified according to the preset age gradient to obtain multiple sample age layers, including sample age layers corresponding to males and sample age layers corresponding to females.

[0041] Individual samples within the same age group were classified by organ system type to obtain classification groups for each organ system within each age group. The corresponding phenotypic aging marker information for each classification group was input into a pre-trained organ system aging model to obtain the biological age of the organ systems of multiple individual samples within each classification group.

[0042] Arrange the biological ages of the organ systems of multiple samples from small to large and divide them into four equal parts. The values ​​at the three split points are the quartiles. The value at the first split point is determined as the first quartile of the biological age of each classification group, recorded as , the value of the third split point is determined as the third quartile of biological age of each classification group, recorded as The difference between the third and first quartiles is also called the interquartile range. This allows the establishment of a reference database of biological age interquartile ranges for each organ system and age group for both men and women.

[0043] After obtaining the reference database, the phenotypic aging status of the organ system of the subject to be evaluated is assessed based on the actual age of the subject to be evaluated, the biological age of the organ system, the first quartile of biological age of each classification group, and the third quartile of biological age.

[0044] Specifically, according to the actual age of the subject to be evaluated, the age group to which the subject to be evaluated belongs is determined; Determine the first quartile and third quartile of biological age corresponding to the organ system of the subject to be evaluated based on the age group to which the subject to be evaluated belongs and the first quartile and third quartile of biological age of each classification group; The phenotypic aging status of the organ system of the subject to be evaluated is assessed based on the biological age of the organ system, the first quartile of biological age corresponding to the organ system, and the third quartile of biological age.

[0045] Search the reference database for the age group to which the subject's actual age belongs. Find the classification group within this age group that has the same type of organ system as the subject's. Determine the first and third quartiles of biological age corresponding to this classification group as the first and third quartiles of biological age corresponding to the subject's organ system. Assess the phenotypic aging status of the subject's organ system using the following formula: ; in, represents the phenotypic aging status of the organ system i of the subject to be evaluated, It represents the biological age of the organ system i of the subject to be evaluated in the age group j to which the actual age belongs, It represents the first quartile of biological age of organ system i in age group j of actual age. It represents the third quartile of biological age of organ system i in age group j of actual age. Indicates that the organ system i of the subject to be evaluated is in a state of decelerated aging, Indicates that the organ system i of the subject to be evaluated is in a normal aging state, It indicates that the organ system i of the subject to be evaluated is in an accelerated aging state.

[0046] As an achievable approach, before obtaining the phenotypic aging marker information of each organ system of the subject to be evaluated, the method further includes: Obtaining sample information of each organ system, wherein the sample information includes three-dimensional original images, two-dimensional original images, original measurement signals, quantitative phenotypic aging marker information, and actual age of each organ system; Preprocessing three-dimensional original images, two-dimensional original images, original measurement signals, and quantitative phenotypic aging marker information; According to the pre-processed three-dimensional original image and actual age, a pre-built three-dimensional convolutional neural network is trained to obtain a first aging model; According to the pre-processed two-dimensional original image and actual age, a pre-built two-dimensional convolutional neural network is trained to obtain a second aging model; According to the pre-processed raw measurement signals and actual age, a pre-built one-dimensional convolutional neural network is trained to obtain the third aging model; According to the pre-processed quantitative trait phenotypic aging marker information and actual age, the fourth aging model is constructed based on the preset machine learning algorithm.

[0047] Obtain sample information of various organ systems from the database and pre-process the sample information. The original three-dimensional images include but are not limited to original brain MRI images, original chest CT images, etc. The process of pre-processing the original three-dimensional images is as follows: Raw image conversion: 3D raw images are in DICOM format and are converted to NIFTI, a widely accepted standard that supports 3D data. NIFTI files typically contain an image data file (.nii) and a header file (.hdr), which retains important metadata information.

[0048] Truncation normalization: Based on professional knowledge, the threshold range of the image is determined, such as the typical CT value range of lung tissue [-1024, 600]. Pixel values ​​are truncated according to the threshold range, that is, values ​​less than the lower limit are set to the lower limit, and values ​​greater than the upper limit are set to the upper limit. The maximum-minimum normalization method is then used for image processing. The calculation formula of the maximum-minimum normalization method is expressed as: ; in, is the value of each pixel in the three-dimensional original image, is the minimum pixel value in the original three-dimensional image. is the maximum pixel value in the original 3D image, It is the normalized pixel value in the original 3D image. After normalization, all pixel values ​​will fall within the range [0,1].

[0049] Resampling: First, set the target number of voxels, such as 112*112*224, then calculate the scaling factor based on the current number of voxels and the target number of voxels, and then use the cubic interpolation method to resample the image to a standardized image of the target number of voxels.

[0050] The two-dimensional original image images include but are not limited to chest X-ray original image images, bone X-ray original image images, etc. The process of preprocessing the two-dimensional original image images is as follows: Raw image conversion: 2D raw image images are usually in DICOM format and need to be converted to PNG image format, which is a lossless compression format suitable for maintaining high-quality images.

[0051] Truncation normalization: Determine the image threshold range based on professional knowledge. For X-ray images, pixel values ​​represent the degree of X-ray absorption by different tissues. For example, the effective pixel range of a lung X-ray image is roughly between [0, 4095]. To highlight lung tissue structure, a range of [0, 3000] might be selected to optimize image quality and reduce unnecessary noise and background interference. Pixel values ​​are truncated according to the threshold range. Values ​​below the lower limit are set to the lower limit, and values ​​above the upper limit are set to the upper limit. The maximum-minimum normalization method described above is then used to adjust the pixel value range to [0, 1].

[0052] Resampling: First, set the target number of voxels, such as 256×256 or 512×512, calculate the scaling factor based on the current number of voxels and the target number of voxels, and then use the bilinear interpolation method to resample the image to a standardized image of the target number of voxels to unify the size differences between different images and reduce computational complexity.

[0053] The original measurement signal includes but is not limited to electrocardiogram (ECG), etc. The process of preprocessing the original measurement signal is as follows: The original ECG measurement signals come in different formats and are therefore uniformly converted to the WFDB format. ECG signals may contain various noises, such as myoelectric noise and baseline drift. The effective frequency range of ECG signals is between 0.5 Hz and 40 Hz, and a bandpass filter is used to remove these noises. Because ECG data from different sources may have different sampling rates, resampling techniques are used to standardize all ECG signals to a fixed sampling rate to ensure consistent frequency resolution. ECG signals are normalized using the maximum-minimum normalization method, scaling the signal values ​​to the [0, 1] range.

[0054] After normalization, the original measurement signal is segmented. Specifically, an event detection-based segmentation method is used to identify and segment the R wave peak of the key event in the original measurement signal. First, the Pan-Tompkins algorithm is used to accurately detect the position of the R wave from the original ECG signal x(t). A bandpass filter is used to remove high-frequency noise and low-frequency baseline drift, and the filtered signal is differentiated to enhance the slope information of the R wave. Assume that the filtered signal is , then the signal obtained after differentiation is , square the differentiated signal , amplify the difference between the R wave and other waveforms, and integrate the signal within a certain time window , T is the integration window length, forming a new signal that is easy to detect the R wave peak; Next, determine the position of the R wave peak in each heart cycle. This can be done by finding the local maximum; assuming that the integrated signal z(t) has been calculated, the position of the i-th R wave can be found as follows: ,in, and After defining the search range and determining the positions of all R waves, the ECG signal can be segmented based on these positions. Taking each R wave as the center, we take 5 seconds before and after as a segment. For the i-th segment, its start and end times are: , , so the i-th segment is represented as , that is, from the original signal Extracted from arrive Data within a time period.

[0055] Since the lengths of different heartbeat cycles may be inconsistent, the lengths of the segments after segmentation may also be different. In order to make all segments suitable for model input, padding or truncation methods are used to achieve a uniform length L. If a segment If the length of a segment is greater than L, you can choose to keep the middle part or the two ends to reach the desired length. If the length of a segment is less than L, you can add a specific value until the length reaches L. Assuming the target length is L, the final processed segment It can be obtained in the following ways: .

[0056] The quantitative phenotypic aging marker information comes from medical and health big data resources, and there will be problems such as outliers and erroneous data in the data set. Therefore, the process of preprocessing the quantitative phenotypic aging marker information is as follows: processing errors, mixed Chinese characters, symbols, etc. in the data, and treating them as missing values; according to the distribution of each marker, combined with the three-quartile range, The percentiles P0.05 and P99.95 were used to determine the appropriate cutoff value range for each marker for outlier processing, and data outside the cutoff value were treated as missing values; data samples with missing values ​​were deleted; finally, the unique identifier was used to eliminate multiple health and medical data of the same person, and only the earliest health and medical data of the same person was retained; for the obtained marker data, in order to avoid systematic errors, all quantitative phenotypic aging markers in the data set needed to be normalized. The method used was maximum-minimum normalization, and the calculation formula was the same as described above.

[0057] After preprocessing the sample information, model training is performed based on the preprocessed sample information. The first aging model is an organ system aging model obtained by training a pre-built 3D convolutional neural network based on the pre-processed 3D original images and actual age. The pre-built 3D convolutional neural network is an enhanced 3D convolutional neural network, an extension of the traditional 3D-Convolutional Neural Networks (3D-CNN). It includes an input layer, multiple convolutional layers, a pooling layer, and integrates an attention mechanism module and a Transformer module. Finally, it is connected to the output layer via a fully connected layer. Integrating the Transformer module into the 3D-CNN architecture can overcome the limitation of the traditional 3D-CNN, which focuses on local feature extraction. Unlike 3D-CNN, which captures local image details through convolution operations, the Transformer utilizes a self-attention mechanism to evaluate and emphasize the relationships between different regions in the image, thereby identifying long-range dependencies, which helps to more accurately identify biological aging-related or abnormal conditions.

[0058] Specifically, the input layer accepts a preprocessed 3D raw image as input. The convolutional layer uses a 3D convolution kernel (typically shaped like depth × height × width) to perform element-wise multiplication and summation on the volume at the corresponding position of the input data, generating a new 3D output data volume to extract image features.

[0059] The pooling layer is a layer that reduces the data dimension and computational complexity while maintaining the key features of the data. 3D-CNN uses a 3D pooling layer, which uses a three-dimensional pooling window to slide on the input feature map and performs maximum pooling or average pooling on the data in each window to generate a new three-dimensional feature map whose size is smaller than the original feature map.

[0060] The attention mechanism module can use either spatial attention or channel attention, or a combination of both. This module can focus on the most critical parts of the input data, significantly increasing the model's focus on important information, thereby improving model performance. The spatial attention mechanism uses each pixel in the current feature map as a unit, assigning a weight to each pixel. This allows the network to focus on important areas of the image and ignore unimportant parts. The channel attention mechanism uses the current feature map as a unit, assigning a weight to each channel. This allows the network to distinguish the importance of different channels, further improving feature representation capabilities.

[0061] The self-attention mechanism of the Transformer module can capture the global features of the image. Suppose the input feature map is , where L is the sequence length, is the feature dimension; in order to introduce position information, position encoding can be added to the input feature map, where the position encoding can be learned or generated based on sine / cosine functions: , where X refers to the input feature map, PE is the position encoding matrix, which is a pre-calculated matrix used to add unique position information to each position in the sequence; It is a feature map with position encoding added. The position encoding PE is added to the original input feature X. Each feature vector not only retains the original content information, but also incorporates the information of the corresponding position. The nonlinear activation function ReLU is introduced after the linear transformation to convert the input feature map into query (Q), key (K) and value (V): , , ,middle , , is the weight matrix, is the dimension of each attention head, Is a learnable bias matrix used to represent the relative position relationship. For each attention head, the attention score is calculated based on the following formula: ; Concatenate the outputs of multiple attention heads: ; Where h is the number of attention heads, is the output weight matrix. The output of the multi-head self-attention mechanism is further transformed through a feedforward network: ;in, are weight and bias parameters. To stabilize training, layer normalization and residual connections are added after the multi-head self-attention mechanism and feedforward network: ; in, represents the input feature matrix, It is the output of the multi-head self-attention mechanism, which captures information in different subspaces through multiple attention heads; Represents the output feature matrix after multi-head self-attention mechanism and layer normalization; represents the layer normalization technology; Y represents the output feature matrix after the multi-head self-attention mechanism, residual connection and layer normalization; represents the output of a feedforward neural network.

[0062] Fully connected layers further learn higher-level abstract representations of the previously extracted features and map these features to the output of the biological age regression prediction task. Fully connected layers can be single or multi-layered. The multidimensional feature map is first flattened into a two-dimensional vector, where the first dimension is the batch_size. This flattened vector is then used as input and linearly transformed through matrix multiplication and bias addition. Nonlinear mapping is performed using activation functions. Common activation functions include ReLU, Sigmoid, and Tanh. To prevent overfitting during model training, regularization techniques and appropriate training strategies are incorporated into each fully connected layer, including weight decay (L2 regularization), dropout, batch normalization, a learning rate scheduler, and early stopping.

[0063] Output layer: The output of the organ system aging model is the organ-specific biological age. It is a regression task, so the output layer is a simple linear layer that can output a continuous value representing the predicted age.

[0064] The organ system aging model is a regression task. Stochastic gradient descent (SGD), Adam, RMSprop, etc. are selected as optimizers to update the network weights in order to minimize the loss function. The loss function used is the mean squared error (MSE), which is specifically: ; where n is the number of images, is the actual age of the i-th image, is the biological age predicted based on the i-th image, as output by the organ system aging model. The smaller the MSE value, the better the accuracy of the prediction model.

[0065] The training process of the first aging model specifically includes: Step S1: construct an enhanced three-dimensional convolutional neural network and initialize the parameters of the enhanced three-dimensional convolutional neural network.

[0066] Step S2: Use a large number of unlabeled 3D original images for self-supervised learning to train a universal feature extractor, and then save its pre-trained weights for subsequent transfer learning.

[0067] Step S3: Load the pre-trained weights as initial weights for the enhanced 3D convolutional neural network. The dataset is divided into training, validation, and test sets, with a ratio of 70%, 15%, and 15%, respectively. The training set samples are fed into the pre-built 3D convolutional neural network to obtain prediction results. The loss function (MSE) is used to calculate the mean square error (MSE). The error is backpropagated based on the loss value, and the model weights are updated to fine-tune the model parameters, ultimately resulting in the first aging model.

[0068] Step S4: Optimal hyperparameters are selected for model tuning of the first aging model through grid search or Bayesian optimization combined with early stopping. Hyperparameters include learning rate, batch size, optimizer type, regularization coefficient, etc.

[0069] Bayesian optimization methods approximate the objective function by constructing a surrogate model and defining an acquisition function based on this model to measure the benefits of potential evaluation points. The acquisition function includes but is not limited to probability improvement, expected improvement, and upper confidence bounds. The surrogate model is an approximate model used to replace the true objective function. Its purpose is to use a simpler model to help understand the behavior of the objective function when the objective function evaluation is expensive or computationally complex. At the end of each epoch, the model performance is evaluated using the validation set, and the model weights that perform best on the validation set are saved. If the validation set loss stops decreasing or starts to increase for multiple consecutive epochs (such as 5 or 10), early stopping is triggered and training is terminated.

[0070] Step S5: Based on the input test sample set, the accuracy performance of the first aging model is evaluated. The evaluation criteria of the estimation algorithm are Pearson correlation coefficient r, mean square error MAE, and determination coefficient .

[0071] The calculation formula of Pearson correlation coefficient is as follows: ; in, represents the actual age corresponding to the i-th sample, is the mean value of x, represents the predicted age corresponding to the i-th sample, yes N is the number of samples, and r represents the strength of the linear association between predicted age and actual age.

[0072] The formula for calculating the mean square error is as follows: ; in, represents the predicted biological age based on the prediction of the Ith sample, represents the actual age of the I-th sample, N is the sample size, MAE is the average error between actual age and biological age.

[0073] The formula for calculating the coefficient of determination is as follows: ; in, represents the biological age predicted based on the i-th sample, represents the actual age of the i-th sample, yes The mean value of , N is the number of samples, Indicates the percentage explained by the regression relationship between biological age and chronological age.

[0074] The second aging model is an organ system aging model trained using a pre-built 2D convolutional neural network (2D-CNN) based on pre-processed 2D raw images and actual age. This pre-built 2D convolutional neural network is an enhanced 2D convolutional neural network that integrates a Transformer module within the architecture of traditional 2D-Convolutional Neural Networks (2D-CNNs). This module overcomes the limitations of traditional 2D-CNNs, which primarily focus on local feature extraction. The Transformer's self-attention mechanism effectively captures long-range dependencies, enabling the model to focus not only on local details but also understand the context of the entire image, helping to more accurately identify features or abnormalities associated with biological aging.

[0075] The pre-built 2D convolutional neural network includes an input layer, multiple convolutional layers, a pooling layer, an attention mechanism module, feature map flattening or rearrangement, and a Transformer module, and is finally connected to the output layer through a fully connected layer.

[0076] Among them, the input layer accepts the preprocessed two-dimensional original image as input, such as a preprocessed lung X-ray image.

[0077] The convolution layer uses a two-dimensional convolution kernel, usually with a shape of height × width, to multiply the volume at the corresponding position of the input data element by element and sum them up to generate a new two-dimensional output data volume to extract image features.

[0078] Pooling layer: To preserve the key features of an image while also understanding it at multiple scales, a 2D pooling layer combined with hybrid pooling is used. This method defines a two-dimensional pooling window and slides it across the input feature map. For each pooling window, the maximum and average values ​​of the pixels within the window are calculated. A weight coefficient α (0 ≤ α ≤ 1) is then assigned to the two values, which are then weighted together. This weight coefficient α is included in the network as a parameter and can be trained like other model parameters. Feature maps processed by hybrid pooling are smaller than the original.

[0079] The attention mechanism module is the same as the attention mechanism module in the pre-built 3D convolutional neural network and will not be described here.

[0080] Flattening or rearranging feature maps: For the original two-dimensional image, a series of feature maps are first extracted through a series of convolutional layers and pooling layers. The feature maps adjusted by the attention mechanism are then flattened or rearranged into a sequence form (i.e., a sequence of length L). This change allows the image information to be input into the Transform for processing in sequence form.

[0081] The Transformer module is introduced, leveraging its self-attention mechanism to capture global image features. The detailed description of the Transformer module, fully connected layers, and output layer is the same as that of the Transformer module in the pre-built 3D convolutional neural network and is not repeated here.

[0082] The training process of the second aging model is basically the same as that of the first aging model, except that the input sample information is different. The two-dimensional original image is used as the input of the pre-built two-dimensional convolutional neural network and substituted into the training process of the first aging model to obtain the second aging model. It will not be repeated here.

[0083] The third aging model is an organ system aging model trained using a pre-built one-dimensional convolutional neural network (1D-CNN) based on pre-processed raw measurement signals and actual age. This pre-built 1D-CNN is an enhanced 1D-CNN that integrates a Transformer module within the traditional 1D-Convolutional Neural Network (1D-CNN) architecture. The convolution kernel of a 1D-CNN slides only in one dimension, effectively capturing local features in time series data. The Transformer model excels at processing sequential data. Unlike the 1D-CNN, which searches for local features through a sliding window, the Transformer model calculates the association weights between elements in the input sequence, better capturing long-range dependencies. Combining these two approaches allows the 1D-CNN to fully leverage its ability to extract local features while leveraging the Transformer to enhance its understanding of global context, thereby improving the model's overall performance.

[0084] The pre-built one-dimensional convolutional neural network includes an input layer, a convolutional layer, a pooling layer, a Transformer module, a fully connected layer, and an output layer.

[0085] Input layer: receives the preprocessed raw measurement signal as input.

[0086] Convolutional layer: A set of learnable convolution kernels slides over the input signal, extracting local features from the input signal and helping the model identify important features in the original measurement signal, such as P waves, QRS complexes, and T waves. Using convolution kernels of different sizes in the convolution layer can capture features of different scales in the ECG signal. Therefore, convolution kernels of lengths of 3, 5, and 7 are used simultaneously, and their outputs are fused, including but not limited to concatenation, element-by-element addition, attention fusion, etc. This can more comprehensively extract information from the signal and help the model identify waveform features of various widths in the original measurement signal. The formula for the convolution kernel is as follows: ; in, Indicates the The output of the i-th convolution kernel in the layer, It is The weight of the convolution kernel of layer i, It is The output of the layer, is the bias, and f is the activation function. Placing a ReLU activation function after each convolutional layer introduces nonlinearity, enabling the network to learn and represent complex patterns and alleviating the vanishing gradient problem. Residual connections are added after some or all convolutional layers. Residual connections help address the vanishing gradient problem encountered during deep network training by directly passing the input to the next layer. For the residual block between layer l and layer l+1, the formula is: , where x is the input and F(x) is the output of the convolutional layer. is the output of the residual block. If the input and output dimensions are inconsistent, additional convolution operations are required to adjust the dimensions.

[0087] Transformer module: The Transformer module captures global dependencies between different locations in the raw measurement signal and is particularly suitable for processing long sequences and complex periodic features in the raw measurement signal. The detailed description of the Transformer module is the same as that of the pre-built 3D convolutional neural network and is not repeated here.

[0088] Pooling layer: The pooling layer helps reduce the size of the feature signal, reduces computational complexity, and improves the robustness of the model to small changes in the input data. 1D-CNN uses an appropriate one-dimensional pooling window size and uses padding to keep the size of the output feature signal unchanged. It performs maximum pooling or average pooling on the data within each window, gradually reducing the size of the feature map.

[0089] Fully connected layer: The fully connected layer integrates the previously extracted local features into a global feature representation and maps these features to the output of the biological age regression prediction task. Fully connected layers can be single or multi-layered. The features are first flattened, then the flattened vector is used as input. Linear transformations are performed via matrix multiplication and bias addition, and nonlinear mapping is performed using the ReLU activation function. To prevent overfitting during model training, regularization techniques and appropriate training strategies are incorporated into each fully connected layer, including weight decay (L2 regularization), dropout, batch normalization, a learning rate scheduler, and early stopping.

[0090] Output layer: The output of the organ-specific aging model is the organ-specific biological age. It is a regression task, so the output layer is a simple linear layer that can output a continuous value representing the predicted biological age.

[0091] The organ system aging model is a regression task. Stochastic gradient descent (SGD), Adam, RMSprop, etc. are selected as optimizers to update the network weights in order to minimize the loss function. The loss function used is the mean square error, and the calculation formula is the same as above.

[0092] The training process of the third aging model is basically the same as that of the first aging model, except that the input sample information is different. The one-dimensional original image is used as the input of the pre-built one-dimensional convolutional neural network and substituted into the training process of the first aging model to obtain the third aging model. No further details will be given here.

[0093] The fourth aging model is an organ system aging model trained using a preset machine learning algorithm based on preprocessed quantitative phenotypic aging marker information and actual age. Preset machine learning algorithms may include support vector machines, progressive gradient regression trees, elastic net regression, k-nearest neighbors, random forests, deep neural networks, and the like.

[0094] Taking the construction of the fourth aging model based on a deep neural network algorithm as an example, the deep neural network model is first established, and the relevant model parameters and hyperparameters, including weights, biases, and the gradient descent learning rate, are initialized. Next, the model pre-training phase begins: information on quantitative phenotypic aging markers is used as input to the deep neural network model, and training is performed layer by layer. The output of the previous neuron serves as the input to all the next neurons, and training is repeated until the initialization parameters of each layer are updated. Finally, the neural network is trained. The error between the model output and the actual age is used to correct the parameters using backpropagation, thereby fine-tuning the model parameters, ultimately obtaining the deep neural network training model. The trained model can then be fine-tuned by selecting the optimal hyperparameters through grid search, including the learning rate α, the number of hidden layer nodes, the mini-batch size, and the momentum parameter β.

[0095] We can also use the random forest algorithm to construct the fourth aging model as an example. The quantitative phenotypic aging marker information and actual age are used as the original training sample set A. For each feature, the model training process includes the following steps: Step 1: Use the bootstrap resampling method to extract with replacement from the original training sample set A, let Ak represent the k-th sample set, and at the same time, the k-th regression tree generates a random vector αk, and use h(Ak,αk) to represent the k-th regression tree model.

[0096] Step 2: Build a decision regression tree for each of the k samples. The decision regression tree uses the minimum mean square error. For each decision regression tree, a portion of features is randomly selected from all features when each node is divided, such as or , and then select the optimal partitioning feature and partition point. Specifically, for any partitioning feature D, the corresponding data set on either side of the partition point s is divided into E1 and E2. The feature and eigenvalue partition point that minimizes the mean squared error of the respective data sets on both sides and minimizes the sum of the mean squared errors of E1 and E2 is the corresponding feature and eigenvalue partition point, where c1 is the mean sample output corresponding to data set E1, and c2 is the mean sample output corresponding to data set E2. Each decision regression tree is grown as far as possible without pruning.

[0097] Step 3: Repeat step 2. Each decision regression tree gives its own prediction results. Finally, k regression trees are built together to form a random forest.

[0098] Step 4: Use the simple averaging method to average the biological ages predicted by all decision tree regressors to obtain the final predicted biological age.

[0099] As an achievable approach, before obtaining the phenotypic aging marker information of each organ system of the subject to be evaluated, the method further includes: Select at least two unimodal models; Randomly select a set of initial weight combinations, and determine the next set of optimal weight combinations based on the preset proxy model and the preset acquisition function; Determine the model weight combination based on the optimal weight combination and a preset validation set, wherein the preset validation set includes sample phenotypic aging marker information of different information types, sample actual age, and sample gender; A multimodal aging model is determined based on at least two unimodal models and a combination of model weights.

[0100] According to the type of phenotypic aging marker information of the organ system, at least two single-modal models are selected. That is, combined training can be performed based on any at least two models among the first aging model, the second aging model, the third aging model, and the fourth aging model to meet the evaluation requirements of the organ system.

[0101] After determining a suitable unimodal model, the weights of the unimodal model are automatically learned using Bayesian optimization. This involves randomly selecting an initial set of weight combinations, where each weight parameter in the initial set sums to 1. A pre-defined surrogate model is constructed based on the weight combinations obtained during the initialization phase and their corresponding performance metrics. A pre-defined acquisition function, including but not limited to expected improvement, probability improvement, or an upper confidence bound, is used to determine the weight parameters for the next evaluation. Based on the pre-defined surrogate model and the pre-defined acquisition function, the next optimal set of weight combinations is determined. This optimal set of weights is used to make predictions on the validation set, and the corresponding mean squared error (MSE) is calculated. The optimal set of weights is then added to the existing training data set, and the pre-defined surrogate model is retrained. The above steps are repeated. If the loss on the pre-defined validation set stops decreasing or begins to increase, early stopping is triggered and training is terminated. Finally, based on the best weight combinations recorded across all iterations, the set of weights that achieves the optimal MSE is selected as the model weight combination. The model weight combination includes the weight parameters for each organ system aging model being trained. Based on at least two unimodal models and model weight combinations, a weighted average algorithm is used to set the logical relationship between multiple unimodal models to obtain a multimodal model, so that the multimodal aging model outputs an accurate biological age.

[0102] Taking the organ system as an example, which includes information on three phenotypic aging markers, the multimodal aging model can calculate the biological age of the organ system according to the following formula: ; in, is the biological age of the ith organ system, is the biological age of the ith organ system predicted by the first unimodal model, is the biological age of the i-th organ system predicted by the second unimodal model, is the biological age of the i-th organ system predicted by the third unimodal model, , and are the optimal weights obtained by Bayesian optimization method.

[0103] The phenotypic aging status assessment results of each organ system are displayed on the human organ diagram.

[0104] After completing the evaluation of the phenotypic aging status of each organ system of the evaluated object, the phenotypic aging status of each organ system and its evaluation details are summarized to generate the evaluation results of the phenotypic aging status of each organ system; the phenotypic aging status evaluation results include the phenotypic aging status of each organ system, the information of the phenotypic aging markers involved in the evaluation and the model modality type, the biological age of the organ system predicted by the organ system aging model, the normal aging status range of the corresponding age group of the organ system in the reference database, and other information.

[0105] The phenotypic aging status assessment results of each organ system are displayed on a human organ map. The human organ map is a graphical representation used to show the location of each organ or system within the human body, allowing users to more intuitively understand the structure, location, and function of the human body. Different colors can be used on the human organ map to represent the phenotypic aging status of each organ system. For example, a green color indicates that the organ system is in a decelerated aging state, a blue color indicates that the organ system is in a normal aging state, and a red color indicates that the organ system is in an accelerated aging state. Click on any organ system on the human organ map to view the assessment details of its phenotypic aging status, such as the information of the phenotypic aging markers involved in the assessment, whether the organ system aging model used is a single-modal model or a multi-modal model, and if it is a multi-modal model, the biological age of the organ system predicted by each modal model is displayed. The biological age of the comprehensive multi-modal organ system is also displayed, and the normal aging state range of the corresponding age group of the organ system in the reference database is also displayed. This is conducive to early identification of high-risk groups in each organ system and to guiding personalized precision health care and aging intervention.

[0106] It should be understood that although Figure 2 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this application, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Figure 2 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0107] Figure 3 This is a schematic diagram of the structure of an organ system phenotype aging assessment device provided in an embodiment of the present application. The device can be set at Figure 1 The terminal 110 in the system shown is used to perform the following operations: Figure 2 The method flow shown in . Figure 3 As shown, the organ system phenotypic aging assessment device 100 may include: an acquisition module 310, a determination module 320, a prediction module 330, and an assessment module 340. It may further include: a training module and a display module. The main functions of each component module are as follows: An acquisition module 310 is used to obtain phenotypic aging marker information of each organ system of the subject to be evaluated and pre-process the phenotypic aging marker information; a determination module 320 for determining a model modality type for processing the phenotypic aging marker information based on the information type of the phenotypic aging marker information and the gender of the subject to be evaluated; A prediction module 330 is configured to predict the biological age of an organ system based on a pre-trained organ system aging model according to the model modality type and the pre-processed phenotypic aging marker information; The evaluation module 340 is used to evaluate the phenotypic aging status of the organ system of the subject to be evaluated based on the actual age of the subject to be evaluated, the biological age of the organ system and pre-stored population sample information.

[0108] As an implementable manner, the model modality type includes a unimodal model and a multimodal model; the determination module 320 is specifically used to: if the information type of the phenotypic aging marker information is one, determine that the model modality type used to process the phenotypic aging marker information is a unimodal model corresponding to the gender of the subject to be evaluated; if the information type of the phenotypic aging marker information is at least two, determine that the model modality type used to process the phenotypic aging marker information is a multimodal model corresponding to the gender of the subject to be evaluated.

[0109] As a feasible method, the pre-stored population sample information includes sample phenotypic aging marker information, sample actual age and sample gender of multiple sample individuals; the evaluation module 340 is specifically used to: stratify the sample actual age according to a preset age gradient based on the sample gender of multiple sample individuals to obtain multiple sample age layers; classify the sample individuals corresponding to each sample age layer in the multiple sample age layers according to the organ system type to obtain the classification group of each organ system in each sample age layer; determine the first quartile of biological age and the third quartile of biological age of each classification group based on the sample phenotypic aging marker information and the classification group based on the pre-trained organ system aging model; evaluate the phenotypic aging status of the organ system of the object to be evaluated based on the actual age of the object to be evaluated, the biological age of the organ system, the first quartile of biological age and the third quartile of biological age of each classification group.

[0110] As an achievable method, the evaluation module 340 is specifically used to: determine the age group of the subject to be evaluated based on the actual age of the subject to be evaluated; determine the first quartile of biological age and the third quartile of biological age corresponding to the organ system of the subject to be evaluated based on the age group of the subject to be evaluated and the first quartile of biological age and the third quartile of biological age of each classification group; and evaluate the phenotypic aging status of the organ system of the subject to be evaluated based on the biological age of the organ system and the first quartile of biological age and the third quartile of biological age corresponding to the organ system.

[0111] As an achievable method, the single-modal model includes a first aging model, a second aging model, a third aging model, and a fourth aging model; the device also includes: a training module for obtaining sample information of various organ systems before obtaining the phenotypic aging marker information of each organ system of the object to be evaluated, wherein the sample information includes three-dimensional original image images, two-dimensional original image images, original measurement signals, quantitative trait phenotypic aging marker information and actual age of various organ systems; preprocessing the three-dimensional original image images, two-dimensional original image images, original measurement signals, and quantitative trait phenotypic aging marker information; training a pre-constructed three-dimensional convolutional neural network based on the preprocessed three-dimensional original image images and actual age to obtain a first aging model; training a pre-constructed two-dimensional convolutional neural network based on the preprocessed two-dimensional original image images and actual age to obtain a second aging model; training a pre-constructed one-dimensional convolutional neural network based on the preprocessed original measurement signals and actual age to obtain a third aging model; and constructing a fourth aging model based on a preset machine learning algorithm based on the preprocessed quantitative trait phenotypic aging marker information and actual age.

[0112] As a feasible method, the training module is also used to select at least two unimodal models before obtaining the phenotypic aging marker information of each organ system of the object to be evaluated; randomly select a set of initial weight combinations, and determine the next set of optimal weight combinations based on a preset proxy model and a preset acquisition function; determine the model weight combination based on the optimal weight combination and a preset validation set, wherein the preset validation set includes sample phenotypic aging marker information of different information types, sample actual age and sample gender; determine a multimodal aging model based on at least two unimodal models and model weight combinations.

[0113] As a feasible method, the device also includes a display module for generating phenotypic aging status evaluation results of each organ system based on the phenotypic aging status of the organ system of the object to be evaluated; and displaying the phenotypic aging status evaluation results of each organ system on a human organ map.

[0114] The same or similar parts between the above embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0115] It should be noted that the embodiments of this application may involve the use of user data. In actual applications, user-specific personal data can be used in the scheme described herein within the scope permitted by applicable laws and regulations, provided that it complies with the applicable laws and regulations of the country where the user is located (for example, with the user's explicit consent, effective notification to the user, and explicit authorization by the user).

[0116] According to an embodiment of the present application, the present application also provides a computer device and a computer-readable storage medium.

[0117] Figure 4 , which is a block diagram of a computer device according to an embodiment of the present application. The term "computer device" is intended to represent various forms of digital computers or mobile devices; digital computers may include desktop computers, portable computers, workstations, personal digital assistants, servers, mainframe computers, and other suitable computers; and mobile devices may include tablet computers, smartphones, wearable devices, and the like.

[0118] like Figure 4 As shown, computer device 400 includes a computing unit 401 , ROM 402 , RAM 403 , a bus 404 , and an input / output (I / O) interface 405 . The computing unit 401 , ROM 402 , and RAM 403 are connected to each other via the bus 404 . The input / output (I / O) interface 405 is also connected to the bus 404 .

[0119] The computing unit 401 can perform various processes in the method embodiments of the present application according to computer instructions stored in a read-only memory (ROM) 402 or computer instructions loaded from a storage unit 408 into a random access memory (RAM) 403. The computing unit 401 can be various general-purpose and / or specialized processing components with processing and computing capabilities. The computing unit 401 can include, but is not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. In some embodiments, the methods provided in the embodiments of the present application can be implemented as a computer software program, which is tangibly contained in a computer-readable storage medium, such as the storage unit 408.

[0120] The RAM 403 may also store various programs and data required for the operation of the computer device 400. Part or all of the computer program may be loaded and / or installed on the computer device 400 via the ROM 402 and / or the communication unit 409.

[0121] The input unit 406, output unit 407, storage unit 408, and communication unit 409 in the computer device 400 can be connected to the I / O interface 405. The input unit 406 can be, for example, a keyboard, a mouse, a touch screen, a microphone, etc.; the output unit 407 can be, for example, a display, a speaker, an indicator light, etc. The computer device 400 can exchange information and data with other devices via the communication unit 409.

[0122] It should be noted that the device may also include other components necessary for normal operation, or may only include components necessary for implementing the solution of this application, rather than all the components shown in the figure.

[0123] Various implementations of the systems and techniques described here can be realized in digital electronic circuitry, integrated circuitry, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations of them.

[0124] The computer instructions for implementing the method of the present application can be written in any combination of one or more programming languages. These computer instructions can be provided to the computing unit 401, so that when the computer instructions are executed by the computing unit 401, such as a processor, the steps involved in the method embodiment of the present application are performed.

[0125] The computer-readable storage medium provided herein may be a tangible medium that may contain or store computer instructions for executing the steps involved in the method embodiments of the present application. The computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, and other forms of storage media.

[0126] The above specific embodiments do not constitute a limitation on the scope of protection of this application. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application shall be included within the scope of protection of this application.

Claims

1. A method for assessing phenotypic aging of an organ system, characterized in that: The method comprises: Acquiring phenotypic aging marker information of each organ system of the subject to be evaluated, and preprocessing the phenotypic aging marker information; determining a model modality type for processing the phenotypic aging marker information according to the information type of the phenotypic aging marker information and the gender of the subject to be evaluated; Predicting the biological age of the organ system based on a pre-trained organ system aging model according to the model modality type and the pre-processed phenotypic aging marker information; The phenotypic aging status of the organ system of the subject to be evaluated is evaluated based on the actual age of the subject to be evaluated, the biological age of the organ system and pre-stored population sample information.

2. The method according to claim 1, characterized in that The model modality types include unimodal models and multimodal models; determining the model modality type for processing the phenotypic aging marker information based on the information type of the phenotypic aging marker information and the gender of the subject to be evaluated includes: If the information type of the phenotypic aging marker information is one, determining that the model modality type used to process the phenotypic aging marker information is a single modality model corresponding to the gender of the subject to be evaluated; If the information type of the phenotypic aging marker information is at least two, the model modality type used to process the phenotypic aging marker information is determined to be a multimodal model corresponding to the gender of the subject to be evaluated.

3. The method according to claim 1, characterized in that The pre-stored population sample information includes sample phenotypic aging marker information, sample actual age, and sample gender of multiple sample individuals; and evaluating the aging status of the organ system of the subject to be evaluated based on the actual age of the subject to be evaluated, the biological age of the organ system, and the pre-stored population sample information, including: stratifying the actual age of the samples according to a preset age gradient based on the sample genders of the multiple sample individuals to obtain multiple sample age strata; Classifying the sample individuals corresponding to each sample age layer in the multiple sample age layers according to the organ system type to obtain the classification group of each organ system in each sample age layer; According to the sample phenotypic aging marker information and the classification group, based on a pre-trained organ system aging model, determining the biological age first quartile and the biological age third quartile of each classification group; The phenotypic aging status of the organ system of the subject to be evaluated is evaluated based on the actual age of the subject to be evaluated, the biological age of the organ system, the first quartile of the biological age of each classification group, and the third quartile of the biological age.

4. The method according to claim 3, characterized in that Assessing the phenotypic aging status of the organ system of the subject to be assessed based on the actual age of the subject to be assessed, the biological age of the organ system, the first quartile of the biological age of each classification group, and the third quartile of the biological age, comprising: Determining the age group to which the subject to be evaluated belongs based on the actual age of the subject to be evaluated; Determining the first quartile of biological age and the third quartile of biological age corresponding to the organ system of the subject to be evaluated according to the age group to which the subject to be evaluated belongs and the first quartile of biological age and the third quartile of biological age of each classification group; The phenotypic aging state of the organ system of the subject to be evaluated is assessed based on the biological age of the organ system, the first quartile of the biological age corresponding to the organ system, and the third quartile of the biological age.

5. The method according to claim 2, characterized in that The single-modality model includes a first aging model, a second aging model, a third aging model, and a fourth aging model; before obtaining phenotypic aging marker information of each organ system of the subject to be evaluated, the method further includes: Acquiring sample information of various organ systems, wherein the sample information includes three-dimensional original images, two-dimensional original images, original measurement signals, quantitative phenotypic aging marker information, and actual age of the various organ systems; Preprocessing the three-dimensional original image, the two-dimensional original image, the original measurement signal, and the quantitative trait phenotypic aging marker information; Training a pre-constructed three-dimensional convolutional neural network according to the pre-processed three-dimensional original image and the actual age to obtain a first aging model; Training a pre-constructed two-dimensional convolutional neural network according to the pre-processed two-dimensional original image and the actual age to obtain a second aging model; Training a pre-constructed one-dimensional convolutional neural network according to the pre-processed raw measurement signal and the actual age to obtain a third aging model; According to the preprocessed quantitative trait phenotypic aging marker information and the actual age, a fourth aging model is constructed based on a preset machine learning algorithm.

6. The method according to claim 5, characterized in that Before obtaining the phenotypic aging marker information of each organ system of the subject to be evaluated, the method further includes: Select at least two unimodal models; Randomly select a set of initial weight combinations, and determine the next set of optimal weight combinations based on the preset proxy model and the preset acquisition function; Determining a model weight combination based on the optimal weight combination and a preset validation set, wherein the preset validation set includes sample phenotypic aging marker information, sample actual age, and sample gender of different information types; A multimodal aging model is determined based on the at least two unimodal models and the model weight combination.

7. The method according to claim 1, characterized in that The method further comprises: generating a phenotypic aging status assessment result of each organ system based on the phenotypic aging status of the organ system of the subject to be assessed; The phenotypic aging status assessment results of each organ system are displayed on a human organ diagram.

8. An organ system phenotypic aging assessment device, characterized in that: The device comprises: an acquisition module, configured to acquire phenotypic aging marker information of each organ system of the subject to be evaluated, and preprocess the phenotypic aging marker information; a determination module, configured to determine a model modality type for processing the phenotypic aging marker information based on the information type of the phenotypic aging marker information and the gender of the subject to be evaluated; a prediction module, configured to predict the biological age of the organ system based on the model modality type and the preprocessed phenotypic aging marker information and a pre-trained organ system aging model; The evaluation module is used to evaluate the phenotypic aging state of the organ system of the subject to be evaluated based on the actual age of the subject to be evaluated, the biological age of the organ system and pre-stored population sample information.

9. A computer device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores computer instructions that can be executed by the at least one processor. The computer instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: The computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 7.

Images