Biological age assessment method and system based on non-destructive body fluid nuclear magnetic resonance signal and whole body magnetic resonance image fusion, electronic equipment and computer readable storage medium

Abstract

The invention provides a biological age assessment method and system based on fusion of a nondestructive body fluid nuclear magnetic resonance signal and a whole body magnetic resonance image, electronic equipment and a computer readable storage medium, and particularly comprises the following steps: collecting original nuclear magnetic resonance spectrogram data of blood and urine, and directly constructing a body fluid signal model by utilizing a full spectrum data point matrix; and inputting the image features and whole body MRI image features into a multi-modal fusion algorithm to realize quantitative calculation of individual biological age and organ aging state positioning. According to the method, a blood and urine combined analysis strategy is adopted, all liquid molecular signals can be captured at a spectrogram level without qualitative or quantitative metabolite, and a molecule-structure cross-scale aging model is established in combination with whole-body MRI structural characteristics. The method can output the overall biological age index, the organ specific aging score and the longitudinal aging rate, and is suitable for health management, disease risk prediction and intervention effect evaluation.

Description

Technical Field

[0001] This invention relates to the fields of medical testing and artificial intelligence technology, specifically to a biological age assessment method, system, electronic device, and computer-readable storage medium based on the fusion of non-destructive body fluid nuclear magnetic resonance signals and whole-body magnetic resonance images. Background Technology

[0002] Biological age reflects the true degree of physiological aging and is a better indicator of an individual's health status than chronological age calculated from birth date. Current biological age assessment methods mainly include single-dimensional schemes based on blood biochemical indicators, genomic / epigenetic indicators, and imaging aging scores. These methods have the following limitations: first, they rely on a limited number of manually selected indicators, making it difficult to cover the complex metabolic and structural changes in the body; second, some tests require complex sample processing or expensive equipment, limiting their widespread application; and third, they lack effective fusion between different modalities of signal, failing to simultaneously utilize fluid metabolic status and organ structural information.

[0003] Nuclear magnetic resonance (NMR) NMR This method allows for non-destructive testing of bodily fluids, directly obtaining high-dimensional spectra containing a large number of chemical shift points, and possesses high repeatability and high standardization potential. Traditional methods often... NMR The signal is converted into the concentrations of several metabolites, and then statistical modeling is performed. This process involves issues related to feature selection and the accumulation of fitting errors. On the other hand, the whole body... MRI It can provide structural and compositional information for multiple organs, including the brain, liver, muscles, and fat, and is closely related to the aging process. However, how to jointly analyze blood and urine at the spectral level remains a challenge. NMR Signals, and further with the whole body MRI There is still a lack of mature technical solutions for multimodal fusion of features to construct a unified biological age assessment model. Summary of the Invention

[0004] Technical Solution: The purpose of this invention is to overcome the problems of limited dimensions, modality separation, and insufficient interpretability of existing biological age assessment indicators, and to propose a method based on blood and urine. NMR A comprehensive analysis system and method for biological age assessment, combining whole-body MRI imaging features, enables precise quantification and dynamic tracking of an individual's overall and organ-specific aging status. Specifically, it provides a biological age assessment method based on the fusion of non-invasive body fluid MRI and whole-body MRI images. The specific steps of this assessment method include: Blood and urine combined spectral characteristics were obtained based on raw nuclear magnetic resonance spectral data of blood and urine. Combined hematuria spectrum characteristics and systemic MRI Image feature input multimodal fusion model; Under the established data fusion mechanism, multi-task deep neural network learning was carried out to establish the mapping relationship between hematuria combined spectrum features, MRI image features and biological age, establish multiple prediction tasks, and carry out and complete model training. Outputs the overall biological age index and relative aging scores for each organ; By using feature importance analysis, the contribution of key spectral segments and key organ regions is obtained, and organ aging heatmaps and spectral segment importance distribution maps are output.

[0005] As an improvement, the importance analysis method includes (1) Metabolospectral feature importance: based on attention weight or gradient backpropagation, identify the chemical shift interval that contributes the most to biological age; (2) Importance of image features: using a method based on Grad-CAM Interpretability methods to obtain MRI Importance map of organ regions Wm Spectral importance vectors corresponding to chemical shift intervals Ws ,Right now Wm This represents the relative contribution weight distribution of each organ or organ region in the biological age prediction task. Ws This indicates the relative contribution of different NMR chemical shift regions to biological age prediction; a cross-modal coupling matrix is ​​constructed. We establish the importance correlation between spectral segments and organ structures, and generate a dual-channel interpretability map of spectral segments and organs based on the coupling matrix, thereby achieving a cross-modal interpretation of the mechanism of biological age contribution.

[0006] As an improvement, the specific steps for obtaining the combined hematuria spectrum characteristics are as follows: (1) Data preprocessing: An adaptive chemical shift calibration algorithm was used to preprocess the 1H-values ​​of blood and urine. NMR The spectrum was processed; specifically, three stable peak regions—0.85–1.00 ppm, 3.20–3.25 ppm, and 4.30–4.70 ppm—were used as multiple anchor points, and a nonlinear drift correction function was constructed using the least squares method. Δppm=a·ppm²+b·ppm+c Nonlinear calibration is performed on the entire spectrum. Parameter a is used to characterize the coefficient of the second-order nonlinear term of chemical shift, reflecting the curvature shift characteristics of the spectrum in the high and low chemical shift regions; parameter b is used to characterize the overall linear shift trend; parameter c is a constant offset term used to compensate for the overall translation error of the entire spectrum, and further achieves fine alignment by combining dynamic time warping. (2) Obtain the combined hematuria spectrum features by setting an encoding strategy mechanism; wherein the encoding strategy mechanism is any one of the following: (2.1) CNN Encoder: Uses a 1D convolutional neural network. ResNet1D or DenseNet1DLocal patterns in blood and urine spectra are extracted separately, and then joint representation learning is performed through a shared weight layer or a mutual information maximization loss function. (2.2) Transformer Encoder: Treats the spectrogram as a sequence input and employs lightweight methods. Vision Transformer or Swin Transformer A variant utilizes a self-attention mechanism to capture long-range dependencies and fuses hematuria signals through cross-modal attention; (2.3) Mixed CNN-Transformer Hybrid encoding: First, local chemical group features are captured through multi-layer 1D convolution, and then the convolution output is flattened into sequence input. Transformer The encoder uses a self-attention mechanism to model the synergistic changes among different metabolic pathways, ultimately outputting a unified joint latent representation.

[0007] (2.4) Region Adaptive Weighted Hybrid Encoder: First, the spectrum is divided into multiple functional segments according to biochemical significance. Each segment is encoded by an independent 1 D-CNN Extract local features; then input the above features. Transformer The module dynamically adjusts the importance of different spectral bands through a region-level adaptive weight vector to obtain a joint latent representation.

[0008] As an improvement, the whole body MRI The steps for obtaining image features include: For the collected whole body or local areas MRI Feature extraction from scanned data: A deep learning segmentation network is used to automatically segment the regions of brain, heart, liver, skeletal muscle, subcutaneous fat and visceral fat, and extract the volume of aorta, regional muscle, bone marrow and fat deposition and lean tissue fat content; Quantitative features: Parameters such as volume, surface area, signal intensity, and fat fraction are extracted for each organ, including abdominal fat fraction, which is often calculated using fat and water maps, and then... ROI Calculate the mean value; plot the region of interest (ROI) on the fat and water map and calculate the fat / total signal ratio to obtain the organ fat fraction.

[0009] As an improvement, the data fusion mechanism includes (1) Early fusion + fully connected regression head: directly splice two modal features, feed them into a 3-layer multilayer perceptron, with random drop-out mechanism and batch normalization, and output continuous biological age values; (2) Dual-branch coding + cross-attention fusion: hematuria branch and MRI Each branch is encoded by an independent sub-network, and then information is exchanged through a cross-attention mechanism, finally returning to the biological age; (3) Dynamic fusion mechanism based on modal prediction uncertainty regulation: acquiring body fluid modalities andMRI The standard deviations σ1 and σ2 of modal predictions are based on The weights ω1 and ω2 of the two modes in the final prediction are adaptively adjusted to improve the stability and robustness of the fusion result.

[0010] As an improvement, the prediction task includes a main task and a sub-task. The main task is overall biological age regression, which outputs continuous values, and the supervision signal is the subject's actual age. The sub-task is organ-specific aging score prediction, which outputs values ​​in the range of 0–1.

[0011] As a specific embodiment of the present invention, the present invention also provides a biological age assessment system based on the fusion of non-destructive body fluid MRI and whole-body MRI images, including a body fluid MRI detection module, a spectrum signal joint analysis module, an MRI image analysis module, a multimodal fusion and biological age modeling module, and an interpretability and visualization module, wherein any two modules can interact with each other. The body fluid nuclear magnetic resonance detection module is used to perform nuclear magnetic resonance detection on blood samples and urine samples respectively to obtain two types of full-spectrum signal data, which are then transmitted to the spectrum signal joint analysis module. The spectral signal joint analysis module is used for preprocessing and aligning the two types of full-spectrum signals to construct hematuria joint spectral features; MRI The image analysis module is used to extract whole-body images. MRI Organ structural features in images; The multimodal fusion and biological age modeling module is used to fuse hematuria combined spectrum features with organ structural features to establish a biological age prediction model. The interpretability and visualization module is used to output biological age results, organ aging scores, and feature importance visualization results.

[0012] As an improvement, the multimodal fusion and biological age modeling module employs a deep neural network, including a first encoding subnetwork for processing hematuria combined spectrum features and a subnetwork for processing... MRI The second encoding subnetwork for features, and the fusion subnetwork for fusing the potential representations of both and outputting biological age prediction results.

[0013] As another specific embodiment of the present invention, the present invention also provides an electronic device, including a processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed, implements the steps of any of the above methods.

[0014] As another specific embodiment of the present invention, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the steps of any of the methods described above.

[0015] Beneficial effects: This invention proposes a method based on blood and urine. NMR Full spectrum joint analysis, combined with whole-body analysis MRI The system and method for biological age assessment based on imaging features involves collecting raw nuclear magnetic resonance (NMR) spectral data from blood and urine, directly constructing a body fluid signal model using a full-spectrum data point matrix, and then comparing it with the whole body... MRI Image features are jointly input into a multimodal fusion algorithm to achieve quantitative calculation of an individual's biological age and localization of organ aging status. This invention employs a combined blood and urine analysis strategy, eliminating the need for qualitative or quantitative analysis of metabolites, to capture whole-body fluid molecule signals at the spectral level, and combine this with whole-body analysis. MRI Structural features are used to establish a molecular-structural cross-scale aging model. This method can output an overall biological age index, organ-specific aging scores, and longitudinal aging rates, and is suitable for health management, disease risk prediction, and intervention effect evaluation.

[0016] The biological age assessment system and method based on the fusion of non-invasive body fluid MRI and whole-body MRI images proposed in this invention have the following advantages compared with existing technologies: 1. By directly utilizing blood and urine NMR Full-spectrum signals enable non-destructive, standardized, high-dimensional characterization of body fluids without relying on metabolites for qualitative or quantitative analysis, thus avoiding information loss due to feature selection.

[0017] 2. A combined blood and urine analysis is used, rather than simply splicing two sets of data, so that the combined spectral characteristics can reflect the coordinated state of the whole body's metabolic and excretion processes.

[0018] 3. Introduced throughout the body MRI By integrating imaging features with information from the molecular level of body fluids and the structural level of organs, a cross-scale biological age assessment model can be established to improve prediction accuracy and stability.

[0019] 4. Through the interpretability module, it is possible to locate spectrum segments and imaging regions that contribute significantly to biological age, assisting in clinical understanding of aging-related pathways and organ targets.

[0020] 5. Supports multi-time point detection, enabling the quantification of individual aging rates and the assessment of the impact of lifestyle or interventions on the aging process. Attached Figure Description

[0021] Figure 1 This is a block diagram of the overall system structure of the present invention.

[0022] Figure 2 This is a schematic diagram of the combined analysis process of blood and urine spectrum signals according to the present invention.

[0023] Figure 3 This is a schematic diagram of the multimodal feature fusion and biological age modeling structure of the present invention.

[0024] Figure 4 This is a schematic diagram of the biological age output and organ aging visualization interface of the present invention. Detailed Implementation

[0025] The technical solutions in the embodiments of the present invention will be clearly and completely described below, so that those skilled in the art can better understand the advantages and features of the present invention, thereby making a clearer definition of the scope of protection of the present invention. The embodiments described in this invention are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0026] The present invention proposes four technical solutions, among which solution one is a biological age assessment system based on the fusion of non-invasive body fluid MRI and whole-body MRI images, see [link to solution]. Figure 1 As shown, the following functional modules are included: 1. Body Fluid MRI Detection Module: Used to separately analyze blood and urine. NMR The detection process yielded two types of full-spectrum signal data point matrices.

[0027] 2. Joint Spectral Signal Analysis Module: Preprocesses and aligns blood and urine spectra, employing an adaptive chemical shift calibration algorithm. A-CSA ) and region adaptive weighted hybrid encoder ( RA-HE We construct a combined hematuria spectrum signal space and achieve joint characterization at the signal level through mutual information maximization or correlation networks.

[0028] 3. MRI Image analysis module: used to read the whole body MRI The image is automatically segmented into multiple target organs and tissues, and structural parameters such as volume, signal intensity, and fat content are extracted to form image feature vectors.

[0029] 4. Multimodal Fusion and Biological Age Modeling Module: Employing deep neural networks, this module integrates hematuria combined spectral features with... MRI Image features are mapped to a unified latent space to establish a biological age prediction model, which outputs an overall biological age index and organ aging score.

[0030] 5. Interpretability and Visualization Module: Used to generate feature importance scores and organ aging maps, showing the contribution of key spectral segments and image regions to biological age prediction; supports visualization of longitudinal aging rates detected at multiple time points.

[0031] As a specific embodiment of the present invention, the nuclear magnetic resonance detection module uses nuclear magnetic resonance spectral sequence to collect full-spectrum data points, and the full-spectrum data points are sampled with a fixed step size within a preset chemical shift range to form a high-dimensional signal vector.

[0032] The joint analysis module for spectral signals includes a preprocessing subunit and a joint coding subunit.

[0033] In the preprocessing stage, this invention proposes an adaptive chemical shift calibration algorithm ( Adaptive Chemical- Shift Alignment, A-CSA Unlike traditional single-point internal standard correction methods, this algorithm utilizes three types of stable chemical environments as anchor regions, including: 0.85–1.00. ppm (Lipid methyl segment); 3.20–3.25 ppm (Cholinergic-related segment); 4.30–4.70 ppm (Shuifeng Neighborhood).

[0034] The module fits the nonlinear drift compensation function using the least squares method. Δppm=a·ppm²+b·ppm+c Nonlinear position calibration is performed on the entire spectrum, combined with dynamic time warping ( DTW Fine alignment of local peak positions can reduce peak position deviation and improve data consistency across batches.

[0035] In the joint encoding stage, this invention employs a region-adaptive weighted hybrid encoder (... Region-Adaptive Hybrid Encoder, RA-HE First, the process A-CSA The processed spectrum was divided into multiple functional regions according to biochemical significance (such as lipid region, branched-chain amino acid region, choline region, organic acid region, etc.), each region consisting of an independent 1 D-CNN Extract local peak features; then use the spliced ​​regional features as sequence input. Transformer Encoder, through the region level Query The system models the synergistic changes among different metabolic pathways. It dynamically learns the importance of each spectral band through a region-adaptive weighting module, enabling the model to focus on aging-related metabolic signals.

[0036] In addition, the module constructs a joint input matrix of blood and urine spectra. X = concat ( x_ blood x_ urine)∈ R N×2 And maximize the objective function using mutual information.

[0037] Strengthening the synergistic expression of two types of body fluid metabolic features enables the joint profile features to more fully reflect the body's aging-related metabolic patterns in the latent space.

[0038] MRI Image analysis module based on Dixon Water-lipid separation sequence or whole body T1WI / T2WI The sequence is used for automatic organ segmentation, and multiple types of structural parameters are extracted. These parameters are not limited to the following examples: (1) Liver volume (mL), fat fraction (%); (2) Visceral fat volume (mL), subcutaneous fat volume (mL); (3) Cross-sectional area of ​​quadriceps femoris (cm²), intramuscular fat infiltration (%); (4) Bone marrow fat fraction of L1–L4 vertebral bodies (%); (5) Average diameter of the aorta (mm).

[0039] The above structural features are all implemented using a deep learning segmentation network (...). U-Net / V-Net Obtain and summarize to form MRI Image feature vectors are used for subsequent multimodal fusion.

[0040] In addition to traditional early-stage fusion and cross-attention fusion methods, this invention further proposes a dynamic fusion mechanism based on modal uncertainty in the multimodal fusion module. Confidence-Modulated Dynamic Fusion, CMDF ).

[0041] Specifically, these are: uncertainty σ1 in the prediction of body fluid modality output, and uncertainty σ2 in the prediction of MRI modality output. The system is based on...

[0042] Automatically adjust the contribution ratio of the two modes in the final prediction. When the data quality of one mode is poor (e.g., ... MRI When motion artifacts occur, or when urine dilution causes a decrease in spectral signal intensity, the system automatically reduces the weight of that mode to improve prediction stability.

[0043] Furthermore, the interpretability and visualization module is used to generate spectral importance distribution maps and organ aging heatmaps to demonstrate the contribution of different spectral segments and different organs to the biological age prediction results.

[0044] See Figure 2As shown, this is the second scheme proposed in this invention: a biological age assessment method based on the fusion of non-destructive body fluid MRI and whole-body MRI images, including the following steps: S1: Collect blood and urine samples from the subject and perform MRI detection to obtain two types of full-spectrum signals; S2: Preprocess and align the two types of full-spectrum signals to construct a combined blood and urine spectrum feature; S3: Collect whole-body MRI images of the subject and extract structural features of multiple organs or tissues; S4: Input the combined blood and urine spectrum feature and the structural features into a multimodal fusion model to output biological age and organ aging scores; S5: Generate feature importance visualization results using an interpretability and visualization module.

[0045] In the evaluation method of this invention, steps S1 to S4 are repeated at different time points to calculate the change in biological age per unit time based on the biological age results at different time points, which serves as an individual aging rate parameter. Further, the biological age results and aging rate parameter are used for health management program development, lifestyle intervention effect evaluation, or disease risk stratification analysis. The preprocessing and alignment in S2 includes phase correction, baseline correction, normalization, and chemical shift alignment of the full-spectrum signals of blood and urine.

[0046] The present invention also provides a third embodiment, and provides an electronic device including a processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed, implements the steps of any of the above methods.

[0047] Option 4: The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above. The following description, in conjunction with specific embodiments of the present invention, further illustrates this invention, but the present invention is not limited to these embodiments. Example 1

[0048] Baseline biological age assessment was performed on the subjects. Venous blood and morning midstream urine were collected from the subjects according to standard procedures and analyzed. NMR Detection and extraction of metabolic profiles from blood and urine. Simultaneously, whole-body samples are acquired. MR Image analysis is performed, and structural parameters are extracted as image features. Metabolic profile features and image features are then input into a pre-trained multimodal neural network to output individual biological age and aging scores for various organs. The specific steps are as follows: 1. Body fluid sample collection: On the day of the subject's physical examination, 10 mL of venous blood and approximately 20 mL of morning midstream urine were collected. No complex chemical separation was required; the samples were directly used for MRI detection after simple centrifugation.

[0049] 2. Nuclear magnetic resonance imaging (NMR) of body fluids and pretreatment: (1) Spectrum acquisition: using 1H- NMRInstruments (e.g., 600) MHz Superconductivity NMR Full spectrum signals from blood and urine were collected separately.

[0050] (2) Phase correction and baseline correction: Remove baseline drift caused by instrument factors.

[0051] (3) Chemical shift alignment: based on A-CSA The algorithm performs chemical shift calibration to compensate for factors. pH Peak position shift caused by temperature or environmental changes.

[0052] (4) Normalization and discretization: Blood samples use internal standards TSP Normalization was performed on urine samples using the creatinine signal to eliminate the influence of sample concentration differences on subsequent analysis. Sampling was conducted at a fixed ppm step size to generate a high-dimensional vector.

[0053] (5) Feature extraction: A region adaptive weighted hybrid encoder is used. First, the features are processed by the encoder. A-CSA The processed spectrum was divided into multiple functional regions (lipid region, branched-chain amino acid region, choline region, organic acid region, etc.) according to biochemical significance, and each region was represented by an independent 1 D-CNN Extract local peak features; then use the spliced ​​regional features as sequence input. Transforme r encoder, through region level Query The system models the synergistic changes among different metabolic pathways. It dynamically learns the importance of each spectral band through a region-adaptive weighting module, enabling the model to focus on aging-related metabolic signals.

[0054] 3. MRI Image detection and preprocessing: (1) Whole body MRI Collection: The subject undergoes one whole-body or localized treatment. MRI Scanning, for example, using Dixon Water-lipid separation sequences can simultaneously produce water images, fat images, and fat fraction maps.

[0055] (2) Feature extraction: using a deep learning segmentation network ( U-Net / V-Net It automatically segments regions such as brain, heart, liver, skeletal muscle, subcutaneous fat, and visceral fat, and extracts the volume of aorta, the volume of regional muscle, bone marrow, and fat deposits, as well as the lean tissue fat content.

[0056] (3) Quantitative characteristics: Parameters such as volume, surface area, signal intensity, and fat fraction are extracted for each organ. Abdominal fat fraction is often calculated using fat and water maps, and then further analyzed in... ROI Calculate the mean value. Plot on a fat and water map. ROI The fat fraction of an organ can be obtained by calculating the ratio of fat to total signal.

[0057] 4. Multimodal fusion and biological age calculation (1) Input construction: combining hematuria combined spectrum features and MRI Image feature input multimodal fusion model.

[0058] (2) Fusion Paradigm: Based on data interaction needs, select any one of the following fusion methods (not limited to the following): Early fusion + fully connected regression head: directly concatenate two-modal features and feed them into 3 layers. MLP (bring Dropout and BatchNorm Output continuous biological age values; Dual-branch coding + cross-attention fusion: hematuria branch and MRI Each branch is encoded by an independent sub-network, and then through a cross-attention mechanism ( Cross-Attention Interacting with each other, the information is ultimately returned to the biological age; In the fusion stage of this embodiment, in addition to the cross-attention multimodal fusion structure, the CMDF dynamic fusion mechanism of this invention is also introduced. This mechanism calculates the prediction uncertainty σ1 of the body fluid modality and... MRI Modal prediction uncertainty σ2, according to

[0059] Dynamically adjust the fusion ratio of the two modes. When MRI When motion artifacts appear, the model automatically reduces their weights, while increasing them when the metabolic spectrum signal has high noise. MRI Modal weights.

[0060] (3) Learning framework: Regardless of which of the above fusion paradigms is adopted, a multi-task learning framework can be combined to simultaneously predict multiple related tasks: Main task: Overall biological age regression (output continuous values, with the supervision signal being the subject's actual age); Subtask: Predict organ-specific aging score (output value in the range of 0–1); Sharing mechanism: The underlying fusion features are shared by all tasks, while the upper layers output them separately; Loss function: The main task adopts MAE or Huber Loss Subtasks adopt MSE The total loss is a weighted sum.

[0061] (4) Model output: The overall biological age index is given and the relative aging score (0-1) is output for each organ.

[0062] Results: The total number of enrolled participants was 1517, aged 30–75 years. Five-fold cross-validation was performed: the optimal configuration (two-branch cross-attention fusion) was determined. CNN-Transformer (Mixed encoding) + CMDF The dynamic fusion mechanism + multi-task learning framework achieves the following: mean absolute error (MAE) MAE 2.1 years old, coefficient of determination (R²) = 0.88. The model prediction of biological age and organ aging score of 30 random subjects is used as an example for demonstration, as shown in Table 1 below, where the higher the score, the more obvious the relative aging of the organ.

[0063] Table 1 shows the model's predicted biological age and organ aging scores as examples. Example 2

[0064] Longitudinal aging rate assessment involves repeating the testing procedure of Example 1 on the same subject at different time points to obtain the biological age results at each time point. By fitting the trend of biological age change over time, the change in biological age per unit time is calculated as an aging rate parameter to assess whether the individual aging process is accelerated or slowed down. Specific implementation steps: 1. Initial testing: Complete body fluid collection according to Example 1. NMR Testing, MRI Scanning and biological age calculation are performed to obtain a baseline biological age.

[0065] 2. Follow-up testing: Repeat the same procedure every 3 to 6 months to ensure that the scanning and testing conditions are consistent.

[0066] 3. Calculate the aging rate: Plot the biological age results at different time points as a function of time, and use linear regression or polynomial fitting to calculate the amount of biological age increase per unit time as the "aging rate". For example, if the biological age increases by 1.0 year after one year, the aging rate is 1.0 year / year, indicating normal aging; if it only increases by 0.2 years, the aging rate is lower than the time age, indicating slowed aging; if the biological age increases by 2 years, the aging rate is higher than the time age, indicating accelerated aging.

[0067] Results: A total of 261 participants underwent two or more tests. As of the time of statistics, the shortest interval was 3 months and the longest interval was 1.5 years. The longitudinal aging rate assessment results are presented using 15 participants with a test interval of more than 1 year as examples; the specific results are shown in Table 2 below.

[0068] Table 2. Results of longitudinal aging rate assessment

[0069] The above results indicate that there are significant differences in aging rates among different populations, which can be further used to assess the effectiveness of lifestyle interventions or disease risks. Example 3

[0070] Model interpretability study: Using labeled training samples, chronological age or biological age obtained through comprehensive clinical assessment was used as a supervisory signal to train a multimodal neural network, enabling it to learn the combined hematuria spectrum characteristics and... MRI The mapping relationship between image features and biological age. After model training, feature importance analysis methods (including attention weighting, gradient weighting, etc.) are used to obtain the contribution of key spectral segments and key organ regions, which are then used to construct organ aging heatmaps and spectral segment importance distribution maps. The specific steps are as follows: 1. Model Training: Model training is completed based on the model training process in Implementation Case 1. 2. Feature Importance Analysis: (1) Importance of metabolospectral features: Based on attention weights or gradient backpropagation, identify the chemical shift intervals that contribute most to biological age.

[0071] (2) Importance of image features: using a method based on Grad-CAM An interpretable method is used to generate heatmaps of various organs. Emphasis is placed on age-predicting regions such as aortic volume, regional muscle, bone marrow, adipose tissue volume, and lean tissue fat content.

[0072] (3) Biomedical significance: The important metabolic profiles and impact characteristics extracted in steps (1) and (2) are combined with existing biomedical knowledge for joint analysis to accurately locate the occurrence of aging problems and provide key biological processes for solving the problems.

[0073] (4) Cross-modal consistency analysis: based on the spectral importance vector obtained in step (1) Ws The organ region importance map obtained in step (2) Wm Constructing a cross-modal coupling matrix C=softmax(Ws·Wmᵀ) ,in C Each element represents the joint contribution strength of a certain chemical shift segment and a certain organ region in biological age prediction. Through analysis of... C Thresholding and visualization are performed to generate a dual-channel interpretable atlas of spectral segments and organs, which is used to show "which spectral segments and which organ structural changes jointly drive biological aging".

[0074] Results: Taking the model feature weights and interpretability analysis results of a subject (actual age 55 years, biological age 56 years) as an example, the features with higher importance include: 1. Blood NMRSpectrum: Chemical shift in the 0.9-1.3 ppm range points to the methyl peak of the aliphatic chain; 2. Urine NMR Spectrum: Chemical shifts in the 3.0-3.2 ppm range indicate characteristic signals of creatine and choline; 3. MRI Imaging: The abdominal aorta, liver fat fraction, and quadriceps volume were the top three areas of attention weighting; These important features can further help researchers understand the decision-making process of the model and provide a theoretical basis for the subsequent development of personalized interventions.

[0075] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A biological age assessment method based on the fusion of non-invasive body fluid MRI and whole-body MRI images, characterized in that, The specific steps of the evaluation method include: Blood and urine combined spectral characteristics were obtained based on raw nuclear magnetic resonance spectral data of blood and urine. Combined hematuria spectrum characteristics and systemic MRI Image feature input multimodal fusion model; Under the established data fusion mechanism, multi-task deep neural network learning was carried out to establish the mapping relationship between hematuria combined spectrum features, MRI image features and biological age, establish multiple prediction tasks, and carry out and complete model training. Outputs the overall biological age index and relative aging scores for each organ; By using feature importance analysis, the contribution of key spectral segments and key organ regions is obtained, and organ aging heatmaps and spectral segment importance distribution maps are output.

2. The biological age assessment method based on the fusion of non-invasive body fluid MRI and whole-body MRI images according to claim 1, characterized in that, Importance analysis methods include (1) Metabolic spectrum feature importance: Based on attention weights or gradient backpropagation, identify the chemical shift intervals that contribute the most to biological age; (2) Importance of image features: using a method based on Grad-CAM Interpretability methods to obtain MRI The importance map Wm of organ regions and the spectral importance vectors corresponding to chemical shift intervals. Ws ,Right now Wm This represents the relative contribution weight distribution of each organ or organ region in the biological age prediction task. Ws This indicates the relative contribution of different NMR chemical shift regions to biological age prediction; a cross-modal coupling matrix is ​​constructed. Establish the importance correlation between spectral segments and organ structures, and generate a dual-channel interpretability map of spectral segments and organs based on the coupling matrix to achieve cross-modal interpretation of the contribution mechanism of biological age.

3. The biological age assessment method based on the fusion of non-invasive body fluid MRI and whole-body MRI images according to claim 1, characterized in that, The specific steps for obtaining combined hematuria and hematuria profile characteristics are as follows: (1) Data preprocessing: An adaptive chemical shift calibration algorithm was used to process the 1H-NMR spectra of blood and urine. Specifically, three stable peak regions of 0.85–1.00 ppm, 3.20–3.25 ppm, and 4.30–4.70 ppm were used as multiple anchor points, and a nonlinear drift correction function was constructed by the least squares method. Δppm = a·ppm² + b·ppm + c Nonlinear calibration is performed on the full spectrum, where the parameters are... a The coefficients of the second-order nonlinear term used to characterize chemical shift reflect the curvature shift characteristics of the spectrum in the high and low chemical shift regions; parameters b Used to characterize the overall linear drift trend; parameters c This is a constant offset term used to compensate for the overall translation error of the entire spectrum, and combined with dynamic time warping to further achieve fine alignment; (2) Obtain the combined hematuria spectrum features by setting an encoding strategy mechanism; wherein the encoding strategy mechanism is any one of the following: (2.1) CNN Encoder: Uses a 1D convolutional neural network. ResNet1D or DenseNet1D Local patterns in blood and urine spectra are extracted separately, and then joint representation learning is performed through a shared weight layer or a mutual information maximization loss function. (2.2) Transformer Encoder: Treats the spectrogram as a sequence input and employs lightweight methods. Vision Transformer or Swin Transformer A variant utilizes a self-attention mechanism to capture long-range dependencies and fuses hematuria signals through cross-modal attention; (2.3) Mixed CNN-Transformer r-mixed encoding: First, through multiple layers of 1 D Convolution captures local chemical group features, and then the convolution output is flattened into a sequence input. Transforme The r encoder uses a self-attention mechanism to model the synergistic changes between different metabolic pathways and ultimately outputs a unified joint potential representation. (2.4) Region Adaptive Weighted Hybrid Encoder: First, the spectrum is divided into multiple functional segments according to biochemical significance. Each segment is encoded by an independent weighted hybrid encoder. 1D-CNN Extract local features; then input the local features. Transformer The module dynamically adjusts the importance of different spectral bands through a region-level adaptive weight vector to obtain a joint latent representation.

4. The biological age assessment method based on the fusion of non-invasive body fluid MRI and whole-body MRI images according to claim 1, characterized in that, whole body MRI The steps for obtaining image features include: For the collected whole body or local areas MRI Feature extraction from scanned data: A deep learning segmentation network is used to automatically segment the regions of brain, heart, liver, skeletal muscle, subcutaneous fat and visceral fat, and extract the volume of aorta, regional muscle, bone marrow and fat deposition and lean tissue fat content; Quantitative features: Parameters such as volume, surface area, signal intensity, and fat fraction are extracted for each organ, including abdominal fat fraction, which is often calculated using fat and water maps, and then... ROI Calculate the mean value; plot the region of interest (ROI) on the fat and water map and calculate the fat / total signal ratio to obtain the organ fat fraction.

5. The biological age assessment method based on the fusion of non-invasive body fluid MRI and whole-body MRI images according to claim 1, characterized in that, Data fusion mechanisms include (1) Early fusion + fully connected regression head: directly splice two modal features, feed them into a 3-layer multilayer perceptron, with random drop-out mechanism and batch normalization, and output continuous biological age values; (2) Dual-branch coding + cross-attention fusion: hematuria branch and MRI Each branch is encoded by an independent sub-network, and then information is exchanged through a cross-attention mechanism, finally returning to the biological age; (3) Dynamic fusion mechanism based on modal prediction uncertainty regulation: acquiring body fluid modalities and MRI Standard deviation of modal prediction σ 1 σ 2 ,according to Adaptively adjust the weights of the two modes in the final prediction. ω 1 ω 2 This is to improve the stability and robustness of the fusion results.

6. The biological age assessment method based on the fusion of non-invasive body fluid MRI and whole-body MRI images according to claim 1, characterized in that, The prediction task includes a main task and a sub-task. The main task is overall biological age regression, which outputs continuous values, and the supervision signal is the subject's actual age. The sub-task is organ-specific aging score prediction, which outputs values ​​in the range of 0–1.

7. A biological age assessment system based on the fusion of non-invasive body fluid MRI and whole-body MRI images, characterized in that, include: Body fluid nuclear magnetic resonance detection module, spectral signal joint analysis module, MRI The image analysis module, the multimodal fusion and biological age modeling module, and the interpretability and visualization module allow for information exchange between any two modules; The body fluid nuclear magnetic resonance detection module is used to perform nuclear magnetic resonance detection on blood samples and urine samples respectively to obtain two types of full-spectrum signal data, which are then transmitted to the spectrum signal joint analysis module. The spectral signal joint analysis module is used for preprocessing and aligning the two types of full-spectrum signals to construct hematuria joint spectral features; MRI The image analysis module is used to extract whole-body images. MRI Organ structural features in images; The multimodal fusion and biological age modeling module is used to fuse hematuria combined spectrum features with organ structural features to establish a biological age prediction model. The interpretability and visualization module is used to output biological age results, organ aging scores, and feature importance visualization results.

8. The biological age assessment system based on the fusion of non-invasive body fluid MRI and whole-body MRI images according to claim 7, characterized in that, The multimodal fusion and biological age modeling module employs a deep neural network, including a first encoding subnetwork for processing combined hematuria spectral features and a subnetwork for processing... MRI The second encoding subnetwork for features, and the fusion subnetwork for fusing the potential representations of both and outputting biological age prediction results.

9. An electronic device comprising a processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed, performs the steps of the method of any one of claims 1-6.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method of claims 1-6.

Images