A multi-organ biological age and disease risk assessment system based on chest CT imaging-based radiomics
The multi-organ biological age and disease risk assessment system based on chest CT radiomics addresses the problem of allochthonous aging of organs and systems, enabling low-cost, non-invasive large-scale population screening and disease risk assessment, and providing organ-targeted aging information and automated clinical reporting.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SICHUAN UNIV
- Filing Date
- 2026-04-02
- Publication Date
- 2026-07-07
AI Technical Summary
Existing technologies cannot effectively address the phenomenon of allometric aging between organs and systems. They are costly, invasive, and have poor model interpretability. Furthermore, they are difficult to perform multi-organ biological age assessment and disease risk assessment in a single low-dose chest CT image.
Design a multi-organ biological age and disease risk assessment system based on chest CT radiomics. The system uses a serial architecture including an input layer, a feature layer, an age prediction layer, a bias correction layer, and a risk assessment layer to generate structured clinical reports using machine learning models and large language models.
It enables low-cost, non-invasive large-scale population screening, provides organ-targeted aging information and disease risk assessment, improves the interpretability and robustness of the model, and supports automated clinical report generation.
Smart Images
Figure CN122348064A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of computer-aided analysis technology of medical images, and relates to a multi-organ biological age and disease risk assessment system based on chest CT radiomics. Background Technology
[0002] Biological age (BA) is used to characterize the true aging degree of an individual's organs and systems. Compared with chronological age (CA), BA more closely reflects the physiological functional state and disease susceptibility of organs. There is an urgent need in clinical and physical examination settings for a low-cost and quantifiable method of aging assessment for early screening and follow-up management.
[0003] Currently, heterochronic aging is a common phenomenon among different organs and systems, meaning that the aging rate of the same subject is not consistent across respiratory, cardiovascular, and musculoskeletal systems. Therefore, a single global age is insufficient to provide actionable organ-targeting information. Existing technical solutions mainly include single global image age models based on end-to-end deep learning, and multimodal fusion models that combine blood indicators and image features.
[0004] However, existing technologies have the following significant shortcomings: Insufficient spatial detail: A single global age model cannot characterize allometric aging and lacks organ- or system-specific detailed information.
[0005] Testing costs and invasiveness: Multimodal assessment solutions rely heavily on multiple testing devices and blood collection procedures, resulting in complex and costly processes. They are also invasive and not suitable for large-scale population screening.
[0006] Weak model interpretability: Although the end-to-end imaging model can obtain the overall age, the traceability of its internal logic is weak, making it difficult to form a chain from features to conclusions that can be clinically reviewed.
[0007] Generalization stability challenge: Due to the differences in data distribution across multiple centers and devices, the model is prone to generalization instability in practical applications, requiring more robust hierarchical modeling and bias correction strategies.
[0008] Lack of clinical closed-loop: The interpretation of results relies heavily on manual writing, lacks a standardized and automated clinical report generation mechanism, and is difficult to quickly connect to the medical system's Picture Archiving and Communication System (PACS).
[0009] Therefore, how to achieve multi-organ and system-level biological age assessment based on a single low-dose chest computed tomography (LDCT) scan, and couple it with a disease risk model validated by a follow-up cohort to automatically generate a structured clinical interpretation report, is a technical problem that urgently needs to be solved in the field of biomedical engineering. Summary of the Invention
[0010] In view of this, the purpose of the present invention is to provide a multi-organ biological age and disease risk assessment system based on chest CT radiomics.
[0011] To achieve the above objectives, the present invention provides the following technical solution: A multi-organ biological age and disease risk assessment system based on chest CT radiomics, wherein the system is designed in a serial architecture according to the data flow direction, and includes, in sequence: The input layer is used to receive the subject's chest computed tomography (CT) radiomics data, calendar age, and gender. The feature layer is used to map the radiomics feature data to an anatomically oriented region of interest (ROI), which includes the heart, blood vessels, lungs, trachea, bones, and muscle tissue. The age prediction layer is used to call a pre-set set of machine learning regression models to infer the standardized radiomics features based on gender stratification, and output 6 organ-level biological ages (BA), 3 system-level biological ages and 1 holistic biological age. The bias correction layer is used to demean the predicted biological age value using a preset correction coefficient to calculate the subject's biological age acceleration (BAA). The risk assessment layer is used to input the biological aging acceleration rate into the built-in Cox proportional hazards regression model based on a longitudinal follow-up cohort, and quantify the individual relative risk multiple (RR) for multiple disease endpoints. The intelligent reporting layer encapsulates the biological age acceleration rate and the relative risk multiple into prompt words, calls a large language model to generate bilingual (Chinese and English) clinical interpretation suggestions, and presents them visually.
[0012] Furthermore, in the age prediction layer, the six organ-level biological ages include heart age, vascular age, lung age, airway age, muscle age, and bone age. The three system-level biological ages include the cardiovascular system age (CardioSys), the respiratory system age (PulmoSys), and the musculoskeletal system age (MuskSys); the one overall biological age is the overall chest age (ChestAge).
[0013] Furthermore, the radiomics features involved in the feature layer are 1218 radiomics features extracted after three-dimensional semantic segmentation of chest CT plain scan sequences. These include morphological features describing the three-dimensional geometric properties of organs, first-order statistical features describing the macroscopic distribution of voxel grayscale, and high-order texture features quantifying the microscopic heterogeneity reconstruction within tissues.
[0014] Furthermore, the age prediction layer executes the following reasoning logic: First, the system classifies the subjects into males or females based on their gender information and matches them with corresponding model weights; Secondly, the input feature vector is subjected to Z-score standardization. Finally, the gradient boosting decision tree algorithm model, which is solidified through a three-stage cascaded training strategy, calculates the biological age at the organ level, system level, and overall level in sequence, with the predicted value of the previous stage serving as the input feature for the next stage.
[0015] Furthermore, the three-stage cascaded training strategy is specifically as follows: In the first stage, the six organ-level biological age models were trained using the cleaned and dimensionality-reduced organ-specific radiomics features as input. In the second stage, the predicted values of heart age and vascular age are used as inputs to train and obtain the cardiovascular system age model, the predicted values of lung age and tracheal age are used as inputs to train and obtain the respiratory system age model, and the predicted values of muscle age and bone age are used as inputs to train and obtain the musculoskeletal system age model. In the third stage, the predicted values of the three system-level biological ages are used as input to train and obtain the one overall chest age model.
[0016] Furthermore, the feature cleaning and dimensionality reduction process in the first stage includes: Features with a missing rate exceeding 40% were removed; Remove near-zero variance features with less than 1% unique values; Fill in the remaining missing values with the mean; Outlier capping was performed using the interquartile range method. We use elastic network regularized regression to screen feature variables with non-zero weights.
[0017] Furthermore, the formula for calculating the biological age acceleration rate (BAA) in the bias correction layer is as follows:
[0018] In the formula, This represents the raw biological age prediction value output by the age prediction layer. Indicates the subject's calendar age. and This represents the correction coefficients obtained by fitting the data to the health validation set beforehand.
[0019] Furthermore, the disease endpoints covered by the risk assessment layer include vascular disease, heart failure and arrhythmia, lung structural disease, respiratory infection, spinal degeneration, osteoporosis and fracture, general weakness, and metabolic syndrome.
[0020] Furthermore, the formula for calculating the individual relative risk multiple (RR) in the risk assessment layer is as follows:
[0021] In the formula, This represents the standard deviation of the hazard ratio per unit corresponding to the disease endpoint. This represents the rate of acceleration of biological age output by the bias correction layer. The standard deviation of the rate of acceleration of biological age.
[0022] Furthermore, the system also includes a data privacy and security protection module. The feature standardization, model inference, and risk calculation are all performed in the local system environment. The intelligent reporting layer only uploads de-identified pure numerical result summaries through an encrypted channel to call the large language model.
[0023] The beneficial effects of this invention are as follows: (1) Only a single low-dose chest computed tomography (LDCT) scan is required to output 10-dimensional biological age (BA) and its acceleration (BAA) at once. No blood collection or multiple devices are required, which significantly reduces the detection cost and patient trauma. It is suitable for large-scale screening and long-term follow-up.
[0024] (2) Through the hierarchical “organ-system-overall” architecture, the system can capture the inconsistency of aging rates between different organs and systems, and provide more targeted organ-targeted aging information than a single global age.
[0025] (3) The system directly converts the abstract biological age acceleration rate (BAA) into the individual relative risk multiple (RR) of specific disease endpoints, covering eight major clinical endpoints such as vascular disease, heart failure, and lung disease, providing clinicians with intuitive and quantitative risk assessment basis.
[0026] (4) Possesses strong model performance and robustness: In the health test set validation, the mean absolute error (MAE) of the overall chest age was as low as 3.33 years, with a correlation of 0.740 with calendar age.
[0027] Cox proportional hazards model validation showed that each dimension of BAA has statistically significant predictive power for the corresponding disease endpoints. For example, the hazard ratio (HR) of the musculoskeletal system accelerated aging rate (MuskSys BAA) to general frailty reached 1.24.
[0028] The integrated Large Language Model (LLM)-driven intelligent reporting layer can automatically convert high-dimensional structured indicators into natural language clinical interpretation suggestions, shortening report generation time and improving the standardization of diagnostic reports.
[0029] (5) Balancing data privacy and system scalability: All core calculations are performed locally, and only de-identified pure numerical summaries are uploaded, effectively avoiding the risk of leakage of patients' original images and privacy information.
[0030] The system supports integration with Picture Archiving and Communication System (PACS) and physical examination systems, and its hierarchical design facilitates future expansion to include more regions of interest (ROI) or disease endpoints.
[0031] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description
[0032] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a system structure diagram of the present invention; Figure 2 This is the front-end interface of a multi-organ biological age and disease risk assessment system based on chest CT radiomics. Detailed Implementation
[0033] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.
[0034] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.
[0035] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.
[0036] 1. System Architecture and Core Modules like Figure 1 As shown, the system of this invention achieves the transformation from raw data to clinical decision support through a clear hierarchical logic: (1) Input and preprocessing: The system receives the subject's structured radiomics features, calendar age, and gender. To eliminate dimensional differences, the system first performs Z-score standardization on the feature vectors.
[0037] (2) Spatial dimension mapping: The feature layer accurately maps the high-dimensional feature pool to 6 core regions of interest (ROIs) to ensure that each subsequent age clock has a clear anatomical significance.
[0038] (3) Three-stage cascaded reasoning core: This is the key technical point of the present invention.
[0039] Phase 1: Predicting organ age of the heart, blood vessels, lungs, trachea, muscles, and bones based on radiomics characteristics (such as morphology and texture features) of a single organ.
[0040] The second stage uses the organ age output from the first stage as input features. For example, a combination of heart age and vascular age can be used to train and predict cardiovascular system age (CardioSys).
[0041] The third stage involves pooling the three system-level predictions from the second stage and training them to obtain the final overall chest age (ChestAge).
[0042] (4) Closed-loop evaluation and output: Bias correction: The original predicted values are converted into biological age acceleration rate (BAA) by correction coefficients to eliminate age bias.
[0043] Risk mapping: Using the built-in Cox proportional hazards model, BAA is projected as a specific disease relative risk multiple (RR).
[0044] Intelligent Interpretation: The system encapsulates quantitative indicators into prompts, calls the Large Language Model (LLM) API to generate natural language reports, and finally presents them in a visual format on the front-end interface.
[0045] This invention provides a multi-organ biological age and disease risk assessment system based on chest CT radiomics. Its core logic lies in quantifying an individual's aging rate through high-dimensional imaging features. For example... Figure 2 As shown, the front-end interface of this invention demonstrates the core interaction logic and output dimensions of the system. The system interface includes a subject basic information area, a 10-dimensional Biological Age Acceleration (BAA) visualization dashboard, a radar chart of eight major disease risks, and a clinical interpretation report area generated by a Large Language Model (LMM).
[0046] The system operates on hardware or a local server environment, and its data flow follows a serial architecture. The input layer first receives the subject's chest computed tomography (CT) radiomics features, chronological age (CA), and gender. The feature layer maps 1218 radiomics features to six core regions of interest (ROIs): heart, blood vessels, lungs, trachea, bone, and muscle tissue.
[0047] In the age prediction layer, the system calls the pre-defined LightGBM gradient boosting decision tree model. This layer follows strict gender-stratified reasoning, that is, it uses independently trained model weights for men and women to avoid differences in physiological baselines. Through three-stage cascaded reasoning, the system outputs six organ-level biological ages (BA), including heart age, vascular age, lung age, airway age, muscle age, and bone age; then it synthesizes three system-level ages: cardiovascular age, respiratory age, and musculoskeletal age; finally, it outputs a chest age.
[0048] The bias correction layer uses correction coefficients fitted on a healthy validation set to mean-reduced the original predicted values and calculate the subject's BAA. This step eliminates the mean convergence phenomenon in the regression model, which overestimates younger subjects and underestimates older subjects, making BAA an independent physiological parameter orthogonal to chronological age. The risk assessment layer incorporates a Cox proportional hazards model built based on longitudinal follow-up data from a cohort of 8,129 subjects. This layer calculates BAA based on the formula... The BAA is projected as the individual relative risk multiple (RR) for eight clinical endpoints, including vascular disease, heart failure, and lung structural disease.
[0049] 2. Example Example 1: Construction and Validation of a Multidimensional Biological Age Prediction Model This example focuses on illustrating the technical solutions regarding the age prediction layer and feature cleaning as described in the claims. The specific workflow is as follows: (1) Data preparation and cleaning: The system reads training set (N=3,948) data from a certain queue. For the feature pool of organs such as the heart, the system automatically performs four cleaning steps: removing features with a missing rate of more than 40%; removing features with a unique value ratio of less than 1%; filling residual missing values with the mean; and using the interquartile range method to cap outliers.
[0050] (2) Feature dimensionality reduction: Elastic Net regularized regression was introduced, and cross-validation was performed when alpha was 0.0, 0.5 and 1.0 to select about 10 core input variables for each organ clock.
[0051] (3) Model training: The LightGBM algorithm was used for 10-fold cross-validation. The learning rate was set to 0.05, the number of leaf nodes was 31, and early stopping was enabled. If the validation error did not decrease for 30 consecutive rounds, training was stopped.
[0052] (4) Performance verification: The performance of the 10-dimensional age clock on the health test set (N=457) was verified. Table 1 shows the performance of the 10-dimensional age clock on the health test set.
[0053] Table 1
[0054] The overall chest age (ChestAge) showed the best performance, with an MAE of 3.33 years, an RMSE of 4.39 years, and a Pearson r of 0.740.
[0055] Bone age and musculoskeletal age also showed extremely high predictive consistency, with correlation coefficients of 0.732 and 0.728, respectively.
[0056] Among organ-level clocks, vascular age (Vessel Age) shows particularly strong predictive performance, with a mean age of 3.71 years (MAE).
[0057] Example 2: Disease Risk Assessment and Intelligent Report Generation Based on BAA This embodiment focuses on illustrating the technical solutions regarding the risk assessment layer and intelligent reporting layer as described in the claims. Its specific workflow is as follows: (1) Individual risk mapping: The system obtains the corrected BAA data of a subject. If the subject's cardiovascular system age acceleration rate (CardioSys BAA) is significantly higher than that of his / her peers, the risk assessment layer inputs this BAA as an exposure variable into the Cox model.
[0058] (2) Calculation of relative risk: As shown in Table 2, the system calculates the RR value (N=8,129; HR is the hazard ratio per SD) based on the longitudinal disease risk coupling validation results: Table 2
[0059] For vascular lesions (A1), the best predictor is CardioSys BAA, with a hazard ratio (HR) of 1.13 for each additional standard deviation (P=0.0001).
[0060] For heart failure / arrhythmia (A2), the best predictor was HeartAge BAA with an HR of 1.10 (P=0.04).
[0061] The most significant risk indicator appeared in general weakness (D1), with a HR of 1.24 (P=0.0001) for MuskSys BAA as a predictor.
[0062] (3) LLM report generation: The intelligent report layer encapsulates the 10-dimensional BAA value, the RR value of the 8 major diseases, and the statistical intensity level into the prompt word template. The template requires the large language model to explain the longitudinal risk in plain language and provide 1-2 action suggestions.
[0063] (4) Output presentation: The system outputs bilingual reports in Chinese and English on the front-end dashboard, realizing a closed loop from image data to clinical interpretation.
[0064] 3. Explanation of attached figures and tables Figure 1 The system's front-end interface displays an interactive interface that includes a 10-dimensional BAA bar chart (used to reflect differences in allometric aging), an RR risk radar chart, and clinical recommendations generated by a large language model.
[0065] Table 1: Performance metrics for 10-dimensional age clock prediction: This table quantifies the prediction accuracy of each level of the model in healthy individuals, demonstrating that the cascaded training architecture can obtain reliable biological age estimates at different anatomical sites.
[0066] Table 2: Cox Regression Risk Coupling Results: This table provides the core parameters for translating BAA into clinical risk, confirming that systemic and organ-level BAA, as exposure variables, can provide statistically significant risk predictions for specific disease spectrum endpoints.
[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A multi-organ biological age and disease risk assessment system based on chest CT radiomics, characterized in that: The system is designed with a serial architecture according to the data flow direction, and includes, in sequence: The input layer is used to receive the subject's chest computed tomography (CT) radiomics data, calendar age, and gender. The feature layer is used to map the radiomics feature data to an anatomically oriented region of interest (ROI), which includes the heart, blood vessels, lungs, trachea, bones, and muscle tissue. The age prediction layer is used to call a pre-set set of machine learning regression models, and to infer the standardized radiomics features based on gender stratification, outputting 6 organ-level biological ages (BA), 3 system-level biological ages, and 1 holistic biological age. A bias correction layer is used to demean the predicted biological age value using a preset correction coefficient and to calculate the subject's biological age acceleration rate (BAA). The risk assessment layer is used to input the biological aging acceleration rate into the built-in Cox proportional hazards regression model based on a longitudinal follow-up cohort, and quantify and output the individual relative risk multiples (RR) for multiple disease endpoints. The intelligent reporting layer encapsulates the biological age acceleration rate and the relative risk multiple into prompt words, calls a large language model to generate bilingual (Chinese and English) clinical interpretation suggestions, and presents them visually.
2. The multi-organ biological age and disease risk assessment system according to claim 1, characterized in that: In the age prediction layer, the six organ-level biological ages include heart age, vascular age, lung age, tracheal age, muscle age, and bone age. The three system-level biological ages include cardiovascular system age, respiratory system age, and musculoskeletal system age; the one overall biological age is the overall chest age.
3. The multi-organ biological age and disease risk assessment system according to claim 1, characterized in that: The radiomics features involved in the feature layer are 1218 radiomics features extracted after three-dimensional semantic segmentation of chest CT plain scan sequences. These include morphological features describing the three-dimensional geometric properties of organs, first-order statistical features describing the macroscopic distribution of voxel grayscale, and high-order texture features quantifying the microscopic heterogeneity reconstruction within tissues.
4. The multi-organ biological age and disease risk assessment system according to claim 1, characterized in that: The age prediction layer performs the following inference logic: First, the system classifies the subjects into males or females based on their gender information and matches them with corresponding model weights; Secondly, the input feature vector is subjected to Z-score standardization. Finally, the gradient boosting decision tree algorithm model, which is solidified through a three-stage cascaded training strategy, calculates the biological age at the organ level, system level, and overall level in sequence, with the predicted value of the previous stage serving as the input feature for the next stage.
5. The multi-organ biological age and disease risk assessment system according to claim 4, characterized in that: The three-stage cascaded training strategy is as follows: In the first stage, the six organ-level biological age models were trained using the cleaned and dimensionality-reduced organ-specific radiomics features as input. In the second stage, the predicted values of heart age and vascular age are used as inputs to train and obtain the cardiovascular system age model, the predicted values of lung age and tracheal age are used as inputs to train and obtain the respiratory system age model, and the predicted values of muscle age and bone age are used as inputs to train and obtain the musculoskeletal system age model. In the third stage, the predicted values of the three system-level biological ages are used as input to train and obtain the one overall chest age model.
6. The multi-organ biological age and disease risk assessment system according to claim 5, characterized in that: The feature cleaning and dimensionality reduction process in the first stage includes: Features with a missing rate exceeding 40% were removed; Remove near-zero variance features with less than 1% unique values; Fill in the remaining missing values with the mean; Outlier capping was performed using the interquartile range method. We use elastic network regularized regression to screen feature variables with non-zero weights.
7. The multi-organ biological age and disease risk assessment system according to claim 1, characterized in that: The formula for calculating the biological age acceleration rate (BAA) in the bias correction layer is as follows: In the formula, This represents the original biological age prediction value output by the age prediction layer. Indicates the subject's calendar age. and This represents the correction coefficients obtained by fitting the data to the health validation set beforehand.
8. The multi-organ biological age and disease risk assessment system according to claim 1, characterized in that: The disease endpoints covered by the risk assessment layer include vascular disease, heart failure and arrhythmia, lung structural disease, respiratory infection, spinal degeneration, osteoporosis and fracture, general weakness, and metabolic syndrome.
9. The multi-organ biological age and disease risk assessment system according to claim 8, characterized in that: The formula for calculating the individual relative risk multiple (RR) in the risk assessment layer is as follows: In the formula, This represents the standard deviation of the hazard ratio per unit corresponding to the disease endpoint. This represents the rate of acceleration of biological age output by the bias correction layer. The standard deviation of the rate of acceleration of biological age.
10. The multi-organ biological age and disease risk assessment system according to claim 1, characterized in that: The system also includes a data privacy and security protection module. The feature standardization, model inference and risk calculation are all performed in the local system environment. The intelligent reporting layer only uploads de-identified pure numerical result summaries through an encrypted channel to call the large language model.