Exercise efficacy prediction method, efficacy prediction model training method, and related devices

Abstract

The application discloses a motion curative effect prediction method, a curative effect prediction model training method and related devices, and relates to the computer field, and includes the following steps: acquiring structural phenotype characteristics, symptoms and motion function indexes of a patient and a current motion prescription, fusing to obtain input characteristics, calling a pre-trained curative effect prediction model to process the input characteristics, and obtaining a curative effect prediction result of the current patient after applying the current motion prescription. The application realizes objective quantitative description of the structural state of the patient by using the structural phenotype characteristics, and can solve the problem of poor prediction accuracy caused by patient differences to a certain extent by jointly using the structural phenotype characteristics and the symptom function indexes for curative effect prediction. The curative effect prediction model disclosed by the application is trained using sample data labeled with a motion rehabilitation curative effect label, the model learns the complex mapping relationship between patient information and motion curative effect, and finally realizes a high-accuracy motion rehabilitation curative effect prediction task by using the model.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method for predicting the therapeutic effect of exercise, a method for training a therapeutic effect prediction model, and related devices. Background Technology

[0002] Chronic obstructive pulmonary disease (COPD) is a chronic inflammatory disease of the airways characterized by persistent airflow limitation. In COPD treatment, medication alone cannot address the systemic functional decline caused by COPD, such as muscle atrophy and decreased physical performance. To address these issues, patients need to rely on exercise prescriptions from their doctors for rehabilitation therapy.

[0003] Currently, when prescribing exercise, doctors often predict the effectiveness of exercise based on the patient's motor function indicators, estimating the rehabilitation effect after the patient uses the exercise prescription. However, due to differences in the pathological and physiological foundations of different patients, there is usually a significant gap between the predicted and actual exercise effects.

[0004] Therefore, how to improve the accuracy of predicting the therapeutic effects of sports rehabilitation has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of the above problems, this application provides a method for predicting the therapeutic effect of exercise, a method for training a predictive model of therapeutic effect, and related devices to improve the accuracy of predicting the therapeutic effect of exercise rehabilitation. The specific solution is as follows:

[0006] The first aspect of this application provides a method for predicting the therapeutic effect of exercise, including:

[0007] Obtain the patient's current symptoms and motor function indicators, as well as the current exercise prescription;

[0008] The structural phenotypic features of the current patient are obtained, which include a feature vector consisting of parameters characterizing the structural state of COPD, and the parameters characterizing the structural state of COPD are determined based on quantitative CT images.

[0009] The input features are obtained by fusing the current patient's structural phenotypic features, symptoms, motor function indicators, and the current exercise prescription.

[0010] The input features are processed by a pre-trained efficacy prediction model to obtain the efficacy prediction result after the current patient applies the current exercise prescription. The efficacy prediction model is trained using sample data labeled with exercise rehabilitation efficacy tags. The sample data is obtained by fusing information from sample patients. The information of the sample patients includes: the sample exercise prescription, and the structural phenotypic features, symptoms, and motor function indicators of the sample patients before applying the sample exercise prescription. The exercise rehabilitation efficacy tags are generated based on the symptoms and motor function indicators of the sample patients before and after applying the sample exercise prescription.

[0011] A second aspect of this application provides a device for predicting the therapeutic effect of exercise, comprising:

[0012] An input feature configuration unit is used to acquire the current patient's symptoms and motor function indicators, as well as the current exercise prescription; acquire the current patient's structural phenotype features, which include a feature vector composed of parameters characterizing the COPD structural state, the parameters characterizing the COPD structural state being determined based on quantitative CT images; and fuse the current patient's structural phenotype features, symptoms and motor function indicators, and the current exercise prescription to obtain input features.

[0013] The exercise efficacy prediction unit is used to call a pre-trained efficacy prediction model to process the input features and obtain the efficacy prediction result after the current patient applies the current exercise prescription. The efficacy prediction model is trained using sample data labeled with exercise rehabilitation efficacy tags. The sample data is obtained by fusing information from sample patients. The sample patient information includes: the sample exercise prescription, and the structural phenotypic features, symptoms, and motor function indicators of the sample patient before applying the sample exercise prescription. The exercise rehabilitation efficacy tags are generated based on the symptoms and motor function indicators of the sample patient before and after applying the sample exercise prescription.

[0014] A third aspect of this application provides a method for training a therapeutic efficacy prediction model, comprising:

[0015] Obtain sample exercise prescriptions, and obtain the structural phenotypic features, symptoms, and exercise function indicators of sample patients before applying the sample exercise prescriptions, as well as the symptoms and exercise function indicators of the sample patients after applying the sample exercise prescriptions; the structural phenotypic features include feature vectors composed of parameters characterizing the structural state of COPD, and the parameters characterizing the structural state of COPD are determined based on quantitative CT images.

[0016] The sample data are obtained by integrating the structural phenotypic characteristics, symptoms, and motor function indicators of the sample patients before applying the sample exercise prescription, as well as the sample exercise prescription.

[0017] Based on the symptoms and motor function indicators of the sample patients before and after applying the sample exercise prescription, an exercise rehabilitation efficacy label is generated, and the sample data is labeled with the exercise rehabilitation efficacy label;

[0018] The pre-configured efficacy prediction model is trained using labeled sample data.

[0019] The fourth aspect of this application provides a training device for a therapeutic efficacy prediction model, comprising:

[0020] The data input unit is used to acquire sample exercise prescriptions, and to acquire the structural phenotypic features, symptoms and exercise function indicators of sample patients before applying the sample exercise prescriptions, as well as the symptoms and exercise function indicators of sample patients after applying the sample exercise prescriptions; the structural phenotypic features include a feature vector composed of parameters characterizing the structural state of COPD, and the parameters characterizing the structural state of COPD are determined based on quantitative CT images.

[0021] The training data configuration unit is used to fuse the structural phenotypic features, symptoms and motor function indicators of the sample patients before applying the sample exercise prescription, as well as the sample exercise prescription, to obtain sample data.

[0022] The tag configuration unit is used to generate exercise rehabilitation efficacy tags based on the symptoms and motor function indicators of the sample patients before and after applying the sample exercise prescription, and to label the sample data with the exercise rehabilitation efficacy tags.

[0023] The model training unit is used to train a pre-configured efficacy prediction model using labeled sample data.

[0024] A fifth aspect of this application provides an electronic device, comprising at least one processor and a memory connected to the processor, wherein:

[0025] The memory is used to store computer programs;

[0026] The processor is used to execute the computer program so that the electronic device can implement the exercise efficacy prediction method of the first aspect above, or implement the efficacy prediction model training method of the third aspect above.

[0027] The sixth aspect of this application provides a computer program product, including computer-readable instructions, which, when executed on an electronic device, cause the electronic device to implement the exercise efficacy prediction method of the first aspect above, or to implement the efficacy prediction model training method of the third aspect above.

[0028] The seventh aspect of this application provides a computer storage medium carrying one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the exercise efficacy prediction method of the first aspect above, or the efficacy prediction model training method of the third aspect above.

[0029] By employing the aforementioned technical solution, this application first obtains the patient's structural phenotypic features, symptoms, motor function indicators, and current exercise prescription. These are then fused to obtain input features. A pre-trained efficacy prediction model is then used to process the input features, yielding a predicted efficacy result for the current patient after applying the current exercise prescription. The structural phenotypic features utilized in this application include feature vectors composed of parameters characterizing the structural state of COPD, thereby achieving an objective quantitative description of the structural state of COPD patients. This objectively reflects the heterogeneous structural damage and respiratory mechanical reserve of COPD patients. Therefore, by combining structural phenotypic features with symptom function indicators for efficacy prediction, the problem of poor efficacy prediction accuracy caused by patient baseline differences can be addressed to some extent. Furthermore, the efficacy prediction model described in this application is trained using sample data labeled with exercise rehabilitation efficacy tags. The pre-trained model determines the complex mapping relationship between structural phenotypic features and exercise rehabilitation efficacy, ultimately achieving a highly accurate exercise rehabilitation efficacy prediction task. Attached Figure Description

[0030] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0031] Figure 1 A schematic diagram of an implementation system architecture provided for an embodiment of this application;

[0032] Figure 2 A flowchart illustrating a method for training a therapeutic efficacy prediction model provided in an embodiment of this application;

[0033] Figure 3 A flowchart illustrating a method for predicting the therapeutic effect of exercise provided in an embodiment of this application;

[0034] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0035] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is only for explaining specific embodiments and is not intended to limit the application. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0036] The applicant's research revealed significant heterogeneity in lung structural damage among different COPD patients. For example, the location of damage, the degree of airway obstruction and gas retention, and the state of respiratory muscles vary greatly among patients. This results in significant differences in the rehabilitation efficacy of exercise even among patients with identical exercise function indicators, when the same exercise prescription is applied. In other words, current predictive methods for exercise rehabilitation efficacy that rely on a single exercise function indicator cannot achieve accurate prediction.

[0037] Building on this, the applicant further discovered that quantitative CT (QCT), widely used in COPD diagnosis, classification, structural assessment, and prognostic analysis, can provide a large number of parameters characterizing a patient's COPD structural status. However, due to the complex nonlinear relationship between COPD structural phenotypes obtained from quantitative CT images and the efficacy of exercise rehabilitation, this relationship cannot be accurately determined by applying single-index analysis techniques or relying on traditional experience, making it difficult to apply COPD structural phenotype characteristics to exercise rehabilitation treatment scenarios.

[0038] To address the aforementioned issues, this application provides a method for predicting exercise efficacy, a method for training an efficacy prediction model, and related devices. By integrating structural phenotypic features characterizing the patient's COPD structural state with symptoms and motor function indicators, the method predicts exercise efficacy, thereby improving the accuracy of predicting the efficacy of exercise rehabilitation for COPD patients.

[0039] The efficacy prediction model training method and exercise efficacy prediction method provided in this application can be applied to, for example... Figure 1 The system architecture shown may include a terminal 100 and a server 200. The server 200 may include one or more servers (…). Figure 1 (This example uses a server as an illustration).

[0040] Either terminal 100 or server 200 can be used independently to execute the methods provided in the embodiments of this application. Alternatively, terminal 100 and server 200 can also be used collaboratively to execute the methods provided in the embodiments of this application. Taking the exercise efficacy prediction method as an example, terminal 100 can be used to receive user-input data, server 200 can be used to process the user-input data to achieve the exercise efficacy prediction task, and terminal 100 can also be used to display the exercise efficacy prediction results. In the embodiments of this application, terminal 100 can be a mobile phone, computer, clinical workstation, etc., and this application does not impose any limitations on this.

[0041] exist Figure 1 In addition, the system may include a memory for storing data, such as user input data and data generated during processing.

[0042] This application provides a method for training a therapeutic efficacy prediction model. Taking the application of this method to a computer device as an example, the computer device can specifically be... Figure 1 The system consists of terminal 100 or a combination of terminal 100 and server 200. (Refer to...) Figure 2 The training method for the efficacy prediction model can specifically include the following steps:

[0043] Step S101: Obtain the sample exercise prescription, and obtain the structural phenotypic characteristics, symptoms and motor function indicators of the sample patient before applying the sample exercise prescription, as well as the symptoms and motor function indicators of the sample patient after applying the sample exercise prescription.

[0044] A sample exercise prescription refers to an exercise prescription issued to a sample patient. The exercise prescription described in this application can indicate the following information: exercise type, exercise intensity, duration of a single exercise session, exercise frequency, and exercise duration period. To adapt to the COPD rehabilitation efficacy prediction scenario and consider the differences in sensitivity of different patients to different types of exercise prescriptions, the exercise prescriptions involved in this application can specifically include at least walking exercise prescriptions and cycling exercise prescriptions; the specific prescription parameters of each exercise prescription can be configured according to the rehabilitation pathway, and this application does not limit the specific prescription parameter values ​​and prescription parameter ranges. For example, one exercise prescription may include: walking at a target pace v for t min (single exercise duration, 20≤t≤40) each time, x times per week (exercise frequency, 3≤x≤5) times, for y (exercise duration period, 8≤y≤12) weeks; another exercise prescription may include: cycling at a target peak power (or expected maximum load) for t min (single exercise duration, 20≤t≤30) each time, x times per week (exercise frequency, 3≤x≤5) times, for y (exercise duration period, 8≤y≤12) weeks.

[0045] It should be noted that in this embodiment, when configuring a training dataset, data from one patient at two time points is acquired. The data before the application of the exercise prescription is referred to as the data at time T0, and the data after the application of the exercise prescription is referred to as the data at time T1. Optionally, the time interval between time T1 and time T0 can be 3 months or 6 months. For example, patient data at time T0 can be acquired before the start of exercise rehabilitation as baseline data, and then patient data at time T1 can be acquired 3 months after the implementation of exercise rehabilitation as reassessment data. Training data is then configured based on the baseline data and reassessment data of the same patient.

[0046] The structural phenotypic features include a feature vector composed of parameters characterizing the structural state of COPD, which are determined based on quantitative CT images. The aforementioned quantitative CT images refer to the patient's chest QCT images, exemplarily, a high-resolution CT scan of the lungs with a slice thickness of approximately 1 mm taken at the end of inspiration.

[0047] Symptom and exercise function indicators comprise two parts: symptom indicators and exercise function indicators. Symptom indicators may include respiratory symptom scores, such as the COPD Assessment Test (CAT) score and the Modified Medical Research Council Dyspnea Scale (mMRC) score. Exercise function indicators may include the results of the 6-minute walk test, such as the 6-minute walk distance (6MWD). Optionally, exercise function indicators may also include the results of the cardiopulmonary exercise test (CPET), such as maximum oxygen uptake (VO2peak) and anaerobic threshold (AT).

[0048] Step S102: Integrate the structural phenotypic characteristics, symptoms, and motor function indicators of the sample patients before applying the sample exercise prescription, as well as the sample exercise prescription, to obtain sample data.

[0049] In this embodiment of the application, the sample data of the sample patients consists of three types of data: structural phenotypic characteristics at time T0, symptoms and motor function indicators at time T0, and exercise prescriptions to be applied after time T0.

[0050] In one possible implementation, the sample data is obtained by fusing the structural phenotypic features, symptoms, and motor function indicators of the sample patients before applying the sample exercise prescription, as well as the sample exercise prescription. This may include:

[0051] The input features are obtained by directly fusing the structural phenotypic features of the sample patients before applying the sample exercise prescription, the symptoms and motor function indicators of the sample patients before applying the sample exercise prescription, and the sample exercise prescription; or...

[0052] The structural phenotypic features, symptoms and motor function indicators, and exercise prescriptions of the sample patients before the application of the sample exercise prescription were encoded, and the encoded structural phenotypic features, encoded symptoms and motor function indicators, and encoded exercise prescriptions were fused together.

[0053] The aforementioned fusion (including direct fusion or post-encoded fusion) can specifically refer to splicing, weighted combination, fusion based on gating or attention mechanisms, or fusion based on feature interactions. For example, sample data can be represented as: (structural phenotypic features + symptoms and motor function indicators) × exercise prescription, reflecting that the effect of the prescription on efficacy is modulated by structural phenotypic and functional state, which helps to determine the heterogeneous efficacy of the prescription and improve the model's individualized predictive ability. Furthermore, different fusion methods can be adopted for parameters within a single variable corresponding to the same variable (structural phenotypic features / symptoms and motor function indicators / exercise prescription). The exercise prescription in the sample data can be represented as: exercise type / other exercise parameters. This application does not limit the specific fusion methods between variables or the specific fusion methods for parameters within each variable.

[0054] It should be noted that the direct fusion method has the advantages of being simple to implement and having less information loss, and is more friendly to the global correlation when modeling the mapping relationship; the fusion method after grouping and encoding can achieve feature dimensionality reduction and noise reduction, and can more naturally deal with the situation where some parameter values ​​are missing.

[0055] Step S103: Based on the symptoms and motor function indicators of the sample patients before and after applying the sample exercise prescription, generate an exercise rehabilitation efficacy label and label the sample data with the exercise rehabilitation efficacy label.

[0056] By comparing the symptoms and exercise function indicators of patients before and after using exercise prescriptions for exercise rehabilitation, we can understand the benefits of exercise for patients and generate exercise rehabilitation efficacy labels accordingly.

[0057] Step S104: Train the pre-configured efficacy prediction model using the labeled sample data.

[0058] This embodiment first integrates the structural phenotypic features, symptoms, and motor function indicators of sample patients before applying the sample exercise prescription, along with the sample exercise prescription itself, to obtain sample data. Then, based on the symptoms and motor function indicators of the sample patients before and after applying the sample exercise prescription, it generates exercise rehabilitation efficacy labels and labels the sample data with these labels, configuring model training data. Subsequently, it uses the labeled sample data to train the pre-configured efficacy prediction model, realizing the correlation modeling of structural phenotypic features, symptoms and motor function indicators, and exercise prescription. It learns the complex mapping relationship between the correlation features and the exercise rehabilitation efficacy response, thereby providing a foundation for achieving a highly accurate exercise rehabilitation efficacy prediction task.

[0059] In one or more embodiments provided in this application, the COPD structural state includes an intrapulmonary structural state and an extrapulmonary structural state.

[0060] Based on this, the process of obtaining the structural phenotypic features of the sample patients before applying the sample exercise prescription may include the following steps:

[0061] Step S201: Quantitatively analyze the quantitative CT images of the sample patients before applying the sample exercise prescription to obtain the COPD structural phenotype parameters of the current patients.

[0062] The quantitative CT images of the sample patients before the application of the sample exercise prescription can refer to the quantitative CT images of the sample patients at time T0.

[0063] Quantitative analysis can specifically include: image preprocessing steps, structure identification and region determination steps, and quantitative calculation steps. Image preprocessing steps may include impregnation, intensity normalization, and artifact removal. Structure identification and region determination steps are used to identify structures / regions such as lung fields, airways, muscle, and adipose tissue. Quantitative calculation steps may include calculations of parameters such as density, area, proportion, wall thickness, and distribution. The specific steps of quantitative analysis can be implemented using rule-based / threshold / morphology-based image processing methods, and / or, AI-assisted automatic analysis methods, and / or, semi-automatic interactive correction methods. AI-assisted automatic analysis methods can be based on automatic segmentation or detection models. It should be noted that quantitative analysis can be implemented based on different computer analysis methods, and this application does not limit the specific algorithm used for quantitative analysis.

[0064] The COPD structural phenotypic parameters may include: whole lung parenchymal parameters, whole lung airway parameters, lung parenchymal and airway parameters of each lung segment, pectoral muscle parameters, spinal muscle parameters, and fat distribution parameters.

[0065] Among them, whole-lung parenchymal parameters, whole-lung airway parameters, and segmental lung parenchymal parameters and airway parameters can characterize the structural state of the lungs and can serve as phenotypes of treatable pulmonary features. For example, lung parenchymal parameters may include the emphysema index LAA%-950, lung density, etc. Furthermore, whole-lung parenchymal parameters may also include statistical values ​​of lung density, such as mean and quantiles, and characteristic parameters reflecting the distribution of the aforementioned parameters throughout the lungs. Airway parameters may include bronchial wall area (WA) percentage (WA%), total diameter ratio (TDR), lung volume index Pi10 (density below -1000 HU), etc. Segmental lung parenchymal parameters and airway parameters refer to stratified statistics of the aforementioned indicators divided by lung lobe / segment. In addition, heterogeneity indicators of the aforementioned parameters may also be included, such as the coefficient of variation and parameters reflecting differences between upper and lower lobes.

[0066] Pectoral muscle parameters, spinal muscle parameters, and fat distribution parameters can characterize the state of extrapulmonary structures and can serve as phenotypes of treatable extrapulmonary features. For example, pectoral muscle parameters may include pectoral muscle area (PMA) and pectoral muscle density (PMD), spinal muscle parameters may include erctor spinal muscle area (ESMA) and erctor spinal muscle density (ESMD), and fat distribution parameters may include visceral adipose tissue area (VATA) and subcutaneous adipose tissue area (SATA).

[0067] Step S202: Based on the COPD structural phenotype parameters of the sample patients, the structural phenotype characteristics of the sample patients are configured.

[0068] Based on step S201, the configuration described in step S202 may refer to configuring several COPD structural phenotypic parameters obtained in step S201 into feature vectors as structural phenotypic features of sample patients.

[0069] In one or more embodiments provided in this application, the COPD structural phenotypic parameters may further include: airway structural features.

[0070] It should be noted that the output of the quantitative analysis step can constitute a QCT quantitative result set, which includes a final quantitative parameter set. This set includes: the aforementioned whole-lung parenchyma parameters, whole-lung airway parameters, lung parenchyma and airway parameters for each lung segment, pectoral muscle parameters, spinal muscle parameters, and fat distribution parameters. The parameters in this set directly serve as primary features constituting the structural phenotypic features. Furthermore, the QCT quantitative result set can also include an intermediate structured result set, which can include airway structural features. Based on the parameters in this set, secondary features for constructing structural phenotypic features can be generated. In other words, the aforementioned quantitative analysis can also include an intermediate structured result generation step, used to generate airway structural features, such as airway tree segmentation results, airway centerlines, branch numbers, airway tree topological relationships, regional mapping relationships for distribution statistics, local statistical tables, and other intermediate structured results. These can also be used to generate lobe / segment zoning annotation results.

[0071] Based on the above, step S202, configuring the structural phenotypic features of the sample patients according to their COPD structural phenotypic parameters, may include:

[0072] Step S301: Extract the branch-level morphological features and mucus plug features of the sample patients based on the airway structure features of the sample patients.

[0073] For example, a graph structure of the airway tree can be generated based on the airway centerline, branch numbers, and airway tree topology output by quantitative analysis, so as to provide a basis for subsequent secondary feature extraction.

[0074] Branch-level morphological characteristics may include: statistical measures of the diameter or wall thickness of each branch lumen, such as mean, quantile, extreme values, etc. In addition, they may include characteristic parameters that reflect the distribution pattern of lumen diameter / wall thickness.

[0075] It should be noted that during quantitative analysis, the location or extent of mucus plugs can be determined through manual annotation, rule-based identification, or artificial intelligence model-based identification. Combining the aforementioned airway structural features or the graph structure of the airway tree, the identified mucus plugs are mapped to airway tree branches or lung segment regions, thereby constructing mucus plug features. Mucus plug features may include mucus plug load characteristics and / or mucus plug distribution characteristics.

[0076] The mucus plug load characteristic reflects the amount of mucus plugs in the entire lung, i.e., the mucus plug load status. For example, it can be obtained through the following steps: Divide the lung into 18 segments according to bronchial anatomy. If a segment contains a mucus plug, score 1 point; if no mucus plug is present, score 0 points, resulting in a total lung mucus plug score (MP), ranging from 0 to 18. Based on this, for standardized characteristics, classify the segments affected by the mucus plug according to their number. Specifically, MP=0 corresponds to no plug, 1≤MP≤3 corresponds to a small number of plugs, and MP≥4 corresponds to a large number of plugs.

[0077] Mucus plug distribution features are used to characterize the branching distribution of embolisms. The extraction process of these features may include: mapping the mucus plug onto airway tree branches based on its location or extent to obtain embolic branches; determining whether the mucus plug affects the main branches; constructing a distribution map structure of embolic branches; and calculating parameters such as the number of embolic branches, the proportion of embolic branches, the degree of embolic aggregation, and the degree of embolic dispersion.

[0078] Step S302: Take the whole lung parenchyma parameters, whole lung airway parameters, lung parenchyma parameters and airway parameters of each lung segment, pectoral muscle parameters, spinal muscle parameters and fat distribution parameters of the sample patients as primary features of the sample patients, take the branch-level morphological features and mucus plug features of the sample patients as secondary features of the sample patients, and configure the structural phenotypic features of the sample patients according to the primary features and secondary features of the sample patients.

[0079] Optionally, when configuring structural phenotypic features, you can use only first-level features, or you can use both first-level and second-level features. Specifically, the configured structural representation features can refer to: a feature vector composed of first-level features, or a feature vector obtained by concatenating first-level and second-level features, or a fused feature vector obtained by performing feature selection, dimensionality reduction, or inlay fusion processing on first-level and second-level features.

[0080] In configuring structural phenotypic features, this embodiment utilizes secondary features in addition to primary features. By deriving or extending feature parameters based on intermediate structured structures, the structured expression of lesion distribution and structural involvement patterns is enhanced.

[0081] In one or more embodiments provided in this application, the sample exercise prescription may include: exercise type information, exercise intensity level determined based on exercise intensity parameter values, duration of a single exercise session, exercise frequency, and exercise duration cycle.

[0082] Specifically, the range of exercise intensity parameters corresponding to any exercise intensity level is as follows:

[0083] A preset fixed range; or,

[0084] Based on the structural phenotypic characteristics of the sample patients before applying the sample exercise prescription and / or the adjusted range of the symptoms and motor function indicators of the sample patients before applying the sample exercise prescription.

[0085] This embodiment uses exercise intensity levels to characterize exercise intensity, achieving a structured representation of exercise prescriptions. The following provides examples of the exercise intensity grading results. Taking a walking exercise prescription as an example, walking training with a target pace of 60%-70% of the individual's comfortable pace can be defined as low-intensity walking training; walking training with a target pace of 70%-80% of the individual's comfortable pace can be defined as moderate-intensity walking training; and walking training with a target pace of 80%-90% of the individual's comfortable pace can be defined as high-intensity walking training. Taking a cycling exercise prescription as an example, cycling training with 50%-60% of peak power can be defined as low-intensity cycling training; cycling training with 60%-70% of peak power can be defined as moderate-intensity cycling training; and cycling training with 70%-80% of peak power can be defined as high-intensity cycling training. In another possible implementation, the intensity thresholds mentioned above can be dynamically adjusted based on the patient's baseline structural phenotype or functional level. For example, the intensity threshold can be lowered for patients with a high proportion of emphysema or a heavy mucus plug load, and raised for patients with good respiratory muscle area and density, so that the intensity of exercise performed or to be performed by the patient is adapted to their own condition. That is, the exercise intensity level in the exercise prescription is the exercise intensity relative to the patient. Low-intensity exercise relative to patients with good respiratory muscle area and density may be high-intensity exercise relative to patients with a high proportion of emphysema or a heavy mucus plug load.

[0086] In one or more embodiments provided in this application, step S103, generating an exercise rehabilitation efficacy label based on the symptoms and motor function indicators of the sample patient before and after applying the sample exercise prescription, may include:

[0087] Step S401: Based on the symptoms and motor function indicators of the sample patients before and after applying the sample exercise prescription, calculate the change parameter value for at least one indicator.

[0088] For example, the change or rate of change of motor function index at time T1 relative to time T0 can be calculated, where the motor function index can be 6MWD; the change or rate of change of symptom index at time T1 relative to time T0 can also be calculated, where the symptom index can be CAT score or mMRC score.

[0089] Step S402: Determine the sports rehabilitation efficacy label based on the calculated change parameter values.

[0090] The exercise rehabilitation efficacy label includes: a change parameter value of at least one indicator, or an efficacy level corresponding to the change parameter value of at least one indicator, or a fusion parameter value calculated from the change parameter values ​​of at least two indicators. In other words, the exercise rehabilitation efficacy label used can be a continuous change / rate of change label, such as Δ6MWD, or an efficacy level determined based on whether the change amount / rate of change reaches a preset change amount / rate of change threshold or based on the change amount / rate of change range to which the change amount / rate of change belongs. The preset efficacy level can include two levels (i.e., effective / ineffective) or multiple levels; this application does not limit this. For example, it can be determined whether Δ6MWD reaches a preset threshold. If it does, the exercise prescription is determined to be effective, and an effective label is configured; otherwise, the exercise prescription is determined to be ineffective, and an ineffective label is configured. As another example, if the decrease in the CAT score is greater than or equal to 2 points or the decrease in the mMRC score is greater than or equal to level 1, the exercise prescription is determined to be effective, and an effective label is configured; otherwise, the exercise prescription is determined to be ineffective, and an ineffective label is configured.

[0091] Furthermore, the exercise rehabilitation efficacy label used in the embodiments of this application can also be a composite label of multiple indicators, such as the changing parameter values ​​of multiple indicators, or the efficacy level determined based on the changing parameter values ​​of multiple indicators. In another possible implementation, it can also be a parameter value that integrates the changing parameter values ​​of multiple indicators, such as a weighted sum of the changing parameter values ​​of multiple indicators.

[0092] In one or more embodiments provided in this application, when the exercise rehabilitation efficacy label is a efficacy level corresponding to the change parameter value of at least one indicator, step S402, determining the exercise rehabilitation efficacy label based on the calculated change parameter value, may include:

[0093] Based on the pre-set range of change parameter values ​​for each indicator corresponding to each therapeutic level, the therapeutic level to which the calculated change parameter value belongs is determined, and this level is used as the therapeutic label for the exercise rehabilitation.

[0094] Specifically, the range of change parameters for any indicator corresponding to any therapeutic efficacy level is as follows:

[0095] A preset fixed range; or,

[0096] Based on the structural phenotypic characteristics of the sample patients before applying the sample exercise prescription and / or the adjusted range of the symptoms and motor function indicators of the sample patients before applying the sample exercise prescription.

[0097] In other words, when determining the level of efficacy, the same criteria can be used for all patients, or a criteria appropriate to the patient can be used.

[0098] For example, assuming that the efficacy label of exercise rehabilitation is determined based on parameter values ​​characterizing the improvement in the 6-minute walk test, Δ6MWD = MWD is calculated according to the following formula. T1 - MWD T0 If Δ6MWD ≥ D1, it is considered an improvement in function and a response / effectiveness of the exercise prescription; otherwise, it is considered that the expected improvement in function has not been achieved and the exercise prescription has not responded / is ineffective. Here, D1 is a preset threshold for the change in 6MWD. D1 can be a pre-set fixed value, such as a parameter value configured according to clinical guidelines, population characteristics, or actual application needs, specifically any number between 25 and 40 (meters); D1 can also be a dynamically adjusted value based on the patient's condition (such as the severity of the patient's baseline symptoms and the type of scale), in which case D1 can be expressed as a·D 10 D 10 The threshold value is a preset fixed threshold. 'a' is an adjustment coefficient that is positively correlated with the patient's condition. The worse the patient's condition, the smaller the value of 'a'. That is, when determining the efficacy level, the threshold for severely ill patients is lower.

[0099] Assuming that the efficacy label of exercise rehabilitation is determined based on symptom scores (such as the CAT score), the improvement in CAT score ΔCAT is calculated according to the following formula: ΔCAT = CAT T1 -CAT T0 If ΔCAT ≥ D2, it is considered as functional improvement and the exercise prescription is responsive / effective; otherwise, it is considered as functional improvement not achieving the expected level and the exercise prescription not responding / ineffective. Here, D2 is a preset threshold for the change in CAT score. D2 can be a fixed value, such as 2; or it can be a dynamically adjusted value based on the patient's condition (such as the severity of the patient's baseline symptoms and the type of scale), in which case D2 can be expressed as b·D. 20 D 20 The threshold is a preset fixed threshold, and b is an adjustment coefficient that is positively correlated with the patient's condition. The worse the patient's condition, the smaller the value of b. That is, when judging the efficacy level, the threshold for judging severe patients is lower.

[0100] Assuming that the exercise rehabilitation efficacy label is determined based on Δ6MWD and ΔCAT, then if Δ6MWD ≥ D1 and ΔCAT ≥ D2, it can be judged as an exercise prescription response; if only one is greater than or equal to the corresponding threshold, it can be judged as a partial exercise prescription response; and if both are less than the corresponding threshold, it can be judged as an exercise prescription non-response. In another possible implementation, Δ6MWD and ΔCAT can be weighted to generate a continuous comprehensive efficacy score, and then the efficacy level corresponding to the comprehensive efficacy score can be determined.

[0101] In one or more embodiments provided in this application, before step S104, training the pre-configured efficacy prediction model with the sample data after applying the labeled data, the following may also be included:

[0102] Step S501: Obtain the structural phenotypic characteristics of the sample patients after applying the sample exercise prescription.

[0103] The structural phenotypic features of sample patients after applying the sample exercise prescription can be determined based on quantitative CT images of the sample patients at time T1. For example, the images can be imported into the system in DICOM format. The steps for obtaining the structural phenotypic features of sample patients after applying the sample exercise prescription can be referred to the description above, and will not be repeated here.

[0104] Step S502: Based on the structural characterization characteristics of the sample patients before and after applying the sample exercise prescription, calculate the change characteristic value of at least one characteristic parameter.

[0105] Step S503: Determine the efficacy prediction auxiliary label based on the calculated change characteristic value, and label the sample data with the efficacy prediction auxiliary label.

[0106] For example, the calculated change feature value can be directly used as an auxiliary label for predicting efficacy; alternatively, the change feature value can be compared with a preset threshold to determine whether the feature value has improved, and the improved / unimproved feature value can be used as an auxiliary label for predicting efficacy. This application does not limit the specific form of the label.

[0107] Based on the above, step S104, training the pre-configured efficacy prediction model using labeled sample data, may include:

[0108] The sample data labeled with the aforementioned sports rehabilitation efficacy tags and efficacy prediction auxiliary tags are used to train the pre-configured efficacy prediction model using multi-task methods.

[0109] Specifically, during the multi-task training, predicting the therapeutic effect of exercise rehabilitation is configured as the primary task, and predicting changes in lung structure is configured as the secondary task.

[0110] This embodiment utilizes the structural phenotypic features at time T1 to assist supervised training, thereby improving the interpretability of efficacy prediction and providing a foundation for achieving high-accuracy efficacy prediction tasks.

[0111] The efficacy prediction model training device provided in the embodiments of this application is described below. The efficacy prediction model training device described below can be referred to in correspondence with the efficacy prediction model training method described above.

[0112] The efficacy prediction model training device provided in this application embodiment may include:

[0113] The data input unit is used to acquire sample exercise prescriptions, and to acquire the structural phenotypic features, symptoms and exercise function indicators of sample patients before applying the sample exercise prescriptions, as well as the symptoms and exercise function indicators of sample patients after applying the sample exercise prescriptions; the structural phenotypic features include a feature vector composed of parameters characterizing the structural state of COPD, and the parameters characterizing the structural state of COPD are determined based on quantitative CT images.

[0114] The training data configuration unit is used to fuse the structural phenotypic features, symptoms and motor function indicators of the sample patients before applying the sample exercise prescription, as well as the sample exercise prescription, to obtain sample data.

[0115] The tag configuration unit is used to generate exercise rehabilitation efficacy tags based on the symptoms and motor function indicators of the sample patients before and after applying the sample exercise prescription, and to label the sample data with the exercise rehabilitation efficacy tags.

[0116] The model training unit is used to train a pre-configured efficacy prediction model using labeled sample data.

[0117] In one implementation, the process by which the label configuration unit generates exercise rehabilitation efficacy labels based on the symptoms and motor function indicators of the sample patient before and after applying the sample exercise prescription may include:

[0118] Based on the symptoms and motor function indicators of the sample patients before and after applying the sample exercise prescription, calculate the change parameter value for at least one indicator;

[0119] The therapeutic efficacy label for exercise rehabilitation is determined based on the calculated change parameter values. The therapeutic efficacy label for exercise rehabilitation includes: the change parameter value of at least one indicator, or the efficacy level corresponding to the change parameter value of at least one indicator, or the fusion parameter value calculated from the change parameter values ​​of at least two indicators.

[0120] In one implementation, when the exercise rehabilitation efficacy label is a efficacy level corresponding to a change parameter value of at least one indicator, the process by which the label configuration unit determines the exercise rehabilitation efficacy label based on the calculated change parameter value may include:

[0121] Based on the pre-set range of change parameter values ​​for each indicator corresponding to each therapeutic level, the therapeutic level to which the calculated transformation parameter value belongs is determined, and this is used as the therapeutic label for the exercise rehabilitation. The range of change parameter values ​​for any indicator corresponding to any therapeutic level is either a pre-set fixed range or a range adjusted based on the structural phenotypic characteristics of the sample patient before applying the sample exercise prescription and / or the range of symptoms and motor function indicators of the sample patient before applying the sample exercise prescription.

[0122] In one implementation, the data input unit can also be used to obtain the structural phenotypic features of the sample patient after applying the sample exercise prescription;

[0123] The label configuration unit can also be used to calculate the change feature value of at least one feature parameter based on the structural characterization features of the sample patients before and after applying the sample exercise prescription; determine the efficacy prediction auxiliary label based on the calculated change feature value; and label the efficacy prediction auxiliary label on the sample data.

[0124] Based on the above, the process by which the model training unit trains a pre-configured efficacy prediction model using labeled sample data can include:

[0125] The sample data labeled with the exercise rehabilitation efficacy tag and the efficacy prediction auxiliary tag are used to train the pre-configured efficacy prediction model in multiple tasks; during the multi-task training, predicting the exercise rehabilitation efficacy is configured as the primary task and predicting lung structural changes is configured as the auxiliary task.

[0126] In one implementation, the sample exercise prescription includes: exercise type information, exercise intensity level determined based on exercise intensity parameter values, duration of a single exercise session, exercise frequency, and exercise duration cycle; wherein, the range of exercise intensity parameters corresponding to any exercise intensity level is: a preset fixed range, or, a range adjusted based on the structural phenotypic characteristics of the sample patient before applying the sample exercise prescription and / or the symptoms and motor function indicators of the sample patient before applying the sample exercise prescription.

[0127] The following describes the method for predicting the therapeutic effect of exercise provided in this application. The method for predicting the therapeutic effect described below can be referred to in conjunction with the method for training the therapeutic effect prediction model described above.

[0128] Figure 3 This is a flowchart illustrating a method for predicting the therapeutic effect of exercise according to an embodiment of this application, combined with... Figure 3 As shown, the method may include:

[0129] Step S601: Obtain the current patient's symptoms and motor function indicators, as well as the current exercise prescription.

[0130] Step S602: Obtain the structural phenotypic features of the current patient.

[0131] The structural phenotypic features include a feature vector composed of parameters characterizing the structural state of COPD, which are determined based on quantitative CT images.

[0132] Step S603: Integrate the current patient's structural phenotypic features, symptoms and motor function indicators, and the current exercise prescription to obtain input features.

[0133] Step S604: Call the pre-trained efficacy prediction model to process the input features and obtain the efficacy prediction result of the current patient after applying the current exercise prescription.

[0134] The efficacy prediction model is trained using sample data labeled with exercise rehabilitation efficacy tags. The sample data is obtained by fusing information from sample patients. The information of the sample patients includes: sample exercise prescriptions, and the structural phenotypic characteristics, symptoms, and motor function indicators of the sample patients before applying the sample exercise prescriptions. The exercise rehabilitation efficacy tags are generated based on the symptoms and motor function indicators of the sample patients before and after applying the sample exercise prescriptions.

[0135] Optionally, the obtained efficacy prediction results may include the efficacy prediction score of the exercise prescription or the response probability of the exercise prescription; in addition, it may also include index values ​​that characterize the confidence level or uncertainty of the output results.

[0136] The specific training process for the efficacy prediction model can be referred to the description above, and will not be repeated here.

[0137] This embodiment first acquires the patient's structural phenotypic features, symptoms, motor function indicators, and current exercise prescription. These are then fused to obtain input features. A pre-trained efficacy prediction model is then called to process the input features, yielding a predicted efficacy result for the current patient after applying the current exercise prescription. The structural phenotypic features utilized in this application include feature vectors composed of parameters characterizing the structural state of COPD, thereby achieving an objective quantitative description of the structural state of COPD patients. This objectively reflects the heterogeneous structural damage and respiratory mechanical reserve of COPD patients. Therefore, by combining structural phenotypic features with symptom and function indicators for efficacy prediction, the problem of poor efficacy prediction accuracy caused by patient baseline differences can be addressed to some extent. Furthermore, the efficacy prediction model described in this application is trained using sample data labeled with exercise rehabilitation efficacy tags. The pre-trained model determines the complex mapping relationship between structural phenotypic features and exercise rehabilitation efficacy, ultimately achieving a highly accurate exercise rehabilitation efficacy prediction task.

[0138] By predicting the outcomes of exercise rehabilitation in advance, doctors can obtain reference information for developing exercise prescriptions, which helps improve the success rate of the first prescription, reduce the possibility of ineffective or inefficient training, and thus improve the actual effectiveness of exercise rehabilitation.

[0139] Considering the differences in patients' sensitivity to different types of exercise prescriptions, multiple types of current exercise prescriptions can be obtained simultaneously, and the efficacy of applying different exercise prescriptions to the same patient can be predicted separately to determine the most suitable exercise mode for the current patient. For example, the obtained current exercise prescriptions may include a walking exercise prescription and a cycling exercise prescription. Then, two input features are constructed based on the two exercise prescriptions, and the efficacy prediction model is called to process the two input features respectively to obtain the efficacy prediction results for the walking exercise prescription and the cycling exercise prescription. Furthermore, after step S604, the method may further include comparing the efficacy prediction results of the walking exercise prescription and the cycling exercise prescription to predict the differences in benefit the current patient receives from different exercise types and determine the more effective exercise type for the current patient.

[0140] In one or more embodiments provided in this application, the COPD structural state includes an intrapulmonary structural state and an extrapulmonary structural state.

[0141] Based on this, step S602, obtaining the structural phenotypic features of the current patient, may include:

[0142] Step S701: Perform quantitative analysis on the quantitative CT images of the current patient to obtain the structural phenotypic parameters of the current patient's COPD.

[0143] The COPD structural phenotypic parameters include: whole lung parenchymal parameters, whole lung airway parameters, lung parenchymal and airway parameters of each lung segment, pectoral muscle parameters, spinal muscle parameters, and fat distribution parameters.

[0144] Step S702: Based on the current patient's COPD structural phenotype parameters, configure the current patient's structural phenotype features.

[0145] In one or more embodiments provided in this application, the COPD structural phenotypic parameters may further include: airway structural features.

[0146] Based on this, step S702, configuring the structural phenotypic features of the current patient according to the current patient's COPD structural phenotypic parameters, may include:

[0147] Step S801: Extract the branch-level morphological features and mucus plug features of the current patient based on the airway structure features of the current patient. The mucus plug features include mucus plug load features and / or mucus plug distribution features.

[0148] Step S802: The whole lung parenchyma parameters, whole lung airway parameters, lung parenchyma parameters and airway parameters of each lung segment, pectoral muscle parameters, spinal muscle parameters and fat distribution parameters of the current patient are used as the primary features of the current patient. The branch-level morphological features and mucus plug features of the current patient are used as the secondary features of the current patient. The structural phenotypic features of the current patient are configured according to the primary features and the secondary features.

[0149] In one or more embodiments provided in this application, step S603, fusing the current patient's structural phenotypic features, symptoms and motor function indicators, and the current exercise prescription, may include:

[0150] The input features are obtained by directly fusing the current patient's structural phenotypic features, symptoms, motor function indicators, and the current exercise prescription; or,

[0151] The structural phenotypic features, symptoms and motor function indicators, and exercise prescription of the current patient are encoded separately, and the encoded structural phenotypic features, symptoms and motor function indicators, and exercise prescription are fused together.

[0152] In one or more embodiments provided in this application, the sample data may also be labeled with efficacy prediction auxiliary tags, which are generated based on the structural phenotypic features of the sample patients before and after applying the sample exercise prescription; the efficacy prediction model is a model trained using the sample data labeled with the exercise rehabilitation efficacy tags and the efficacy prediction auxiliary tags in a multi-task training manner, and in the multi-task training process of the efficacy prediction model, predicting the exercise rehabilitation efficacy is configured as the primary task, and predicting lung structural changes is configured as the auxiliary task.

[0153] Based on the above, the output of the efficacy prediction model can also include the prediction results of structural phenotypic features, such as the mucus plug load characteristics and muscle state characteristics after applying the exercise prescription, thereby improving the interpretability of exercise efficacy prediction.

[0154] Optionally, after step S604, the method may further include: after a preset duration (e.g., 3 months / 6 months), acquiring the current patient's follow-up data, and iteratively optimizing the efficacy prediction model based on the current patient's baseline data, follow-up data, and actual exercise prescription.

[0155] The follow-up data may include the patient's symptoms and motor function indicators after applying the actual exercise prescription, as well as the patient's structural phenotypic characteristics after applying the actual exercise prescription.

[0156] The exercise efficacy prediction device provided in the embodiments of this application is described below. The exercise efficacy prediction device described below can be referred to in correspondence with the exercise efficacy prediction method described above.

[0157] The exercise therapy prediction device provided in this application embodiment may include:

[0158] An input feature configuration unit is used to acquire the current patient's symptoms and motor function indicators, as well as the current exercise prescription; acquire the current patient's structural phenotype features, which include a feature vector composed of parameters characterizing the COPD structural state, the parameters characterizing the COPD structural state being determined based on quantitative CT images; and fuse the current patient's structural phenotype features, symptoms and motor function indicators, and the current exercise prescription to obtain input features.

[0159] The exercise efficacy prediction unit is used to call a pre-trained efficacy prediction model to process the input features and obtain the efficacy prediction result after the current patient applies the current exercise prescription. The efficacy prediction model is trained using sample data labeled with exercise rehabilitation efficacy tags. The sample data is obtained by fusing information from sample patients. The sample patient information includes: the sample exercise prescription, and the structural phenotypic features, symptoms, and motor function indicators of the sample patient before applying the sample exercise prescription. The exercise rehabilitation efficacy tags are generated based on the symptoms and motor function indicators of the sample patient before and after applying the sample exercise prescription.

[0160] In one possible implementation, the COPD structural state includes intrapulmonary structural state and extrapulmonary structural state; the process by which the input feature configuration unit acquires the structural phenotypic features of the current patient may include:

[0161] Quantitative analysis was performed on the quantitative CT images of the current patient to obtain the COPD structural phenotypic parameters of the current patient. The COPD structural phenotypic parameters include: whole lung parenchyma parameters, whole lung airway parameters, lung parenchyma and airway parameters of each lung segment, pectoral muscle parameters, spinal muscle parameters, and fat distribution parameters.

[0162] Based on the current patient's COPD structural phenotype parameters, the current patient's structural phenotype features are configured.

[0163] In one possible implementation, the COPD structural phenotypic parameters may further include: airway structural features; the process by which the input feature configuration unit configures the structural phenotypic features of the current patient based on the current patient's COPD structural phenotypic parameters may include:

[0164] Based on the airway structure of the current patient, the branch-level morphological features and mucus plug features of the current patient are extracted, and the mucus plug features include mucus plug load features and / or mucus plug distribution features;

[0165] The total lung parenchyma parameters, total lung airway parameters, lung parenchyma parameters and airway parameters of each lung segment, pectoral muscle parameters, spinal muscle parameters, and fat distribution parameters of the current patient are used as the primary features of the current patient. The branch-level morphological features and mucus plug features of the current patient are used as the secondary features of the current patient. The structural phenotypic features of the current patient are configured based on the primary features and the secondary features.

[0166] In one possible implementation, the process by which the input feature configuration unit integrates the current patient's structural phenotypic features, symptoms and motor function indicators, and the current exercise prescription may include:

[0167] The input features are obtained by directly fusing the current patient's structural phenotypic features, symptoms, motor function indicators, and the current exercise prescription; or,

[0168] The structural phenotypic features, symptoms and motor function indicators, and exercise prescription of the current patient are encoded separately, and the encoded structural phenotypic features, symptoms and motor function indicators, and exercise prescription are fused together.

[0169] This application also provides an electronic device in its embodiments. (See reference...) Figure 4 The diagram illustrates a structural schematic suitable for implementing the electronic device in the embodiments of this application. The electronic device in the embodiments of this application may include, but is not limited to, fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 4 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0170] like Figure 4 As shown, the electronic device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 1, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 2 or a program loaded from a storage device 8 into a random access memory (RAM) 3, to implement any of the exercise efficacy prediction methods or efficacy prediction model training methods provided in the embodiments of this application. When the electronic device is powered on, the RAM 3 also stores various programs and data required for the operation of the electronic device. The processing unit 1, ROM 2, and RAM 3 are interconnected via a bus 4. An input / output (I / O) interface 5 is also connected to the bus 4.

[0171] Typically, the following devices can be connected to I / O interface 5: input devices 6 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 7 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 8 including, for example, memory cards, hard drives, etc.; and communication devices 9. Communication device 9 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have instead.

[0172] This application also provides a computer program product including computer-readable instructions. When the computer-readable instructions are run on an electronic device, the electronic device enables the electronic device to implement any of the exercise efficacy prediction methods or efficacy prediction model training methods provided in this application.

[0173] This application also provides a computer-readable storage medium that carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the exercise efficacy prediction methods or efficacy prediction model training methods provided in this application.

[0174] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0175] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0176] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

Claims

1. A method for predicting the therapeutic effect of exercise, characterized in that, include: Obtain the patient's current symptoms and motor function indicators, as well as the current exercise prescription; The structural phenotypic features of the current patient are obtained, which include a feature vector consisting of parameters characterizing the structural state of COPD, and the parameters characterizing the structural state of COPD are determined based on quantitative CT images. The input features are obtained by integrating the current patient's structural phenotypic features, symptoms, motor function indicators, and the current exercise prescription. The input features are processed by a pre-trained efficacy prediction model to obtain the efficacy prediction result after the current patient applies the current exercise prescription. The efficacy prediction model is trained using sample data labeled with exercise rehabilitation efficacy tags. The sample data is obtained by fusing information from sample patients. The information of the sample patients includes: the sample exercise prescription, and the structural phenotypic features, symptoms, and motor function indicators of the sample patients before applying the sample exercise prescription. The exercise rehabilitation efficacy tags are generated based on the symptoms and motor function indicators of the sample patients before and after applying the sample exercise prescription.

2. The method for predicting the therapeutic effect of exercise according to claim 1, characterized in that, The COPD structural state includes the intrapulmonary structural state and the extrapulmonary structural state; Obtaining the structural phenotypic features of the current patient includes: Quantitative analysis of the current patient's quantitative CT images was performed to obtain the COPD structural phenotypic parameters of the current patient. The COPD structural phenotypic parameters include: whole lung parenchyma parameters, whole lung airway parameters, lung parenchyma and airway parameters of each lung segment, pectoral muscle parameters, spinal muscle parameters, and fat distribution parameters. Based on the current patient's COPD structural phenotype parameters, the current patient's structural phenotype features are configured.

3. The method for predicting the therapeutic effect of exercise according to claim 2, characterized in that, The COPD structural phenotypic parameters also include: airway structural features; Based on the current patient's COPD structural phenotype parameters, the structural phenotype features of the current patient are configured, including: Based on the airway structure of the current patient, the branch-level morphological features and mucus plug features of the current patient are extracted, and the mucus plug features include mucus plug load features and / or mucus plug distribution features; The total lung parenchyma parameters, total lung airway parameters, lung parenchyma parameters and airway parameters of each lung segment, pectoral muscle parameters, spinal muscle parameters, and fat distribution parameters of the current patient are used as the primary features of the current patient. The branch-level morphological features and mucus plug features of the current patient are used as the secondary features of the current patient. The structural phenotypic features of the current patient are configured based on the primary features and the secondary features.

4. The method for predicting the therapeutic effect of exercise according to any one of claims 1-3, characterized in that, The integration of the current patient's structural phenotypic characteristics, symptoms, motor function indicators, and the current exercise prescription includes: The input features are obtained by directly fusing the current patient's structural phenotypic features, symptoms, motor function indicators, and the current exercise prescription; or, The structural phenotypic features, symptoms and motor function indicators, and exercise prescription of the current patient are encoded separately, and the encoded structural phenotypic features, symptoms and motor function indicators, and exercise prescription are fused together.

5. A method for training a therapeutic efficacy prediction model, characterized in that, include: Obtain sample exercise prescriptions, and obtain the structural phenotypic features, symptoms, and exercise function indicators of sample patients before applying the sample exercise prescriptions, as well as the symptoms and exercise function indicators of the sample patients after applying the sample exercise prescriptions; the structural phenotypic features include feature vectors composed of parameters characterizing the structural state of COPD, and the parameters characterizing the structural state of COPD are determined based on quantitative CT images. The sample data are obtained by integrating the structural phenotypic characteristics, symptoms, and motor function indicators of the sample patients before applying the sample exercise prescription, as well as the sample exercise prescription. Based on the symptoms and motor function indicators of the sample patients before and after applying the sample exercise prescription, an exercise rehabilitation efficacy label is generated, and the sample data is labeled with the exercise rehabilitation efficacy label; The pre-configured efficacy prediction model is trained using labeled sample data.

6. The method for training a therapeutic efficacy prediction model according to claim 5, characterized in that, Based on the symptoms and motor function indicators of the sample patients before and after applying the sample exercise prescription, an exercise rehabilitation efficacy label is generated, including: Based on the symptoms and motor function indicators of the sample patients before and after applying the sample exercise prescription, calculate the change parameter value for at least one indicator; The therapeutic efficacy label for exercise rehabilitation is determined based on the calculated change parameter values. The therapeutic efficacy label for exercise rehabilitation includes: the change parameter value of at least one indicator, or the efficacy level corresponding to the change parameter value of at least one indicator, or the fusion parameter value calculated from the change parameter values ​​of at least two indicators.

7. The method for training a therapeutic efficacy prediction model according to claim 6, characterized in that, When the exercise rehabilitation efficacy label is a efficacy level corresponding to the change parameter value of at least one indicator, the exercise rehabilitation efficacy label is determined based on the calculated change parameter value, including: Based on the pre-set range of change parameter values ​​for each indicator corresponding to each therapeutic level, the therapeutic level to which the calculated transformation parameter value belongs is determined, and this is used as the therapeutic label for the exercise rehabilitation. The range of change parameter values ​​for any indicator corresponding to any therapeutic level is either a pre-set fixed range or a range adjusted based on the structural phenotypic characteristics of the sample patient before applying the sample exercise prescription and / or the range of symptoms and motor function indicators of the sample patient before applying the sample exercise prescription.

8. The method for training a therapeutic efficacy prediction model according to claim 6, characterized in that, Before training the pre-configured efficacy prediction model using labeled sample data, the following steps are also included: Obtain the structural phenotypic features of the sample patients after applying the sample exercise prescription; Based on the structural characterization features of the sample patients before and after applying the sample exercise prescription, calculate the change feature value for at least one feature parameter; Based on the calculated change characteristic values, auxiliary labels for efficacy prediction are determined, and the sample data are labeled with the auxiliary labels for efficacy prediction. The process of training a pre-configured efficacy prediction model using labeled sample data includes: The sample data labeled with the exercise rehabilitation efficacy tag and the efficacy prediction auxiliary tag are used to train the pre-configured efficacy prediction model in multiple tasks; during the multi-task training, predicting the exercise rehabilitation efficacy is configured as the primary task and predicting lung structural changes is configured as the auxiliary task.

9. The method for training a therapeutic efficacy prediction model according to any one of claims 5-8, characterized in that, The sample exercise prescription includes: exercise type information, exercise intensity level determined based on exercise intensity parameter values, duration of a single exercise session, exercise frequency, and exercise duration cycle; Wherein, the range of exercise intensity parameters corresponding to any exercise intensity level is: a preset fixed range, or, a range adjusted based on the structural phenotypic characteristics of the sample patient before applying the sample exercise prescription and / or the symptoms and exercise function indicators of the sample patient before applying the sample exercise prescription.

10. An electronic device, characterized in that, It includes at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program so that the electronic device can implement the exercise efficacy prediction method as described in any one of claims 1 to 4, or implement the efficacy prediction model training method as described in any one of claims 5 to 9.

Images