Medical severe risk prediction method based on physical sign data monitored by smart watch

Abstract

The invention discloses a medical critical risk prediction method based on physical sign data monitored by a smart watch, and relates to the technical field of critical monitoring, and the method comprises the steps: constructing a multi-dimensional physical sign fusion analysis model, collecting dynamic physiological parameters in real time through the smart watch, and updating a personalized health baseline; based on the personalized health baseline, detecting a pathological fluctuation mode in the dynamic physiological parameters through a time sequence convolutional network; inputting the identified abnormal waveform features into a risk prediction model optimized by transfer learning, and calculating a coupling risk index of the multi-organ system; establishing a three-level response early warning mechanism based on the coupling risk index of the multi-organ system; and generating an intervention scheme based on the coupling risk index of the multi-organ system. Compared with the prior art, the intelligent wearable device has the advantages that the multi-organ risk conduction model is combined, the severe risk is accurately predicted, a risk identification-early warning-disposal dynamic management system is formed, and the clinical practical value of the intelligent wearable device in a pre-hospital emergency scene is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of critical care monitoring technology, specifically to a method for predicting the risk of critical medical conditions based on vital sign data monitored by a smartwatch. Background Technology

[0002] Current wearable medical monitoring systems face significant technical bottlenecks in the field of critical illness risk prediction: traditional methods generally employ static threshold alarm mechanisms, lacking the ability to dynamically model individualized physiological baselines, resulting in insufficient accuracy in identifying potential pathological features; simultaneously, existing technical frameworks are mostly based on single organ function assessments, failing to effectively model the synergistic effects and risk transmission mechanisms between multiple physiological systems, thus limiting the reliability of early warnings for complex diseases. Furthermore, most systems stop at the risk alert level, failing to construct a real-time feedback intelligent intervention loop, and thus unable to provide dynamically optimized diagnostic and treatment pathway support for clinical decision-making.

[0003] With the deepening application of artificial intelligence technology, existing intelligent early warning systems still have key shortcomings: vital sign analysis algorithms based on general models struggle to adapt to individual physiological differences, easily leading to feature distortion during cross-scenario data migration; time-series risk assessment modules largely rely on static analysis within fixed time windows, lacking sufficient sensitivity to capture sudden abnormal physiological parameter patterns; and intervention strategy generation mechanisms are disconnected from real-time physiological feedback, lacking dynamic control capabilities for multi-system functional interactions, severely restricting the timeliness and accuracy of early intervention for critical illnesses. These technological limitations restrict the clinical practical value of intelligent wearable devices in pre-hospital emergency care scenarios. Summary of the Invention

[0004] To address the aforementioned technical challenges, this paper provides a method for predicting the risk of critical illness in medical settings based on vital sign data monitored by smartwatches. This solution constructs a personalized dynamic health baseline, combines a multi-organ risk transmission model with a three-tiered intelligent early warning mechanism, and achieves accurate prediction of critical illness risk. Transfer learning is employed to optimize temporal feature extraction, significantly improving the sensitivity of identifying sudden abnormal physiological patterns. Through organ coupling index calculation and augmented reality intervention, a dynamic management system of risk identification, early warning, and treatment is formed, effectively enhancing the clinical practical value of smart wearable devices in pre-hospital emergency care scenarios.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: Medical critical illness risk prediction methods based on smartwatch-monitored vital sign data include: A multi-dimensional vital sign fusion analysis model is constructed, and dynamic physiological parameters are collected in real time through a smartwatch and a personalized health baseline is updated. The personalized health baseline is obtained through adaptive learning of historical stable period data. Based on a personalized health baseline, a temporal convolutional network is used to detect pathological fluctuation patterns in dynamic physiological parameters and identify abnormal waveform features including QT interval variations and blood oxygenation gradient drops. The identified abnormal waveform features are input into the risk prediction model optimized by transfer learning to calculate the coupling risk index of the multi-organ system. The coupling risk index of the multi-organ system uses the recurrent neural network module to capture the risk transmission relationship between organs. Based on the coupling risk index of multiple organ systems, a three-level response and early warning mechanism is established: when the 24-hour risk slope exceeds the first threshold, a local early warning is triggered; when it exceeds the second threshold, remote medical data sharing is initiated; and when the third threshold is reached, the continuous vital signs enhanced monitoring mode is activated. Based on the coupling risk index of multi-organ systems, an intervention plan containing target organ protection strategies is generated, and the risk mitigation path and key vital sign control targets are visualized through an augmented reality interface.

[0006] Preferably, the step of detecting pathological fluctuation patterns in dynamic physiological parameters using a temporal convolutional network based on a personalized health baseline specifically includes: A waveform segmentation algorithm enhanced with attention mechanism is used to locate the spatiotemporal features of ST segment elevation in electrocardiogram signals; Dynamic time warping technology is used to match blood oxygen saturation waveforms with a pre-set pathological pattern library; Sympathetic nerve activation features of respiratory entropy values ​​were extracted using a frequency domain energy redistribution algorithm. Based on the abnormal detection output of various physiological parameters, a multidimensional abnormal feature description vector is formed through feature fusion.

[0007] Preferably, the step of inputting the identified abnormal waveform features into the risk prediction model optimized by transfer learning to calculate the coupling risk index of the multi-organ system specifically includes: The risk prediction model includes an abnormal waveform feature matching module and a multi-organ coupling module; The abnormal waveform feature matching module matches the input abnormal waveform features with the waveform features of each critical medical disease to obtain the basic feature matching degree of each critical medical disease. The multi-organ coupling module determines the origin weight of each critical medical disease based on the matching degree of the basic features of each critical medical disease and the inducing relationship between each critical medical disease. Based on the origin weight of each critical medical disease and the corresponding body organ, it determines the coupling risk index of each body organ.

[0008] Preferably, the multi-organ coupling module specifically includes: Based on medical experience, the causative relationships of multiple serious medical conditions were identified. Based on the induced relationships of multiple critical medical diseases, a directed graph of critical disease induction is constructed. The nodes in the directed graph of critical disease induction represent critical diseases, and the directed line segments in the graph represent the induced relationships between critical medical diseases. Based on organ-directed graphs, a formula for calculating the origin of critical medical diseases is constructed. The parameters in the formula are the induction probabilities between critical medical diseases, which are obtained through transfer learning optimization. The matching degree of the basic features of each critical medical disease is input into the calculation formula of the origin index of critical medical diseases for at least one iteration to obtain the origin weight of each critical medical disease. The coupling risk index of each body organ is obtained by summing the origin weights of all critical medical diseases corresponding to each organ.

[0009] Preferably, the specific steps for optimizing the induced probability among the aforementioned critical medical diseases through transfer learning are as follows: Users are categorized based on their physical characteristics data; The medical history of serious illnesses corresponding to user categories is used as sample analysis data. If medical critical illness i is directional to medical critical illness j, then the probability of medical critical illness i inducing medical critical illness j can be obtained by comparing the number of medical records in the sample analysis data that simultaneously contain medical critical illness i and medical critical illness j with the number of medical records containing medical critical illness i.

[0010] Preferably, the multi-organ system coupling risk index generates an intervention plan that includes target organ protection strategies, and the risk mitigation path and key vital sign control targets are visualized through an augmented reality interface. Based on the directed graph induced by severe diseases, and combined with the coupling risk index of various body organs, key organs are identified as intervention target organs. Intervention plans containing target organ protection strategies are generated based on reinforcement learning, and the comprehensive risk reduction effect under different intervention plans is simulated to select the optimal intervention plan. Under the optimal intervention plan, the user's physiological parameters are dynamically monitored, and the intervention plan is adjusted in real time.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention constructs a multi-disease induction relationship analysis framework based on medical prior knowledge and data-driven approaches. It achieves precise localization of primary diseases by dynamically modeling the cascading risk transmission paths between diseases. Specifically, by constructing a directed graph structure for severe disease induction, it transforms discrete pathological features into a visualized risk propagation network, overcoming the limitations of traditional isolated assessments of single diseases. It designs an iterative algorithm based on primary weights, combined with a cross-disease induction probability matrix optimized through transfer learning, and analyzes the core pathogenic factors in multi-disease concurrent scenarios through a backpropagation mechanism. Based on the dual constraints of induction relationship strength and organ function compensatory capacity, it automatically generates a sequence of intervention strategies targeting primary diseases. By revealing the cascading effects and main pathogenic factors among potential diseases, it significantly improves the accuracy of interventions for complex diseases, especially in clinical scenarios with multi-system interactions. It enables earlier identification of core pathogenic factors and the development of prospective protective strategies, effectively solving the misjudgment problem caused by neglecting disease synergy in traditional methods. Attached Figure Description

[0012] Figure 1 Here is a flowchart of the medical critical illness risk prediction method based on smartwatch monitoring vital sign data proposed in Example 1; Figure 2 This is a flowchart illustrating the working steps of the multi-organ coupling module proposed in Example 3; Figure 3 The flowchart of the method for generating an intervention plan that includes a target organ protection strategy, as proposed in Example 4; Figure 4 This is an architecture diagram of the electronic devices in this solution; Figure 5 This is a schematic diagram of the computer-readable storage medium structure in this scheme.

[0013] The numbers on the map are: 500 - Electronic device; 501 - Bus; 502 - CPU; 503 - ROM; 504 - RAM; 505 - Communication port; 506 - Input / output component; 507 - Hard disk; 508 - User interface; 600 - Computer-readable storage medium. Detailed Implementation

[0014] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0015] Example 1: Reference Figure 1 As shown, the method for predicting the risk of severe medical illness based on vital sign data monitored by a smartwatch includes: A multi-dimensional vital sign fusion analysis model is constructed, and dynamic physiological parameters are collected in real time through a smartwatch and personalized health baselines are updated. The personalized health baselines are obtained through adaptive learning of historical stable period data. An adaptive sliding window mechanism is employed, combined with a variational autoencoder, to learn the representation of a user's historical health status. A dynamic time warping algorithm is used to eliminate individual differences in physiological rhythms. Compared to the traditional fixed threshold method, this approach, by establishing a probability distribution model that evolves over time, can distinguish between normal physiological fluctuations and early pathological shifts, effectively overcoming baseline drift caused by motion artifacts and environmental interference, and providing a dynamic reference benchmark for subsequent analysis. Based on a personalized health baseline, a temporal convolutional network is used to detect pathological fluctuation patterns in dynamic physiological parameters, identifying abnormal waveform features including QT interval variations and sudden drops in blood oxygenation gradients. The specific steps include: A waveform segmentation algorithm enhanced with attention mechanism is used to locate the spatiotemporal features of ST segment elevation in electrocardiogram signals; Dynamic time warping technology is used to match blood oxygen saturation waveforms with a pre-set pathological pattern library; Sympathetic nerve activation features of respiratory entropy values ​​were extracted using a frequency domain energy redistribution algorithm. Based on the abnormal detection output of various physiological parameters, a multidimensional abnormal feature description vector is formed through feature fusion.

[0016] The temporal convolutional network employs a causal dilated convolutional structure, combined with an attention mechanism to construct a multi-level feature pyramid. Residual connection modules are designed to enhance waveform detail capture capabilities, and deformable convolutions are used to adaptively match different morphologies of pathological waveforms. This architecture maintains the local correlation of time series data while achieving subclinical-level abnormal waveform detection through multi-scale feature fusion, improving the granularity of identifying hidden pathological changes compared to traditional peak detection algorithms.

[0017] The identified abnormal waveform features are input into the risk prediction model optimized by transfer learning to calculate the coupling risk index of the multi-organ system. The coupling risk index of the multi-organ system uses the recurrent neural network module to capture the risk transmission relationship between organs. Based on the coupling risk index of multiple organ systems, a three-level response and early warning mechanism is established: when the 24-hour risk slope exceeds the first threshold, a local early warning is triggered; when it exceeds the second threshold, remote medical data sharing is initiated; and when the third threshold is reached, the continuous vital signs enhanced monitoring mode is activated. The first threshold is based on the statistical distribution of the personalized health baseline and is set as the boundary value of the confidence interval that deviates from the baseline mean by three times the standard deviation. The second threshold is dynamically calculated based on the intensity of risk interaction between organ systems. Specifically, it is the ratio of the maximum interaction risk value of each organ pair to the baseline risk value, and then multiplied by the disease-specific adjustment coefficient. The interaction risk value reflects the intensity of pathological transmission between organs. The third threshold adopts a time window sliding update mechanism, which incorporates both the rate of change of the risk index and the declining trend of physiological compensatory ability in its calculation. The sensitivity of the early warning is adaptively adjusted through dynamic weight allocation.

[0018] Based on the coupling risk index of multi-organ systems, an intervention plan containing target organ protection strategies is generated, and the risk mitigation path and key vital sign control targets are visualized through an augmented reality interface.

[0019] The intervention plan generation uses an integrated reinforcement learning framework to simulate organ function response curves for different treatment strategies based on medical experience through a virtual digital twin system. The augmented reality interface employs light field projection technology to construct a 3D anatomical model, overlaying risk heatmaps onto corresponding organ regions and visually displaying multi-system risk transmission pathways through dynamic streamlines. Healthcare professionals can adjust intervention parameters via gesture interaction, and the system provides real-time feedback on the predicted achievement of vital sign control targets.

[0020] Example 2: Based on Example 1, this embodiment further specifies the steps for inputting the identified abnormal waveform features into the risk prediction model optimized by transfer learning to calculate the coupling risk index of the multi-organ system: The risk prediction model includes an abnormal waveform feature matching module and a multi-organ coupling module; The abnormal waveform feature matching module matches the input abnormal waveform features with the waveform features of various critical medical diseases to obtain the basic feature matching degree of each critical medical disease. The basic feature matching degree of critical medical diseases is calculated using cosine similarity. Specifically, in some preferred embodiments, the abnormal waveform feature matching module employs deep metric learning to optimize the feature embedding space, and uses a triplet loss function to constrain the intra-class compactness and cross-class separability of waveform features of similar diseases. Furthermore, a dynamic weight allocation mechanism can be designed to assign adaptive weight coefficients to key indicators such as QT interval morphology and blood oxygenation recovery rate based on the differences in the contribution of different pathological features to disease diagnosis, thereby improving the clinical relevance of cosine similarity calculation.

[0021] The multi-organ coupling module determines the origin weight of each critical medical disease based on the matching degree of its basic features and the induced relationship between them. Based on the origin weight of each critical medical disease and the corresponding body organ, it determines the coupling risk index of each body organ. The input of the multi-organ coupling module is the matching degree of its basic features. Based on the induced relationship between each critical medical disease, it performs correlation analysis to obtain the origin weight of each critical medical disease.

[0022] Example 3 Reference Figure 2 As shown, based on Embodiment 2, this embodiment further proposes the following working steps for the multi-organ coupling module: Based on medical experience, the causative relationships of multiple serious medical conditions were identified. Based on the induced relationships of multiple critical medical diseases, a directed graph of critical disease induction is constructed. The nodes in the directed graph of critical disease induction represent critical diseases, and the directed line segments in the graph represent the induced relationships between critical medical diseases. Specifically, if the occurrence of critical illness i leads to the occurrence of critical illness j, then a directed line segment from critical illness i to critical illness j is constructed in the directed graph of critical illness induction. Based on organ-directed graphs, a formula for calculating the origin of critical medical diseases is constructed. The parameters in the formula are the induction probabilities between critical medical diseases, which are obtained through transfer learning optimization. Specifically, in this embodiment, the formula for calculating the origin index of critical medical diseases is constructed using an improved PAGERANK algorithm, and the specific formula is as follows: In the formula, Let the basic characteristics of the medical critical illness be the matching degree. To directly match weights, Let i be the matching score for the i-th critical medical condition. Let j be the set of critical medical conditions that the i-th critical medical condition points to, and j be the set of critical medical conditions that the j-th critical medical condition points to. The j-th element in for The matching score of the j-th element in the dataset. For the i-th critical medical condition The induced probability of the j-th element in the matrix. For pointing The set of critical medical conditions with the j-th element in the dataset, where K is... The j-th element in the medical critical illness set is the k-th element. for The j-th element in the medical critical illness set is the k-th element of the set of causes. The induced probability of the j-th element in the matrix. Specifically, for the related influencing factors, and satisfy In some preferred embodiments, The value range is 0.6-0.8. The value range is 0.2-0.4; Since critical medical conditions often exhibit induced relationships, such as when a patient with chronic heart failure develops acute pulmonary edema, increased pulmonary circulation resistance can lead to increased right ventricular afterload and decompensation of right ventricular function. Therefore, when identifying and assessing primary diseases, it is necessary to fully explore the induced associations between these critical illnesses. In this embodiment, an improved PAGERANK algorithm is used to quantify this induced relationship. By revealing the cascade effects and main causative agents among potential diseases, the accuracy of interventions for complex conditions is significantly improved. Especially in clinical scenarios involving the interaction of multiple systems, it is possible to identify core pathogenic factors earlier and formulate prospective protective strategies.

[0023] The matching degree of the basic features of each critical medical disease is input into the calculation formula of the origin index of critical medical diseases for at least one iteration to obtain the origin weight of each critical medical disease. The coupling risk index of each body organ is obtained by summing the origin weights of all critical medical diseases corresponding to each organ.

[0024] This paper innovatively integrates medical prior knowledge and data-driven analysis in disease-induced relationship modeling, constructing a dynamically evolving disease association network through a cross-institutional medical knowledge graph. An improved PageRank algorithm introduces a bidirectional propagation mechanism, considering not only explicit induced relationships between diseases, such as heart failure leading to renal function deterioration, but also capturing potential compensatory effects, such as the reverse compensatory effect of respiratory failure on the circulatory system. During weight iteration, key pathogenic hub nodes are automatically identified, such as the core role of sepsis in multiple organ failure, and the induced probability parameters are dynamically adjusted based on real-time physiological feedback, overcoming the limitations of traditional static association models.

[0025] In the organ risk index calculation phase, an anatomical topological constraint strategy is employed, combined with the physiological coupling characteristics of organ systems, such as the portal hypertension conduction pathway in hepatorenal syndrome and the temporal decay curve of compensatory capacity, and the deceleration rate of systolic function after myocardial ischemia, to perform multi-dimensional fusion analysis on the primacy weights. The resulting coupled risk index not only quantifies the current degree of organ damage but also predicts the temporal evolution trend of multi-system decompensation, such as the critical time window for the progression of acute kidney injury to uremia. This multi-scale risk assessment mechanism enables the system to penetrate superficial symptoms and accurately locate the core pathogen in complex clinical scenarios, such as acute exacerbations of chronic diseases, providing spatiotemporal dynamic decision support for targeted intervention. Furthermore, in this embodiment, the method steps for optimizing the induction probability among serious medical diseases through transfer learning are as follows: Based on users’ physical characteristics data, users are classified. In the classification process, users are aggregated and classified based on prior knowledge such as age and basic medical history. The medical history of serious illnesses corresponding to user categories is used as sample analysis data. If medical critical illness i is directional to medical critical illness j, then the probability of medical critical illness i inducing medical critical illness j can be obtained by comparing the number of medical records in the sample analysis data that simultaneously contain medical critical illness i and medical critical illness j with the number of medical records containing medical critical illness i.

[0026] This embodiment significantly improves the individualized adaptability of severe illness risk prediction by constructing dynamic user profiles. Traditional methods use static population probability models, neglecting key individual differences such as patient age stratification and underlying diseases, leading to biased cross-population induced probability estimations. This embodiment establishes a multi-dimensional feature-driven user clustering model, achieving accurate grouping based on composite indicators such as physiological compensatory capacity and chronic disease burden index, and dynamically analyzes the disease evolution feature spectrum of different populations using a medical knowledge graph. By real-time association of historical medical data of user categories, the system can capture hidden cross-disease association patterns within specific groups. For example, the unique pathological link of diabetic patients with cardiorenal syndrome generates a clinically interpretable dynamic induced probability matrix. This association strength calculation method based on real-world evidence not only overcomes the limitations of traditional statistical methods in small-sample rare disease association analysis, but also adapts to the differences in treatment models among different medical institutions, providing a high-confidence decision-making basis for individualized severe illness early warning, effectively avoiding the risk of misjudgment due to assumptions of group commonality, and demonstrating important value in the optimal allocation of medical resources, making early intervention for high-risk groups more targeted.

[0027] Example 4: Reference Figure 3As shown, in this embodiment, an intervention plan containing target organ protection strategies is generated based on the coupling risk index of multi-organ systems, and the risk mitigation path and key vital sign control targets are visualized through an augmented reality interface. Specifically, these include: Based on the directed graph induced by severe diseases, and combined with the coupling risk index of various body organs, key organs are identified as intervention target organs. Intervention plans containing target organ protection strategies are generated based on reinforcement learning, and the comprehensive risk reduction effect under different intervention plans is simulated to select the optimal intervention plan. Under the optimal intervention plan, the user's physiological parameters are dynamically monitored, and the intervention plan is adjusted in real time.

[0028] By integrating multi-organ risk transmission mechanisms with intelligent decision-making algorithms, this approach overcomes the limitations of traditional single-objective intervention strategies, achieving precision and dynamic treatment for acute and critical illnesses. Traditional methods, neglecting the cascading effects between diseases, often lead to a "treat the symptom, not the cause" dilemma of localized treatment. This approach innovatively constructs an organ system interaction map, identifying key target organs that are both the core of the current injury and the hub of multi-system failure based on a coupling risk index, such as the pivotal role of microcirculatory disturbances in septic shock. A virtual physiological simulation environment is built using a deep reinforcement learning framework to simulate the compensatory response network of multi-organ functions under different intervention strategies, autonomously exploring the optimal target organ protection sequence, such as prioritizing the improvement of myocardial perfusion rather than simply raising blood pressure. A dynamic adjustment mechanism introduces a real-time risk propagation prediction model, continuously monitoring the organ function recovery trajectory after intervention, such as the predictive effect of urine output change rate on renal perfusion improvement, automatically correcting treatment parameters and providing early warnings of secondary injury risks, such as the probability of acute pulmonary edema induced by massive fluid resuscitation. This closed-loop intelligent intervention system significantly improves the systematic nature of multi-system critical care and can simultaneously achieve the dual goals of short-term life support and long-term organ function protection in complex clinical scenarios.

[0029] Furthermore, the method according to the embodiments of this application can also be achieved by means of... Figure 4 The architecture of the electronic device shown is used to implement this. For example... Figure 4 As shown, the electronic device 500 may include a bus 501, one or more CPUs 502, ROM 503, RAM 504, a communication port 505 connected to a network, an input / output component 506, a hard disk 507, etc. The storage device in the electronic device 500, such as the ROM 503 or the hard disk 507, may store a medical critical illness risk prediction method based on smartwatch monitoring vital sign data provided in this application. The electronic device 500 may also include a user interface 508. Of course, Figure 4 The architecture shown is merely exemplary and can be omitted as needed when implementing different devices. Figure 4 One or more components in the illustrated electronic device.

[0030] Figure 5 This is a schematic diagram of a computer-readable storage medium structure provided in one embodiment of this application. Figure 5 The diagram illustrates a computer-readable storage medium 600 according to one embodiment of this application. The computer-readable storage medium 600 stores computer-readable instructions. When executed by a processor, the computer-readable instructions can perform a medical critical illness risk prediction method based on vital sign data monitored by a smartwatch, as described above with reference to the accompanying drawings, according to an embodiment of this application. The storage medium 600 includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0031] In summary, the advantages of this invention are as follows: It constructs a multi-disease induction relationship analysis framework based on prior medical knowledge and data-driven approaches, achieving precise localization of primary diseases through dynamic modeling of cascading risk transmission paths between diseases. Specifically, by constructing a directed graph structure for severe disease induction, it transforms discrete pathological features into a visualized risk propagation network, overcoming the limitations of traditional isolated assessments of single diseases. It designs an iterative algorithm based on primary weights, combined with a cross-disease induction probability matrix optimized through transfer learning, and analyzes the core pathogenic factors in multi-disease concurrent scenarios through a backpropagation mechanism. Based on the dual constraints of induction relationship strength and organ function compensatory capacity, it automatically generates intervention strategy sequences targeting primary diseases. By revealing the cascading effects and main pathogenic factors among potential diseases, it significantly improves the accuracy of interventions for complex diseases, especially in clinical scenarios with multi-system interactions, enabling earlier identification of core pathogenic factors and the development of prospective protective strategies, effectively solving the misjudgment problem caused by neglecting disease synergy in traditional methods.

[0032] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for predicting the risk of severe medical illness based on vital sign data monitored by a smartwatch, characterized in that, include: A multi-dimensional vital sign fusion analysis model is constructed, and dynamic physiological parameters are collected in real time through a smartwatch and a personalized health baseline is updated. The personalized health baseline is obtained through adaptive learning of historical stable period data. Based on a personalized health baseline, a temporal convolutional network is used to detect pathological fluctuation patterns in dynamic physiological parameters and identify abnormal waveform features including QT interval variations and blood oxygenation gradient drops. The identified abnormal waveform features are input into the risk prediction model optimized by transfer learning to calculate the coupling risk index of the multi-organ system. The coupling risk index of the multi-organ system uses the recurrent neural network module to capture the risk transmission relationship between organs. Based on the coupling risk index of multiple organ systems, a three-level response and early warning mechanism is established: when the 24-hour risk slope exceeds the first threshold, a local early warning is triggered; when it exceeds the second threshold, remote medical data sharing is initiated; and when the third threshold is reached, the continuous vital signs enhanced monitoring mode is activated. Based on the coupling risk index of multi-organ systems, an intervention plan containing target organ protection strategies is generated, and the risk mitigation path and key vital sign control targets are visualized through an augmented reality interface.

2. The method for predicting the risk of severe medical illness based on vital sign data monitored by a smartwatch according to claim 1, characterized in that, The method of detecting pathological fluctuation patterns in dynamic physiological parameters using a temporal convolutional network based on a personalized health baseline specifically includes: A waveform segmentation algorithm enhanced with attention mechanism is used to locate the spatiotemporal features of ST segment elevation in electrocardiogram signals; Dynamic time warping technology is used to match blood oxygen saturation waveforms with a pre-set pathological pattern library; Sympathetic nerve activation features of respiratory entropy values ​​were extracted using a frequency domain energy redistribution algorithm. Based on the abnormal detection output of various physiological parameters, a multidimensional abnormal feature description vector is formed through feature fusion.

3. The method for predicting the risk of severe medical illness based on vital sign data monitored by a smartwatch according to claim 2, characterized in that, The step of inputting the identified abnormal waveform features into the risk prediction model optimized by transfer learning to calculate the coupling risk index of the multi-organ system specifically includes: The risk prediction model includes an abnormal waveform feature matching module and a multi-organ coupling module; The abnormal waveform feature matching module matches the input abnormal waveform features with the waveform features of each critical medical disease to obtain the basic feature matching degree of each critical medical disease. The multi-organ coupling module determines the origin weight of each critical medical disease based on the matching degree of the basic features of each critical medical disease and the inducing relationship between each critical medical disease. Based on the origin weight of each critical medical disease and the corresponding body organ, it determines the coupling risk index of each body organ.

4. The method for predicting the risk of severe medical illness based on vital sign data monitored by a smartwatch according to claim 3, characterized in that, The multi-organ coupling module specifically includes: Based on medical experience, the causative relationships of multiple serious medical conditions were identified. Based on the induced relationships of multiple critical medical diseases, a directed graph of critical disease induction is constructed. The nodes in the directed graph of critical disease induction represent critical diseases, and the directed line segments in the graph represent the induced relationships between critical medical diseases. Based on organ-directed graphs, a formula for calculating the origin of critical medical diseases is constructed. The parameters in the formula are the induction probabilities between critical medical diseases, which are obtained through transfer learning optimization. The matching degree of the basic features of each critical medical disease is input into the calculation formula of the origin index of critical medical diseases for at least one iteration to obtain the origin weight of each critical medical disease. The coupling risk index of each body organ is obtained by summing the origin weights of all critical medical diseases corresponding to each organ.

5. The method for predicting the risk of severe medical illness based on vital sign data monitored by a smartwatch according to claim 4, characterized in that, The specific steps for optimizing the induced probability among the aforementioned critical medical diseases through transfer learning are as follows: Users are categorized based on their physical characteristics data; The medical history of serious illnesses corresponding to user categories is used as sample analysis data. If medical critical illness i is directional to medical critical illness j, then the probability of medical critical illness i inducing medical critical illness j can be obtained by comparing the number of medical records in the sample analysis data that simultaneously contain medical critical illness i and medical critical illness j with the number of medical records containing medical critical illness i.

6. The method for predicting the risk of severe medical illness based on vital sign data monitored by a smartwatch according to claim 1, characterized in that, The multi-organ system-based coupling risk index generates an intervention plan that includes target organ protection strategies, and visualizes the risk mitigation path and key vital sign control targets through an augmented reality interface. Based on the directed graph induced by severe diseases, and combined with the coupling risk index of various body organs, key organs are identified as intervention target organs. Intervention plans containing target organ protection strategies are generated based on reinforcement learning, and the comprehensive risk reduction effect under different intervention plans is simulated to select the optimal intervention plan. Under the optimal intervention plan, the user's physiological parameters are dynamically monitored, and the intervention plan is adjusted in real time.

7. An electronic device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the medical critical illness risk prediction method based on smartwatch monitoring vital sign data as described in any one of claims 1-6.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the medical critical illness risk prediction method based on smartwatch monitoring vital sign data as described in any one of claims 1-6.

Images