Wearable device-based varicocele cloud early warning and intervention method

Abstract

This invention discloses a cloud-based early warning and intervention method for varicocele based on wearable devices, belonging to the field of wearable device medical monitoring technology. The specific steps of the method are as follows: collecting time-series data from wristwatches, abdominal pressure, and groin sensor patches according to disease course tags, and encrypting and uploading the data to the cloud; obtaining a standardized dataset after timestamp alignment and noise filtering, analyzing risk characteristics, calculating integrals, weighted risk values, and risk contribution rates through multiple algorithms, retaining high-risk samples, and pushing early warning or assessment messages to doctors' terminals based on different population groups; this invention utilizes multiple types of wearable devices to collect multi-dimensional time-series data, encrypts and transmits the data, cleans it in the cloud, analyzes the risk characteristics of both mechanical and neural regulation pathways, completes risk quantification and decomposition through a series of algorithms and retains high-risk samples, and outputs results based on a composite multimodal time-series model to differentiate population groups, establishing a hierarchical push mechanism to achieve home-based, non-intrusive monitoring, and assisting in risk early warning, postoperative assessment, and clinical intervention.

Description

Technical Field

[0001] This invention relates to the field of wearable device medical monitoring technology, specifically a cloud-based early warning and intervention method for varicocele based on wearable devices. Background Technology

[0002] Varicocele is a common vascular disease in andrology, and its progression can adversely affect reproductive function. Clinically, patients are divided into two main groups: high-risk individuals who have not undergone surgery and those undergoing post-operative follow-up. During clinical diagnosis and treatment, it is necessary to continuously monitor the patient's daily physical behavior, intra-abdominal pressure, superficial blood flow, and autonomic nervous system-related physiological changes to assess the disease's progression and post-operative recovery. Currently, clinical assessment primarily relies on outpatient visits and ultrasound examinations, but these examinations are limited to specific time windows within the hospital, only providing short-term static test results. With the continuous development of wearable sensing technology and cloud data processing technology, using wearable devices to collect time-series data on daily physiological behavior and combining this data with cloud platforms for long-term health monitoring has become an important development direction in the medical field. This provides a feasible technical foundation for continuous home monitoring and risk identification of varicocele, and also creates conditions for conducting targeted risk assessments for different disease stages.

[0003] In traditional treatment models, the diagnosis and postoperative recovery assessment of varicocele heavily rely on in-hospital imaging examinations and subjective doctor consultations. Risk factors related to patients' daily physiological behaviors at home cannot be continuously captured. After leaving the hospital, key factors affecting the condition, such as postural habits, fluctuations in intra-abdominal pressure, and changes in autonomic nervous system status, lack effective data collection methods, making it impossible to establish continuous and complete monitoring records. Conventional monitoring methods can only obtain instantaneous status at the time of medical visit, making it difficult to capture high-risk events that occur dynamically within a short period, and unable to distinguish between the disease risks brought about by mechanical factors and neuromodulatory factors. Furthermore, postoperative follow-up lacks the ability to reuse data from past high-risk events, risk assessment lacks the ability to adapt to dynamic disease progression, and risk results are difficult to break down the contribution levels of different risk sources. The early warning feedback chain has limitations; after risk is detected, it is impossible to simultaneously provide tiered alerts to patients and attending physicians, hindering timely home intervention and clinical intervention. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a cloud-based early warning and intervention method for varicocele based on wearable devices. This solution targets high-risk individuals who have not undergone surgery and post-operative follow-up users. It utilizes various wearable devices to collect physiological and behavioral time-series data. After encryption and uploading to the cloud, the data undergoes preprocessing to obtain biomechanical and neuromodulation risk characteristics. Risk quantification is achieved through dual-path time-series integral aggregation, adaptive nonlinear weighting of disease progression, and dual-path risk normalization decomposition algorithms. High-risk event time-series samples are saved for post-operative risk calculation. Based on a composite multimodal time-series model, differentiated risk results are output, and corresponding prompts and messages are pushed to user mobile devices and doctor terminals according to risk level.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a cloud-based early warning and intervention method for varicocele based on wearable devices, the specific steps of which are as follows: S1. Determine the data collection configuration based on the user's disease course label, and collect time-series data from smartwatches of high-risk individuals who have not undergone surgery, smartwatches of postoperative follow-up users, abdominal pressure sensor belts, and groin sensor patches; encrypt and transmit the raw time-series data to the cloud. S2. Perform timestamp alignment on cloud-based time-series data, filter out artifacts and jump data caused by motion disturbances and poor contact, and generate a standardized time-series dataset; S3. Based on the standardized time series dataset, analyze the mechanical risk features and the neural modulation risk features, output the fused risk feature set, and call the dual-path time series integral aggregation algorithm to calculate the integral values ​​of the two types of risk features; S4. Input the fusion risk feature set and integral value into the composite multimodal time series model, call the disease course adaptive nonlinear weighted algorithm to calculate the weighted risk value of the fusion risk feature set, call the dual-path risk normalization decomposition algorithm to calculate the normalized proportion of mechanical risk contribution and neuromodulation risk contribution, and store the time series samples of high-risk events for postoperative use. S5. For high-risk individuals who have not undergone surgery, a warning message will be pushed to the bound doctor's terminal in the early warning scenario; for high-risk users who are followed up after surgery, a standardized assessment message will be pushed to the attending physician's terminal in the postoperative assessment scenario.

[0006] Furthermore, in S1, the following data are uploaded by the smartwatch for high-risk individuals who have not undergone surgery: time-domain parameters of heart rate variability, frequency-domain parameters of heart rate variability, skin conductivity values, triaxial acceleration data, triaxial angular velocity data, cumulative duration of standing position, cumulative duration of sitting position, and cumulative duration of bending position. For postoperative follow-up users, the following data are uploaded by the smartwatch: time-domain parameters of heart rate variability, frequency-domain parameters of heart rate variability, skin conductivity values, triaxial acceleration data, triaxial angular velocity data, cumulative duration of standing position, cumulative duration of sitting position, and cumulative duration of bending position. The following data are also uploaded by the abdominal pressure sensor belt: second-by-second abdominal pressure time-series data. Finally, the following data are uploaded by the groin sensor patch: time-series data of surface blood flow pulsation amplitude and time-series data of surface blood flow pulsation interval.

[0007] Furthermore, in S3, the mechanical risk characteristics include the cumulative duration of abdominal pressure exceeding a preset threshold per unit time, the total duration of standing position per unit time, the total duration of sitting position per unit time, the total duration of bending position per unit time, the cumulative duration of surface blood flow pulsation amplitude exceeding a benchmark value per unit time, and the cumulative duration of surface blood flow pulsation interval deviating from a benchmark value per unit time.

[0008] Furthermore, in S3, the neural regulation risk characteristics include the offset of the time-domain parameter of heart rate variability relative to the individual baseline, the offset of the frequency-domain parameter of heart rate variability relative to the individual baseline, the offset magnitude of the skin conductivity value relative to the individual baseline, and the duration of the offset of the autonomic nerve tension parameter relative to the steady-state baseline per unit time.

[0009] Furthermore, in S3, the mathematical expression of the dual-path time-series integral aggregation algorithm is:

[0010] In the formula, The total integral value of the two types of risk characteristics, This represents the path integral value characteristic of mechanical risk. The integral value of the neural regulatory risk characteristic path. To observe the duration of the sliding window, For the current cloud moment, For timing variables within the window, This represents the dynamic normalization coefficient for mechanical risk characteristics. This represents the dynamic normalization coefficient of the neural regulatory risk characteristics. for The value of the mechanical risk characteristic at any given time. for The value of the risk characteristic of neural regulation at any time.

[0011] Furthermore, in S4, the composite multimodal time series model outputs an onset warning indicator with a duration range of 15 to 45 minutes for high-risk individuals who have not undergone surgery; the composite multimodal time series model outputs a short-term postoperative recurrence probability value and a medium- to long-term valve re-injury risk quantification value for postoperative follow-up individuals. The short-term postoperative time range is 0 to 6 months after surgery, and the medium- to long-term time range is 6 months to 3 years after surgery.

[0012] Furthermore, in S4, the high-risk event time series sample storage operation marks the complete time series data within a preset time period before and after the warning identification trigger time as high-risk event samples and writes them to the user's exclusive storage area in the cloud for persistent storage; when the user's disease course label is updated to the postoperative follow-up status, the cloud automatically retrieves all high-risk event time series samples in the corresponding user's exclusive storage area and incorporates them into the input dataset for postoperative recurrence risk calculation.

[0013] Furthermore, in S4, the mathematical expression for the disease course adaptive nonlinear weighted algorithm is:

[0014] In the formula, For weighted risk values, The total integral value of the two types of risk characteristics, This is the global smoothing adjustment coefficient. This represents the time interval between the completion of the surgery and its completion.

[0015] Furthermore, in S4, the mathematical expression for the dual-path risk normalization decomposition algorithm is:

[0016] In the formula, The normalized proportion of contribution to mechanical risk. The normalized percentage contribution to the risk of neural regulation This represents the path integral value characteristic of mechanical risk. The integral value of the neural regulatory risk characteristic path. This is the total integral value of the two types of risk characteristics.

[0017] Furthermore, in S5, based on the weighted risk value The risk level is divided into low-risk, medium-risk, and high-risk levels. When the risk level is low, push reminders to users' mobile devices to control standing time, sitting time, and bending over. When the risk level is medium, a preset number of pelvic floor relaxation exercise sequences and breathing regulation exercise sequences will be pushed to the user's mobile device. When the risk level is high, for high-risk individuals who have not undergone surgery, a warning message containing the user identifier, risk trigger time, and risk value is pushed to the bound doctor's terminal in the warning scenario; for postoperative follow-up users, a standardized assessment message is pushed to the attending physician's terminal in the postoperative assessment scenario. The standardized assessment message contains the user identifier, risk level value, risk source classification, and time-series characteristic statistical data, and a risk warning message is pushed to the corresponding user's mobile terminal at the same time.

[0018] Compared with existing technologies, this cloud-based early warning and intervention method for varicocele based on wearable devices has the following beneficial effects: 1. This invention utilizes various wearable sensing devices to collect multi-dimensional, time-series physiological and behavioral data. It differentiates data collection logic between high-risk individuals who have not undergone surgery and post-operative follow-up users. Combined with encrypted transmission at the device end and cloud-based data cleaning and processing, it ensures data transmission security and quality. Based on a dual-path analysis approach of biomechanical risk characteristics and neural regulatory risk characteristics, it performs time-series integral aggregation processing on these two types of risk characteristics. This allows it to simultaneously capture the biomechanical risks caused by postural abdominal pressure and the regulatory risks caused by autonomic nervous system fluctuations, overcoming the limitations of single-indicator assessment. Through adaptive nonlinear weighting processing and risk normalization decomposition, it can dynamically adjust risk weights for different disease stages and extract the contribution rate of different risk sources. Time-series samples of high-risk events are retained for reuse in post-operative risk assessment, providing complete original event evidence for subsequent risk judgment and achieving a multi-dimensional, objective, and quantitative assessment of varicocele risk.

[0019] 2. This invention utilizes a composite multimodal time-series model to achieve differentiated output for different population groups. It provides early warning indicators for high-risk individuals who have not undergone surgery, and quantifies the short-term recurrence probability and medium- to long-term valve re-injury risk for post-operative follow-up users, establishing a hierarchical push response mechanism. For low, medium, and high risk levels, it pushes behavioral control prompts and action adjustment sequences to users' mobile devices, and early warning messages and standardized assessment reports to doctors' terminals, establishing a two-way information channel between users and medical staff. This solution enables continuous, unobtrusive home monitoring, achieving dynamic risk identification without frequent in-person medical examinations. By combining risk warnings, post-operative assessments, and intervention prompts, it helps users promptly adjust their behavior and facilitates attending physicians' understanding of patients' dynamic risk status, supporting clinical follow-up and early intervention. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0021] Figure 1 This is an overall flowchart of the present invention; Figure 2 This is a data flow diagram for dual-path risk feature extraction and sliding window integral aggregation; Figure 3 This is a logic diagram for the computation of composite multimodal time series models and the storage of high-risk event samples. Detailed Implementation

[0022] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0023] Example 1: like Figure 1 As shown, this invention provides a cloud-based early warning and intervention method for varicocele based on wearable devices. It utilizes various wearable sensing devices to collect temporal physiological positional data, abdominal pressure data, and groin surface blood flow data. By integrating mechanical risk characteristics and neural regulatory risk characteristics, and relying on a dual-path temporal integral aggregation algorithm, a disease course adaptive nonlinear weighting algorithm, a dual-path risk normalization decomposition algorithm, and a composite multimodal temporal model, it distinguishes between two groups: high-risk individuals who have not undergone surgery and postoperative follow-up users. This enables stratified risk assessment, tiered user-end prompts, and doctor-end message pushes, completing cloud-based early warning and postoperative follow-up assessment and intervention for varicocele.

[0024] The method described in this embodiment specifically includes: Based on the user's disease course tags, the data collection configuration is determined, and time-series data of smartwatches are collected from high-risk individuals who have not undergone surgery, smartwatches of postoperative follow-up users, abdominal pressure sensor belts, and groin sensor patches. The raw time-series data is then encrypted on the device and transmitted to the cloud. Perform timestamp alignment on the time-series data received from the cloud, filter out artifacts and jump data caused by motion disturbances and poor contact, and generate a standardized time-series dataset; Based on the standardized time series dataset, the mechanical risk features and neural modulation risk features are analyzed, and a fused risk feature set is output. The dual-path time series integral aggregation algorithm is called to calculate the integral values ​​of the two types of risk features. The fusion risk feature set and integral value are input into the composite multimodal time series model. The disease course adaptive nonlinear weighted algorithm is called to calculate the weighted risk value of the fusion risk feature set. The dual-path risk normalization decomposition algorithm is called to calculate the normalized proportion of mechanical risk contribution and neuromodulation risk contribution. High-risk event time series samples are stored for postoperative use. For high-risk individuals who have not undergone surgery, warning messages are pushed to the bound doctor's terminal in the early warning scenario; for high-risk users who are followed up after surgery, standardized assessment messages are pushed to the attending physician's terminal in the postoperative assessment scenario; at the same time, according to low, medium and high risk levels, corresponding control prompts or risk warning messages are pushed to the user's mobile terminal.

[0025] The process of implementing this invention involves collecting multi-dimensional temporal signals of body position, autonomic nervous system, abdominal pressure, and local blood flow in the groin area using multiple wearable sensing terminals, and extracting two feature systems in parallel: mechanical risk features and neuromodulation risk features. A dual-path temporal integral aggregation algorithm is used to perform temporal integral operations on the mechanical and neuromodulation risk features within a sliding observation window, obtaining independent integral results for each path and a total integral value. Combined with the user's surgical history label, a disease-adaptive nonlinear weighted algorithm is used to introduce the time interval after surgery for nonlinear decay correction, outputting a weighted risk value that fits the user's disease status. A dual-path risk normalization decomposition algorithm is used to separate the contribution rates of the two risk categories. The composite multimodal temporal model outputs differentiated results for users with different disease history labels: for high-risk individuals who have not undergone surgery, an attack warning indicator is output for the next 15-45 minutes; for postoperative follow-up users, quantitative values ​​for short-term recurrence probability (0-6 months) and medium-to-long-term valve re-injury risk (6 months-3 years) are output. The cloud-based system marks and saves the raw time-series data before and after the risk trigger moment as high-risk event time-series samples. When the user's disease course label is switched to postoperative follow-up status, the historical high-risk event time-series samples are automatically included in the recurrence risk calculation. The system performs tiered intervention according to three risk levels: low, medium, and high. For low-risk users, it pushes daily posture duration control reminders; for medium-risk users, it pushes pelvic floor relaxation and breathing regulation exercise sequences; and for high-risk users, in addition to pushing reminders, it also pushes warning messages or standardized assessment reports to the corresponding doctor's terminal, completing a two-way intervention process from the cloud to the user and from the cloud to the doctor's terminal.

[0026] The process involves determining the data collection configuration based on user disease course tags, collecting time-series data from smartwatches of high-risk individuals who have not undergone surgery, smartwatches of post-operative follow-up users, abdominal pressure sensor belts, and groin sensor patches; and encrypting and transmitting the raw time-series data to the cloud, including: Read the user's medical history tags stored in the cloud. If the medical history tag indicates a high-risk group who has not undergone surgery, enable the data acquisition channel of the smartwatch and disable the acquisition channels of the abdominal pressure sensor belt and groin sensor patch. If the medical history tag indicates a postoperative follow-up user, enable all acquisition channels of the smartwatch, abdominal pressure sensor belt, and groin sensor patch. For high-risk individuals who have not undergone surgery, the following time-series data were collected from smartwatches: heart rate variability time-domain parameters, heart rate variability frequency-domain parameters, skin conductivity values, triaxial acceleration data, triaxial angular velocity data, cumulative duration of standing, sitting, and bending positions. For postoperative follow-up users, the following time-series data were collected from smartwatches: heart rate variability time-domain parameters, heart rate variability frequency-domain parameters, skin conductivity values, triaxial acceleration data, triaxial angular velocity data, cumulative duration of standing, sitting, and bending positions; intra-second abdominal pressure time-series data from the abdominal pressure sensor belt; and time-series data of surface blood flow pulsation amplitude and interval from the groin sensor patch. The raw time-series data is encrypted using a symmetric encryption algorithm on the device side. The encryption key is agreed upon in advance with the cloud server. The encrypted raw time-series data is transmitted to the cloud server via the cellular mobile network. After receiving the data, the cloud server performs decryption verification. If the verification is successful, the data is stored in the raw data storage area. If the decryption verification fails, the current batch of data is discarded and a retransmission command is sent to the wearable device.

[0027] Specifically, the system first reads the patient history tags stored in the cloud under the user's account. These tags are categorized into two types: high-risk individuals without surgery and post-operative follow-up users. Different tags correspond to different acquisition channel switching strategies to avoid power waste caused by continuous acquisition from unnecessary sensing devices. High-risk individuals without surgery can obtain all time-series parameters related to body position and autonomic nervous system solely through a smartwatch. Post-operative follow-up users need to wear an abdominal pressure sensing band and a groin sensing patch to acquire information on abdominal pressure and local blood flow in the groin area. The smartwatch continuously outputs time-domain and frequency-domain parameters of heart rate variability, skin conductivity, triaxial acceleration, and angular velocity, while also calculating the cumulative duration of different body positions. The abdominal pressure sensing band outputs the abdominal pressure time-series at a sampling rate of one second. The groin sensing patch continuously captures the amplitude and interval time-series of blood flow pulsation. All sensing terminals perform symmetric encryption locally, with the key pre-negotiated between the device and the cloud server during the manufacturing process. The device will not send physiological time-series data out in plaintext. Encrypted data packets are uploaded to the cloud server via the cellular network. The cloud server decrypts the packets using an agreed-upon key and verifies the integrity of the data packets and the validity of the timestamps. If the verification is successful, the data is stored in the original data storage area. If decryption fails, data packets are lost, or timestamps are incorrect, the cloud server issues a retransmission command, and the device re-uploads the batch of time-series data.

[0028] During the data acquisition configuration process, the user account retrieves the disease course tag indicating a postoperative follow-up user. The cloud sends acquisition configuration commands to the wearable device group, activating data acquisition on the smartwatch, abdominal pressure sensor belt, and groin sensor patch. The smartwatch outputs sampled data including heart rate variability (NN interval, RMSSD), frequency domain LF and HF parameters, skin conductivity, triaxial acceleration, and angular velocity. It also provides statistics on the cumulative standing time (320 seconds), sitting time (210 seconds), and bending time (45 seconds) within 10 minutes. The abdominal pressure sensor belt outputs a second-by-second abdominal pressure sequence. The groin sensor patch outputs the amplitude and interval of surface blood flow pulsation. All raw time-series data is encrypted on the device using an agreed-upon symmetric key, generating encrypted data packets that are then uploaded to the cloud server via a 4G cellular network. The cloud server receives data packets and performs decryption verification to check the integrity of the data packets and the absence of timestamp sequence jumps. If the verification passes, the batch of time-series data is written to the original data storage area. If a bit error occurs during the transmission of a data packet and the decryption verification fails, the cloud sends a retransmission command to the device group, and the device reads the local cache and re-uploads the time-series data.

[0029] The process of performing timestamp alignment on cloud-based time-series data, filtering out artifacts and jump data caused by motion disturbances and poor contact, and generating a standardized time-series dataset includes: Read multi-source time-series data of the same user in the original data storage area, use the unified cloud system time as the benchmark, perform timestamp alignment on each time-series record output by smartwatch, abdominal pressure sensor belt, and groin sensor patch, and map the sampling time of different devices to the same unified time axis; Set signal jump thresholds and reasonable ranges for signal amplitude to identify motion disturbance samples caused by violent limb swinging and to identify artifacts and jump samples caused by loose or detached sensor wear. The identified artifacts and abrupt anomalies are removed. Short-term missing time segments are filled by interpolation of adjacent valid time segments. Long-term missing segments are marked with missing identifiers and are not filled by interpolation. After completing timestamp alignment, outlier filtering, and missing marker processing, the time series is regularized according to a fixed sampling interval to generate a standardized time series dataset.

[0030] Different wearable hardware sampling clocks operate independently. The local clocks of smartwatches, abdominal pressure sensor bands, and groin sensor patches may have slight offsets. Therefore, it's necessary to use the cloud server system time as a unified benchmark to map the time series records of different devices onto the same timeline, aligning timestamps to ensure that body position, abdominal pressure, blood flow, and physiological neurological parameters can be matched and calculated at the same time. Based on clinical experience, signal jump thresholds and reasonable ranges for physiological signal amplitudes are preset. When a sudden change in signal amplitude exceeds the jump threshold within a short period, it is judged as a motion disturbance; when the signal amplitude exceeds the reasonable upper and lower limits of human physiological parameters, it is judged as a sensor contact malfunction artifact jump. Identified abnormal samples are directly discarded. For short-term data gaps of 1-3 seconds, interpolation is used to complete the gaps using adjacent valid time series data. For long-term signal gaps exceeding 5 seconds, no interpolation is performed; instead, a missing marker is directly added to the corresponding position in the dataset to avoid introducing false features through interpolation. The time series is normalized to a unified sampling step size, outputting a standardized time series dataset for subsequent risk feature analysis.

[0031] During time-series data preprocessing, multi-source time-series data from the original storage area of ​​the same postoperative follow-up user are read. The local clock of the smartwatch is offset by 1.2 seconds compared to the cloud reference, and the local clock of the abdominal pressure sensor is offset by -0.8 seconds. Timestamp alignment is completed with the cloud system time, and all time series are mapped to the same time axis. A sudden jump in the triaxial angular velocity values ​​is detected in a 2-second window, and at the same time, the abdominal pressure value instantaneously exceeds the upper limit of human physiological function. This is determined to be a large-scale limb movement disturbance and a brief contact detachment of the sensor band, and the time series samples in this window are removed. There is a 2-second missing groin blood flow data segment, which is filled by interpolation using adjacent valid samples. A subsequent 8-second segment with no output from the abdominal pressure sensor signal is marked as a long-term missing segment, and no interpolation is performed. All time series are normalized according to a uniform sampling step size of 1 second, and the standardized time-series dataset for this user is output.

[0032] The process involves analyzing mechanical risk features and neural modulation risk features based on a standardized time-series dataset, outputting a fused risk feature set, and then using a dual-path time-series integral aggregation algorithm to calculate the integral values ​​of the two types of risk features, including: Traverse the standardized time-series dataset and extract mechanical risk features. The mechanical risk features include the cumulative duration of abdominal pressure exceeding the preset threshold per unit time, the total duration of standing position per unit time, the total duration of sitting position per unit time, the total duration of bending position per unit time, the cumulative duration of surface blood flow pulsation amplitude exceeding the benchmark value per unit time, and the cumulative duration of surface blood flow pulsation interval deviating from the benchmark value per unit time. Traverse the standardized time-series dataset and extract neural regulation risk features; neural regulation risk features include the offset of heart rate variability time-domain parameters relative to the individual baseline, the offset of heart rate variability frequency-domain parameters relative to the individual baseline, the offset of skin conductivity values ​​relative to the individual baseline, and the duration of the offset of autonomic nerve tension parameters relative to the steady-state baseline per unit time. The analysis will combine all mechanical risk features and neural regulation risk features to form a fusion risk feature set. The two-path time series integration aggregation method is invoked, and the duration of the observation sliding window is... Within, respectively to Momentary mechanical risk characteristic value , Time-based neural regulation risk characteristic value Perform time-series integration to obtain the path integral score of the mechanical risk characteristics. Neuromodulation risk characteristic path integral score Then sum them up to get the total integral score of the two types of risk characteristics. The mathematical expression for the dual-path time-series integral aggregation method is:

[0033] In the formula, The total score for the two types of risk characteristics; The integral score for the mechanical risk characteristic path; The integral score of the neural regulation risk characteristic path; To observe the duration of the sliding window; This refers to the current cloud-based moment; For timing variables within the window; This is the dynamic normalization coefficient for mechanical risk characteristics; The dynamic normalization coefficient for the risk characteristics of neural regulation; for The value of the mechanical risk characteristic at any given moment; for The value of the risk characteristic of neural regulation at any time.

[0034] Dynamic normalization coefficient of mechanical risk characteristics Calculate the expression:

[0035] Dynamic normalization coefficient of neural regulatory risk characteristics Calculate the expression:

[0036] In the formula, This represents the maximum value of the mechanical risk characteristic within the sliding window; Establish a steady-state baseline for individual user mechanical risk; A maximum value range is preset for the mechanical risk characteristics; This represents the maximum value of the neural modulation risk characteristic within the sliding window; Establish a baseline for the steady-state risk of individual user neuromodulation; A maximum value range is preset for the risk characteristics of neural regulation.

[0037] Dynamic normalization coefficient of mechanical risk characteristics The coefficient is calculated by subtracting the user's individual steady-state baseline mechanical risk from the current maximum value of the sliding window's mechanical risk characteristic, and then dividing by the preset maximum value range of the mechanical risk characteristic. This coefficient is used to map mechanical risk characteristics of different dimensions to a unified numerical range, eliminating calculation biases caused by differences in the dimensions of different physiological indicators; the dynamic normalization coefficient of neural regulation risk characteristics. The coefficient is calculated by subtracting the user's individual neural regulation risk steady-state baseline from the current maximum value of the sliding window neural regulation risk feature, and then dividing by the preset maximum value range of the neural regulation risk feature. This coefficient achieves dimensional unification of multi-source features of neural regulation, so that the integral results of the mechanical path and the neural regulation path have a numerical basis for addition and comparison.

[0038] The mechanical risk dimension primarily captures changes in the mechanical load on the spermatic vein caused by intra-abdominal pressure, postural behavior, and fluctuations in local groin blood flow; the neuroregulatory risk dimension captures changes in vascular regulatory function caused by fluctuations in autonomic nervous system state. The system pre-stores the user's personal steady-state baseline, which is generated by statistically analyzing multiple sets of historical data collected from the user in a static, relaxed state. The individual baseline is periodically updated based on the user's long-term data. Various mechanical risk characteristic indicators per unit time are statistically analyzed from the standardized time-series dataset. Simultaneously, the offset and duration of each neurologically related parameter relative to the personal steady-state baseline are calculated, and these are summarized to form a fused risk feature set. A fixed observation sliding window duration is set. Within this time window, time-series integration is performed on the mechanical risk feature sequence and the neural regulation risk feature sequence, and then normalized by dividing by the window duration. The integration results of the two paths are obtained respectively, and the sum is used to obtain the total integral value. This total integral represents the overall risk accumulation level of the user within the current window.

[0039] When performing risk feature analysis and dual-path time series integral aggregation calculation, the duration of the observation sliding window is set. The standardized time-series dataset of the postoperative follow-up user was read, and the following statistics were obtained within a 30-minute window: cumulative duration of abdominal pressure exceeding the threshold: 420 seconds; total standing time: 960 seconds; total sitting time: 650 seconds; total bending time: 120 seconds; cumulative duration of surface blood flow pulsation amplitude exceeding the baseline: 310 seconds; cumulative duration of surface blood flow pulsation interval deviating from the baseline: 260 seconds. Mechanical risk features were extracted. The user's individual steady-state baseline was read, and the time-domain and frequency-domain parameter offsets of heart rate variability, skin conductivity offset amplitude, and duration of autonomic nerve tension offset were calculated. Neuromodulation risk features were extracted and merged to obtain a fusion risk feature set.

[0040] Maximum value of mechanical risk characteristics within this window User individual mechanical risk steady-state baseline The maximum value range of the mechanical risk characteristics is preset. Maximum value of neural regulation risk characteristics User-specific neural regulatory risk homeostasis baseline The maximum value range of the neural regulation risk characteristics is preset. Substituting into the coefficient calculation formula yields... -0.20) / 1.00=0.64, -0.24) / 1.00=0.48; Perform integration on the time-series features within a 30-minute sliding window to calculate the path integral value of the mechanical risk feature. Integral value of neural regulatory risk characteristic path According to the formula Calculate the total integral value of the two types of risk characteristics. ,like Figure 2 As shown.

[0041] The process of inputting the fused risk feature set and integral values ​​into a composite multimodal time-series model, calling a disease course adaptive nonlinear weighted algorithm to calculate the weighted risk value of the fused risk feature set, calling a dual-path risk normalization decomposition algorithm to calculate the normalized proportions of mechanical risk contribution and neuromodulation risk contribution, and storing high-risk event time-series samples for postoperative use includes: This will integrate the risk feature set and the path integral value of the mechanical risk feature. Integral value of neural regulation risk characteristic pathway Total score Input the composite multimodal time series model; the composite multimodal time series model outputs an onset warning indicator with a duration range of 15 to 45 minutes for high-risk unoperated individuals; the composite multimodal time series model outputs short-term postoperative recurrence probability values ​​and medium- to long-term valve re-injury risk quantification values ​​for postoperative follow-up individuals, with the short-term time range being 0 to 6 months after surgery and the medium- to long-term time range being 6 months to 3 years after surgery; The adaptive nonlinear weighted algorithm for disease course is used to calculate the weighted risk value of the fused risk feature set. ; The mathematical expression for the adaptive nonlinear weighted algorithm for disease progression is:

[0042] Global smoothing adjustment coefficient Calculate the expression:

[0043] In the formula, Weighted risk value; The total integral value of the two types of risk characteristics; This is the global smoothing adjustment coefficient; This refers to the time interval between the completion of the surgery; The basic attenuation coefficient; Influence factors of historical high-risk events; This represents the percentage of high-risk events occurring within the statistical period.

[0044] Global smoothing adjustment coefficient The basic attenuation coefficient is calculated by multiplying the basic attenuation coefficient by 1 and subtracting the product of the historical high-risk event impact factor and the proportion of high-risk events occurring within the statistical period; The impact factor of historical high-risk events was determined based on clinical postoperative follow-up sample datasets. As a preset constant, the percentage of high-risk events occurring within a statistical period is calculated by dividing the actual number of high-risk events within the statistical period by the maximum allowed number of high-risk events within the statistical period. This calculation method ensures a smooth global adjustment coefficient. It can dynamically change based on the frequency of a user's historical high-risk events; the more frequent the historical high-risk events, the better. The smaller the value, the slower the risk decay rate, thus reflecting the continued impact of historical high-risk behaviors on postoperative risk assessment. The normalized proportion of mechanical risk contribution is calculated using a dual-path risk normalization decomposition algorithm. Normalized proportion of risk contribution from neural regulation The mathematical expression for the dual-path risk normalization decomposition algorithm is:

[0045] In the formula, Normalized percentage of contribution to mechanical risk; The normalized percentage of the contribution to the risk of neuromodulation; This represents the integral value of the mechanical risk characteristic path. The integral value of the neural regulatory risk characteristic path; The total integral value of the two types of risk characteristics; When weighted risk value If the high-risk threshold is reached, a high-risk event is triggered. The high-risk event time series sample storage operation marks the complete time series data within a preset time period before and after the warning identification trigger time as high-risk event samples and writes them to the user's dedicated cloud storage area for persistent storage. When the user's disease course label is updated to postoperative follow-up status, the cloud automatically retrieves all high-risk event time series samples in the corresponding user's dedicated storage area and incorporates them into the input dataset for postoperative recurrence risk calculation.

[0046] After receiving the feature set and integration results, the composite multimodal temporal model differentiates the results based on disease progression labels. For high-risk individuals who have not undergone surgery, it outputs a warning sign of onset within the next 15-45 minutes, indicating a risk of disease worsening in the short term. For postoperative follow-up users, it outputs quantitative values ​​for short-term recurrence probability (0-6 months) and medium-to-long-term valve re-injury risk (6 months-3 years). An adaptive nonlinear weighted algorithm for disease progression incorporates the time interval since surgery completion. The total risk integral is corrected through an exponential decay function; the global smoothing adjustment coefficient... Pre-configured to characterize the rate of postoperative risk decay over time. A dual-path risk normalization decomposition algorithm breaks down the total risk into the proportion of mechanical and neuromodulation-related sources, facilitating subsequent message differentiation of risk sources. When the weighted risk value reaches a high-risk threshold, high-risk event saving logic is triggered, saving complete time-series data before and after the trigger as high-risk event samples. If the user's disease course label is subsequently changed to postoperative follow-up, the cloud automatically reads all historical high-risk event time-series samples for that user and uses them as part of the model input for recurrence risk calculation. Historical high-risk behaviors will affect postoperative assessment. If the value approaches 0, the normalized proportion calculation is not performed, and the risk contribution rate is directly marked as invalid.

[0047] During the implementation of model computation, weighting, risk decomposition, and high-risk sample storage, the user's disease course label was "postoperative follow-up user," indicating the time interval since the completion of the surgery. Day; Set the basic attenuation coefficient Historical high-risk event impact factor The percentage of high-risk events occurring within the statistical period Substitute into the global smoothing adjustment coefficient calculation formula ,calculate - ; will integrate risk feature sets, , , Inputting a composite multimodal time-series model, the model outputs a short-term postoperative recurrence probability of 27.3% and a medium- to long-term valve re-injury risk quantification value of 0.41. Substituting these values ​​into the adaptive nonlinear weighted algorithm formula... ,calculate Substituting into the dual-path risk normalization decomposition algorithm, , That is, the contribution rate of mechanical risk is 62%, and the contribution rate of neural regulation risk is 38%. The preset high-risk judgment threshold is 0.70, and the current weighted risk value is... If the value exceeds a threshold, a high-risk event is triggered. Complete time-series data for 10 minutes before and after the trigger time is captured, marked as high-risk event samples, and written to the user's dedicated cloud storage area. Subsequently, if the user switches from the "non-surgical high-risk group" disease course label to the postoperative follow-up status, the cloud automatically reads all historical high-risk event time-series samples from the user's dedicated storage area and sends them to the composite multimodal time-series model to participate in the recurrence risk calculation process.

[0048] For high-risk individuals who have not undergone surgery, the aforementioned measures include pushing warning messages to the bound doctor's terminal in early warning scenarios; and for high-risk users undergoing postoperative follow-up, pushing standardized assessment messages to the attending physician's terminal in postoperative assessment scenarios, including: Based on weighted risk value The risk level is divided into low-risk, medium-risk, and high-risk levels. For low-risk areas, push notifications to users' mobile devices to control standing time, sitting time, and bending over. For medium-risk users, preset sets of pelvic floor relaxation and breathing regulation exercise sequences are pushed to their mobile devices. In high-risk warning scenarios, a warning message containing user identifier, risk trigger time, and risk value is pushed to the bound doctor's terminal; in high-risk postoperative assessment scenarios, a standardized assessment message is pushed to the attending physician's terminal. The standardized assessment message contains user identifier, risk level value, risk source classification, and time-series characteristic statistics, and a risk warning message is pushed to the corresponding user's mobile terminal simultaneously.

[0049] The system pre-configures risk level threshold ranges, mapping weighted risk values ​​to three risk levels: low, medium, and high. Low risk indicates a relatively mild level of accumulated risk, with intervention primarily focused on daily behavioral management reminders, prompting users to control the duration of standing, prolonged sitting, and bending. Medium risk indicates an increased risk level, pushing standardized rehabilitation exercise sequences to guide users in performing pelvic floor relaxation and breathing regulation to reduce venous pressure load. High risk indicates that the accumulated risk has reached a warning level, distinguishing between warning scenarios (high-risk individuals without surgery) and post-operative assessment scenarios (post-operative follow-up users). In warning scenarios, a concise warning message is pushed to the bound doctor's terminal, informing the user of their identity, trigger time, and risk value. In post-operative assessment scenarios, a standardized assessment report is issued, including a risk source classification (mechanical risk-dominated or neuromodulation risk-dominated) and time-series characteristic statistical data, facilitating multi-dimensional information reference for the attending physician in diagnosis and treatment. Simultaneously, regardless of the high-risk scenario, a risk alert message is simultaneously pushed to the user's mobile device, informing the user of their current risk status and recommending timely follow-up visits.

[0050] During the implementation of tiered intervention message push, a pre-defined risk classification range is used: a weighted risk value less than 0.3 indicates low risk; 0.3-0.7 indicates medium risk; and greater than 0.7 indicates high risk. Case 1: A high-risk user who did not undergo surgery had a weighted risk value of 0.22, classifying them as low risk. The user received a message on their mobile device: "Please pay attention to controlling your daily standing and sitting time, and reduce frequent bending movements." Case 2: Another high-risk user who did not undergo surgery had a weighted risk value of 0.51, classifying them as medium risk. Their mobile device received a push notification with three sets of pelvic floor relaxation exercise sequences and corresponding breathing regulation exercise sequences, along with instructions on the duration of each exercise. Case 3: The aforementioned postoperative follow-up user's weighted risk value... If the weighted risk value is approximately 0.754, it is considered high-risk and falls under the post-operative assessment scenario. A standardized assessment message is pushed to the attending physician's terminal, carrying the user's unique identifier, the risk level value of 0.754, and the risk source classification: mechanical risk contribution rate of 62% and neuromodulation risk contribution rate of 38%. It also includes statistical data on the temporal characteristics of intra-window abdominal pressure, body position, blood flow, and heart rate variability. Simultaneously, a risk warning message is pushed to the user's mobile device, indicating a high current risk level and suggesting a follow-up consultation with the attending physician. If the weighted risk value for a high-risk user who has not undergone surgery is 0.79, it falls under the warning scenario. An warning message is pushed to the bound doctor's terminal, containing the user identifier, risk trigger time, and risk value of 0.79. A high-risk warning is also pushed to the user's mobile device.

[0051] Example 2

[0052] The method is basically the same as in Example 1, except that it further includes: Periodically read all high-risk event time-series samples from the user's dedicated cloud storage area; The frequency of high-risk events within a statistical unit period; The frequency of high-risk events is used as an auxiliary input to feed into the composite multimodal time series model to correct the onset warning indicators, the short-term postoperative recurrence probability value, and the medium- and long-term valve re-injury risk quantification value.

[0053] Specifically, a dedicated storage area persistently stores all historical high-risk event time-series samples, and the system statistically analyzes the trigger frequency of high-risk events on a weekly or monthly basis. A higher frequency of high-risk events indicates that the user repeatedly experiences high-risk physiological and postural states, corresponding to an increased potential for disease aggravation or postoperative recurrence. The statistically obtained high-risk event frequency is used as an auxiliary feature input into a composite multimodal time-series model to correct the model's output. For high-risk individuals who have not undergone surgery, the confidence weight of the onset warning indicator is increased. For users undergoing postoperative follow-up, the quantitative values ​​for short-term postoperative recurrence probability and medium- to long-term valve re-damage risk are increased, enabling the model output to consider both current window risk and the cumulative effect of historical long-term high-risk behaviors, further improving the reliability of cloud-based early warning and assessment.

[0054] When performing the model output result correction process, the cloud-based dedicated storage area of ​​a high-risk user who has not undergone surgery is read, and it is found that a total of 7 high-risk events were triggered in the past 30 days, with a high-risk event frequency of 7 times / 30 days. This frequency is used as an auxiliary feature and fed into the composite multimodal time series model, increasing the confidence weight of the 15-45 minute onset warning label output by the model. For a postoperative follow-up user, the dedicated storage area is read, and it is found that 5 high-risk events occurred in the past 90 days. The frequency information is input into the composite multimodal time series model, and the quantitative values ​​of the postoperative short-term recurrence probability and the medium- and long-term valve re-injury risk are adjusted upwards. The risk assessment optimization is completed by comprehensively considering historical high-risk behaviors. Figure 3 As shown.

[0055] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A cloud-based early warning and intervention method for varicocele based on wearable devices, characterized in that, The specific steps of this method are as follows: S1. Determine the data collection configuration based on the user's disease course label, and collect time-series data from smartwatches of high-risk individuals who have not undergone surgery, smartwatches of postoperative follow-up users, abdominal pressure sensor belts, and groin sensor patches; encrypt and transmit the raw time-series data to the cloud. S2. Perform timestamp alignment on cloud-based time-series data, filter out artifacts and jump data caused by motion disturbances and poor contact, and generate a standardized time-series dataset; S3. Based on the standardized time series dataset, analyze the mechanical risk features and the neural modulation risk features, output the fused risk feature set, and call the dual-path time series integral aggregation algorithm to calculate the integral values ​​of the two types of risk features; S4. Input the fusion risk feature set and integral value into the composite multimodal time series model, call the disease course adaptive nonlinear weighted algorithm to calculate the weighted risk value of the fusion risk feature set, call the dual-path risk normalization decomposition algorithm to calculate the normalized proportion of mechanical risk contribution and neuromodulation risk contribution, and store the time series samples of high-risk events for postoperative use. S5. For high-risk individuals who have not undergone surgery, push warning messages to the bound doctor's terminal in the warning scenario; For high-risk users during postoperative follow-up, standardized assessment messages are pushed to the attending physician's terminal during postoperative evaluation.

2. The cloud-based early warning and intervention method for varicocele based on wearable devices according to claim 1, characterized in that, In S1, the following data are uploaded by the smartwatch for high-risk individuals who have not undergone surgery: time-domain parameters of heart rate variability, frequency-domain parameters of heart rate variability, skin conductivity values, triaxial acceleration data, triaxial angular velocity data, cumulative duration of standing position, cumulative duration of sitting position, and cumulative duration of bending position. The following data are uploaded by the smartwatch for postoperative follow-up users: time-domain parameters of heart rate variability, frequency-domain parameters of heart rate variability, skin conductivity values, triaxial acceleration data, triaxial angular velocity data, cumulative duration of standing position, cumulative duration of sitting position, and cumulative duration of bending position. The following data are uploaded by the abdominal pressure sensor belt: time-series data of intra-abdominal pressure. The following data are uploaded by the groin sensor patch: time-series data of surface blood flow pulsation amplitude and time-series data of surface blood flow pulsation interval.

3. The cloud-based early warning and intervention method for varicocele based on wearable devices according to claim 1, characterized in that, In S3, the mechanical risk characteristics include the cumulative duration of abdominal pressure exceeding a preset threshold per unit time, the total duration of standing position per unit time, the total duration of sitting position per unit time, the total duration of bending position per unit time, the cumulative duration of surface blood flow pulsation amplitude exceeding a benchmark value per unit time, and the cumulative duration of surface blood flow pulsation interval deviating from a benchmark value per unit time.

4. The cloud-based early warning and intervention method for varicocele based on wearable devices according to claim 3, characterized in that, In S3, the neuromodulation risk characteristics include the offset of the time-domain parameter of heart rate variability relative to the individual baseline, the offset of the frequency-domain parameter of heart rate variability relative to the individual baseline, the magnitude of the offset of the skin conductivity value relative to the individual baseline, and the duration of the offset of the autonomic nerve tension parameter relative to the steady-state baseline per unit time.

5. The cloud-based early warning and intervention method for varicocele based on wearable devices according to claim 4, characterized in that, In S3, the mathematical expression of the dual-path time-series integral aggregation algorithm is: In the formula, The total integral value of the two types of risk characteristics. This represents the path integral value characteristic of mechanical risk. The integral value of the neural regulatory risk characteristic path. To observe the duration of the sliding window, For the current cloud moment, For timing variables within the window, This represents the dynamic normalization coefficient for mechanical risk characteristics. This represents the dynamic normalization coefficient of the neural regulatory risk characteristics. for The value of the mechanical risk characteristic at any given time. for The value of the risk characteristic of neural regulation at any time.

6. The cloud-based early warning and intervention method for varicocele based on wearable devices according to claim 1, characterized in that, In S4, the composite multimodal time series model outputs an onset warning indicator with a duration of 15 to 45 minutes for high-risk individuals who have not undergone surgery; the composite multimodal time series model outputs a short-term postoperative recurrence probability value and a medium- to long-term valve re-injury risk quantification value for postoperative follow-up individuals. The short-term postoperative time range is 0 to 6 months after surgery, and the medium- to long-term time range is 6 months to 3 years after surgery.

7. The cloud-based early warning and intervention method for varicocele based on wearable devices according to claim 6, characterized in that, In S4, the high-risk event time series sample storage operation marks the complete time series data within a preset time period before and after the warning identification trigger time as high-risk event samples and writes them into the user's exclusive storage area in the cloud for persistent storage. When a user's disease course label is updated to postoperative follow-up status, the cloud automatically retrieves all high-risk event time series samples from the corresponding user's exclusive storage area and incorporates them into the input dataset for calculating the postoperative recurrence risk.

8. The cloud-based early warning and intervention method for varicocele based on wearable devices according to claim 7, characterized in that, In S4, the mathematical expression of the disease course adaptive nonlinear weighted algorithm is: In the formula, For weighted risk values, The total integral value of the two types of risk characteristics, This is the global smoothing adjustment coefficient. This represents the time interval between the completion of the surgery and its completion.

9. The cloud-based early warning and intervention method for varicocele based on wearable devices according to claim 8, characterized in that, In S4, the mathematical expression for the dual-path risk normalization decomposition algorithm is: In the formula, The normalized proportion of contribution to mechanical risk. The normalized percentage contribution to the risk of neural regulation This represents the path integral value characteristic of mechanical risk. The integral value of the neural regulatory risk characteristic path. This is the total integral value of the two types of risk characteristics.

10. The cloud-based early warning and intervention method for varicocele based on wearable devices according to any one of claims 1-9, characterized in that, In step S5, based on the weighted risk value The risk level is divided into low-risk, medium-risk, and high-risk levels. When the risk level is low, push reminders to users' mobile devices to control standing time, sitting time, and bending over. When the risk level is medium, a preset number of pelvic floor relaxation exercise sequences and breathing regulation exercise sequences will be pushed to the user's mobile device. When the risk level is high, for high-risk individuals who have not undergone surgery, a warning message containing the user identifier, risk trigger time, and risk value is pushed to the bound doctor's terminal in the warning scenario; for postoperative follow-up users, a standardized assessment message is pushed to the attending physician's terminal in the postoperative assessment scenario. The standardized assessment message contains the user identifier, risk level value, risk source classification, and time-series characteristic statistical data, and a risk warning message is pushed to the corresponding user's mobile terminal at the same time.

Images