Postoperative recurrence early warning optimization method and system based on he4 dominance

Abstract

Based on HE4-dominated postoperative recurrence early warning optimization method and system, this invention belongs to the field of intelligent medical diagnostic auxiliary technology. It acquires HE4 and CA125 data of the target object at discrete time points before and after the target time point, extracts reference benchmarks for each, and generates a status suspension command when the HE4 data at the target time point meets the triggering condition relative to the benchmark. It also calculates the CA125 time delay change rate within the observation window, performs backtracking confirmation in conjunction with the CA125 benchmark, outputs an early warning signal, and dynamically updates the triggering condition based on feedback on the accuracy of the early warning. This method can fundamentally eliminate systemic false positives caused by heterogeneous expression time delays, significantly advance the recurrence early warning time window without increasing blood sampling frequency, and achieve accurate early recurrence capture.

Description

Technical Field

[0001] This application belongs to the field of intelligent medical diagnostic auxiliary technology, specifically a postoperative recurrence early warning optimization method and system based on HE4. Background Technology

[0002] Postoperative recurrence monitoring of ovarian cancer is crucial for patient prognosis. Existing long-term follow-up systems generally employ a combined detection strategy of serum tumor markers, among which the simultaneous combined determination of human epididymal protein 4 (HE4) and cancer antigen 125 (CA125) is the most widely used technical baseline. Current protocols are often based on the simultaneous validation assumption: if substantial tumor recurrence occurs, HE4 and CA125 will necessarily exceed the upper limit of the normal reference range simultaneously at the same testing time point.

[0003] However, this assumption overlooks the heterotonic expression patterns of microrecurrences. HE4 has a small molecular weight and a sensitive secretion pathway, allowing for a significant spike in blood concentration even in the early stages of recurrence. In contrast, CA125, as a high-molecular-weight glycoprotein, requires a certain tumor burden to accumulate, resulting in an objective phase lag of 1 to 2 sampling cycles. When HE4 shows abnormal signals at an early stage, the existing system, due to the normal CA125 value at the same time, forcibly identifies and clears this early recurrence signal as a false positive.

[0004] This systematic false positive caused by timing mismatch significantly reduces the sensitivity of early recurrence detection, delaying the discovery of a large number of real recurrence cases and missing the early intervention window. Therefore, the defect of the existing technology is that, under long-period discrete sampling conditions, due to the rigid assumption of synchronous verification of the execution section, it fails to identify the time delay phase difference characteristics of the different frequency expression of the primary and secondary markers, resulting in the continuous false positives of early real recurrence signals by the lagging auxiliary indicators. Summary of the Invention

[0005] The purpose of this application is to provide an optimized method and system for postoperative recurrence early warning based on HE4, which can fundamentally eliminate systemic false positives caused by the time delay of heterofrequency expression, and significantly advance the recurrence early warning time window without increasing the blood sampling frequency, so as to achieve accurate capture of early recurrence.

[0006] The objective of this application can be achieved through the following technical solution: Firstly, a postoperative recurrence early warning optimization method based on HE4, comprising the following steps:

[0007] Acquire HE4 data and CA125 data of the target object at a first discrete time point, a target time point, and a second discrete time point, wherein the first discrete time point is before the target time point and the second discrete time point is after the target time point;

[0008] Based on the HE4 data and CA125 data at the first discrete time point, the HE4 reference benchmark and CA125 reference benchmark are extracted respectively.

[0009] When the relative deviation characteristics between the HE4 data at the target time point and the HE4 reference benchmark meet the preset trigger conditions, a status suspension command is generated.

[0010] In response to the state suspension command, an observation time window containing a second discrete time point is generated based on the target time point, and the delay change rate of the CA125 data within the observation time window is extracted;

[0011] Based on the time delay change rate and the CA125 reference benchmark, the state suspension command is backtracked and verified, the verification result is output and the corresponding target warning signal is generated.

[0012] After generating the target warning signal, feedback optimization parameters characterizing the accuracy of the warning are obtained, and the preset triggering conditions are updated accordingly.

[0013] Secondly, the postoperative recurrence early warning and optimization system based on HE4 includes the following modules:

[0014] The sampling module is used to acquire HE4 data and CA125 data of the target object at a first discrete time point, a target time point, and a second discrete time point, wherein the first discrete time point is before the target time point and the second discrete time point is after the target time point.

[0015] The reference extraction module is used to extract the HE4 reference reference and the CA125 reference reference based on the HE4 data and CA125 data at the first discrete time point, respectively.

[0016] The instruction generation module is used to generate a status suspension instruction when the relative deviation characteristics between the HE4 data at the target time point and the HE4 reference benchmark meet the preset trigger conditions.

[0017] The rate of change extraction module is used to respond to the state suspension command, generate an observation time window containing a second discrete time point according to the target time point, and extract the rate of change of the delay of the CA125 data within the observation time window;

[0018] The output module is used to perform backtracking verification processing on the state suspension command based on the time delay change rate and the CA125 reference benchmark, output the verification result and generate the corresponding target warning signal;

[0019] The feedback optimization module is used to obtain feedback optimization parameters that characterize the accuracy of the warning after the target warning signal is generated, and to update the preset triggering conditions based on the parameters.

[0020] Thirdly, a computer storage medium storing computer-executable instructions, which, when executed, implement the postoperative recurrence early warning optimization method based on HE4 as described in the first aspect.

[0021] Compared with the prior art, the beneficial effects of this application are:

[0022] This invention addresses the technical shortcomings of existing synchronous verification mechanisms that lead to the false rejection of early recurrence signals by constructing a novel architecture of asymmetric sliding window and reverse timing verification. By generating a status suspension instruction when HE4 is abnormal, the veto power of the current CA125 on the initial alarm signal is blocked, fundamentally eliminating the systemic false rejection caused by inter-frequency expression delay.

[0023] The CA125 assessment method was reconstructed from absolute concentration comparison to cross-period kinetic slope analysis. Whether it exceeds the historical natural fluctuation rate serves as confirmatory evidence. Relapse can be confirmed based on abnormal accumulation trends even before the absolute CA125 concentration exceeds the standard, significantly improving diagnostic sensitivity. This invention also constructs a closed-loop optimization mechanism to ensure the system's robustness across different hospitals, instruments, and patient groups. Without increasing blood sampling frequency or detection indicators, the relapse early warning sensitivity time window can be shifted forward simply by reconstructing the time-series processing logic of existing data, demonstrating extremely high clinical practical value. Attached Figure Description

[0024] Figure 1 This is a schematic diagram illustrating the steps of the HE4-based postoperative recurrence early warning optimization method of this application;

[0025] Figure 2 This is a schematic diagram of the module of the HE4-based postoperative recurrence early warning optimization system of this application. Detailed Implementation

[0026] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components described and shown in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but only to illustrate selected embodiments of this application.

[0027] Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following figures indicate similar items. Therefore, once an item has been defined in one figure, it does not need to be further defined and explained in subsequent figures. The terms first, second, etc. are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0028] In traditional ovarian cancer postoperative follow-up systems, the combined early warning logic of HE4 and CA125 assumes causal equivalence for abnormal elevations of the two markers at the same time point, and considers concurrent exceedances of both markers as a necessary condition for triggering recurrence warnings. This logic introduces serious systemic biases when dealing with the early biological behavior of small recurrent lesions.

[0029] Taking a typical scenario as an example: After surgery, the patient enters a stable period and undergoes serum biomarker testing every 3 months. At a certain follow-up visit (target time point)... In the study, HE4 levels showed a significant jump compared to the patient's individualized baseline, exceeding the individualized trigger threshold. However, CA125 levels during the same period remained within the normal reference range because the tumor burden was still small and had not yet reached the critical threshold for triggering a large influx of CA125 into the bloodstream.

[0030] Under the existing system's cross-sectional synchronous verification logic, the initial warning signal of HE4 in this period was negated by the normal CA125 value, and the system output a judgment of no recurrence risk, thus losing a valuable opportunity for early intervention. It was not until the next follow-up 3 months or even 6 months later that CA125 finally increased, and both exceeded the standard, at which point the system triggered an early warning. By this time, the tumor volume had increased significantly, and the difficulty of treatment increased dramatically.

[0031] If the aforementioned issues are not addressed, very early true relapse signals will continue to be systematically misrepresented, causing relapse monitoring to operate in a passive mode, waiting for CA125 to catch up with HE4, thus failing to leverage HE4's early warning value as a leading indicator. This rigid cross-sectional synchronization logic essentially downgrades multi-marker combined detection to single CA125 detection, resulting in both a waste of the clinical value of HE4 testing and a delay in patients' optimal treatment time, forming a persistent industry-wide bottleneck.

[0032] To address the aforementioned issues, this invention first considers how to decouple the judgment logic of primary and secondary markers along the time axis. The core idea is to no longer require the two markers to exhibit synchronous anomalies on the same cross-section, but instead treat the initial jump of HE4 as an independent trigger signal. Simultaneously, the judgment rule for CA125 is completely reconstructed, no longer focusing on its current absolute value, but instead on its dynamic accumulation trend within subsequent observation time windows. If CA125 exhibits a positive accumulation slope exceeding the upper limit of natural fluctuations within the window, this is used as a basis for retrospective confirmation. The initial warning at time HE4 is a genuine pathological jump, thus achieving a causal logical loop between the primary and secondary markers in the temporal dimension.

[0033] Therefore, such as Figure 1 As shown, this application provides an optimization method for postoperative recurrence early warning based on HE4, including the following steps:

[0034] Acquire HE4 data and CA125 data of the target object at a first discrete time point, a target time point, and a second discrete time point, wherein the first discrete time point is before the target time point and the second discrete time point is after the target time point;

[0035] Based on the HE4 data and CA125 data at the first discrete time point, the HE4 reference benchmark and CA125 reference benchmark are extracted respectively.

[0036] When the relative deviation characteristics between the HE4 data at the target time point and the HE4 reference benchmark meet the preset trigger conditions, a status suspension command is generated.

[0037] In response to the state suspension command, an observation time window containing a second discrete time point is generated based on the target time point, and the delay change rate of the CA125 data within the observation time window is extracted;

[0038] Based on the time delay change rate and the CA125 reference benchmark, the state suspension command is backtracked and verified, the verification result is output and the corresponding target warning signal is generated.

[0039] After generating the target warning signal, feedback optimization parameters characterizing the accuracy of the warning are obtained, and the preset triggering conditions are updated accordingly.

[0040] In another implementation, such as Figure 2 As shown, this application also provides a postoperative recurrence early warning and optimization system based on HE4, including the following modules:

[0041] The sampling module is used to acquire HE4 data and CA125 data of the target object at a first discrete time point, a target time point, and a second discrete time point, wherein the first discrete time point is before the target time point and the second discrete time point is after the target time point.

[0042] The reference extraction module is used to extract the HE4 reference reference and the CA125 reference reference based on the HE4 data and CA125 data at the first discrete time point, respectively.

[0043] The instruction generation module is used to generate a status suspension instruction when the relative deviation characteristics between the HE4 data at the target time point and the HE4 reference benchmark meet the preset trigger conditions.

[0044] The rate of change extraction module is used to respond to the state suspension command, generate an observation time window containing a second discrete time point according to the target time point, and extract the rate of change of the delay of the CA125 data within the observation time window;

[0045] The output module is used to perform backtracking verification processing on the state suspension command based on the time delay change rate and the CA125 reference benchmark, output the verification result and generate the corresponding target warning signal;

[0046] The feedback optimization module is used to obtain feedback optimization parameters that characterize the accuracy of the warning after the target warning signal is generated, and to update the preset triggering conditions based on the parameters.

[0047] Specifically, the sampling module is used to acquire HE4 and CA125 data of the target object at a first discrete time point, a target time point, and a second discrete time point. The first discrete time point is before the target time point, and the second discrete time point is after the target time point. The sampling module is the data entry point for the system to interface with the LIS, which can be implemented through medical information exchange standard protocols such as HL7 or FHIR. Its function is to ensure that raw data from different biochemical testing devices enter the system processing flow in a unified format, providing the raw data foundation for all subsequent calculation steps.

[0048] The benchmark extraction module is used to extract HE4 and CA125 reference benchmarks based on the HE4 and CA125 data at the first discrete time points, respectively. This module is the core source of the system's individualized adaptability. Specifically, it is implemented by performing mean, variance statistics, and adjacent slope traversal algorithms on historical stable period data. Its function is to isolate individual patient differences from the data level, ensuring that subsequent judgment logic is based on the patient's own healthy baseline rather than a population reference range, thus completely eliminating the risk of misjudgment caused by differences in basal metabolic rate.

[0049] The instruction generation module generates a status suspension instruction when the relative deviation characteristics of the HE4 data at the target time point from the HE4 reference benchmark meet preset trigger conditions. The instruction generation module is a key control node that breaks the synchronous cross-section verification logic. Specifically, it achieves precise triggering through a dual-layer filtering mechanism (absolute transition judgment combined with relative transition rate noise filtering). Its function is to proactively reconstruct the system's processing rules for auxiliary indicators in the form of control signals, ensuring that the initial HE4 alarm signal is not rejected due to the normal CA125 reading in the current period, thus opening a legitimate channel for subsequent look-ahead verification from a timing logic perspective.

[0050] The rate of change extraction module, in response to the state suspension command, generates an observation time window containing a second discrete time point based on the target time point, and extracts the rate of change of CA125 data between the target time point and the second discrete time point within the observation time window. This rate of change extraction module is the data quantification engine of the temporal dynamics verification mechanism of this invention. Specifically, it is implemented through the delineation of an asymmetric window with the target time point as the starting point and the dynamic time span as the width, and the calculation of discrete derivatives with the difference in real timestamps as the denominator. Its function is to replace the traditional threshold comparison binary label with the continuous variable of physical accumulation rate, providing physically meaningful quantitative evidence for retrospective verification.

[0051] The output module is used to perform backtracking verification processing on the state suspension command based on the time delay change rate and the CA125 reference benchmark, output the verification result, and generate the corresponding target early warning signal. The output module is the final executor of the system decision logic. Specifically, it is implemented by comparing the time delay change rate with the CA125 reference benchmark (the upper limit of the historical natural fluctuation rate). Its function is to use causal reasoning to reversely confirm the authenticity of the initial warning of HE4 at the target time point by utilizing the subsequent dynamic performance of CA125 within the observation window, and generate an early warning signal with time priority using the backtracking timestamp when a recurrence is confirmed, which is then sent to the doctor's workstation through the HIS system interface.

[0052] The feedback optimization module is used to obtain feedback optimization parameters characterizing the accuracy of the warning after the target warning signal is generated, and to update the preset triggering conditions based on these parameters. The feedback optimization module is a self-evolving closed-loop controller for the system. Specifically, it achieves this by interfacing with PACS and EMR to obtain imaging gold standard labels, statistically analyzing the number of false positives and false negatives, calculating the sensitivity deviation coefficient, and performing penalty compensation updates. Its function is to enable the system's judgment sensitivity to be dynamically calibrated based on factors such as the hospital environment, instrument accuracy, and patient group characteristics, achieving continuous evolution from initial fixed parameters to personalized optimal parameters.

[0053] To eliminate the static influence of differences in patients' basal metabolic rates and the background noise of the testing instruments, individualized absolute reference benchmarks and normal fluctuation limits are extracted from patients' historical real data to provide a unique and legitimate reference coordinate system for subsequent judgment steps.

[0054] This invention retrieves HE4 and CA125 data from a laboratory information system at multiple consecutive first discrete time points during the postoperative stable period of the target subject. It should be noted that the first discrete time point defined in this invention refers to a historical follow-up time point prior to the target time point, typically excluding data from the stable period after the initial 3-month wound healing phase post-surgery. At least three consecutive sampling sessions are required to ensure the representativeness of the statistical results.

[0055] In extracting the HE4 reference baseline, the system performs the following calculations on the HE4 detection values ​​at multiple consecutive first discrete time points: First, the mean and variance of the patient's HE4 data are calculated respectively. The former represents the steady-state concentration of HE4 in the patient's relapse-free state, and the latter quantifies the patient's own non-specific concentration oscillation amplitude; then, combined with the system's preset confidence parameters, the HE4 reference baseline is calculated according to the individualized confidence formula.

[0056] This formula uses the mean as a central baseline and a multiple of the variance as the allowable fluctuation bandwidth to generate an individualized abnormality trigger threshold specific to the patient. In words: the HE4 reference baseline equals the HE4 mean plus the product of the confidence parameter and the HE4 variance, where the confidence parameter is typically calibrated to between 2 and 3 based on the accuracy of the biochemical equipment. The mathematical formula is as follows:

[0057] ;

[0058] in, Indicates HE4 reference datum. This represents the HE4 mean at multiple consecutive first discrete time points. This represents the variance of HE4. This represents the preset confidence level parameter.

[0059] In extracting the CA125 reference benchmark, the system's processing logic is completely different from that of the HE4 reference benchmark, reflecting the core design concept of this invention: instead of using an absolute concentration threshold for auxiliary markers, its kinetic rate characteristics are extracted. Specifically, the system iterates through all pairs of adjacent time nodes in multiple consecutive first discrete time points, and calculates the slope of the CA125 data change between each pair of adjacent nodes, i.e., the change in CA125 concentration per unit time.

[0060] Subsequently, the slope of the largest change calculated among all adjacent node pairs is determined as the CA125 reference baseline for that patient. This reference baseline physically represents the upper limit of the natural rate of fluctuation in CA125 concentration due to daily physiological fluctuations in a patient in a relapse-free, healthy state. It is the only legitimate benchmark for determining whether pathological accumulation of CA125 has occurred in subsequent retrospective confirmatory steps. The output of this step is a personalized baseline feature package specific to the corresponding patient, including the HE4 mean, HE4 variance, HE4 reference baseline, and CA125 reference baseline.

[0061] To accurately capture the very early HE4 high-frequency release signal caused by tiny recurrent lesions, and to actively block the synchronous rejection path of the CA125 on the HE4 initial alarm signal in traditional logic by generating a state suspension instruction, thus creating the preconditions for subsequent timing dynamics verification.

[0062] This invention receives a combination of in vitro serum detection data collected at the current follow-up time point, i.e., the target time point, including HE4 data and CA125 data at the target time point, and simultaneously calls the individualized baseline feature package output by the above steps, as well as the inherent error lower limit parameters of the biochemical equipment pre-stored in the system.

[0063] The core judgment logic consists of a two-layer filtering mechanism. The first layer is absolute transition judgment: the HE4 data at the target time point is compared with the HE4 reference baseline to determine whether the current HE4 detection value exceeds the upper limit of the individualized normal fluctuation bandwidth. The second layer is relative transition rate calculation and secondary noise filtering: to prevent misjudgments due to extremely small absolute numerical fluctuations in patients with low baseline concentrations, the relative deviation of the HE4 data at the target time point from the mean is further calculated, i.e., the relative transition rate. In words: the relative transition rate is equal to the target time point HE4 data minus the HE4 mean, then divided by the HE4 mean. The mathematical formula is as follows:

[0064] ;

[0065] in, Indicates the relative transition rate. This refers to the HE4 data at the target time point. This represents the mean of HE4.

[0066] A suspected abnormal jump signal in HE4 can be confirmed and a status suspension command can be generated only if both of the above conditions are met simultaneously: the HE4 data at the target time point is greater than the HE4 reference baseline, and the relative transition rate is greater than the preset inherent error lower limit parameter.

[0067] The state suspension instruction serves two key functions: first, it marks the target time point as an abnormal state to be confirmed within the system, while simultaneously caching the timestamp and CA125 data at that moment as a benchmark anchor point for subsequent calculations; second, it forcibly disables the system's absolute value comparison logic for the CA125 data at the target time point. This means that regardless of whether the absolute concentration of CA125 in the current period exceeds the normal reference range, it cannot be used as a basis for rejecting the HE4 initial alarm signal. This power deprivation mechanism is one of the key breakthroughs that distinguishes this invention from existing technologies. It cuts off the synchronous section rejection path and opens a channel for time-series dynamics verification.

[0068] To address the objective time difference in the release of biomarkers into the bloodstream, an asymmetric prospective observation interval was established on the time axis. Furthermore, the decision operator for the auxiliary indicator CA125 was completely switched from current absolute concentration comparison to cross-period kinetic slope comparison, thus establishing a new rule framework for data processing within the validation window.

[0069] This invention, upon receiving the status suspension command output in the above steps, uses the target time point where a suspected abnormal jump in HE4 occurs as the starting time boundary. Based on the medically recognized CA125 blood release hysteresis cycle, the termination time boundary is calculated by adding the starting time boundary to a preset time span, thereby forming a forward-looking continuous time interval, i.e., the observation time window.

[0070] The preset time span is not a fixed constant, but is dynamically and adaptively calculated based on the mutation intensity of HE4 data at the target time point. In the simplified implementation scheme, the preset time span can be set to a fixed medical experience value, such as 90 days, which covers 1 to 2 subsequent routine follow-up cycles.

[0071] After the observation time window is generated, the system synchronously completes the judgment rule reset: taking over the processing of the auxiliary marker CA125. Within this window, any newly entered CA125 data of the target patient will no longer be judged on whether it exceeds the upper limit of the normal reference range. Instead, it will be forcibly routed to the change trend rate calculation logic, providing a data flow basis for the extraction of the time delay change rate in subsequent steps.

[0072] In the preferred embodiment, the preset time span is not a fixed value, but a dynamic calculation result. Its physical basis stems from the heterogeneity of tumor proliferation rate: when HE4 only slightly exceeds the standard, it indicates a small recurrent tumor burden and indolent proliferation; CA125 needs a longer time to accumulate to a definitive level, thus requiring a wider observation window. Conversely, when HE4 experiences a dramatic surge, it indicates high-frequency tumor proliferation; the delayed release of CA125 will quickly follow, and in this case, there is no need for a long wait, but rather a compressed window for urgent confirmation. This causal relationship determines that the dynamic time span should be inversely coupled to the HE4 mutation intensity.

[0073] The specific calculation process is as follows: First, obtain the preset baseline time span and the minimum preset time span. The former represents the limit observation period after the second body fluid characteristic data enters the bloodstream, and the latter represents the minimum physical interval limit for the medical system to perform continuous sampling. Then, calculate the ratio of the HE4 reference baseline to the HE4 data at the target time point to obtain the attenuation coefficient representing the intensity of abnormal mutation. Next, multiply the attenuation coefficient by the baseline time span to obtain the initial calculated time span. Finally, compare the initial calculated time span with the minimum preset time span and take the maximum value of the two to determine the target dynamic time span.

[0074] In words: The dynamic time span equals the baseline time span multiplied by the ratio of the HE4 reference baseline to the HE4 data at the target time point. However, if this product is less than the minimum preset time span, the minimum preset time span takes precedence. The engineering significance of introducing a minimum time span lower bound is to prevent the mathematically calculated window length from approaching zero when HE4 data experiences extreme spikes, thus breaching the minimum physical interval constraint for actual medical follow-up. The mathematical formula is as follows:

[0075] ;

[0076] in, Indicates the dynamic time span of the target. This indicates the minimum allowed preset time span. This indicates the preset baseline time span. Indicates HE4 reference datum. This represents the HE4 data at the target time point. The prerequisite for triggering the status suspension instruction is... Greater than Therefore, the ratio It must be located between (0,1), which mathematically guarantees that the dynamic time span will never exceed the baseline time span, reflecting the inverse proportional relationship between the HE4 mutation intensity and the waiting window.

[0077] Once the observation time window is activated and enters the listening delay state, the system receives the corresponding second discrete time point data whenever the target patient undergoes their next routine follow-up test within the window. This invention defines the second discrete time point as the follow-up time node located after the target time point and falling within the observation time window. The system first performs a timeliness check: determining whether the new sampling time node is within the termination time boundary; if it exceeds the window boundary, it is determined that the process has timed out, and the current verification round is terminated; if it is within the window, the following calculations continue.

[0078] The calculation of the time delay change rate strictly uses the actual time difference as the denominator, rather than a pre-assumed fixed sampling interval. Specifically, the system acquires the latest CA125 data at the second discrete time point within the observation time window, combines it with the CA125 data at the target time point as a reference anchor, and uses the actual timestamps of the two time points. By calculating the ratio of the increase in CA125 concentration to the time increment, the physical accumulation slope of CA125 during this period is obtained. In words: the time delay change rate equals the difference between the CA125 data at the second discrete time point and the CA125 data at the target time point, divided by the actual time interval between the two time points. The mathematical formula is as follows:

[0079] ;

[0080] in, Indicates the rate of change of time delay. This represents the latest CA125 data at the second discrete time point within the observation time window. This represents CA125 data at the target time point. This represents the timestamp of the second discrete time point. The timestamp representing the target time point.

[0081] The engineering significance of using the actual time difference as the denominator lies in the fact that, in actual clinical follow-up, patients often have their follow-up examinations several days earlier or later due to factors such as travel inconvenience or sudden illness. If the denominator uses a preset standard sampling interval (such as 90 days) instead of the actual interval, it will lead to systematic biases in the rate calculation. Using the difference in actual timestamps can perfectly accommodate irregular follow-up behaviors of patients and ensure the accuracy of physical rate calculation.

[0082] To assess the actual development trend of the auxiliary indicator CA125 within the observation time window, to verify the authenticity of the HE4 initial warning signal at the very early stage of the target time point, and to generate corresponding confirmation results and target warning signals accordingly.

[0083] This invention receives the aforementioned output delay change rate, CA125 reference benchmark (i.e., the upper limit of the natural fluctuation rate of CA125 in the patient's historical state), and the timestamp of the target time point, and executes nonlinear reverse backtracking comparison logic to form two mutually exclusive scenario determinations:

[0084] Scenario A (First Confirmation Result – True Pathological Accumulation): If the system determines that the CA125 time-delay change rate is greater than the CA125 reference baseline, it indicates that even though the absolute concentration of CA125 at the second discrete time point may still be lower than the conventional alarm upper limit, it has already shown an irreversible accumulation trend in the blood exceeding the natural fluctuation limit. Based on this, the system concludes, according to objective causal laws, that this continuous pathological accumulation dynamic confirms that the abnormal increase of HE4 at the target time point is driven by true tumor recurrence, rather than occasional physiological fluctuations.

[0085] At this point, the system maintains the state suspension command, outputs the first confirmation result, generates a target early warning signal indicating a recurrence risk for the target individual, and forcibly anchors and writes back the logical effective timestamp of this early warning signal to a very early target time point, rather than the second discrete time point at the time of confirmation. This timestamp backtracking mechanism is crucial—it ensures that the earliest detection time of the recurrence signal is clearly marked in the early warning information received by the doctor, rather than the confirmation time after passive waiting. This accurately records the early warning effect at the archival level and facilitates subsequent clinical follow-up tracking and feedback optimization.

[0086] Scenario B (Second Confirmation Result – Single Physiological Noise): If the system determines that the CA125 time delay change rate is less than or equal to the CA125 reference baseline (including cases where the calculated result is negative), it indicates that CA125 does not show a pathological accumulation trend, and the increase in HE4 at the target time point is due to benign occasional physiological fluctuations, instrument detection noise, or transient interference caused by non-tumor factors (such as inflammation, hormonal level changes, etc.). At this time, the system releases the state suspension command, restores the corresponding target time point to the normal state, outputs the second confirmation result, generates a target warning signal indicating that the target subject has no risk of recurrence, and clears the alarm.

[0087] In summary, the transition at the target time point HE4 serves as the trigger switch, the dynamic accumulation slope of CA125 within the observation time window is used as the verification feature, and whether the slope exceeds the upper limit of historical natural fluctuations is used as the confirmation criterion. Finally, a time-priority warning signal is output through a backtracking anchoring method. The entire logic chain, from triggering to confirmation, is based on the principle of time-series release of physical concentrations and has complete causal validity.

[0088] To establish a closed-loop feedback optimization mechanism based on the gold standard of clinical imaging, the system's false positive and false negative performance over a period of time was statistically analyzed to calculate feedback optimization parameters. Based on this, the inherent error lower limit parameter was adaptively updated so that the system's judgment sensitivity dynamically converged to the optimal equilibrium point under different deployment environments.

[0089] This invention obtains feedback optimization parameters and, after each target warning signal is generated, interfaces with a hospital's PACS (Picture Archiving and Communication System) or electronic medical record system to obtain the target's imaging diagnosis report. This report reflects the clinical gold standard determination of whether the target has postoperative recurrence. Based on historical records within a preset testing period (e.g., the past 12 months), the system statistically analyzes the following three core quantities:

[0090] First, the total number of target warning signals generated within the preset testing period is recorded as the total number. Second, the number of times a target warning signal indicates a risk of recurrence in the target subject, but the imaging diagnosis report reflects no postoperative recurrence in the target subject, is recorded as the first error number, i.e., the number of false positives. Third, the number of times a target warning signal indicates no risk of recurrence in the target subject, but the imaging diagnosis report reflects postoperative recurrence in the target subject, is recorded as the second error number, i.e., the number of false negatives. These three numbers together constitute the feedback optimization parameters.

[0091] Then, the sensitivity deviation coefficient is calculated by dividing the difference between the number of first errors and the number of second errors by the total number of tests. This yields the sensitivity deviation coefficient, which characterizes the direction of the system's decision deviation. In words: the sensitivity deviation coefficient equals the difference between the number of first errors and the number of second errors, divided by the total number of tests. The mathematical formula is as follows:

[0092] ;

[0093] in, Indicates the sensitivity deviation coefficient. Indicates the number of the first error. Indicates the number of the second error. This represents the total quantity. The physical meaning of this formula is clear: if A value greater than zero indicates that the system is overly sensitive (proliferating false positives), and the current inherent error lower limit parameter is too low and needs to be raised; if A value less than zero indicates that the system is sluggish (proliferating false negatives), and the current inherent error lower limit parameter is too high, requiring downward compression; if A value of zero indicates that the system is at its optimal equilibrium point under the current environment and requires no adjustment. The reason for using the difference method instead of the ratio method is for engineering robustness: when the number of second errors is zero, the ratio method will lead to a division-by-zero crash, while the difference method can output a valid real number regardless of whether either side is zero, ensuring the stable operation of the system under any extreme conditions.

[0094] During the update phase of the inherent error lower limit parameter, the system performs a fusion operation between the currently effective inherent error lower limit parameter and the penalty compensation term to generate an updated error lower limit parameter. Specifically, the sensitivity deviation coefficient is multiplied by a pre-calibrated iteration step size weight constant and then added to a constant to obtain a compensation multiplier; the current inherent error lower limit parameter is multiplied by the compensation multiplier to obtain the initial calculated error lower limit parameter; subsequently, the initial calculated error lower limit parameter is compared with the preset engineering limit lower threshold and engineering limit upper threshold, respectively, and values ​​exceeding the limit range are truncated. The truncated result is determined as the updated error lower limit parameter. In words: the updated error lower limit parameter is equal to the result of multiplying the current inherent error lower limit parameter by the compensation multiplier and then truncating it on both sides, ensuring that it is not lower than the engineering limit lower threshold and not higher than the engineering limit upper threshold. The mathematical formula is as follows:

[0095] ;

[0096] in, This indicates that the lower limit of the error parameter is being updated. This represents the current inherent error lower bound parameter. This represents the preset update step size factor. Indicates the sensitivity deviation coefficient. This represents the minimum permissible inherent error lower limit parameter. This represents the lower limit parameter of the maximum permissible inherent error.

[0097] In this formula, the double-sided truncation mechanism is a crucial engineering design to prevent system crashes. Without upper and lower bound protection, in extreme cases (such as a period with an extremely high false positive rate), the formula might update the inherent error lower bound parameter to a physically meaningless or absurd value greater than 1, or even to a negative value, causing the relative transition rate filtering logic to completely fail. Upper and lower bound truncation ensures that the parameters always operate within a physically reasonable range, giving the entire feedback optimization mechanism strong engineering robustness. After the update, the system overwrites the newly calculated updated error lower bound parameter into system memory. During the next review of the target object, the aforementioned relative transition rate filtering logic will use this adaptively optimized parameter for comparison, thereby achieving closed-loop iterative optimization of the entire early warning system.

[0098] The complete workflow of this invention will be described in detail below with specific numerical examples.

[0099] Assuming a post-operative ovarian cancer patient has entered a stable phase, the system retrieves six discrete follow-up monitoring data points from the LIS (Limited Intraoperative System) over the past eight months (excluding the first three months post-surgery) as the first discrete time point dataset. The mean HE4 score for this patient is then calculated. The variance of HE4 was 68.5 pmol / L. The confidence level is 4.2 pmol / L. If we choose 2, then the HE4 reference datum is... The concentration was 76.9 pmol / L. For the extraction of the CA125 reference standard, the system traversed six adjacent sampling points, calculating the slope of the CA125 concentration change between each pair of adjacent points. The maximum slope occurred between the 4th and 5th follow-ups, when CA125 increased from 13.2 U / mL to 14.8 U / mL over a time interval of 91 days, with a slope of 0.0176 U / (mL·day). This value was determined as the CA125 reference standard. The inherent error lower limit parameter is currently set at 5%.

[0100] At the 9th follow-up (target time point) The system received the patient's test data 88 days after the last sampling. It is 83.6 pmol / L. The value was 15.1 U / mL. Execution criteria: 83.6 is greater than 76.9, satisfying the absolute transition condition; relative transition rate... The value is 22.0%, which is much greater than the inherent error lower limit parameter of 5%, satisfying the relative transition condition. With both conditions met, the system generates a state suspension command, and... Marked as an abnormal pending verification state, and cached simultaneously. With timestamp This serves as an anchor point, and the absolute value of CA125 in the current period (15.1 U / mL, still within the normal range) is used to veto the signal. If, in the existing system, 15.1 U / mL does not exceed the normal reference upper limit of CA125 (usually 35 U / mL), then the initial alarm signal of HE4 will be vetoed.

[0101] In the calculation of dynamic time span, the baseline time span Set to 100 days, minimum preset time span Set to 30 days. The initial dynamic time span is estimated to be approximately 92 days. Since 92 days is greater than 30 days, the final dynamic time span ΔT is determined to be 92 days, i.e., the observation time window is [ The system has entered a listening delay state.

[0102] exist Approximately 89 days later (the second discrete time point) (Within the observation window), the patient completed their 10th follow-up visit. The concentration was 18.7 U / mL. Timeliness verification was performed. The observation period is 89 days, which falls within the observation time window. The rate of change of time delay is then calculated. It is approximately equal to 0.0404 U / (mL·day), which is greater than the CA125 reference standard of 0.0176 U / (mL·day), thus satisfying the conditions of scenario A.

[0103] The system determined that: Although in While the absolute CA125 level of 18.7 U / mL was still below the usual alarm limit of 35 U / mL, its accumulation rate significantly exceeded the patient's historical natural fluctuation limit, confirming... The abnormal increase in HE4 at time point is driven by actual tumor recurrence. The system outputs the first confirmatory result, generates a target early warning signal indicating a risk of recurrence in the target subject, and writes back the logical effective timestamp of the early warning signal. The warning signal was then sent to the doctor's workstation via the HIS interface, prompting the doctor to arrange further imaging examinations for the patient.

[0104] In contrast, in existing simultaneous cross-sectional verification systems, this patient would only trigger an alert when the absolute CA125 level exceeded 35 U / mL. Based on its accumulation rate of 0.0404 U / (mL·day), it would take approximately (35−18.7) / 0.0404, roughly 403 days (about 13 months), for the level to rise from 18.7 U / mL to 35 U / mL. By then, the tumor burden would be extremely heavy. This invention successfully advances the alert time by approximately 4 to 5 months, fully demonstrating the significant superiority of the temporal dynamics verification mechanism over simultaneous cross-sectional verification.

[0105] After 12 months of system operation, a total of 48 tests were performed, with 7 first-order errors and 3 second-order errors. The sensitivity deviation coefficient at this point is approximately 0.083. This indicates that the system is slightly oversensitive, and the inherent error lower limit parameter needs to be slightly increased. The iteration step size weight is then used. It is 0.5, currently. It is 5%, if 3%, If it is 15%, then The positive rate was 5.21%. By overriding the system parameters, the first-level threshold was raised in subsequent follow-ups to reduce false positives.

[0106] Through the above complete embodiments, the present invention achieves a leap from cross-sectional synchronous false positives to temporal dynamic confirmation. Without increasing the blood sampling frequency or introducing new detection indicators, it significantly advances the sensitivity time window of recurrence early warning, successfully capturing very early true recurrence signals that were originally masked by synchronous rules. This provides ovarian cancer patients with earlier and more accurate intervention opportunities. Specifically, the present invention can achieve the following technical effects:

[0107] By generating a status suspension command when an abnormal HE4 transition is detected at the target time point, the veto power of the current CA125 absolute value on the HE4 initial alarm signal is actively blocked, fundamentally eliminating the systemic false alarm problem caused by the time delay of the marker's different frequency expression, so that the very early real recurrence signal is no longer erased by the synchronization rule.

[0108] By extracting the rate of change of CA125 within the observation time window, the judgment operator for auxiliary indicators is reconstructed from the traditional absolute concentration comparison to a cross-cycle dynamic slope comparison. Whether this slope exceeds the CA125 reference benchmark is used as the sole valid confirmatory criterion, thus achieving a complete closed loop of causal logic. This mechanism can confirm recurrence solely based on its abnormal accumulation trend, even when the absolute concentration of CA125 has not exceeded the conventional alarm limit, significantly improving diagnostic sensitivity.

[0109] A dynamic time span calculation mechanism is introduced, inversely coupling the length of the observation time window with the initial mutation intensity of HE4. When HE4 experiences a sharp surge, the system automatically compresses the window length to achieve emergency confirmation; when HE4 slightly exceeds the limit, the system automatically extends the window to fully wait for the accumulation response of CA125.

[0110] A closed-loop adaptive optimization mechanism based on the gold standard feedback in imaging was constructed. By statistically analyzing the number of false positives and false negatives within a preset testing period, the sensitivity deviation coefficient is calculated, and a penalty compensation calculation is performed on the inherent error lower limit parameter accordingly, achieving dynamic self-correction of the threshold. This mechanism enables the system to automatically adapt to different hospital environments, different instrument accuracies, and different patient groups, exhibiting strong engineering robustness.

[0111] Without increasing the frequency of blood collection for patients or introducing new and expensive detection indicators, the sensitivity time window for recurrence warning can be shifted forward by 1 to 2 follow-up cycles simply by reconstructing the time-series processing logic of existing HE4 and CA125 data. This has extremely high clinical practical value and potential for promotion.

[0112] In another embodiment, this application also provides a computer storage medium storing computer-executable instructions that, when executed, implement the HE4-based postoperative recurrence early warning optimization method.

[0113] The above embodiments are only used to illustrate the technical methods of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of this application without departing from the spirit and scope of the technical methods of this application.

Claims

1. A postoperative recurrence early warning optimization method based on HE4, characterized in that, Includes the following steps: Acquire HE4 data and CA125 data of the target object at a first discrete time point, a target time point, and a second discrete time point, wherein the first discrete time point is before the target time point and the second discrete time point is after the target time point; Based on the HE4 data and CA125 data at the first discrete time point, the HE4 reference benchmark and CA125 reference benchmark are extracted respectively. When the relative deviation characteristics between the HE4 data at the target time point and the HE4 reference benchmark meet the preset trigger conditions, a status suspension command is generated. In response to the state suspension command, an observation time window containing a second discrete time point is generated based on the target time point, and the delay change rate of the CA125 data within the observation time window is extracted; Based on the time delay change rate and the CA125 reference benchmark, the state suspension command is backtracked and verified, the verification result is output and the corresponding target warning signal is generated. After generating the target warning signal, feedback optimization parameters characterizing the accuracy of the warning are obtained, and the preset triggering conditions are updated accordingly.

2. The postoperative recurrence early warning optimization method based on HE4 as described in claim 1, characterized in that, The process of extracting the HE4 and CA125 reference bases includes: Obtain the mean of HE4 data at multiple consecutive first discrete time points. and variance Combined with preset confidence parameters Obtain the corresponding HE4 reference benchmark ; Extract the slope of change between any two adjacent first discrete time points in the continuous plurality of first discrete time points, and take the slope of change with the largest value as the corresponding CA125 reference benchmark.

3. The postoperative recurrence early warning optimization method based on HE4 as described in claim 2, characterized in that, The process of generating a status suspension instruction includes: Based on HE4 data at the target time point and the mean of the HE4 data at multiple consecutive first discrete time points To obtain the relative transition rate at the target time point ; When the relative transition rate is greater than the preset inherent error lower limit parameter, and the HE4 data at the target time point is greater than the HE4 reference benchmark, it is determined that the preset triggering condition is met, and a status suspension instruction is generated to mark the target time point as an abnormal state.

4. The postoperative recurrence early warning optimization method based on HE4 as described in claim 3, characterized in that, The process of extracting the rate of change of time delay includes: The target time point is used as the starting time boundary. The starting time boundary is numerically added to the preset time span to obtain the ending time boundary. The continuous time interval formed by the starting time boundary and the ending time boundary is used as the observation time window. The observation time window contains at least one second discrete time point, and the latest second discrete time point within the observation time window is obtained. The mean of CA125 data Combined with the target time point CA125 data Obtain the rate of change of time delay within the corresponding observation time window. .

5. The postoperative recurrence early warning optimization method based on HE4 as described in claim 4, characterized in that, The preset time span The HE4 reference base and target time point HE4 data, along with a preset reference time span, are used. and the minimum allowed preset time span Calculated; 。 6. The postoperative recurrence early warning optimization method based on HE4 as described in claim 1, characterized in that, The process of generating a target early warning signal includes: If the time delay change rate is greater than the CA125 reference benchmark, the state suspension instruction is maintained, and a first confirming result characterizing the abnormal cause of the target time point as real pathological accumulation is output. When the first confirming result is output, a target warning signal indicating that the target object has a risk of recurrence is generated. If the time delay change rate is less than or equal to the CA125 reference benchmark, the state suspension command is released, the corresponding target time point is restored to the normal state, and a second confirmation result characterizing the abnormal cause of the target time point as a single physiological noise is output. When the second confirmation result is output, a target warning signal indicating that the target object does not have a risk of recurrence is generated.

7. The postoperative recurrence early warning optimization method based on HE4 as described in claim 3, characterized in that, The process of obtaining feedback optimization parameters includes: After each target warning signal is generated, an imaging diagnostic report of the target object is obtained. This imaging diagnostic report reflects whether the target object has postoperative recurrence, and the total number of target warning signals generated within the preset testing cycle is obtained. ; The number of times within the same preset testing cycle, where the target warning signal indicates a risk of recurrence in the target subject, but the imaging diagnosis report reflects no postoperative recurrence in the target subject, is used as the first error count. ; The number of times the target warning signal indicates that the target object has no risk of recurrence within the same preset testing cycle, but the imaging diagnosis report reflects that the target object has postoperative recurrence, is recorded as the second error number. The total number, the number of first errors, and the number of second errors are all part of the feedback optimization parameters.

8. The postoperative recurrence early warning optimization method based on HE4 as described in claim 7, characterized in that, The process of updating preset trigger conditions includes: Obtain the sensitivity deviation coefficient within the preset testing period. When the sensitivity deviation coefficient is not zero, the current inherent error lower limit parameter in the preset triggering condition is applied. Perform an update to obtain the updated intrinsic error lower bound parameter. ,in, This represents the preset update step size factor. and These are the minimum and maximum allowable lower bound parameters of inherent error, respectively.

9. A postoperative recurrence early warning and optimization system based on HE4, characterized in that, Includes the following modules: The sampling module is used to acquire HE4 data and CA125 data of the target object at a first discrete time point, a target time point, and a second discrete time point, wherein the first discrete time point is before the target time point and the second discrete time point is after the target time point. The reference extraction module is used to extract the HE4 reference reference and the CA125 reference reference based on the HE4 data and CA125 data at the first discrete time point, respectively. The instruction generation module is used to generate a status suspension instruction when the relative deviation characteristics between the HE4 data at the target time point and the HE4 reference benchmark meet the preset trigger conditions. The rate of change extraction module is used to respond to the state suspension command, generate an observation time window containing a second discrete time point according to the target time point, and extract the rate of change of the delay of the CA125 data within the observation time window; The output module is used to perform backtracking verification processing on the state suspension command based on the time delay change rate and the CA125 reference benchmark, output the verification result and generate the corresponding target warning signal; The feedback optimization module is used to obtain feedback optimization parameters that characterize the accuracy of the warning after the target warning signal is generated, and to update the preset triggering conditions based on the parameters.

10. A computer storage medium storing computer-executable instructions, characterized in that, When the computer-executable instructions are executed, they implement the postoperative recurrence early warning optimization method based on HE4 as described in any one of claims 1-8.

Images