A method for developing an HPV long-term treatment strategy for prevention and treatment of cervical cancer

Abstract

The application discloses a kind of long-term treatment strategy making methods for cervical cancer prevention and treatment of HPV, belong to medical information processing technical field.The method includes: obtaining the digital HPV health record containing multiple HPV typing detection results, cervical cytology history record and individual physiological characteristic parameters;Three-dimensional treatment situation cube is constructed by virus infection state subspace, cervical tissue state subspace and treatment intervention state subspace;In the cube, state backtracking and forward-looking deduction algorithm based on time axis are used, virus clearance probability field and lesion outcome probability field are calculated;High-risk persistent infection space-time region and low-risk self-healing tendency space-time region are demarcated;Current patient state is projected to the cube, if fall into high-risk region, then trigger long-term treatment strategy generation process including multiple treatment stages, output intervention opportunity, intensity and state transition condition of each stage.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, specifically to a method for developing a long-term HPV treatment strategy for the prevention and treatment of cervical cancer. Background Technology

[0002] HPV infection management in the field of cervical cancer prevention and treatment relies on regular screening and timely clinical intervention. Currently, treatment strategies for HPV-positive patients are primarily based on single test results and static clinical guidelines. Physicians provide empirical follow-up or treatment recommendations based on the current HPV type, semi-quantitative viral load, and cervical cytology grade, referring to treatment guidelines. This decision-making model simplifies the continuously evolving pathophysiological process over time into isolated sections, failing to reflect the dynamic characteristics of the long-term interaction between the virus and the host. Existing technologies have significant shortcomings. Firstly, the state representation dimension is too singular, failing to integrate the multi-type mixed infection state of the virus, the progressive development trajectory of cervical tissue lesions, and the adjustment of treatment interventions over time into a unified analytical space. This results in fragmented clinical information from different sources during decision-making. Secondly, there is a lack of quantitative projection capabilities for future disease outcomes. Current treatment pathways cannot calculate the probability distribution of spontaneous clearance or continued progression of specific infection and lesion combinations in the future spatiotemporal context based on the patient's individual historical state sequence. This makes it difficult for clinicians to identify patients who appear low-risk but actually possess potential persistent infection characteristics, and also makes it difficult to distinguish individuals with a truly high probability of self-healing, resulting in a lack of spatiotemporal precision in the allocation of treatment resources. To address these shortcomings, the technical challenge is how to integrate discrete, multidimensional follow-up data into a continuous form that can characterize the synergistic evolutionary relationship among infection, tissue, and intervention, and within this form, achieve quantitative calculation of the probability distribution of disease outcomes and precise definition of high-risk spatiotemporal boundaries. Summary of the Invention

[0003] This invention provides a method for formulating a long-term HPV treatment strategy for cervical cancer prevention and treatment. The aim is to integrate multi-dimensional HPV genotyping, cervical cytology, and individual physiological characteristic data into a unified treatment status analysis framework. Through joint modeling and spatiotemporal extrapolation of the three states of infection, tissue, and intervention, it realizes the quantitative calculation of the probability field of disease progression and the dynamic division of risk areas, thereby automatically generating a long-term treatment strategy that includes the timing, intensity, and metastasis conditions of staged interventions.

[0004] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a method for formulating a long-term HPV treatment strategy for cervical cancer prevention and treatment, comprising the following steps: obtaining a digital HPV health record of the target patient, wherein the digital HPV health record contains multiple HPV genotyping test results, cervical cytology history records, and individualized physiological characteristic parameters; based on the digital HPV health record, constructing a three-dimensional treatment situation cube spanned by a virus infection state subspace, a cervical tissue state subspace, and a treatment intervention state subspace, which can unify the patient's multidimensional disease course information into a single spatiotemporal analysis framework; in the three-dimensional treatment situation cube, using a time-axis-based state retrospective and prospective extrapolation algorithm, calculating the virus clearance probability field and A lesion outcome probability field is used to achieve dual-channel probabilistic prediction of the virus clearance prospect and the direction of cervical lesion development. Based on the virus clearance probability field and the lesion outcome probability field, at least one high-risk persistent infection spatiotemporal region and at least one low-risk self-healing tendency spatiotemporal region are delineated to achieve spatiotemporal partitioning of risk levels. The current patient status is projected onto the three-dimensional treatment situation cube to determine whether it falls into the high-risk persistent infection spatiotemporal region or its neighboring region. When it falls into the region, a long-term treatment strategy generation process containing multiple treatment stages is triggered, so that the treatment decision is directly linked to the real-time risk space. The long-term treatment strategy includes the intervention timing, intervention intensity, and state transition conditions between each treatment stage, so that the treatment has temporality and adaptability.

[0005] By constructing a three-dimensional treatment situation cube and using a probability field for state deduction, this invention can dynamically couple HPV infection, cervical lesion evolution, and treatment intervention, accurately identify spatiotemporal regions with a high risk of persistent infection, and automatically trigger long-term treatment strategies including virus clearance induction, continuous infection control, and consolidation follow-up monitoring. This achieves a technological leap from passive follow-up to proactive planning, and from single-point-of-time assessment to dynamic intervention throughout the entire disease course.

[0006] As a preferred technical solution of the present invention, the acquisition of the target patient's digital HPV health record specifically includes: extracting the target patient's historical medical records from a medical information system, and separating the timestamp field, HPV genotype field, semi-quantitative viral load field, TBS grading field of cervical liquid-based thin-layer cytology, and patient age field; dividing the HPV genotype field into high-risk, suspected high-risk, and low-risk groups according to the carcinogenic risk level, and assigning a type risk weight coefficient to each group; mapping the TBS grading field to multiple discrete state levels from normal to squamous cell carcinoma according to the severity of the lesion, and assigning a state level score to each discrete state level; dividing the patient age field into multiple age risk intervals according to the age distribution curve of cervical cancer incidence, with each age risk interval corresponding to an age risk correction factor; and aligning the timestamp field, the type risk weight coefficient, the semi-quantitative viral load field, the state level score, and the age risk correction factor along the time axis to form the digital HPV health record. This approach standardizes multi-source heterogeneous clinical data into structured profiles with risk weights, laying the data foundation for subsequent accurate modeling.

[0007] Furthermore, constructing a three-dimensional treatment status cube composed of a virus infection state subspace, a cervical tissue state subspace, and a treatment intervention state subspace specifically includes: establishing a three-dimensional coordinate system, where the first dimension corresponds to the virus infection state subspace, the second dimension corresponds to the cervical tissue state subspace, and the third dimension corresponds to the treatment intervention state subspace; in the virus infection state subspace, setting a virus type axis and a viral load axis, where each coordinate point on the virus type axis represents an HPV type or type combination, and each coordinate point on the viral load axis represents a viral load level range; in the cervical tissue state subspace, setting cytological scores... The system comprises a grade axis and a lesion extent axis. Each coordinate point on the cytological grading axis represents a TBS grade, and each coordinate point on the lesion extent axis represents the number or percentage range of cervical epithelial involvement. In the treatment intervention state subspace, an intervention method axis and an intervention intensity axis are defined. Each coordinate point on the intervention method axis represents a treatment or intervention method, and each coordinate point on the intervention intensity axis represents the frequency or dosage level of that intervention method. Each discrete grid point in the three-dimensional coordinate system is defined as a basic unit of treatment status, corresponding to a specific combination of viral infection status, cervical tissue status, and treatment intervention status. This cubic structure finely covers all possible clinical status combinations a patient may be in, providing a complete spatial carrier for precise calculation of the probability field.

[0008] As a preferred embodiment of the present invention, the step of employing a time-axis-based state backtracking and prospective extrapolation algorithm to calculate the virus clearance probability field and the lesion outcome probability field specifically includes: extracting the actual state sequence of the target patient at multiple historical time points from the digital HPV health record; projecting each state in the actual state sequence onto the three-dimensional treatment situation cube to obtain a historical state trajectory; statistically analyzing the residence time distribution and subsequent state transition distribution of each basic treatment situation unit under the condition of no additional intervention along the time direction of the historical state trajectory to form a state transition probability matrix; based on the state transition probability matrix, extrapolating multiple preset time steps forward from the basic treatment situation unit where the current patient state is located, and calculating the probability value of reaching each basic treatment situation unit at each future time step; organizing all the probability values ​​obtained at all future time steps according to their coordinate positions in the three-dimensional treatment situation cube to obtain the virus clearance probability field and the lesion outcome probability field, wherein the virus clearance probability field corresponds to the probability distribution of viral load dropping below the detection threshold in the virus infection state subspace, and the lesion outcome probability field corresponds to the probability distribution of cytological grade dropping to normal or low-grade lesion in the cervical tissue state subspace. This process uses real historical trajectories to drive state transition learning, and the generated dynamic probability field can be continuously updated as the disease progresses, so that the prediction results always fit the patient's actual condition.

[0009] Preferably, the state transition probability matrix adopts an online learning method based on Bayesian updates. Each time a new HPV genotyping result or cervical cytology record is obtained, the corresponding transition probability is posteriorly corrected, thereby enabling the probability field to have adaptive evolution capabilities and continuously improving prediction accuracy as the amount of data increases.

[0010] In delineating risk areas, a preferred approach is as follows: In the virus clearance probability field, extract all basic units of treatment status with virus clearance probabilities below a first preset threshold, and mark the spatiotemporal continuous regions occupied by these basic units of treatment status as the high-risk persistent infection spatiotemporal regions; In the lesion outcome probability field, extract all basic units of treatment status with lesion outcome probabilities above a second preset threshold, and mark the spatiotemporal continuous regions occupied by these basic units of treatment status as the low-risk self-healing tendency spatiotemporal regions; Obtain the set of boundary coordinates of the high-risk persistent infection spatiotemporal regions in the three-dimensional treatment status cube, and the set of boundary coordinates of the low-risk self-healing tendency spatiotemporal regions in the three-dimensional treatment status cube; Perform spatial overlay analysis on the set of boundary coordinates of the high-risk persistent infection spatiotemporal regions and the set of boundary coordinates of the low-risk self-healing tendency spatiotemporal regions to obtain the overlapping and mutually exclusive parts of the two regions, so as to take into account the overlapping intervals of different outcome directions when formulating treatment strategies and formulate differentiated intervention plans.

[0011] More preferably, the first preset threshold and the second preset threshold are adaptively adjusted based on the immune status index and age risk correction factor in the individualized physiological characteristic parameters of the target patient, thereby making the division of risk areas highly individualized and avoiding misjudgment caused by using fixed thresholds.

[0012] Another preferred implementation method for determining whether a patient's status falls into a high-risk area includes the following steps: obtaining the target patient's current HPV genotyping test results, current cervical cytology test results, and current treatment intervention status; mapping these three pieces of information to the three dimensions of the three-dimensional treatment status cube to obtain the current status coordinate point; sequentially comparing the coordinate values ​​of the current status coordinate point in each dimension with the coordinate interval of the high-risk persistent infection spatiotemporal region in that dimension; if the coordinate values ​​in all three dimensions fall within the corresponding coordinate intervals, the current status coordinate point is determined to be located within the high-risk persistent infection spatiotemporal region; if the coordinate value in any dimension does not fall within the corresponding coordinate interval, the minimum Euclidean distance from the current status coordinate point to the boundary of the high-risk persistent infection spatiotemporal region is further calculated, and this minimum Euclidean distance is compared with a distance threshold; if the minimum Euclidean distance is less than the distance threshold, the current status coordinate point is determined to be located in the vicinity of the high-risk persistent infection spatiotemporal region. This scheme not only identifies patients who have entered the high-risk area but also provides early warnings for patients who are about to enter the high-risk area, achieving proactive intervention.

[0013] In a preferred embodiment, the process of generating a long-term treatment strategy includes: using the projection range of the high-risk persistent infection spatiotemporal region onto the treatment intervention state subspace as a reference range for treatment intensity, and using the projection range of the low-risk self-healing tendency spatiotemporal region onto the treatment intervention state subspace as a lower limit reference value for treatment intensity, thereby defining upper and lower boundaries with clear risk basis for the intervention intensity; retrieving candidate strategy templates that match the spatial morphological characteristics of the high-risk persistent infection spatiotemporal region from a preset treatment strategy knowledge base, wherein the candidate strategy templates include strategy identifiers, phase division rules, phase triggering conditions, and phase exit conditions; using the current patient state coordinate point as the starting point, and any coordinate point in the low-risk self-healing tendency spatiotemporal region as the target endpoint set, using a shortest path search algorithm to search for multiple candidate treatment paths from the starting point to the target endpoint set in the three-dimensional treatment state cube; sorting the multiple candidate treatment paths in ascending order of path length, and selecting a preset number of candidate treatment paths at the top of the sorting as candidate schemes for the long-term treatment strategy. This process uses the spatial geometric relationship of the risk area as a constraint to automatically plan the optimal intervention sequence for migrating from the current high-risk state to a low-risk self-healing state, taking into account both treatment effectiveness and medical resource consumption.

[0014] In the above scheme, the preset treatment strategy knowledge base preferably includes multiple historical successful treatment cases of patients and their corresponding three-dimensional treatment status cube trajectories. By calculating the dynamic time warping distance between the current patient status trajectory and the trajectory of the successful treatment cases, matching candidate strategy templates are retrieved, thereby improving the clinical reproducibility of strategy recommendations.

[0015] Furthermore, the phase division rules, phase triggering conditions, and phase exit conditions in the candidate strategy template specifically include: the phase division rules divide the long-term treatment strategy into a virus clearance induction phase, a continuous infection control phase, and a consolidation follow-up monitoring phase according to time sequence, with each phase associated with a phase target state range and a maximum allowed stay time; the phase triggering conditions include exit action triggering conditions of the previous phase and time triggering conditions independent of the previous phase, and the current phase begins execution when either of them is met, ensuring the flexibility and safety of phase switching; the phase exit conditions include target state achievement exit conditions and time exhaustion forced exit conditions, and the current phase exits normally and enters the next phase when the state range in the target state achievement exit condition covers the current patient state coordinate point, and the current phase reaches the maximum allowed stay time but the target state achievement exit condition is still not met, the current phase forcibly exits and triggers an exception handling branch, thereby ensuring treatment effectiveness while avoiding indefinite delay.

[0016] In terms of path search, another preferred approach is as follows: Each basic unit of the treatment state in the three-dimensional treatment state cube is defined as a node in a graph structure. The state transition relationship between two adjacent basic units of the treatment state is defined as a directed edge in the graph structure. Each directed edge is associated with a transition probability from the source node to the target node and a transition time cost. Starting from the node where the starting point is located, a heuristic search algorithm based on Dijkstra's principle is used to search for the shortest path to any node in the target endpoint set, with the goal of minimizing the cumulative transition time cost. The sequence of nodes traversed by this shortest path and the corresponding sequence of directed edges are recorded. After finding the first shortest path, a penalty factor is temporarily added to the transition cost of each directed edge on the first shortest path, and the heuristic search algorithm is executed again to obtain the second shortest path. This process is repeated until a preset number of candidate treatment paths are obtained. The sequence of nodes traversed by each candidate treatment path is mapped back to the treatment intervention state subspace to obtain the sequence of treatment intervention intensity to be executed at each time node, which serves as the treatment strategy scheme corresponding to the candidate treatment path. This search mechanism can generate multiple near-optimal paths that differ from each other, providing clinicians with flexible treatment options.

[0017] As a further improvement of the present invention, after selecting a preset number of candidate treatment paths with high ranking as candidate solutions for the long-term treatment strategy, the following steps are also included: obtaining the target patient's compliance score for historical treatment interventions, using the compliance score as a correction factor to normalize and adjust the treatment intervention intensity sequence in each candidate treatment path to obtain an individual-adapted treatment intervention intensity sequence; binding the individual-adapted treatment intervention intensity sequence with time nodes in the candidate treatment paths to generate a treatment task list with timestamps, where each entry in the treatment task list includes an execution time window, intervention method code, and intervention intensity parameter; outputting the treatment task list to the user interface of the clinical decision support system, and displaying the temporal relationship of each treatment stage in the form of a Gantt chart in the user interface; receiving confirmation or modification instructions from clinicians, locking the final treatment strategy according to the confirmation or modification instructions, and storing the final treatment strategy in the target patient's digital HPV health record as historical reference data for the next strategy formulation. This step incorporates adherence into the strategy adjustment process and achieves a closed-loop integration of physician experience and algorithm recommendations through visual interaction, significantly improving the clinical feasibility and continuous optimization capabilities of long-term treatment strategies.

[0018] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0019] By constructing a three-dimensional treatment situation cube composed of a virus infection state subspace, a cervical tissue state subspace, and a treatment intervention state subspace, HPV type combinations, viral load levels, cytological grades, lesion extent, intervention methods, and intensities are mapped to discrete grid points in a unified coordinate system, forming standardized basic units of treatment situations. This structure compresses heterogeneous data originally scattered across multiple medical records into computable spatiotemporal coordinates. The interaction process between the virus and host tissue, as well as the intervention effect, can be continuously expressed within the same framework, no longer relying on static comparisons of single test values. Under this cube structure, the co-evolutionary relationship between different HPV type combinations and different cervical lesion grades under specific intervention conditions is explicitly represented, eliminating the risk of misjudgment caused by the fragmentation of information dimensions in traditional methods. Using a time-axis-based state backtracking and look-ahead extrapolation algorithm, the state transition probability matrix of each basic unit of treatment situation is extracted from the historical state trajectory, and the probability values ​​of reaching each unit in the future are extrapolated step by step from the previous unit, generating a virus clearance probability field and a lesion outcome probability field. This mechanism transforms disease course prediction from qualitative, empirical judgment to quantitative probability calculation based on an individual's historical state sequence, enabling the differentiation of risk areas that appear similar but exhibit drastically different underlying evolutionary trends. In delineating high-risk persistent infection spatiotemporal regions and low-risk spontaneous recovery tendency spatiotemporal regions, the spatial continuity of the probability field ensures that the region boundaries are not fixed cutoff values, but rather formed by the natural drop in probability density in three-dimensional space. This incorporates the Euclidean distance between the current patient's coordinates and the region boundary into the judgment logic, not only identifying states already within high-risk areas but also capturing potential risks in areas adjacent to the boundary, achieving proactive identification of dangerous states. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0021] Figure 1 This is a flowchart illustrating the methodology for developing long-term HPV treatment strategies for cervical cancer prevention and control.

[0022] Figure 2 This is a flowchart of the digital HPV health record generation process;

[0023] Figure 3 This is a schematic diagram of the process of constructing a three-dimensional treatment situation cube and extrapolating the probability field;

[0024] Figure 4It is a flowchart for the spatiotemporal delineation and status assessment of high-risk persistent infection and low-risk self-healing tendency based on individualized probability fields;

[0025] Figure 5 This is a flowchart for generating long-term treatment strategies;

[0026] Figure 6 It is a distribution map of the cosine similarity of spatial morphological feature vectors of the candidate treatment strategy knowledge base. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] See Figure 1 This invention provides a method for developing a long-term HPV treatment strategy for cervical cancer prevention and treatment. The overall implementation scheme includes: acquiring a digital HPV health record for the target patient, containing multiple HPV genotyping test results, cervical cytology history, and individualized physiological characteristic parameters. Based on the digital HPV health record, a three-dimensional treatment situation cube is constructed, consisting of a virus infection state subspace, a cervical tissue state subspace, and a treatment intervention state subspace. Within the three-dimensional treatment situation cube, a time-axis-based state backtracking and prospective extrapolation algorithm is used to calculate the virus clearance probability field and the lesion outcome probability field. Based on the virus clearance probability field and the lesion outcome probability field, at least one high-risk persistent infection spatiotemporal region and at least one low-risk self-healing tendency spatiotemporal region are delineated in the three-dimensional space. The current patient state is projected onto the three-dimensional treatment situation cube, and it is determined whether the current state falls into a high-risk persistent infection spatiotemporal region. If it does, a long-term treatment strategy generation process containing multiple treatment stages is triggered. This long-term treatment strategy includes the intervention timing, intervention intensity, and state transition conditions between each treatment stage.

[0029] Example 1:

[0030] In specific implementation, please refer to Figure 2The process of obtaining a digital HPV health record for a target patient begins by extracting the patient's historical medical records from the medical information system. These records contain structured test data and unstructured text reports generated from multiple visits. By parsing CDA documents or HL7 messages exported from the hospital information system, each medical record related to cervical cancer screening or HPV testing is extracted. From these records, the timestamp field, HPV genotype field, semi-quantitative viral load field, TBS grading field from cervical liquid-based thin-layer cytology, and patient age field are extracted.

[0031] For the timestamp field, the test date or sample collection date from each test report is extracted and converted into a uniform time format. For the HPV genotype field, the list of positive genotypes corresponding to the "HPV DNA Typing" item in the test results is identified. If a report contains multiple types, all types are recorded in the HPV genotype field, separated by delimiters. For the semi-quantitative viral load field, the viral load test value is extracted, and the raw quantitative result is mapped to a pre-defined level identifier within several viral load level ranges according to the testing method. For example, a chemiluminescence immunoassay value below 1.0 RLU / CO is mapped to a low viral load level; a value between 1.0 RLU / CO and 10.0 RLU / CO is mapped to a medium viral load level; and a value above 10.0 RLU / CO is mapped to a high viral load level. For the TBS grading field of cervical liquid-based thin-layer cytology, the diagnostic conclusion is extracted from the cytology report, the terminology descriptions are identified, and they are standardized into a TBS grading string. For the patient age field, the date of birth is extracted from the patient's basic information, and the actual age is calculated by combining it with the examination date corresponding to the timestamp field.

[0032] After obtaining the above fields, the HPV genotype field is divided into high-risk, suspected high-risk, and low-risk groups according to the carcinogenic risk level, and a type risk weight coefficient is assigned to each group. Specifically, the high-risk group includes HPV16, HPV18, HPV31, HPV33, HPV35, HPV39, HPV45, HPV51, HPV52, HPV56, HPV58, HPV59, and HPV68; the suspected high-risk group includes HPV26, HPV53, HPV66, HPV73, and HPV82; and the low-risk group includes other detected types not listed in the high-risk and suspected high-risk groups. If a test result contains types from multiple groups, the highest type risk weight coefficient among all groups is taken as the type risk weight coefficient for that test result. The type risk weighting coefficients were determined through retrospective cohort studies. Based on the average sustained infection risk in the population, the type risk weighting coefficients for high-risk types were set at 1.0, for suspected high-risk types at 0.5, and for low-risk types at 0.1.

[0033] In processing the TBS grading field of cervical liquid-based thin-layer cytology, the TBS grading field is mapped to multiple discrete status levels from normal to squamous cell carcinoma according to the severity of the lesion, and a status level score is assigned to each discrete status level. The discrete status levels, from mild to severe, are: no intraepithelial lesion or malignant cells, atypical squamous cells of indeterminate significance, low-grade squamous intraepithelial lesion, atypical squamous cells where high-grade lesion cannot be ruled out, high-grade squamous intraepithelial lesion, and squamous cell carcinoma. The corresponding status level scores are assigned as 0, 1, 2, 3, 4, and 5, respectively. If both biopsy pathology and cytology results are available in a single visit, the biopsy pathology result is used as the final grading basis; if only cytology results are available, the cytology grading result is used.

[0034] In processing the patient age field, it was divided into multiple age risk intervals according to the cervical cancer incidence age distribution curve, with each interval corresponding to an age risk correction factor. The cervical cancer incidence age distribution curve was plotted based on the age-specific incidence rates of cervical cancer in various countries or regions from epidemiological survey data, with the age group with the highest incidence rate taken as the baseline interval. The age risk intervals were divided as follows: under 22 years old, 22 to 29 years old, 30 to 35 years old, 36 to 45 years old, 46 to 55 years old, and over 55 years old. The 22 to 29 year old interval corresponds to the rapid increase in cervical cancer incidence, the 30 to 35 year old interval corresponds to the plateau period, the 36 to 45 year old interval corresponds to the peak incidence period, the 46 to 55 year old interval corresponds to the slow decline period, and under 22 years old and over 55 years old correspond to the low points of incidence. An age-based risk correction factor was assigned to each age range: 0.7 for those under 22, 0.9 for those aged 22-29, 1.0 for those aged 30-35, 1.3 for those aged 36-45, 1.1 for those aged 46-55, and 0.8 for those over 55. The values ​​of the age-based risk correction factors were determined through multivariate regression analysis, using the average persistent infection risk in the 30-35 age group as a reference, and the risk ratios for other age groups as factor values.

[0035] After determining the values ​​of the above fields, the timestamp field, type risk weight coefficient, viral load semi-quantitative field, status grade score, and age risk correction factor are aligned along the timeline to form a digital HPV health record. The timeline alignment method is as follows: using the timestamp field as the horizontal axis, all field records separated from a single visit are arranged chronologically. If a certain test record is missing at a certain time point, the corresponding field is left blank, and linear interpolation is performed based on the values ​​of adjacent time points during subsequent use. The aligned digital HPV health record is stored in the form of a structured data table. Each row corresponds to a time point, and the columns include the timestamp, type risk weight coefficient, viral load semi-quantitative field grade identifier, status grade score, age risk correction factor, and a composite risk correction factor obtained by multiplying the type risk weight coefficient and the age risk correction factor.

[0036] Example 2:

[0037] In specific implementation, please refer to Figure 3 The process of constructing a three-dimensional treatment state cube begins with establishing a three-dimensional coordinate system. The first dimension of the three-dimensional coordinate system corresponds to the viral infection state subspace, the second dimension corresponds to the cervical tissue state subspace, and the third dimension corresponds to the treatment intervention state subspace.

[0038] In the viral infection state subspace, a viral type axis and a viral load axis are set up. Coordinate points on the viral type axis are divided according to HPV types or type combinations, with single-type infection and multi-type co-infection listed as independent coordinate points. Each type coordinate point is identified by its international standard number. Coordinate points for multi-type co-infection are represented by a combination of the international standard numbers of the constituent types arranged in ascending order and connected by underscores. The coordinate points on the viral type axis are arranged in descending order of carcinogenic risk: types in the high-risk group, types in the suspected high-risk group, and types in the low-risk group. Within the same type group, they are arranged in ascending order of type number. Each coordinate point on the viral load axis represents a viral load level range. The viral load levels are divided into three ranges: low viral load corresponds to a detection value below 1.0 RLU / CO, medium viral load corresponds to a detection value between 1.0 RLU / CO and 10.0 RLU / CO, and high viral load corresponds to a detection value above 10.0 RLU / CO.

[0039] Within the cervical tissue state subspace, a cytological grading axis and a lesion extent axis are established. The coordinate points on the cytological grading axis are divided according to the TBS grading system, including six coordinate points: no intraepithelial lesion or malignant cells, atypical squamous cells of indeterminate significance, low-grade squamous intraepithelial lesion, atypical squamous cells where high-grade lesion cannot be ruled out, high-grade squamous intraepithelial lesion, and squamous cell carcinoma, arranged in order from mild to severe. The coordinate points on the lesion extent axis represent the number of quadrants of cervical epithelial involvement. The division method is as follows: the cervix is ​​divided into four quadrants clockwise, and the value is based on the number of quadrants spanned by the lesion area seen during colposcopy. The coordinate points on the lesion extent axis are, in order, zero quadrants, one quadrant, two quadrants, three quadrants, and four quadrants. Zero quadrants indicate no identifiable lesion area, and four quadrants indicate that the lesion involves all visible cervical epithelium.

[0040] Within the treatment intervention state subspace, an intervention mode axis and an intervention intensity axis are set. Coordinate points on the intervention mode axis represent different treatments or interventions. The complete set of interventions includes no intervention, expectant observation, topical interferon preparations, photodynamic therapy, loop electrosurgical excision procedure (LEEP), cold knife conization, and total hysterectomy. Coordinate points on the intervention mode axis are arranged in ascending order of intervention trauma. Coordinate points on the intervention intensity axis represent the execution frequency or dosage level of the corresponding intervention, set according to the different categories of interventions. For topical interferon preparations, coordinate points on the intervention intensity axis are divided into three levels: once every other day, once daily, and twice daily; for photodynamic therapy, coordinate points on the intervention intensity axis are divided into three levels: single irradiation, repeated irradiation at one-week intervals, and repeated irradiation at two-week intervals; for surgical interventions, coordinate points on the intervention intensity axis represent a single execution.

[0041] Each discrete grid point in the three-dimensional coordinate system is defined as a basic unit of treatment status. A basic unit of treatment status is uniquely determined by a coordinate point in the virus infection state subspace, a coordinate point in the cervical tissue state subspace, and a coordinate point in the treatment intervention state subspace. This is represented in the form of a triplet. ,in The first subspace representing the virus infection state The combination of virus type and viral load corresponding to each coordinate point The first subspace representing the state of cervical tissue The combination of cytological grade and lesion extent corresponding to each coordinate point The first state in the treatment intervention state subspace The combination of intervention methods and intensity corresponding to each coordinate point.

[0042] After the three-dimensional treatment status cube is constructed, a time-axis-based state retrospective and prospective extrapolation algorithm is used to calculate the virus clearance probability field and the lesion outcome probability field. The actual state sequence of the target patient at multiple historical time points is extracted from the digital HPV health record. Each state in the actual state sequence is projected onto the three-dimensional treatment status cube to obtain a historical state trajectory. The projection method is as follows: For each historical time point, the type risk weight coefficient, the level identifier of the semi-quantitative viral load field, the state level score, and the lesion extent information are read from the digital HPV health record. The type risk weight coefficient and the level identifier of the semi-quantitative viral load field are mapped to the corresponding coordinate points in the virus infection state subspace; the state level score and the lesion extent information are mapped to the corresponding coordinate points in the cervical tissue state subspace; and the actual treatment intervention record received at that time point is mapped to the corresponding coordinate points in the treatment intervention state subspace. These three coordinate points together determine a basic unit of treatment status. All basic units of treatment status corresponding to all historical time points are connected in chronological order to form a historical state trajectory.

[0043] Along the time axis of the historical state trajectory, the dwell time distribution and subsequent state transition distribution of each basic treatment posture unit under the condition of no additional intervention are statistically analyzed to form a state transition probability matrix. For a given basic treatment posture unit, the duration of occupation by the historical state trajectory in different time intervals is recorded, and the empirical distribution of dwell time is calculated. Simultaneously, the frequency of transitions from the historical state trajectory starting from this basic treatment posture unit to other basic treatment posture units in the next time step is recorded. After normalizing the frequency, the one-step transition probability from this basic treatment posture unit to other basic treatment posture units is obtained. The one-step transition probabilities from all basic treatment posture units are collected to form a complete state transition probability matrix.

[0044] Based on the state transition probability matrix, starting from the basic unit of the treatment posture where the current patient state is located, multiple preset time steps are extrapolated forward, and the probability value of reaching each basic unit of the treatment posture at each future time step is calculated. The preset time step length is set to three months. The extrapolation calculation adopts an iterative recursive method of Markov chain state probabilities, and the recursive formula is:

[0045]

[0046] in, Indicates the forward extrapolation to the th The probability vector of the patient's state in each basic treatment posture unit at a preset time step. The dimension is equal to the total number of basic units of the treatment situation in the three-dimensional treatment situation cube. The first in element Indicates the first At a preset time step, the patient's state is at the [number]th [time step]. The probability value of each basic unit of treatment posture. Integers that start counting from 0. The maximum value corresponds to the maximum forward extrapolation step size, which is set to 36, corresponding to a forward extrapolation period of nine years. This is the initial probability vector. The element corresponding to the basic unit of the current patient's treatment status takes a value of 1, while the other elements take a value of 0. Here is the state transition probability matrix. The number of rows and columns are both equal to the total number of basic units of the treatment posture. The Middle Line number Column elements Indicates from the first The basic unit of treatment posture is transferred one step at a time to the first The probability value of each basic unit of treatment posture.

[0047] The probability values ​​calculated over all future time steps are organized according to their coordinate positions in the three-dimensional treatment status cube to obtain the virus clearance probability field and the lesion outcome probability field. The virus clearance probability field corresponds to the probability distribution of viral load decreasing to below the detection threshold in the virus infection status subspace, with below the detection threshold corresponding to the low viral load interval. For a fixed position in the three-dimensional treatment status cube determined by the virus type coordinates, cervical tissue status coordinates, and treatment intervention status coordinates, all future time steps are traversed, and the probability values ​​of the virus infection status subspace coordinates at that fixed position that are in the low viral load interval are extracted at each future time step. The maximum probability value over time is taken as the value of the virus clearance probability field at that fixed position. The lesion outcome probability field corresponds to the probability distribution of cytological grade decreasing to normal or low-grade lesions in the cervical tissue status subspace. Normal corresponds to the coordinate points on the cytological grade axis where no intraepithelial lesions or malignant cells are seen, and low-grade lesions correspond to the coordinate points on the cytological grade axis where low-grade squamous intraepithelial lesions are seen. For a fixed position in the three-dimensional treatment status cube that is jointly determined by the coordinates of the virus infection status, the cervical tissue status, and the treatment intervention status, traverse all future time steps, extract the probability value of the cytological grade coordinate point at the fixed position as no intraepithelial lesion, malignant cells, or low-grade squamous intraepithelial lesion, and take the maximum value in time as the value of the lesion outcome probability field at the fixed position.

[0048] For maintaining the state transition probability matrix, an online learning approach based on Bayesian updates is employed. Each time a new HPV genotyping result or cervical cytology record is obtained, the corresponding transition probability is posteriorly corrected. The transition probability of each row in the state transition probability matrix is ​​considered as a parameter of a multinomial distribution, whose conjugate prior distribution is a Dirichlet distribution. During initialization, a symmetric Dirichlet prior distribution is set for each row of transition probability vectors, with the lumped parameter uniformly set to 1.0, representing no prior information. Each time a new state transition sample record is acquired, the row index of the starting treatment state basic unit of that transition is determined, the target endpoint count in the observation count of the corresponding multinomial distribution for that row is incremented by one, the posterior mean of the transition probability vector for that row is recalculated, and the element values ​​of the corresponding row in the state transition probability matrix are updated using the posterior mean. After the update, the element value of each row in the state transition probability matrix is ​​the mean of the posterior Dirichlet distribution corresponding to that row.

[0049] Example 3:

[0050] In specific implementation, please refer to Figure 4The process of delineating high-risk persistent infection spatiotemporal regions from the virus clearance probability field is as follows: Each basic unit of the treatment posture in the three-dimensional treatment posture cube is traversed, and the virus clearance probability value corresponding to that basic unit is read from the virus clearance probability field. This virus clearance probability value is compared with a first preset threshold. If the virus clearance probability value is lower than the first preset threshold, the basic unit of the treatment posture is marked as a high-risk persistent infection candidate unit. After all basic units of the treatment posture have been marked, a three-dimensional connected component marking algorithm is used to merge adjacent units marked as high-risk persistent infection candidate units into the same connected component. Each connected component is considered a high-risk persistent infection spatiotemporal region. Adjacent is defined as the distance between two basic units of the treatment posture along any coordinate axis in the three-dimensional coordinate system, which is one grid step.

[0051] The process of delineating low-risk self-healing tendency spatiotemporal regions from the lesion outcome probability field is as follows: Each basic unit of the treatment state in the three-dimensional treatment state cube is traversed, and the lesion outcome probability value corresponding to that basic unit is read from the lesion outcome probability field. This lesion outcome probability value is compared with a second preset threshold. If the lesion outcome probability value is higher than the second preset threshold, the basic unit of the treatment state is marked as a low-risk self-healing tendency candidate unit. A three-dimensional connected component labeling algorithm is used to merge adjacent units marked as low-risk self-healing tendency candidate units into the same connected component, and each connected component serves as a low-risk self-healing tendency spatiotemporal region.

[0052] After obtaining the spatiotemporal regions of high-risk persistent infection and low-risk self-healing tendency, the boundary coordinate set of the high-risk persistent infection spatiotemporal region in the three-dimensional treatment state cube is obtained. For a connected domain of a high-risk persistent infection spatiotemporal region, the minimum and maximum coordinates of all basic units of the treatment state within the connected domain are extracted in the dimensions of the viral infection state subspace, the cervical tissue state subspace, and the treatment intervention state subspace. The boundary coordinate set of a connected domain is represented as a three-dimensional interval, which consists of closed intervals in each of the three dimensions. If there are multiple connected domains of high-risk persistent infection spatiotemporal regions, the three-dimensional intervals of each connected domain are included in the boundary coordinate set of the high-risk persistent infection spatiotemporal region. The boundary coordinate set of the low-risk self-healing tendency spatiotemporal region is obtained in the same way, that is, the minimum and maximum coordinates in the three dimensions are extracted for each low-risk self-healing tendency spatiotemporal region connected domain to form the corresponding three-dimensional interval. The three-dimensional intervals of all connected domains are summarized into the boundary coordinate set of the low-risk self-healing tendency spatiotemporal region.

[0053] Spatial overlay analysis is performed on the boundary coordinate set of the high-risk persistent infection spatiotemporal region and the boundary coordinate set of the low-risk self-healing tendency spatiotemporal region. For each three-dimensional interval in the boundary coordinate set of the high-risk persistent infection spatiotemporal region, traverse each three-dimensional interval in the boundary coordinate set of the low-risk self-healing tendency spatiotemporal region, and calculate the intersection of the two three-dimensional intervals in three dimensions. If the intersection in all three dimensions is non-empty, calculate the common part of the two three-dimensional intervals, which is an overlapping region. The remaining part after subtracting the overlapping region from the three-dimensional interval of the high-risk persistent infection spatiotemporal region is recorded as the mutually exclusive part of the high-risk persistent infection spatiotemporal region; the remaining part after subtracting the overlapping region from the three-dimensional interval of the low-risk self-healing tendency spatiotemporal region is recorded as the mutually exclusive part of the low-risk self-healing tendency spatiotemporal region. The set of all overlapping regions constitutes the overlapping part of the two regions, and the set of all mutually exclusive parts constitutes the mutually exclusive part of the two regions.

[0054] The first and second preset thresholds are adaptively adjusted based on the immune status indicators and age risk correction factors in the individualized physiological characteristic parameters of the target patient. In practice, the most recent peripheral blood T lymphocyte subset analysis record of the target patient is extracted from the individualized physiological characteristic parameter section of the digital HPV health record. The absolute counts of CD4-positive T lymphocytes and CD8-positive T lymphocytes are read from this record, and the ratio of the absolute count of CD4-positive T lymphocytes to the absolute count of CD8-positive T lymphocytes is calculated and recorded as the CD4 / CD8 ratio for this test. The normal reference range for the CD4 / CD8 ratio provided by the clinical laboratory is obtained, with a lower limit of 0.71, a median of 1.5, and an upper limit of 2.5. The target patient's CD4 / CD8 ratio is divided by the median of the normal reference range, 1.5, to obtain the initial value of the immune status indicator. The initial value of the immune status indicator is clamped within the interval [0.5, 1.5], taking 1.5 if greater than 1.5 and 0.5 if less than 0.5. The clamped value is used as the final immune status indicator. The age risk correction factor is directly read from the record in the digital HPV health record that is closest to the timestamp of the current immune test. If there is no record with a corresponding time point, linear interpolation is used to obtain it.

[0055] The adaptive adjustment method of the first preset threshold is as follows: set the first preset threshold baseline value to 0.3, multiply the first preset threshold baseline value by the immune status index and then by the age risk correction factor to obtain the initial adjustment value of the first preset threshold, clamp the initial adjustment value of the first preset threshold within the range [0.1, 0.6], and use it as the final first preset threshold.

[0056] The adaptive adjustment method of the second preset threshold is as follows: set the second preset threshold baseline value to 0.7, calculate the product of the immune status index and the age risk correction factor, and record it as the composite adjustment factor. Multiply the second preset threshold baseline value by 2.0 and subtract the difference of the composite adjustment factor to obtain the initial adjustment value of the second preset threshold. Clamp the initial adjustment value of the second preset threshold within the interval [0.5, 0.9] as the final second preset threshold.

[0057] After delineating the high-risk persistent infection spatiotemporal region and the low-risk spontaneous recovery tendency spatiotemporal region, it is determined whether the current patient's status falls within the high-risk persistent infection spatiotemporal region. The target patient's current HPV genotyping results, current cervical cytology results, and current treatment intervention status are obtained. The current HPV genotyping results are mapped onto the viral type and viral load axes of the viral infection status subspace to obtain coordinate points in the viral infection status subspace; the current cervical cytology results and the lesion extent assessed by colposcopy are mapped onto the cytology grading and lesion extent axes of the cervical tissue status subspace to obtain coordinate points in the cervical tissue status subspace; the current treatment intervention status is mapped onto the intervention method and intervention intensity axes of the treatment intervention status subspace to obtain coordinate points in the treatment intervention status subspace. The coordinate points of the current status are jointly determined by the coordinate points of the three subspaces, and the three-dimensional coordinate vector of the current status coordinate point is denoted as... .

[0058] The coordinates of the current state point in each dimension are compared sequentially with the coordinate intervals of the high-risk persistent infection spatiotemporal region in the corresponding dimension. If the boundary coordinate set of the high-risk persistent infection spatiotemporal region contains multiple three-dimensional intervals, each three-dimensional interval is compared one by one. For a given three-dimensional interval, the following steps are performed: Whether it falls within the closed interval of the virus infection state subspace dimension of this three-dimensional interval. Whether it falls within the closed interval of the cervical tissue state subspace dimension of this three-dimensional interval. Does the current state coordinate point fall within the closed interval of the treatment intervention state subspace dimension of this three-dimensional interval? If all three judgment results are yes, then the current state coordinate point is determined to be located within the high-risk persistent infection spatiotemporal region, and the judgment process ends.

[0059] If, after traversing all three-dimensional intervals in the boundary coordinate set of the high-risk persistent infection spatiotemporal region, no three dimensions simultaneously fall within the set, then the minimum Euclidean distance from the current state coordinate point to the boundary of the high-risk persistent infection spatiotemporal region is further calculated. The formula for calculating the minimum Euclidean distance is:

[0060]

[0061] in, This represents the minimum Euclidean distance, expressed in grid step size. A three-dimensional coordinate vector representing the current state coordinate point. ; The set of boundary points represents the spatiotemporal region of high-risk persistent infection. The set of boundary points consists of the coordinate vectors of the basic unit of treatment status in all connected domains of the spatiotemporal region of high-risk persistent infection, where at least one adjacent unit is not in the region. Represents the set of boundary points any coordinate vector in ; Representing vectors with vector The Euclidean norm between them is calculated using the following formula: ; This represents the minimum Euclidean distance among all boundary points.

[0062] The calculated The distance is compared with a preset distance threshold. The distance threshold is set to a grid step size of 1.0, based on the fact that the Euclidean distance between any two adjacent basic units of the treatment posture in the 3D treatment posture cube is 1.0 in any coordinate axis direction. Therefore, any point not located inside the high-risk persistent infection spatiotemporal region, if its minimum distance to the region boundary is less than 1.0, indicates that the point is adjacent to the surface of the high-risk persistent infection spatiotemporal region and belongs to the adjacent position outside the region. If the value is less than 1.0, the current state coordinates are determined to be located in the vicinity of a high-risk persistent infection spatiotemporal region; if... If the value is greater than or equal to 1.0, then the current state coordinate point is determined to be neither located within the high-risk persistent infection spatiotemporal region nor in its adjacent region.

[0063] Example 4:

[0064] In specific implementation, please refer to Figure 5Once it is determined that the current patient's condition falls within the high-risk persistent infection spatiotemporal region, a long-term treatment strategy encompassing multiple treatment phases is generated. The projection interval of the high-risk persistent infection spatiotemporal region onto the treatment intervention state subspace is extracted as a reference range for treatment intensity. The extraction method is as follows: obtain the closed interval of each three-dimensional interval in the boundary coordinate set of the high-risk persistent infection spatiotemporal region on the treatment intervention state subspace dimension, and take the union of all closed intervals; this union constitutes the reference range for treatment intensity. The reference range for treatment intensity represents the set of treatment intervention intensity values ​​covered by the high-risk persistent infection spatiotemporal region, expressed as coordinate values ​​on the intervention intensity axis. The projection interval of the low-risk self-healing tendency spatiotemporal region onto the treatment intervention state subspace is extracted as a lower limit reference value for treatment intensity. The extraction method is as follows: obtain the closed interval of each three-dimensional interval in the boundary coordinate set of the low-risk self-healing tendency spatiotemporal region on the treatment intervention state subspace dimension, and take the endpoint value with the smallest coordinate value among all closed intervals; this minimum endpoint value is used as the lower limit reference value for treatment intensity.

[0065] Candidate strategy templates are retrieved from a pre-defined treatment strategy knowledge base. This knowledge base is stored as a relational database table on the backend server of the clinical decision support system. Each record in the knowledge base corresponds to a successful treatment case for a historical patient and includes the following fields: case identifier, spatial morphological feature vector of the high-risk persistent infection spatiotemporal region, candidate strategy template identifier, strategy identifier, phase division rules, phase triggering conditions, phase exit conditions, and a three-dimensional treatment posture cube trajectory sequence. The spatial morphological feature vector of the high-risk persistent infection spatiotemporal region is composed of seven components: the span value of the region in the viral infection state subspace dimension, the span value in the cervical tissue state subspace dimension, the span value in the treatment intervention state subspace dimension, the total number of basic treatment posture units contained within the region, and the first-order moment coordinates of the region in the three dimensions. The three-dimensional treatment posture cube trajectory sequence is a sequence of coordinate vectors of basic treatment posture units arranged in chronological order.

[0066] When retrieving matching records from the treatment strategy knowledge base, the spatial morphological feature vector of the high-risk persistent infection spatiotemporal region of the current patient is calculated and denoted as the current spatial morphological feature vector. Traverse the spatial morphological feature vector of each record in the treatment strategy knowledge base. ,calculate and The cosine similarity between them. The formula for calculating cosine similarity is:

[0067]

[0068] in, Represents the current spatial morphological feature vector Spatial morphological feature vector of a record in the treatment strategy knowledge base Cosine similarity between them The value of is in the range of [-1, 1], and the closer the value is to 1, the more consistent the directions of the two vectors are. This represents the spatial morphological feature vector of the high-risk persistent infection region for the current patient. It is a seven-dimensional real vector, and the value of each component is determined by the specific values ​​of the aforementioned seven components. Let the spatial morphological feature vector be a certain record in the treatment strategy knowledge base. It is also a seven-dimensional real vector; Representing vectors with vector The inner product; Representing vectors The Euclidean norm; Representing vectors The Euclidean norm of the algorithm is used. All records with a cosine similarity higher than 0.85 are selected, and the candidate strategy templates associated with these records are used as the retrieved candidate strategy template set. If there are more than 50 records with a cosine similarity higher than 0.85, they are sorted in descending order of cosine similarity, and the first 50 records are selected.

[0069] For each candidate strategy template in the retrieved candidate strategy template set, the dynamic time warping distance between the current patient state trajectory and the corresponding 3D treatment posture cube trajectory sequence in the treatment strategy knowledge base is further calculated. The current patient state trajectory is composed of the coordinate vectors of the basic treatment posture units in the 3D treatment posture cube corresponding to all historical states of the current patient from the initial consultation time to the current time, arranged in chronological order. The calculation of the dynamic time warping distance adopts the standard dynamic time warping algorithm, using Euclidean distance as the local distance metric, setting the window constraint parameter to 10% of the trajectory length, and using a symmetric step mode for the recursion of the cumulative distance matrix. The five candidate strategy templates with the smallest dynamic time warping distance are output as the final matched candidate strategy templates. Each candidate strategy template includes a strategy identifier, phase division rules, phase triggering conditions, and phase exit conditions.

[0070] The phase division rules in the candidate strategy template divide long-term treatment strategies into three phases in chronological order: the virus clearance induction phase, the sustained infection control phase, and the consolidation and follow-up monitoring phase. The target state range associated with the virus clearance induction phase is: the viral load axis coordinate point in the virus infection status subspace is in the low viral load range, and the cytological grading axis coordinate point in the cervical tissue status subspace is no intraepithelial lesion, malignant cells, or low-grade squamous intraepithelial lesion. The maximum allowable duration for the virus clearance induction phase is six months. The target state range associated with the sustained infection control phase is: the viral load axis coordinate point in the virus infection status subspace is in the low viral load range, and the cytological grading axis coordinate point in the cervical tissue status subspace is no intraepithelial lesion or malignant cells. The maximum allowable duration for the sustained infection control phase is twelve months. The target status range for the consolidation follow-up monitoring phase is as follows: in two consecutive follow-ups, the coordinates of the viral load axis in the viral infection status subspace are both in the low viral load range, and the coordinates of the cytology grading axis in the cervical tissue status subspace are both free of intraepithelial lesions or malignant cells. The maximum allowable duration for the consolidation follow-up monitoring phase is twenty-four months.

[0071] The phase triggering conditions include the exit action triggering condition of the previous phase and a time triggering condition independent of the previous phase. The exit action triggering condition of the previous phase is defined as follows: when the previous phase exits normally due to meeting the exit conditions for the target state, the start of the current phase is automatically triggered. The time triggering condition independent of the previous phase is defined as follows: starting from the overall start time of the long-term treatment strategy, a pre-set mandatory start time threshold for the current phase is established. When the system time reaches the mandatory start time threshold, the start of the current phase is triggered regardless of whether the previous phase exited normally. The mandatory start time threshold for the virus clearance induction phase is set to the overall start time of the long-term treatment strategy; the mandatory start time threshold for the continuous infection control phase is set to six months after the overall start time of the long-term treatment strategy; and the mandatory start time threshold for the consolidation follow-up monitoring phase is set to eighteen months after the overall start time of the long-term treatment strategy. The current phase begins execution when either the exit action triggering condition or the time triggering condition of the previous phase is met.

[0072] The exit conditions for each stage include the exit condition for achieving the target state and the forced exit condition for time exhaustion. The state range in the exit condition for achieving the target state covers the current patient's state coordinates. When the exit condition for achieving the target state is met, the current stage exits normally and proceeds to the next stage. Specifically, the determination method is as follows: obtain the target state range associated with the current stage, compare the current patient's state coordinates with the target state range. If the coordinates of the current patient's state coordinates in the viral infection state subspace dimension fall within the viral load range specified by the target state range, and the coordinates in the cervical tissue state subspace dimension fall within the cytological grading range specified by the target state range, then the exit condition for achieving the target state is deemed met. The forced exit condition for time exhaustion is defined as: accumulating the duration of the current stage execution. When the accumulated duration reaches the maximum allowable dwell time associated with the current stage, if the exit condition for achieving the target state is still not met, the current stage forcibly exits and triggers an exception handling branch. The exception handling branch involves re-inputting the current patient's state coordinates into the three-dimensional treatment situation cube, re-judging the high-risk persistent infection spatiotemporal region, and re-triggering the long-term treatment strategy generation process.

[0073] See Figure 6 In the figure, the horizontal axis represents cosine similarity, with a value range of [-1, 1], and the vertical axis represents the frequency of records corresponding to the cosine similarity. The bar chart is filled with gray to represent the distribution of cosine similarity. The dashed line in the legend is the cosine similarity threshold line, with a threshold of 0.85, and the vertical dashed line is located at 0.85 on the horizontal axis.

[0074] The gray histogram shows that the cosine similarity of most records is concentrated between -1.0 and 0.2, with frequencies fluctuating between 10 and 38, exhibiting a right-skewed distribution. The middle section, with negative cosine similarity and values ​​close to zero, shows higher frequencies, reflecting the existence of diverse spatial morphological feature vector matching records in the database. The histograms showing cosine similarity greater than 0.85 have significantly fewer frequencies, with only a few records to the right of the dashed line, indicating that strategy template records meeting high similarity matching criteria are relatively scarce.

[0075] This diagram specifically corresponds to the process of retrieving candidate strategy templates from the treatment strategy knowledge base in Example 4. The cosine similarity on the horizontal axis represents the spatial morphological feature vector of the high-risk persistent infection region for the current patient. Spatial morphological feature vectors of each record in the knowledge base The results of cosine similarity calculation are shown in the figure. Using 0.85 as a preset threshold, a set of candidate strategy templates with high matching degree is selected, providing a preliminary data foundation for further filtering using dynamic time-normalized distance.

[0076] This distribution map reflects the similarity distribution characteristics among spatial morphological feature vectors in the knowledge base, providing a basis for setting the threshold for candidate strategy template retrieval in Example 4. The threshold of 0.85 is located at the right tail of the cosine similarity distribution, ensuring high relevance of the retrieval results and accuracy of strategy template matching, which helps to generate more personalized and effective long-term treatment strategies.

[0077] Example 5:

[0078] In practice, starting from the current patient state coordinates, any coordinate point in the low-risk self-healing tendency spatiotemporal region is used as the target endpoint set. A shortest path search algorithm is employed to search for multiple candidate treatment paths from the starting point to the target endpoint set within the 3D treatment situation cube. Each basic unit of treatment situation in the 3D treatment situation cube is defined as a node in a graph structure. The total number of nodes in the graph structure equals the number of discrete grid points in the 3D treatment situation cube, and each node is identified by a unique triplet coordinate. The state transition relationship between two adjacent basic units of treatment situation is defined as a directed edge in the graph structure. Two adjacent basic units of treatment situation refer to two basic units of treatment situation that are one grid step apart along any coordinate axis in the 3D coordinate system. For each pair of adjacent basic units of treatment situation, a directed edge is established from the source node to the target node, and the direction of the directed edge is consistent with the one-step transition direction recorded in the state transition probability matrix. If the one-step transition probability from a certain basic unit of treatment situation to an adjacent basic unit in the state transition probability matrix is ​​zero, then no directed edge is established for this pair of adjacent units in the graph structure.

[0079] Each directed edge associates a transition probability from the source node to the target node and a transition time cost. The transition probability is the value of the corresponding row and column in the state transition probability matrix, representing the probability of transitioning from the source node to the target node in one step. The transition time cost is set as the average dwell time of the basic treatment state unit where the source node is located in the state transition probability matrix. The average dwell time is calculated as follows: for the basic treatment state unit where the source node is located, the total duration of the basic treatment state unit occupied in the historical state trajectory is counted, and the arithmetic mean of all durations is taken. If the basic treatment state unit has never been occupied in the historical state trajectory, the average dwell time is assigned the length of a preset time step, i.e., three months. The unit of the transition time cost is months.

[0080] Starting from the node where the starting point is located, a heuristic search algorithm based on Dijkstra's principle is used to search for the shortest path to any node in the target endpoint set. During algorithm initialization, an open set and a closed set are created. The open set stores nodes to be examined, and the closed set stores nodes that have already been examined. The starting point node is added to the open set, and its cumulative transfer time cost is initialized to zero. The target endpoint set consists of nodes corresponding to all basic units of treatment states within the low-risk self-healing tendency spatiotemporal region.

[0081] In each iteration, the node with the minimum cumulative transfer time cost is selected from the open set as the current node under consideration, and then moved from the open set to the closed set. If the current node under consideration belongs to the target endpoint set, the search terminates, and the node backtracks from the current node to the starting node along the parent node pointer, recording the sequence of nodes traversed and the corresponding directed edge sequence. This sequence of nodes and directed edges constitutes the shortest path from the starting point to the target endpoint. If the current node under consideration does not belong to the target endpoint set, all outgoing edges of the current node in the graph structure are traversed. For each adjacent node pointed to by an outgoing edge, the cumulative transfer time cost from the starting point through the current node to the adjacent node is calculated. The cumulative transfer time cost equals the cumulative transfer time cost of the current node plus the transfer time cost associated with the directed edges from the current node to the adjacent node. If an adjacent node is neither in the open set nor the closed set, add it to the open set, record its parent node as the currently examined node, and record its cumulative transition time cost. If the adjacent node is already in the open set, and the newly calculated cumulative transition time cost is less than the cumulative transition time cost already recorded in the open set, update the adjacent node's cumulative transition time cost and update its parent node pointer to the currently examined node. Repeat this iterative process until the target endpoint is found or the open set is empty.

[0082] After finding the first shortest path, record the sequence of nodes traversed by the first shortest path. And directed edge sequence Find the first shortest path. A penalty factor is temporarily added to the transition time cost of each directed edge. The penalty factor is 1.5 times the original transition time cost of the directed edge. This value is set based on the following principle: by penalizing used edges, subsequent searches are forced to prioritize exploring different topological branches under similar cumulative costs, while avoiding excessive penalties that could drastically worsen the cumulative cost of subsequent paths. The heuristic search algorithm is then executed again on the penalized graph structure to obtain the second shortest path, and the sequence of nodes traversed by the second shortest path is recorded. And directed edge sequence Each time a new candidate path is found, a penalty factor is temporarily added to the transition time cost of all directed edges on that path. The search continues, repeating this process until the number of candidate treatment paths reaches a preset limit of 10, or the open set is empty and no new path to the target endpoint set can be found. After each search, the temporarily modified transition time cost of all directed edges is restored to its original value, and a penalty factor is applied to the directed edges of the next path to ensure that the penalty for each path search only applies to the current iteration.

[0083] The node sequence traversed by each candidate treatment path is mapped back to the treatment intervention state subspace to obtain the treatment intervention intensity sequence to be executed at each time node. The mapping method is as follows: For each node in the candidate treatment path node sequence, the intervention mode axis coordinate value and intervention intensity axis coordinate value of the treatment intervention state subspace in the corresponding basic unit of treatment status are read, and these values ​​are combined into a treatment intervention intensity entry. All treatment intervention intensity entries are arranged in the order of the node sequence to form the treatment intervention intensity sequence corresponding to the candidate treatment path. The path length of each candidate treatment path is calculated, defined as the sum of the transition time costs of all directed edges on the path. Multiple candidate treatment paths are sorted in ascending order of path length, and the top 5 candidate treatment paths are selected as candidate solutions for long-term treatment strategies.

[0084] After outputting candidate treatment plans, the adherence score of the target patient to historical treatment interventions is obtained. The adherence score is obtained by extracting the target patient's completion records of all past treatment tasks from the digital HPV health record, calculating the number of treatment tasks completed on time and the total number of treatment tasks that should have been completed, and then recording this ratio as the original adherence rate. Multiplying the original adherence rate by 100 and linearly mapping it to the 0-1 interval yields the adherence score; that is, the adherence score equals the original adherence rate. If the target patient has no historical treatment records, the adherence score is assigned a default value of 0.8.

[0085] Adherence scores were used as a correction factor to normalize the treatment intervention intensity sequence in each candidate treatment path. The adjustment method was as follows: For each treatment intervention intensity item in the sequence, the corresponding intervention method axis coordinate value was read to determine whether the intervention method belonged to a non-surgical intervention requiring active patient intervention. Non-surgical interventions requiring active patient intervention included no intervention, expectant observation, topical interferon preparations, and photodynamic therapy. If the intervention method belonged to a non-surgical intervention requiring active patient intervention, the intervention intensity axis coordinate value of that item was multiplied by the reciprocal of the adherence score to obtain the adjusted intervention intensity axis coordinate value. For several interventions that belonged to surgical interventions, including loop electrosurgical excision procedure (LEEP), cold knife conization, and total hysterectomy, the intervention intensity axis coordinate value remained unchanged. The adjusted intervention intensity axis coordinate value was clamped within the maximum coordinate value of the intervention intensity axis corresponding to that intervention method to obtain the individualized treatment intervention intensity sequence.

[0086] The individual-fitted treatment intervention intensity sequence is bound to time nodes in the candidate treatment path. The time nodes in the candidate treatment path are determined by the cumulative transfer time cost corresponding to the node sequence. The time corresponding to the starting point is set to time zero, and the node sequence... The time corresponding to each node represents the cumulative transfer time cost from the starting point to that node. The intervention method code and adjusted intervention intensity parameter at each time node are combined into an entry to generate a timestamped treatment task list. Each entry in the treatment task list contains an execution time window, an intervention method code, and an intervention intensity parameter. The execution time window is set to a time interval extending one week before and after the time node.

[0087] The treatment task list is output to the user interface of the clinical decision support system. The user interface displays the temporal relationship of each treatment phase in the form of a Gantt chart. The horizontal axis of the Gantt chart represents time, and the vertical axis is arranged in rows according to the candidate treatment path numbers. Different colored bars in each row represent the duration of different treatment phases: blue bars for the virus clearance induction phase, orange bars for the sustained infection control phase, and green bars for the consolidation follow-up monitoring phase. The corresponding intervention code and intervention intensity parameter are labeled on each bar. The user interface provides a detailed information pop-up window for each candidate treatment path, displaying the path length, the target state range for each phase, and the maximum allowed dwell time.

[0088] The system receives confirmation or modification instructions from clinicians. Clinicians can select a candidate treatment path and trigger a confirmation instruction via a click operation in the user interface, or trigger a modification instruction by dragging the horizontal boundaries of the Gantt chart or modifying the intervention intensity parameter input box. The user interface backend recalculates the modified time nodes and intervention intensity sequence based on the modifications in the instruction, generates a modified treatment task list, and refreshes the Gantt chart display. Upon receiving a confirmation instruction, the final treatment strategy selected by the confirmation instruction is locked. The locked final treatment strategy includes all entries in the treatment task list, the target state range of each stage, stage triggering conditions, and stage exit conditions. The final treatment strategy is stored as structured data in the target patient's digital HPV health record. The stored fields include strategy identifier, generation timestamp, strategy version number, treatment task list, stage division information, and state transition condition information, serving as historical reference data for the next strategy formulation.

[0089] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for developing a long-term HPV treatment strategy for cervical cancer prevention and treatment, characterized in that, Includes the following steps: Obtain the target patient's digital HPV health record, which includes multiple HPV genotyping test results, cervical cytology history, and individualized physiological characteristic parameters; Based on the digital HPV health record, a three-dimensional treatment status cube is constructed, consisting of a virus infection status subspace, a cervical tissue status subspace, and a treatment intervention status subspace. In the three-dimensional treatment situation cube, a state backtracking and prospective deduction algorithm based on the time axis is used to calculate the virus clearance probability field and the lesion outcome probability field; Based on the virus clearance probability field and the lesion outcome probability field, at least one high-risk spatiotemporal region of persistent infection and at least one low-risk spatiotemporal region of self-healing tendency are delineated. The current patient status is projected onto the three-dimensional treatment situation cube to determine whether it falls into the high-risk persistent infection spatiotemporal region. If so, a long-term treatment strategy generation process containing multiple treatment stages is triggered. The long-term treatment strategy includes the timing and intensity of intervention for each treatment phase, as well as the conditions for state transitions between phases.

2. The method for formulating a long-term HPV treatment strategy for cervical cancer prevention and control according to claim 1, characterized in that, The acquisition of the target patient's digital HPV health record specifically includes: Extract the target patient's historical medical records from the medical information system, and separate the timestamp field, HPV genotype field, viral load semi-quantitative field, TBS grade field of cervical liquid-based thin-layer cytology examination, and patient age field from the historical medical records. The HPV genotype field is divided into high-risk type group, suspected high-risk type group and low-risk type group according to the carcinogenic risk level, and a type risk weight coefficient is assigned to each group; The TBS grading field is mapped to multiple discrete status levels from normal to squamous cell carcinoma according to the severity of the lesion, and a status level score is assigned to each discrete status level. The patient age field is divided into multiple age risk intervals according to the cervical cancer incidence age distribution curve, and each age risk interval corresponds to an age risk correction factor. The timestamp field, the type risk weight coefficient, the viral load semi-quantitative field, the status level score, and the age risk correction factor are aligned along the time axis to form the digital HPV health record.

3. The method for developing a long-term HPV treatment strategy for cervical cancer prevention and control according to claim 1, characterized in that, The construction of a three-dimensional treatment situation cube, consisting of a virus infection state subspace, a cervical tissue state subspace, and a treatment intervention state subspace, specifically includes: Establish a three-dimensional coordinate system, where the first dimension corresponds to the virus infection state subspace, the second dimension corresponds to the cervical tissue state subspace, and the third dimension corresponds to the treatment intervention state subspace. In the virus infection state subspace, a virus type axis and a virus load axis are set. Each coordinate point on the virus type axis represents an HPV type or combination of types, and each coordinate point on the virus load axis represents a virus load level range. In the cervical tissue state subspace, a cytology grading axis and a lesion extent axis are set. Each coordinate point on the cytology grading axis represents a TBS grading level, and each coordinate point on the lesion extent axis represents the number of quadrants or percentage ranges of cervical epithelial involvement. In the treatment intervention state subspace, an intervention method axis and an intervention intensity axis are set. Each coordinate point on the intervention method axis represents a treatment or intervention method, and each coordinate point on the intervention intensity axis represents the execution frequency or dosage level of the intervention method. Each discrete grid point in the three-dimensional coordinate system is defined as a basic unit of treatment status, and each basic unit of treatment status corresponds to a specific combination of viral infection status, cervical tissue status and treatment intervention status.

4. The method for developing a long-term HPV treatment strategy for cervical cancer prevention and control according to claim 3, characterized in that, The algorithm employs a time-axis-based state backtracking and look-ahead deduction to calculate the virus clearance probability field and the lesion outcome probability field, specifically including: Extract the actual state sequence of the target patient at multiple historical time points from the digital HPV health record, and project each state in the actual state sequence onto the three-dimensional treatment status cube to obtain the historical state trajectory. Along the time direction of the historical state trajectory, the dwell time distribution and subsequent state transition distribution of each basic unit of treatment status under the condition of no additional intervention are statistically analyzed to form a state transition probability matrix; Based on the state transition probability matrix, starting from the basic unit of the treatment situation where the current patient is located, multiple preset time steps are extrapolated forward, and the probability value of reaching each basic unit of the treatment situation at each future time step is calculated. The probability values ​​calculated over all future time steps are organized according to their coordinate positions in the three-dimensional treatment state cube to obtain the virus clearance probability field and the lesion outcome probability field. The virus clearance probability field corresponds to the probability distribution of viral load dropping below the detection threshold in the virus infection state subspace, and the lesion outcome probability field corresponds to the probability distribution of cytological grade dropping to normal or low-grade lesion in the cervical tissue state subspace.

5. The method for formulating a long-term HPV treatment strategy for cervical cancer prevention and control according to claim 4, characterized in that, The state transition probability matrix adopts an online learning method based on Bayesian updates. Each time a new HPV typing test result or cervical cytology examination record is obtained, the corresponding transition probability is posteriorly corrected.

6. The method for formulating a long-term HPV treatment strategy for cervical cancer prevention and control according to claim 4, characterized in that, The delineation of at least one high-risk spatiotemporal region of persistent infection and at least one low-risk spatiotemporal region of spontaneous recovery specifically includes: In the virus clearance probability field, all basic units of treatment status with virus clearance probability values ​​lower than a first preset threshold are extracted, and the spatiotemporal continuous regions occupied by these basic units of treatment status are marked as the high-risk spatiotemporal regions of persistent infection. In the disease outcome probability field, all basic units of treatment status with disease outcome probability values ​​higher than the second preset threshold are extracted, and the spatiotemporal continuous regions occupied by these basic units of treatment status are marked as the low-risk self-healing tendency spatiotemporal regions. Obtain the set of boundary coordinates of the high-risk persistent infection spatiotemporal region in the three-dimensional treatment situation cube, and the set of boundary coordinates of the low-risk self-healing tendency spatiotemporal region in the three-dimensional treatment situation cube; By performing a spatial overlay analysis on the boundary coordinate set of the high-risk persistent infection spatiotemporal region and the boundary coordinate set of the low-risk self-healing tendency spatiotemporal region, the overlapping and mutually exclusive parts of the two regions are obtained.

7. The method for formulating a long-term HPV treatment strategy for cervical cancer prevention and control according to claim 6, characterized in that, The first preset threshold and the second preset threshold are adaptively adjusted based on the immune status index and age risk correction factor in the individualized physiological characteristic parameters of the target patient.

8. The method for formulating a long-term HPV treatment strategy for cervical cancer prevention and control according to claim 6, characterized in that, The determination of whether it falls within the high-risk persistent infection spatiotemporal region specifically includes: The target patient's current HPV typing test results, current cervical cytology test results, and current treatment intervention status are obtained. The above three pieces of information are mapped to the three dimensions of the three-dimensional treatment status cube to obtain the current status coordinate point. The coordinates of the current state coordinate point in each dimension are compared with the coordinate interval of the high-risk persistent infection spatiotemporal region in that dimension. If the coordinates of the current state coordinate point in each of the three dimensions fall within the corresponding coordinate interval, it is determined that the current state coordinate point is located inside the high-risk persistent infection spatiotemporal region. If the coordinate value in any dimension does not fall within the corresponding coordinate interval, the minimum Euclidean distance from the current state coordinate point to the boundary of the high-risk persistent infection spatiotemporal region is further calculated, and the minimum Euclidean distance is compared with a distance threshold. If the minimum Euclidean distance is less than the distance threshold, it is determined that the current state coordinate point is located in the vicinity of the high-risk persistent infection spatiotemporal region.

9. A method for developing a long-term HPV treatment strategy for cervical cancer prevention and control according to claim 1, characterized in that, The process of triggering the generation of a long-term treatment strategy comprising multiple treatment phases specifically includes: The projection range of the high-risk persistent infection spatiotemporal region onto the treatment intervention state subspace is used as the reference range of treatment intensity, and the projection range of the low-risk self-healing tendency spatiotemporal region onto the treatment intervention state subspace is used as the lower limit reference value of treatment intensity. From a pre-defined treatment strategy knowledge base, candidate strategy templates that match the spatial morphological characteristics of the high-risk persistent infection spatiotemporal region are retrieved. The candidate strategy templates include strategy identifiers, phase division rules, phase triggering conditions, and phase exit conditions. Using the current patient status coordinates as the starting point and any coordinate point in the low-risk self-healing tendency spatiotemporal region as the target endpoint set, the shortest path search algorithm is used to search for multiple candidate treatment paths from the starting point to the target endpoint set in the three-dimensional treatment situation cube. The candidate treatment paths are sorted in ascending order of path length, and a predetermined number of the top-ranked candidate treatment paths are selected as candidate solutions for the long-term treatment strategy.

10. A method for developing a long-term HPV treatment strategy for cervical cancer prevention and control according to claim 9, characterized in that, The preset treatment strategy knowledge base contains multiple historical successful treatment cases of patients and their corresponding three-dimensional treatment status cube trajectories. Matching candidate strategy templates are retrieved by calculating the dynamic time warping distance between the current patient status trajectory and the trajectory of the successful treatment cases.

Images