Multi-dimensional classification-based stroke home rehabilitation guidance method and system

Abstract

The invention relates to the field of medical information processing, in particular to a stroke home rehabilitation guidance method and system based on multi-dimensional classification. The method comprises the following steps: acquiring basic conditions of a stroke patient, and classifying disease characteristics, neurological function defect parts, severity and complications to form a classification result; executing patient group mapping based on a classification result, and establishing corresponding key monitoring related indexes; change collection and change analysis are carried out according to key monitoring related indexes, and stage rehabilitation scheme suggestions are generated; and carrying out abnormal judgment and online guidance according to the stage rehabilitation scheme suggestions, forming adjustment suggestions, and writing back the adjustment suggestions. According to the invention, the classification result is used as an organization basis throughout the whole process, so that a consistent closed-loop processing link is formed by monitoring, analysis, guidance and write-back, and the continuity, traceability and collaborative stability of the scheme generation and adjustment process in the stroke home rehabilitation guidance process are improved.

Description

Technical Field

[0001] This invention relates to the field of medical information processing, and in particular to a method and system for guiding home-based rehabilitation of stroke based on multidimensional classification. Background Technology

[0002] In the field of medical information processing, existing solutions typically focus on collecting basic information about stroke patients and key monitoring indicators. These solutions combine classification, treatment recommendations, and remote follow-up to form a closed-loop management system. However, this approach suffers from limitations such as the difficulty in creating a master index that runs throughout the entire process, the difficulty in tracing and verifying the processing standards at different stages, and the difficulty in ensuring consistency in online guidance. Existing methods largely rely on static classification or empirical rules to categorize stroke patients' basic information. The relationship between classification results and subsequent key monitoring indicators and phased rehabilitation program recommendations lacks unified constraints. Under stroke home rehabilitation guidance, inconsistencies easily arise between the establishment and processing of key monitoring indicators and the mapping of ten patient groups, and unclear input boundaries exist between phased rehabilitation program recommendations, anomaly detection, and online guidance processing. This makes it difficult to achieve stable implementation of adjustment recommendations and their rewriting. Regarding the combined processing of classification results, ten patient groups, key monitoring indicators, and phased rehabilitation program recommendations, existing technologies generally have shortcomings in areas such as the solidification and version registration of mapping rules, the structured recording of anomaly detection and online guidance processing, and the consistent maintenance of write-back status. This makes it difficult to form a continuous and consistent process in the application scenario of stroke home rehabilitation guidance, encompassing data collection, classification processing, mapping processing, change collection and analysis, anomaly detection and online guidance processing, and write-back. Consequently, it is difficult to align the verification criteria between cross-phase programs and online guidance, and it is difficult to establish a traceable link between the adjustment suggestions after write-back and the existing phased rehabilitation program recommendations. This leads to unstable process connections and insufficient auditability of records during the implementation of home rehabilitation services and medical collaborative treatment. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a stroke home rehabilitation guidance method based on multidimensional classification, comprising:

[0004] S100. Obtain the basic information of stroke patients, classify them according to disease characteristics, location of neurological deficit, severity, and comorbidities, and obtain the classification results.

[0005] S200. Based on the classification results, perform 10 patient group mapping processing to generate 10 patient groups;

[0006] S300. Based on the aforementioned 10 patient groups, key monitoring-related indicators are established and processed to generate key monitoring-related indicators;

[0007] S400. Based on the key monitoring indicators, collect and analyze changes to generate phased rehabilitation plan recommendations.

[0008] S500: Based on the aforementioned phased rehabilitation plan recommendations, perform anomaly detection and online guidance processing, generate adjustment suggestions, and write them back.

[0009] Furthermore, the process of classifying and treating disease characteristics, locations of neurological deficits, severity, and comorbidities includes:

[0010] The disease feature classification process includes enumerating and normalizing stroke types, normalizing the onset time on a time axis to generate a disease duration field, and outputting classification labels containing stroke type, onset time, recurrence information, and a description of the current rehabilitation stage.

[0011] The classification and processing of neurological deficit sites includes structured analysis, key phrase extraction and conflict resolution for four types of neurological deficits: language, motor, cognitive and psychological, and finally mapping the final phrases to corresponding deficit site classification labels.

[0012] The severity classification process includes determining the source priority and merging the severity descriptions of the doctor's terminal and the mobile terminal, and outputting the severity level or the pending confirmation mark.

[0013] Comorbidity classification processing includes structurally splitting comorbidity descriptions to generate comorbidity entries containing entry sources and confirmation status, and verifying the association between past medication information and comorbidity entries.

[0014] Furthermore, the process of mapping 10 patient groups includes:

[0015] The mapping processing for the 10 patient groups includes:

[0016] The classification results are subjected to field decomposition, standardization, and consistency checks, including source consistency checks and logical consistency checks.

[0017] The preset mapping rule set is loaded from the storage module. The mapping rule set is maintained by version number. Each rule entry consists of a combination of four types of enumerated values: disease feature classification label, neurological deficit site classification label, severity level, and comorbidity classification label.

[0018] Rule matching is performed using a two-level matching chain, which consists of: Level 1 matching, which uses the classification labels of the neurological deficit sites and severity levels in the classification results to form a combination key and selects a set of candidate rule entries from the mapping rule set; and Level 2 matching, which uses the disease feature classification labels and comorbidity classification labels in the classification results to perform cross-filtering on the set of candidate rule entries to determine the final rule entries.

[0019] When the classification result has a pending confirmation marker, a mapping confidence status field is attached to the 10 patient groups output by the mapping process.

[0020] Furthermore, the process of generating 10 patient groups includes:

[0021] Based on the final rule entries, 10 patient groups are generated. The 10 patient groups include a patient group identifier, a mapping confidence state field, and an association index field. The association index field is used to record the version number of the classification result on which the generation is based and the version number of the mapping rule set.

[0022] The generated 10 patient groups and the associated index field are written into the patient group file of the storage module to form a patient group file index.

[0023] Furthermore, the process of establishing and processing key monitoring indicators includes:

[0024] The establishment and processing of the key monitoring indicators includes:

[0025] The group identifier, mapping confidence state field, and association index field of the 10 patient groups are parsed from the patient group files.

[0026] Based on the group identifier, a set of rules for key monitoring indicators is loaded from the storage module for matching. The set of rules for key monitoring indicators is maintained in the form of a version number.

[0027] When a match is successful, a key monitoring related indicator structure is generated based on the matched rule entries. The key monitoring related indicator structure includes an indicator field list and a collection frequency configuration. The indicator field list consists of a subset of health data and training data fields corresponding to the language, motor, cognitive, and psychoneurological function deficits. The collection frequency configuration includes a collection frequency switching rule based on the training time period marker to switch between rest and exercise or rehabilitation training states.

[0028] The generated key monitoring indicator structure and the rule version number on which it is based are written into the monitoring configuration file of the storage module.

[0029] Furthermore, the process of change acquisition and change analysis includes:

[0030] The data transmission unit executes the collection scheduling according to the collection frequency configuration in the key monitoring indicators, switches the collection frequency according to the training time period mark, and performs field existence verification and field value consistency verification on the reported health data and training data, generating change collection records containing collection timestamp, data source, missing test mark and review mark, which are organized by the storage module in units of collection window;

[0031] The data processing unit reads change collection records associated with the key monitoring indicators from the storage module, performs a combined processing chain of time sequence comparison, intra-window comparison, and inter-window comparison, generates change analysis records containing change direction labels, fluctuation labels, and mutation labels, and performs isolation processing on data with missing measurement marks or verification marks.

[0032] Furthermore, the process of generating key monitoring indicators includes:

[0033] Based on the change analysis records and the patient group file index, a stage switching determination is made according to a preset set of stage switching rule entries to obtain a stage marker; based on the stage marker and the patient group file index, rehabilitation nursing, rehabilitation training, and psychological nursing suggestion entries are matched and assembled from the suggestion template library maintained by the storage module to form a stage suggestion entry set; a suggestion version record containing the patient group file index, the version number of the key monitoring related indicators, the stage marker, and the stage switching basis field is registered for the stage suggestion entry set to generate a stage rehabilitation plan suggestion.

[0034] Furthermore, the process of anomaly detection and online guidance includes:

[0035] The anomaly detection and online guidance process includes:

[0036] Anomaly Detection and Handling: The data processing unit, based on the proposed phased rehabilitation plan and its associated version records, change analysis records, and change collection records, and according to the anomaly detection conditions loaded from the storage module and bound to the patient group profile index, performs suggestion consistency verification and execution feedback verification. The suggestion consistency verification verifies the correspondence between the proposed phased rehabilitation plan and its generation basis, and the execution feedback verification verifies the consistency between the training records submitted by the mobile terminal and the rehabilitation training suggestion items in the proposed phased rehabilitation plan. When the verification fails, a structured anomaly status is generated, including an anomaly type field, an anomaly source field, an association index field, and a processing status field. The association index field includes at least the patient group profile index, the version number of the key monitoring related indicators, and the suggestion version record number. The anomaly status is then written to the storage module.

[0037] Online guidance processing: The online guidance module generates an online guidance task based on the abnormal state and pushes it to the doctor's terminal; it receives structured adjustment suggestions returned by the doctor's terminal, which include fields for adjustment content, applicable stage, applicable patient group, and review and signature.

[0038] Furthermore, the process of generating and writing back adjustment suggestions includes:

[0039] The adjustment suggestion input is merged with the phased rehabilitation plan suggestion to generate an adjustment suggestion record. The adjustment suggestion record includes an adjustment suggestion version number that is associated with the suggestion version record number, a write-back status field, and the associated index field. The adjustment suggestion record is written to the storage module, the processing status field of the abnormal status is updated, and the adjustment suggestion record is synchronously written back to the home management system and the mobile terminal through the data transmission unit.

[0040] Furthermore, a stroke home rehabilitation guidance system based on multidimensional classification includes: a data acquisition module, a data processing module, a remote monitoring module, an online guidance module, and a storage module; the modules are connected in sequence to implement the method described in any of the above-mentioned embodiments.

[0041] The key innovations of this invention include:

[0042] (1) Based on the classification results, perform 10 patient group mapping processing, organize the classification results into a unified reference object for subsequent steps, and generate 10 patient groups for key monitoring related indicator establishment processing, change collection and change analysis processing, and abnormal judgment and online guidance processing.

[0043] (2) In the abnormality judgment and online guidance processing link, the phased rehabilitation plan suggestion is set as the entry object for abnormality judgment, and after the adjustment suggestion is generated, a write-back is performed so that the adjustment suggestion and the phased rehabilitation plan suggestion form a traceable update relationship in the same link.

[0044] (3) Focus on the 10 patient groups to organize key monitoring indicators, establish processing and change collection and change analysis processing, bind the key monitoring indicators and the phased rehabilitation plan recommendations to the same patient group semantics, and form a data organization method that runs through the main process from classification results to phased rehabilitation plan recommendations.

[0045] The following are its main beneficial effects:

[0046] (1) For the home rehabilitation guidance scenario for stroke, the classification results are transformed into a unified reference object for subsequent steps through the mapping processing of 10 types of patient groups. This ensures that the establishment, processing and change collection and change analysis of key monitoring indicators follow a consistent data entry and calling standard during operation, reducing the impact of cross-step reference ambiguity on the consistency of the link.

[0047] (2) In response to the problem of unclear connection between the phased rehabilitation plan suggestions and online guidance in the existing scheme, the anomaly judgment and online guidance processing are carried out around the phased rehabilitation plan suggestions, and the suggestions are written back after the adjustment suggestions are generated, so that the online guidance is transformed from interactive output into adjustment suggestion data that can be read and reviewed by subsequent steps, reducing the link breakpoints caused by the disconnect between suggestions and guidance.

[0048] (3) In response to the problem that the key monitoring indicators and the grouping strategy are difficult to link in the existing scheme, the key monitoring indicators are established and processed based on the 10 patient groups, and the phased rehabilitation plan suggestions are generated in the change collection and change analysis processing. This makes the monitoring configuration, change analysis and plan suggestions under the same semantic constraints of the same patient group, and reduces the configuration drift caused by different processing links acting independently.

[0049] (4) After the adjustment suggestions are made and written back, the phased rehabilitation plan suggestions will have the basis for reading the adjustment suggestions in subsequent operation, so that the change collection and change analysis and processing and the anomaly judgment and online guidance processing form a closed-loop update path, which alleviates the problem of repeated judgment and repeated guidance caused by the inconsistency of writing back in the prior art.

[0050] (5) During the continuous operation of stroke home rehabilitation guidance, the classification results, the 10 patient groups, the key monitoring indicators, the phased rehabilitation plan suggestions and the adjustment suggestions are organized and transmitted along the same main process, making it easier to verify and trace the source relationship of cross-phase data objects, and reducing the instability of collaborative handling caused by the difficulty in tracing and verifying the processing scope in the existing technology. Attached Figure Description

[0051] Figure 1 A flowchart illustrating a stroke home rehabilitation guidance method based on multidimensional classification, provided in an embodiment of this application;

[0052] Figure 2 This is a structural block diagram of a stroke home rehabilitation guidance system based on multidimensional classification, provided in an embodiment of this application. Detailed Implementation

[0053] Example 1: Refer to Figure 1 This is a flowchart illustrating a stroke home rehabilitation guidance method based on multidimensional classification provided by an embodiment of the present invention. The process may include at least steps S100-S500:

[0054] S100. Obtain the basic information of stroke patients, classify them according to disease characteristics, location of neurological deficit, severity, and comorbidities, and obtain the classification results.

[0055] S200. Based on the classification results, perform 10 patient group mapping processing to generate 10 patient groups;

[0056] S300. Based on the aforementioned 10 patient groups, key monitoring-related indicators are established and processed to generate key monitoring-related indicators;

[0057] S400. Based on the key monitoring indicators, collect and analyze changes to generate phased rehabilitation plan recommendations.

[0058] S500: Based on the aforementioned phased rehabilitation plan recommendations, perform anomaly detection and online guidance processing, generate adjustment suggestions, and write them back.

[0059] S100. Obtain the basic information of stroke patients, classify them according to disease characteristics, location of neurological deficit, severity, and comorbidities, and obtain the classification results.

[0060] Specifically, this step is triggered when a stroke patient registers in the home management system or when a doctor initiates a home rehabilitation follow-up. The home management system receives basic information submitted by the stroke patient's mobile terminal and receives basic information entered or updated by the doctor's terminal. The basic information describes the stroke patient's basic information set during the home rehabilitation phase, including identification information, onset-related information, medical history information, current status information, and medical treatment information. The identification information is used to establish a unique association for the stroke patient in the home management system. Onset-related information includes the onset time, stroke type, and previous stroke history. Medical history information includes previous diagnoses and medications. Current status information includes the neurological deficits in language, motor, cognitive, and psychological aspects. Medical treatment information includes discharge instructions and rehabilitation training precautions. To facilitate the use of the basic information for classifying disease characteristics, neurological deficit locations, severity, and comorbidities in subsequent steps, this step first involves the data acquisition module processing the submissions from the mobile terminal and the doctor's terminal. This processing includes field completeness verification, field consistency verification, duplicate submission removal, and missing item marking. The field completeness verification focuses on the essential fields of the basic information, including stroke type, onset time, neurological deficit status, severity description, and comorbidity description. If missing items are found, the home management system generates a supplementary data entry task and pushes it to the mobile terminal, while simultaneously registering the missing items in the storage module. The system marks and supplements the status; verifies the consistency of field definitions by addressing the differences in expression of the same field between mobile and doctor terminals. It uses a preset field mapping table to merge synonymous expressions into a unified field value and records the value before and after mapping; it uses a time window and content hashing strategy to deduplicate duplicate submissions. When the mobile terminal submits the same content consecutively within a short period of time, it retains the last submission and registers the deduplication record in the storage module; it marks missing items to constrain the availability of subsequent classification processing. Before missing items are supplemented, the home management system restricts the classification branch in which the missing item is involved in the calculation to enter the "pending confirmation" state and retains the pending confirmation mark in the classification results. Furthermore, to ensure continuous operation and auditability during the home-based rehabilitation phase, after the access processing is completed, the storage module establishes a version record for the basic information. The version record includes the submission source, submission time, list of updated fields, and review status. The review status is confirmed by the doctor's terminal or automatically confirmed by the home management system based on consistency rules. The consistency rules are based on the degree of consistency of the same field in different sources. When the difference exceeds the preset consistency threshold, the field is marked as "requires review" and the doctor's terminal review process is triggered. At the same time, the review task and review conclusion are registered in the storage module, thereby forming a traceable source link and evolution link.

[0061] After completing the access and version recording of the basic information, this step enters the processing chain of disease feature classification, neurological deficit site classification, severity classification, and comorbidity classification. The disease features are used to characterize the disease manifestations and course elements of stroke patients, including stroke type, onset time, recurrence information, and current rehabilitation stage description; the stroke type and onset time come from the onset-related information in the basic information, the recurrence information comes from the previous stroke history in the basic information, and the current rehabilitation stage description comes from the discharge instructions or follow-up records on the doctor's terminal. The disease feature classification process is performed by the data processing unit. First, it enumerates and normalizes stroke types, aligning the natural language descriptions from the mobile terminal with the standard terminology from the doctor's terminal. Next, it normalizes the onset time along a timeline, unifying different date and time formats to the home management system's time base and generating a disease duration field. Then, it performs consistency verification on recurrence information; if the past stroke history is inconsistent with the doctor's terminal record, a review task is triggered, and a "recurrence pending confirmation" marker is temporarily stored. Finally, combining the disease duration with the current rehabilitation stage description, disease feature classification labels are output. The term "neurological deficit site" describes the corresponding location of the stroke-induced neurological deficit in the functional domain. This step does not introduce new external terms but limits the neurological deficit site to a set of deficit sites corresponding to language, motor, cognitive, and psychological functions, and restricts its source to the neurological deficit situation described in the basic case. The classification of neurological deficit sites is performed by the data processing unit, which performs structured analysis on four types of neurological deficits: language, motor, cognitive, and psychological. It aligns the assessment descriptions from the doctor's terminal with the self-assessment descriptions from the mobile terminal. The alignment process includes description segmentation, key phrase extraction, and conflict resolution. Description segmentation divides the long text into four sub-segments according to the paragraph boundaries corresponding to language, motor, cognitive, and psychological deficits. Key phrase extraction extracts a set of phrases representing the degree and manifestation of the deficit from each sub-segment and registers them as candidate phrases. Conflict resolution determines the final phrase based on the doctor's terminal priority rule when candidate phrases contradict each other, while retaining the covered phrase as a control record. Subsequently, the data processing unit maps the final phrases to neurological deficit site classification labels and combines the four types of classification labels to form the substructure of the neurological deficit site classification result.The severity level is used to describe the severity grading of the stroke patient's current state. This step limits the input of severity level to two sources: severity description from the doctor's terminal and severity description from the mobile terminal. The severity classification processing is performed by the data processing unit. It first determines the source priority, taking the doctor's terminal as the primary source and the mobile terminal as the secondary source. Then, it performs severity description merging processing, merging expressions such as "mild," "moderate," and "severe" into preset severity levels. For fields with a "requires review" status, it performs branch processing. If the review is not completed, the severity level is temporarily stored as "pending confirmation," and the triggering reason is recorded. At the same time, the data processing unit performs consistency verification between the severity level and the classification label of the neurological function deficit site. When there is a significant contradiction between the severity level and the deficit phrases of language, motor, cognitive, and psychological functions, it triggers the doctor's terminal review and writes the contradiction record into the storage module. The comorbidities are used to describe the set of other diagnoses and risk factors coexisting in stroke patients during the home rehabilitation phase. The input sources are the previous diagnoses and medications in the basic information, as well as comorbidity descriptions supplemented by the doctor's terminal. Comorbidity classification is performed by the data processing unit, which first performs structured decomposition of the comorbidity descriptions, breaking them down into several comorbidity entries, and establishing the entry source, entry time, and entry confirmation status for each comorbidity entry. Next, it performs association processing on the previous medication information, verifying the association between previous medications and comorbidity entries. If inconsistencies are found between medications and comorbidities, a review task is triggered, and the reason for the inconsistency is recorded. Subsequently, the data processing unit generates comorbidity classification labels based on the confirmation status of the comorbidity entries, and registers "pending confirmation" and "confirmed" entries separately for subsequent steps to access according to boundary constraints. In this step, during the four classification processes described above, the algorithm module performs consistency checks and rule verification on the intermediate results of the classification process according to preset rehabilitation algorithms and rules. These preset rehabilitation algorithms and rules are manifested in this step as a set of classification rules and a set of conflict resolution rules. The set of classification rules defines which classification label to output under what input combinations for disease characteristics, neurological deficit locations, severity, and comorbidities. The set of conflict resolution rules outputs a "requires review" or "pending confirmation" marker when there are inconsistencies in the source or logical contradictions. Both the set of classification rules and the set of conflict resolution rules have version numbers registered by the storage module and are associated with the version record of this step, thus forming a traceable version chain for the classification process. This avoids situations where the same input outputs inconsistent results in different versions, leading to a lack of evidence, during subsequent evolution.Understandably, in the above processing chain, disease characteristics, neurological deficit location, severity, and comorbidities constitute the minimum set of core parameters for this step. The absence of any one of these will result in the classification result being marked as "pending confirmation" and triggering supplementary recording or review. On the other hand, the field mapping table, time axis normalization rules, consistency threshold, and conflict handling rules are extended configuration items at the implementation level. They are maintained by the home management system in the storage module and evolve with version records. Different configurations are allowed in different deployment scenarios, but the field structure of the classification result remains consistent.

[0062] In terms of engineering implementation, stroke patients enter the home rehabilitation stage after discharge and complete the initial registration through a mobile terminal. The registration content includes the stroke type, onset time, language, motor, cognitive and psychological corresponding neurological function deficits, severity descriptions and comorbidity descriptions. At the same time, the doctor's terminal enters the discharge medical orders and severity descriptions during follow-up visits, and supplements the descriptions of comorbidities. The data acquisition module triggers this step when content is submitted via a mobile terminal in the home management system. First, it verifies the completeness of the fields and marks any missing items. Then, the data processing unit enumerates and normalizes stroke types, normalizes the onset time along a timeline and generates a disease duration field, performs structured analysis on language, motor, cognitive, and psychological neurological deficits and outputs classification labels for the neurological deficit sites, merges severity descriptions and outputs severity levels, and structurally splits comorbidity descriptions and outputs comorbidity classification labels. During the above output process, the algorithm module performs consistency checks and rule verification. If inconsistencies are found between the severity descriptions on the mobile terminal and the doctor's terminal, a review task is triggered, and the severity is set to "pending confirmation." If inconsistencies are found between past medications and comorbidity entries, the reason for the inconsistency is recorded, and a review is triggered. Ultimately, the home management system combines disease characteristic classification labels, neurological deficit site classification labels, severity levels, and comorbidity classification labels to form a classification result. The classification result is written into the version record of the storage module and used as the input product for subsequent steps. The classification result is sent to the "Classification Result" of S200 to perform mapping processing of 10 patient groups, supporting the closed-loop link of subsequent 10 patient groups, key monitoring indicators, phased rehabilitation plan suggestions, abnormal judgment and online guidance processing.

[0063] Summary of the technical effects of this step: This step establishes a classification and processing chain based on disease characteristics, location of neurological deficits, severity, and comorbidities. It incorporates the source, conflict, and version records of the classification processing into the same operational process, ensuring that the classification results have a traceable field structure and review boundaries. The classification results form an input-output connection with the subsequent mapping processing of 10 patient groups, supporting the establishment and processing of key monitoring indicators and the generation of phased rehabilitation plan suggestions for each patient group. The marking of missing items and the record of review tasks during the classification process form a continuously running trigger condition chain, maintaining the consistency and auditability of the home management system throughout its evolution.

[0064] S200. Based on the classification results, perform 10 patient group mapping processing to generate 10 patient groups;

[0065] Specifically, this step is triggered after the home management system completes S100 and registers the classification results in the storage module. The classification results serve as the sole input to the data processing module, which consists of a data processing unit, an algorithm module, and a storage module. The data processing unit is responsible for performing structured parsing and consistency checks on the classification results. The algorithm module is responsible for matching and conflict handling of mapping rules. The storage module is responsible for registering version records, mapping evidence, and output products of the mapping process. The classification results are used to characterize the set of classification labels for stroke patients across four dimensions: disease characteristics, neurological deficit location, severity, and comorbidities. Disease characteristics include stroke type, onset time, recurrence information, and current rehabilitation stage description. Neurological deficit location includes classification labels for deficit locations corresponding to language, motor, cognitive, and psychological aspects. Severity includes severity levels or a "pending confirmation" marker. Comorbidities include comorbidity entries and their confirmation status. Understandably, the 10 patient group mapping process described in this step is the process of converting the classification results into 10 preset patient groups. The 10 patient groups are a patient group classification structure formed by using the four classification labels of the classification results as index keys. It is used in subsequent steps to establish and process key monitoring indicators and generate phased rehabilitation plan suggestions according to the patient groups. Therefore, this step imposes constraints on the availability boundaries of the input fields. The boundary constraints include missing item marking constraints and review status constraints: when the classification results have a "pending confirmation" mark or a "requires review" status, the home management system still performs mapping processing but outputs 10 patient groups with a mapping confidence status field, and writes the mapping confidence status field into the storage module for the call boundary of subsequent steps; when the classification results are missing any one of the disease characteristics, neurological function deficit sites, severity, or comorbidities, the data processing unit puts the classification results into the "mapping pending record" status and triggers the mobile terminal to complete the record task, while temporarily storing the classification results as mapping candidate records, without entering the 10 patient group determination output link.

[0066] In terms of processing links, the data processing unit first performs field decomposition and standardization on the classification results. Field decomposition breaks down the classification results into four substructures according to four dimensions: disease characteristics, neurological deficit location, severity, and comorbidities, and extracts the classification labels and status markers in each substructure. Standardization unifies the spelling, enumeration values, and null value expressions of the classification labels, merges synonym expressions into unified enumeration values, and distinguishes and records null values ​​and missing items using missing item markers. Furthermore, to support the stable operation of subsequent mapping rules, the data processing unit performs consistency checks on the classification results. These checks include source consistency checks and logical consistency checks: Source consistency checks verify whether the classification labels for disease characteristics, neurological deficit sites, severity, and comorbidities all have corresponding version records and submission sources. If a version record or source identifier is missing, the classification result is placed in a "source missing" state and a doctor's terminal review is triggered. Logical consistency checks verify whether there are contradictory combinations between severity and neurological deficit sites. If the severity is high but the language, motor, cognitive, and psychological phrases related to the neurological deficit site all show mild descriptions, a logical contradiction is registered, and the mapping confidence status field is set to low confidence, while a doctor's terminal review is triggered. All records of the above consistency checks are written to the mapping evidence record in the storage module. The mapping evidence record includes the check time, check items, check conclusions, and review status, providing a basis for subsequent steps to trace the formation of the 10 patient groups when anomaly detection and online guidance are needed.

[0067] After completing the input structuring and consistency checks, the algorithm module proceeds to the loading and matching process of mapping rules for 10 patient groups. These 10 patient group mapping rules are a preset rule set. The preset rule set uses the "combination pattern of classification results" as the matching condition and outputs "10 patient group numbers." The combination pattern of classification results is composed of disease feature classification labels, neurological deficit location classification labels, severity levels, and comorbidity classification labels. To avoid introducing new terminology or complex models in the specification, this step implements the mapping rule set as a set of rule entries. Each rule entry includes a matching condition field, a conflict priority field, and an output field. The matching condition field consists of a combination of enumerated values ​​from four dimensions. The conflict priority field is used to determine the priority order when multiple rules are satisfied simultaneously. The output field is the group identifier for the 10 patient groups. Furthermore, the set of rule entries is maintained by the storage module using version numbers. These version numbers increment as the doctor's terminal updates the mapping rules. Each update registers the updated field list, update time, and update source in the storage module, and the updated set of rule entries is retained alongside the old version as a comparison. This ensures an auditable version chain for the mapping process during system evolution. Understandably, the minimum set of four dimensions of classification labels are essential for constituting the 10 patient groups in the mapping rule set: disease characteristic classification label, neurological deficit location classification label, severity level, and comorbidity classification label. The absence of any dimension will result in a non-closed rule match, leading to the entry into the mapping candidate record. The conflict priority field, mapping confidence state field, and mapping evidence record are extended fields used to ensure that the 10 patient groups can still be output and backtracking evidence is retained even when rules overlap or the input state is unstable.

[0068] In the rule matching phase, the algorithm module performs line-by-line matching on the four-dimensional combination patterns of the classification results. The matching process employs a two-level matching chain, starting with coarse-grained matching and then fine-grained matching. The first-level matching is used to quickly filter candidate rule entries, constructing a candidate set based on the classification labels of the neurological deficit site and the severity level. The second-level matching is used to determine the final rule entry within the candidate set by combining the disease feature classification labels and the comorbidity classification labels. The first-level matching is implemented as follows: the algorithm module reads the classification labels of the language, motor, cognitive, and psychological deficit sites and combines them with the severity level to form a deficit-severity combination key. Then, it searches the rule entry set for rule entries that match this combination key and forms a candidate rule entry set. When the candidate rule entry set is empty, the algorithm module places the classification result in a "rule missing" state and triggers the doctor's terminal to record the mapping rule, while simultaneously adding the classification result to the mapping candidate record. When there are multiple candidate rule entries, the second-level matching is initiated. The secondary matching is implemented as follows: the algorithm module extracts disease feature classification labels and comorbidity classification labels from the classification results and performs cross-filtering within the candidate rule entry set. When a unique matching entry exists, it is determined as the final rule entry. When multiple matching entries still exist, the algorithm module sorts them according to the conflict priority field and selects the entry with the highest priority. At the same time, the unselected entries are recorded as conflict entries and written to the storage module. When no matching entry exists but the candidate rule entry set is not empty, the algorithm module places the classification result in the "combination not covered" state and triggers doctor terminal review or rule update. The above triggering and recording are all part of the automation level and trigger condition chain of this step. After receiving the doctor terminal review conclusion or rule update, the home management system triggers this step again to recalculate the mapping candidate records and writes the recalculation process into the mapping evidence record, thereby maintaining the continuous operation of the mapping process.

[0069] During the output generation phase, the algorithm module generates 10 patient groups based on the final rule entries and establishes an association record between these 10 patient groups and the classification results. The field structure of these 10 patient groups includes a group identifier, a group description field, a mapping confidence state field, and an association index field. The group identifier identifies the patient group category to which the stroke patient belongs; the group description field reiterates the disease characteristics, neurological deficit location, severity, and comorbidity combination patterns corresponding to the mapping rule entries; the mapping confidence state field indicates whether the mapping is affected by states such as "pending confirmation," "requires review," or "logical contradiction"; and the association index field records the version number of the classification results and the version number of the mapping rule set. Further, the storage module writes the 10 patient groups into a patient group file and associates the mapping evidence record and conflict entry record as attachments to the patient group file. The patient group file establishes a one-to-one association with the stroke patient in the home management system and is passed as a referenced object in subsequent processes. Understandably, once the 10 patient groups are generated, they are used as key products in subsequent steps. The 10 patient groups are recorded as output field names when this step is completed, and are called as inputs in the "10 patient groups" position in the next step S300 to perform the key monitoring related indicator establishment process, thereby closing the index link of "classification results - 10 patient groups - key monitoring related indicators" during the operation.

[0070] In an engineering implementation, after stroke patients are discharged, they submit basic information via a mobile terminal, and doctors supplement the descriptions of severity and comorbidities via their terminals. The home management system completes S100, generates classification results, and writes them to the storage module. Subsequently, the home management system automatically triggers this step. The data processing unit performs field decomposition and standardization on the classification results, and executes source consistency checks and logical consistency checks. When the severity description is found to be in a "pending confirmation" state, the algorithm module still performs rule matching but sets the mapping confidence state field to low confidence and synchronously writes this state to the mapping evidence record. The algorithm module loads the current version of the rule entry set from the storage module, filters the candidate rule entry set based on the neurological deficit location classification label and severity level, and then combines the disease feature classification label and comorbidity classification label to determine the final rule entries and generate 10 patient groups. When the candidate rule entry set is empty, the home management system generates a rule supplementation task, pushes it to the doctor's terminal, and adds the classification result to the mapping candidate record. After the doctor's terminal completes the rule supplementation, this step is triggered again for recalculation and outputs the 10 patient groups. Finally, the storage module writes the 10 patient groups into the patient group file and associates the classification result version number with the rule version number. At the same time, it transmits the 10 patient groups as the output of this step to the "10 patient groups" in S300, and enters the operation link for key monitoring related indicator establishment and processing.

[0071] In summary, the technical effects of this step are as follows: This step converts the classification results into 10 patient groups and forms a traceable patient group profile under the constraints of the mapping rule set, conflict handling, and version records; when the input has a "pending confirmation" or "requires verification" status, this step outputs the mapping confidence status field and retains the mapping evidence record, so that the subsequent key monitoring related indicator establishment and processing has a clear calling boundary; this step forms an index link of "classification results - 10 patient groups" and sends the 10 patient groups into the "10 patient groups" of S300, supporting the closed-loop process of binding key monitoring related indicators according to patient groups.

[0072] S300. Based on the aforementioned 10 patient groups, key monitoring-related indicators are established and processed to generate key monitoring-related indicators;

[0073] Specifically, this step is triggered after the home management system completes S200 and registers the 10 patient groups in the patient group file. The 10 patient groups serve as the input source for this step and enter the data processing module. The data processing module still consists of a data processing unit, an algorithm module, and a storage module. The data processing unit is responsible for extracting patient group index information from the 10 patient groups and organizing a structured generation process for key monitoring-related indicators. The algorithm module is responsible for matching, conflict handling, and consistency verification of the rules for establishing key monitoring-related indicators. The storage module is responsible for registering the version records, generated evidence, and subsequent call relationships of the key monitoring-related indicators. The key monitoring-related indicators are a set of monitoring indicators that correspond one-to-one with the 10 patient groups. The set of monitoring indicators is used to collect and analyze changes in the health data and training data of stroke patients during the home rehabilitation phase, thereby generating phased rehabilitation program suggestions. Therefore, the key monitoring-related indicator establishment process in this step is limited to a "differentiated establishment process driven by the patient group index" and maintains a traceable correlation with the classification results during operation. Understandably, the minimum set of key monitoring indicators consists of four subsets of monitoring indicators corresponding to the neurological deficit sites: language, motor, cognitive, and psychological. The severity and comorbidities are considered when selecting and prioritizing these subsets. Language, motor, cognitive, and psychological indicators are established terms in this embodiment, serving as organizational dimensions for the monitoring indicator subsets. Severity and comorbidities act as constraints in the selection and frequency configuration of the monitoring indicator subsets, while disease characteristics serve as time benchmarks and stage description constraints in configuring the monitoring cycle. This step does not introduce new terminology; instead, it defines the monitoring indicators as "indicator fields that can be obtained and recorded by the home management system and doctor's terminal during follow-up and home rehabilitation," and establishes a binding relationship between "patient groups and monitoring indicator fields" through rule entries.

[0074] In terms of the processing chain, the data processing unit first performs structured parsing on the 10 patient groups. This structured parsing includes reading the group identifier, disassembling the group description field, reading the mapping confidence state field, and parsing the association index field. The group identifier serves as the primary index for subsequent rule matching. The group description field is used to trace the basis for the patient group's formation based on disease characteristics, neurological deficit locations, severity, and comorbidity patterns. The mapping confidence state field determines whether this step requires triggering a doctor's terminal review process. The association index field is used to read the classification result version record and mapping rule set version number corresponding to the patient group from the storage module. Furthermore, the data processing unit performs input boundary constraint checks during the parsing process. When the mapping confidence state field is in a low-confidence state or has a "review required" state, the home management system still generates key monitoring indicators, but registers the monitoring confidence state field in these indicators and triggers a doctor's terminal confirmation task. After the confirmation task is completed, this step is triggered again to recalculate the generated key monitoring indicators and update the version record, thus forming a closed-loop chain of automated operation and triggering conditions. The records of the aforementioned boundary constraint checks are written into the evidence generation record of the storage module. The evidence generation record includes the triggering reason, triggering time, confirmation status, and recalculation count, which serves as the basis for establishing retrospective monitoring indicators when generating subsequent phased rehabilitation plan suggestions, determining anomalies, and providing online guidance.

[0075] During the loading and matching phase of the rules for establishing key monitoring-related indicators, the algorithm module reads the set of rules for establishing key monitoring-related indicators from the storage module. This set of rules is a preset set of rule entries, with a group identifier as the matching condition and the key monitoring-related indicator structure as the output. The key monitoring-related indicator structure consists of an indicator field list, a collection frequency configuration, a collection cycle configuration, a status marker configuration, and a recording requirement configuration. The indicator field list specifies the health and training data fields that need to be collected and recorded during the home rehabilitation phase. The collection frequency configuration describes the rules for switching the data collection frequency between resting and exercise or rehabilitation training states. The collection cycle configuration describes the organization of the collection window by day or by follow-up cycle. The status marker configuration describes the registration methods for missing test markers, abnormal data markers, and review markers. The recording requirement configuration describes the recording granularity and version recording method of the data transmission unit and storage module for the collected data. Understandably, the minimum set of elements in the above structure that constitutes the core improvement of this embodiment is the indicator field list and the collection frequency configuration. These two directly determine the input data set and time organization method used for change collection and change analysis in subsequent S400. The collection cycle configuration, status marker configuration, and record requirement configuration are preferred extended functions used to improve the data continuity and traceability during the home rehabilitation phase, but they do not affect the basic conditions for the indicator field list and collection frequency configuration to serve as inputs for subsequent processing. To meet the constraint of not introducing new terminology, the indicator field list is organized in this step as "health data and training data fields corresponding to language, movement, cognition, and psychology." The health data fields come from daily records and follow-up records reported by the mobile terminal and entered by the doctor's terminal. The training data fields come from training records and training history during the home rehabilitation training process. All fields have corresponding entries in the field mapping table of the home management system and are registered with version numbers by the storage module to avoid subsequent steps being unable to call due to field name drift.

[0076] During the rule matching phase, the algorithm module retrieves corresponding rule entries based on the group identifier and outputs candidate key monitoring indicators. When a unique matching rule entry exists, the algorithm module directly generates key monitoring indicators. When multiple matching rule entries exist, the algorithm module enters the conflict handling process. This process prioritizes the neurological deficit location, severity, and comorbidities in the group description field, following a priority order: "neurological deficit location first, severity second, comorbidity constraint." First, the classification labels for the four deficit categories (language, motor, cognitive, and psychological) are compared, and the rule entry with the most consistent coverage is selected. If there are still tied rule entries, the rule entry with stricter coverage is selected based on the severity level. If there are still tied rule entries, the rule entry containing the comorbidity constraint field is selected based on the comorbidity classification label. This prioritization logic is implemented as a configurable conflict priority field in the algorithm module, and the version number is registered by the storage module for version management when updating rules on the doctor's terminal. Furthermore, when the group description field contains "pending confirmation" or "requires verification" markers, the algorithm module retains more candidate indicator fields in the conflict handling chain and sets the monitoring confidence status field to low confidence, while triggering a confirmation task on the doctor's terminal. After the confirmation task is completed, the home management system re-executes this step for the patient group, using the confirmed classification result version record as the input source, thereby converging uncertainty within a traceable confirmation chain. If the algorithm module does not find a matching rule entry, it places the patient group in a "rule missing" state and triggers the doctor's terminal to supplement the key monitoring-related indicators to establish rules, while simultaneously registering the rule missing record and supplementation status in the storage module. After the supplementation is completed, the home management system automatically triggers recalculation and updates the version record of the key monitoring-related indicators, forming a continuously operating automated closed loop.

[0077] Regarding the practical path for generating key monitoring indicators, this step publicly discloses the generation of the indicator field list and the generation of the collection frequency configuration as core action links. For the indicator field list, the algorithm module outputs four subsets of indicator fields—language, movement, cognition, and psychology—based on the patient group index, and establishes a field source tag for each subset. The field source tag distinguishes between fields reported by mobile terminals and fields entered by doctors' terminals. Based on the field source tag, the data processing unit filters the field availability. The field availability filtering is based on whether the data collection module and remote monitoring module bound to the home management system have a collection path. When a field lacks a collection path in the current deployment scenario, the field is set to "pending binding" and a configuration task is triggered. The configuration task is pushed by the home management system to the operation and maintenance configuration interface or the doctor's terminal configuration interface. This step is recalculated after the data collection module binding is completed. For the data acquisition frequency configuration, the algorithm module reads the data acquisition frequency switching rules corresponding to the rest state, exercise, or rehabilitation training state from the rule entries and implements them as an executable data acquisition scheduling configuration in the data processing unit. The determination of the rest state and exercise or rehabilitation training state is derived from the time period markers of the training records and training history on the mobile terminal. When the training record shows entry into a training time period, the data acquisition scheduling configuration switches to a higher data acquisition frequency; when the training record shows exit from a training time period, the data acquisition scheduling configuration switches to a lower data acquisition frequency. To avoid introducing new terminology, this step treats the training records and training history as existing record structures in the home management system. These are submitted by the mobile terminal and registered as version records by the storage module. The algorithm module forms training time period markers by reading the start and end times from the training records and writes these training time period markers as trigger conditions for the data acquisition frequency switching rules into the data acquisition frequency configuration. Furthermore, when the mobile terminal does not submit training records or the training history has missing test markers, the data acquisition scheduling configuration enters the default data acquisition frequency and writes the missing test markers into the status marker configuration of key monitoring indicators, thereby imposing boundary constraints on the input quality of change acquisition and change analysis in subsequent S400 steps.

[0078] Regarding record and version management, the storage module establishes version records for the key monitoring indicators generated in this step. These records include the patient group profile index, rule version number, generation time, generation source, and recalculation count. Generation evidence records, rule missing records, conflict entry records, and confirmation task records are also associated with these version records. Understandably, this step's version management strategy employs a dual-index approach of "patient group profile index and rule version number," allowing new versions of key monitoring indicators generated after rule updates for the same patient group to coexist with older versions and maintain a clear evolution path. When the S400 subsequently collects and analyzes changes, the home management system prioritizes the latest version of the key monitoring indicators while retaining a retrospective entry point for older versions. This facilitates tracing the input basis for previous phased rehabilitation plan recommendations during anomaly detection and online guidance processing. Furthermore, to adapt to the continuous operation of the home-based rehabilitation phase, this step triggers automatic recalculation when an update to the patient group profile is detected. The triggering conditions for the update of the patient group profile include the update of the classification result version record, the update of the mapping rule set version number, and the confirmation of task completion by the doctor's terminal. After the trigger, the home management system re-executes this step and generates a new version of the key monitoring related indicators, so that the key monitoring related indicators maintain a consistent index relationship with the 10 patient groups.

[0079] In an engineering implementation, stroke patients are mapped to a specific patient group during the home rehabilitation phase, forming a patient group profile. The home management system automatically triggers this step after the follow-up record is submitted in the nighttime batch processing window or on the doctor's terminal. The data processing unit reads the group identifier, group description field, mapping confidence state field, and association index field from the patient group profile and retrieves the corresponding classification result version record from the storage module. The algorithm module loads key monitoring indicators to establish a rule set and matches the group identifier, outputting four categories of indicator fields: language, movement, cognition, and psychology. It then filters and prioritizes the indicator field subsets based on severity and comorbidities. The data processing unit organizes the filtered indicator field list and the collection frequency switching rules into a collection scheduling configuration and synchronizes the collection scheduling configuration to the data transmission unit in the data acquisition module and the remote monitoring module, enabling subsequent changes in collection frequency to automatically switch according to the training time period marking. When the mobile terminal fails to submit training records, resulting in a missing test in the training time period marking, the collection scheduling configuration enters the default collection frequency and registers the missing test mark in the status marking configuration. Finally, the storage module writes the key monitoring indicators into the monitoring configuration file and registers the version record. At the same time, it records the key monitoring indicators as the output field name of this step in the system and sends the key monitoring indicators to the "Key Monitoring Indicators" of S400 as input, entering the change acquisition and change analysis processing link.

[0080] In summary, the technical effects of this step are as follows: This step establishes key monitoring indicators corresponding to the 10 patient groups and incorporates the indicator field list and collection frequency configuration as the core generation link into the same operation process, forming a differentiated monitoring configuration driven by the patient group index; this step writes the generation basis and evolution process of the monitoring configuration into the storage module through rule entries, conflict handling, and version records, so that subsequent change collection and change analysis processing have a clear input structure and backtracking basis; the key monitoring indicators generated in this step are entered as the output product into the "Key Monitoring Related Indicators" of S400, supporting the continuous closed-loop operation of phased rehabilitation plan suggestion generation, abnormal judgment, and online guidance processing.

[0081] S400. Based on the key monitoring indicators, collect and analyze changes to generate phased rehabilitation plan recommendations.

[0082] Specifically, this step is triggered after the home management system completes S300 and registers the key monitoring indicators in the monitoring configuration file. The key monitoring indicators serve as the input source for this step and enter the collaborative link between the remote monitoring module and the data processing module. The remote monitoring module includes a data transmission unit, a remote access unit, and an online guidance module, while the data processing module includes a data processing unit, an algorithm module, and a storage module. Understandably, the change acquisition and change analysis processing in this step consists of two tightly coupled operational links. The first operational link, on the data transmission unit side, completes data acquisition, time organization, missing data marking, and record storage around the key monitoring indicators, forming a change acquisition record that can be analyzed and retrieved. The second operational link, on the data processing unit and algorithm module side, performs change analysis and stage switching determination around the change acquisition record, generating a staged rehabilitation plan suggestion bound to the 10 patient groups, and writes the staged rehabilitation plan suggestion into the storage module for continued retrieval by S500. In this step, the proposed phased rehabilitation plan is defined as a "set of phased recommendations." This set of recommendations consists of three categories: rehabilitation nursing, rehabilitation training, and psychological nursing. Each category of recommendations has structured fields that can be displayed in the home management system and reviewed by the doctor's terminal. Furthermore, each recommendation has a traceable correlation index with the key monitoring indicators, allowing subsequent anomaly detection and online guidance to trace back to the basis for the recommendations.

[0083] Regarding the input sources for the change acquisition chain, the key monitoring indicators include a list of indicator fields and a collection frequency configuration. The indicator field list defines the health data fields and training data fields that need to be collected and recorded, and the collection frequency configuration defines the switching rules for data collection frequency between resting and exercise or rehabilitation training states. The health data fields originate from daily records and follow-up records reported by mobile terminals and entered by doctors' terminals. The training data fields originate from training records and training history submitted by mobile terminals. The resting state and exercise or rehabilitation training state are determined by the training time period markers in the training records. Furthermore, to ensure stable operation of the change acquisition link in a home environment, this step uses "acquisition timestamp, data source, data verification, missing data marker, and review marker" as the minimum set of fields for the change acquisition record. The acquisition timestamp is used to establish the time sequence in subsequent change analysis; the data source is used to distinguish between the input source from the mobile terminal and the doctor's terminal; the data verification is used to check the format and range of the reported data; the missing data marker is used to indicate situations where no data was obtained within a certain acquisition cycle; and the review marker is used to indicate situations where the doctor's terminal requests a review of the data's authenticity or consistency. These minimum set of fields are generated by the data transmission unit at the time of acquisition and written to the storage module. They are essential core parameter sets for generating the recommended phased rehabilitation plan in this step. The acquisition cycle configuration, status marker configuration, and record requirement configuration are preferred extended fields that can be used to refine the acquisition window and supplementary recording process, but they do not affect the basic conditions for the minimum set of fields to constitute a change acquisition record.

[0084] Regarding the processing flow of the variable acquisition link, the data transmission unit executes acquisition scheduling according to the acquisition frequency configuration and organizes multiple reporting records within the same acquisition cycle by time. Specifically, the data transmission unit switches the acquisition frequency according to the training time period marker in the training record. When the training time period marker is within the training time period, the data transmission unit requests the mobile terminal to report the health data field and training data field corresponding to the indicator field list at a higher acquisition frequency. When the training time period marker is outside the training time period, the data transmission unit acquires the health data field at a lower acquisition frequency and only retains a summary record of the training history for the training data field. Further, the data transmission unit performs data verification processing on each reporting record. The data verification processing includes field existence verification and field value consistency verification. Field existence verification is used to verify whether the fields in the indicator field list appear in the reporting record. Missing fields are marked with a missing test mark and trigger a mobile terminal supplementary recording reminder. Field value consistency verification is used to verify whether there is a contradiction in repeated reporting near the same acquisition timestamp. If a contradiction exists, a review mark is registered and the contradiction record is pushed to the remote access unit of the doctor's terminal for review. Understandably, the above-mentioned supplementary recording reminder and review push belong to the automated trigger condition chain of this step. The trigger condition for supplementary recording reminder is that the missing test mark appears and the data has not been recorded within the preset supplementary recording window. The trigger condition for review push is that the field value consistency check fails or there is an inconsistency with the follow-up record. After the trigger, the system continues to perform data collection scheduling, but in the subsequent change analysis, the data with the review mark is downweighted or isolated, and the isolation record is written to the storage module for easy retrospective by the online guidance module.

[0085] To transform "change acquisition" into input that can be analyzed and invoked, this step creates change acquisition records in the storage module. These records are organized using the patient group profile index and indicator field list as primary keys, and sorted by the acquisition timestamp as a time-series record structure. Specifically, the storage module adds a related index field to each record submitted by the data transmission unit. This related index field includes the patient group profile index, the version number of the key monitoring indicator, and the acquisition frequency configuration version number, thus enabling the data collected from the same patient after the key monitoring indicator version is updated to distinguish between versions and avoid mixing. Furthermore, the storage module triggers an acquisition window closing operation at the end of each day or after the follow-up record is submitted. This operation solidifies the change acquisition records for that day or the current follow-up period into "acquisition window records," and records the start time, end time, summary of missing test markers, and summary of review markers for each acquisition window record. When a key monitoring indicator version is updated, the storage module archives the old version of the acquisition window records and opens a new acquisition window record for the new version, thereby forming a versioned and auditable acquisition chain. Understandably, the collection window record belongs to the preferred extended structure, but its versioning and sealing strategy and associated index fields are the key foundation for the stable operation of subsequent change analysis, and are one of the essential core operating mechanisms for generating phased rehabilitation plan suggestions in this step.

[0086] Regarding the input sources for the change analysis process, the data processing unit reads the change collection records corresponding to the key monitoring indicators from the storage module, and simultaneously reads the patient group file index and the version number of the key monitoring indicators, thus strictly limiting the change analysis to the same patient group and the same monitoring indicator version. The change analysis process uses "change" as the analysis object. Change refers to the differentiated performance of the same indicator field within adjacent collection timestamps or the same collection window. Change can manifest in three basic forms: trend change, fluctuation change, and abrupt change. Trend change describes a continuous increase or decrease; fluctuation change describes repeated fluctuations within an interval; and abrupt change describes a significant deviation within a short period. To avoid introducing new computer terminology, this step implements the change analysis as a combined processing link of "time-sequence comparison, intra-window comparison, and inter-window comparison": time-sequence comparison calculates the direction of change between adjacent collection timestamps and registers direction labels; intra-window comparison compares the maximum, minimum, and current values ​​of the indicator field within the same collection window record and registers fluctuation labels; and inter-window comparison compares window summaries between adjacent collection window records and registers trend labels. Furthermore, the change analysis process employs boundary constraint strategies when encountering missing and verification markers. These strategies include missing isolation and verification isolation: Missing isolation means that during the time period covered by the missing marker, no change direction label is output; instead, the missing status is output and written to the change analysis record. Verification isolation means that data with verification markers are not included in the window summary calculation, and the verification isolation record is written to the change analysis record. Recalculation is triggered only after the doctor's terminal has completed the verification. The triggering conditions and recording methods of the above boundary constraint strategies are all completed by the data processing unit and written to the storage module, which is one of the key mechanisms for the stable operation of this step.

[0087] In terms of stage switching determination and generation of staged rehabilitation plan suggestions, the algorithm module reads change analysis records and combines them with the index relationships of the 10 patient groups to complete the generation chain of "stage switching determination - suggestion item selection - suggestion item assembly - suggestion version recording". Understandably, stage switching determination in this step is limited to "a determination process based on change analysis records and patient group indexes". This determination process is implemented by a preset set of stage switching rule items. The set of stage switching rule items uses group identifiers and change labels of indicator fields as matching conditions, and outputs stage markers. The stage markers describe the current rehabilitation stage and are bound to the set of staged suggestion items. To avoid introducing new terminology, the stage markers in this step are represented as an enumeration of values ​​such as "stage one, stage two, stage three", and the version number is registered by the storage module. The doctor's terminal can update the set of stage switching rule items and form a rule version chain. Furthermore, when determining the stage switch, the algorithm module prioritizes using the change labels of indicator fields that are strongly correlated with the patient group index. The strong correlation means that the indicator field belongs to the subset of the four categories of indicator fields (language, motor, cognitive, and psychological) that are consistent with the defect location of the patient group. For example, when the patient group file shows that motor deficit is the main defect location, the algorithm module prioritizes using the trend label and change label of the motor indicator field to participate in the stage switch determination. When the change label is unavailable due to missing test isolation or verification isolation of the strongly correlated indicator field, the algorithm module uses the fluctuation label of other indicator fields in the same acquisition window record as a substitute and registers the substitute source in the change analysis record for subsequent online guidance module to backtrack.

[0088] Regarding the generation structure of phased rehabilitation plan recommendations, the algorithm module selects recommendation items from the recommendation template library based on phase markers and patient group indexes and assembles them into a set of phased recommendation items. The recommendation template library is maintained by the storage module and managed by version numbers. The library contains three types of item sets: rehabilitation nursing recommendation items, rehabilitation training recommendation items, and psychological nursing recommendation items. Each recommendation item includes a recommendation content field, an applicable phase field, an applicable patient group field, and a review status field. Understandably, the minimum essential set for generating phased rehabilitation plan recommendations in this step consists of the recommendation content field and the applicable phase field. The recommendation content field is used for display in the home management system and for review on the doctor's terminal. The applicable phase field is used to match the phase markers. The applicable patient group field and the review status field are extended fields used to limit the scope of application and mark the review process when the recommendation template library is reused across patient groups. Specifically, when selecting suggestion items, the algorithm module first matches the applicable stage field with the current stage marker, and then matches the applicable patient group field with the group identifier within the same stage item set to obtain a candidate suggestion item set. If the candidate suggestion item set is empty, the doctor's terminal is triggered to supplement the suggestion template, and this situation is written to the suggestion missing record in the storage module. If there are multiple candidate suggestion items, they are sorted according to the defect location label and severity constraint field in the suggestion items, and priority is given to selecting items that are consistent with the defect location of the patient group and have stricter severity constraints. The unselected items are registered as conflict item records. Further, to be consistent with the description of "providing rehabilitation plan suggestions in stages" in this embodiment, the algorithm module registers the "stage start time" and "stage switching basis" fields for the assembled stage suggestion item set. The stage switching basis field references the change label summary in the change analysis record, which is used as the basis for subsequent anomaly judgment and online guidance processing to trace the generation of suggestions. After the set of phased recommendations is generated, the storage module establishes a recommendation version record for it. The recommendation version record includes the patient group file index, the version number of key monitoring indicators, the phase mark, the generation time and the review status. The phased rehabilitation plan recommendations are written into the home management system for display on mobile terminals. At the same time, the review task is synchronized to the remote access unit of the doctor's terminal. After the review task is completed, the review status field is updated in the storage module and the necessary recalculation of the set of recommendations is triggered in this step.

[0089] Regarding automation and triggering conditions, the change acquisition link in this step is automatically executed by the data transmission unit according to the acquisition frequency configuration. Triggering conditions include changes in training time period markers, submission of follow-up records, and closure of the acquisition window. The change analysis link is triggered by the data processing unit when the acquisition window closes or a preset analysis cycle is reached. Triggering conditions include completion of acquisition window recording, changes in review marker status, and updates to the versions of key monitoring indicators. The stage switching determination and generation of staged rehabilitation plan suggestions are triggered by the algorithm module after the change analysis record is formed. Triggering conditions include updates to the change tag summary and changes in stage markers. To maintain auditability during system evolution, the storage module jointly registers the acquisition frequency configuration version number, the key monitoring indicator version number, the stage switching rule version number, and the suggestion template library version number, and writes the set of input versions for each run into the run record. This allows the S500 to accurately locate the rule version and acquisition data version on which the suggestion generation is based when performing anomaly determination and online guidance processing. Understandably, the above joint version registration is a key mechanism for the stable operation of this step and an indispensable component of the core link of "patient group-bound staged suggestion generation."

[0090] In the engineering implementation, stroke patients submit daily records via mobile terminals during the home rehabilitation phase and training records after training. Doctors supplement follow-up records during follow-up visits and process review push notifications. The home management system, based on the collection frequency of key monitoring indicators, requests health data fields and training data fields from the mobile terminal at a higher frequency during the training period, and collects health data fields at a lower frequency outside the training period while retaining a training history summary. The data transmission unit performs field existence checks and field value consistency checks on the reported records. If a field is missing, a missing test marker is registered and a supplementary recording reminder is pushed; if contradictory records are found, a review marker is registered and pushed to the doctor's terminal. The storage module stores the daily data as a collection window record and triggers change analysis at night. The data processing unit performs time-series comparison, intra-window comparison, and inter-window comparison to generate change analysis records, and implements missing test and review isolation for missing test and review markers. The algorithm module reads the change analysis records and outputs stage markers according to the stage switching rule item set. Then, it selects rehabilitation nursing, rehabilitation training, and psychological nursing suggestion items from the suggestion template library and assembles them into a stage suggestion item set. The storage module registers the suggestion version record and synchronizes it to the home management system and doctor terminal. Finally, the stage rehabilitation plan suggestion generated in this step is recorded in the system as an output field name and is called as the input of the S500's "stage rehabilitation plan suggestion" to enter the operation link of anomaly judgment and online guidance processing.

[0091] In summary, the technical effects of this step are as follows: This step transforms key monitoring indicators into actionable change collection records, and forms traceable change analysis records under the boundary constraints of missing test markers and review markers; under the constraints of the patient group index and the set of stage switching rule entries, this step outputs stage markers, which are then assembled from the suggestion template library to form staged rehabilitation plan suggestions, ensuring that the generated suggestions are consistent with the patient group and monitoring data version; the staged rehabilitation plan suggestions output in this step are entered into the "Staged Rehabilitation Plan Suggestions" of S500, supporting the subsequent abnormal judgment and online guidance to form a continuous closed loop.

[0092] S500: Based on the aforementioned phased rehabilitation plan recommendations, perform anomaly detection and online guidance processing, generate adjustment suggestions, and write them back;

[0093] Specifically, this step is triggered after the home management system completes S400 and the phased rehabilitation plan suggestion is generated in the storage module. The phased rehabilitation plan suggestion serves as the direct input to this step, entering the collaborative link between the data processing module and the remote monitoring module. Simultaneously, the storage module provides the patient group file index, key monitoring indicator version numbers, collection window records, change analysis records, and operation records associated with the phased rehabilitation plan suggestion, ensuring that anomaly detection and online guidance processing share the same associated index caliber. Understandably, the anomaly detection in this step is limited to the process of "verifying the consistency between the formation basis and execution feedback of the phased rehabilitation plan suggestion and outputting an anomaly status." The anomaly status is an intermediate product of this step. The anomaly status establishes a correspondence between the patient group file index and the phased rehabilitation plan suggestion and is used to drive task generation in the online guidance module, review interaction on the doctor's terminal, and the closure of the write-back action. In this step, the online guidance process is defined as "the process of doctor terminal interaction and adjustment suggestions based on the abnormal state and the phased rehabilitation plan suggestions". The adjustment suggestions are the final output of this step. The adjustment suggestions are written back to the home management system and the version number is registered in the storage module, so that the subsequent loop execution of the change analysis of S400 and the phased rehabilitation plan suggestion generation link can obtain a consistent input.

[0094] Regarding input sources and prior references, the proposed phased rehabilitation plan is generated by S400 and written into the storage module in the form of a set of phased suggestion items. The set of phased suggestion items includes rehabilitation nursing suggestion items, rehabilitation training suggestion items, and psychological nursing suggestion items. The suggestion version record registers the patient group file index, the version number of key monitoring indicators, the phase mark, the generation time, and the review status. The acquisition window record and the change analysis record are also formed by the change acquisition and change analysis processing of S400 and stored in the storage module. The change analysis record includes change direction label, fluctuation label, mutation label, missing status, and review status. The phase mark and the change analysis record have a reference relationship and together constitute the basis for suggestion generation. Furthermore, this step also reads the anomaly judgment conditions bound to the patient group file index from the storage module. The anomaly judgment conditions are registered using the same index caliber as the 10 patient groups. The content consists of "indicator field range, change label combination, stage marker range, missing test status constraint, review status constraint, and follow-up record constraint". Among them, the indicator field range limits the subset of key monitoring-related indicators covered by the anomaly judgment, the change label combination limits the combination of change direction label, fluctuation label and mutation label under which the abnormal status is output, the stage marker range limits the applicable stage of the anomaly judgment, the missing test status constraint and the review status constraint limit the judgment path when the missing test status and the review status occur, and the follow-up record constraint limits the verification relationship between the follow-up records entered by the doctor terminal and the daily records reported by the mobile terminal. The above field set is an indispensable minimum set for anomaly judgment. The minimum set is loaded by the data processing unit at runtime and the version number is registered in the runtime record to facilitate the subsequent retrospective online guidance processing based on the judgment caliber.

[0095] In terms of the anomaly detection process, the data processing unit performs two types of processing on the phased rehabilitation plan recommendations: recommendation consistency verification and feedback verification. The verification results are then grouped into the anomaly status. Specifically, the recommendation consistency verification checks the correspondence between the phased rehabilitation plan recommendations and their generation basis. The data processing unit extracts the patient group profile index, phase marker, and key monitoring indicator version numbers from the recommendation version record. It also extracts the change direction label, fluctuation label, and mutation label corresponding to the phase marker from the change analysis record. Subsequently, it matches the change label combination with the phase marker range based on the anomaly detection conditions. If the match is not found, the anomaly status is registered and written to the storage module. If the match is found, the data processing unit continues to verify whether the recommendation content fields in the phased recommendation item set are consistent with the recommendation template library version corresponding to the patient group profile index. If the recommendation template library version is inconsistent with the recommendation version record, the anomaly status is registered and the source of "version inconsistency" is marked in the operation record. The execution feedback verification verifies the training records and daily records submitted by the mobile terminal. The data processing unit extracts the training history summary and review status from the collection window records and aligns the training history summary with the training frequency and training time fields in the rehabilitation training suggestion items. When the training history summary shows missing training records and the missing test status in the collection window records meets the missing test status constraints, the data processing unit registers the abnormal status and triggers a supplementary recording reminder on the mobile terminal. When the training history summary shows that the training frequency and training time are inconsistent with the suggestion content fields and the review status meets the review status constraints, the data processing unit registers the abnormal status and triggers a review task on the doctor's terminal. Understandably, the missing test status and review status do not directly replace abnormal judgment, but rather serve as boundary constraints entering the abnormal judgment conditions. After loading the abnormal judgment conditions, the data processing unit first determines whether the missing test status constraints and review status constraints are met, and then enters the matching path of change label combination and stage marker range to avoid unclear sources of abnormal status due to missing test status or review status.

[0096] Regarding the structure for generating abnormal states, the storage module performs structured registration of abnormal states output by the data processing unit. These abnormal states include an abnormality type field, an abnormality source field, an association index field, and a processing status field. The abnormality type field distinguishes between version inconsistencies, mismatched change tags, inconsistent training histories, inconsistent follow-up records, and data entry timeouts. The abnormality source field records a summary of the change analysis record or the collection window record that triggered the abnormality determination. The association index field includes the patient group file index, the version number of key monitoring indicators, the suggested version record number, and the collection window record number. The processing status field indicates whether the abnormal state has entered online guidance processing and whether adjustment suggestions have been generated. This set of fields constitutes the minimum essential set for generating online guidance processing tasks. The storage module synchronously writes the running record when registering abnormal states, enabling the online guidance module to retrieve pending abnormal states based on the processing status field and form a task queue on the doctor's terminal. In contrast, the abnormality display summary and mobile terminal prompt text are extended fields used to improve terminal presentation consistency but do not affect the basic logic of abnormal states driving online guidance processing.

[0097] Regarding the processing chain of online guidance, the online guidance module of the remote monitoring module generates online guidance tasks based on the abnormal status and distributes them to the doctor's terminal through the remote access unit. Simultaneously, the phased rehabilitation plan suggestions, change analysis record summaries, and collection window record summaries related to the task are displayed as task attachments. Specifically, the online guidance module retrieves abnormal statuses when polling the processing status field or under event triggering conditions during abnormal status writing. When the abnormal type field is version inconsistency or change tag mismatch, the online guidance module pushes a "suggestion review task" to the doctor's terminal. The task attachment includes the suggestion version record, suggestion template library version number, phase marker, and change tag summary. When the abnormal type field is training history inconsistency or data entry timeout, the online guidance module pushes a "training record review task" to the doctor's terminal. The task attachment includes a training history summary, missing test status, and review status. When the abnormal type field is follow-up record inconsistency, the online guidance module pushes a "follow-up record review task" to the doctor's terminal. The task attachment includes a follow-up record summary and the corresponding collection window record summary. The doctor's terminal checks the task attachments and enters processing opinions in the remote access unit interface. In this step, the processing opinions are limited to "adjustment suggestion input". The adjustment suggestion input includes adjustment content field, applicable stage field, applicable patient group field, and review and signature field. The adjustment content field points to the modification content of the proposed phased rehabilitation plan. The applicable stage field and applicable patient group field are used to limit the scope of application of the adjustment suggestion. The review and signature field is used to mark the time and identity of the doctor's terminal in completing the online guidance processing. The above set of fields is the minimum set that is indispensable for generating adjustment suggestions. If the doctor's terminal only enters the adjustment content field and does not enter the applicable stage field or applicable patient group field, the online guidance module writes the stage mark and patient group file index as default values ​​into the adjustment suggestion input and marks the source as "default inheritance" in the running record.

[0098] Regarding the generation and write-back of adjustment suggestions, the data processing module receives the adjustment suggestion input submitted by the doctor's terminal and generates the adjustment suggestion. The adjustment suggestion forms an adjustment suggestion record in the storage module and completes the write-back closure. Specifically, the algorithm module performs difference merging processing on the adjustment suggestion input and the phased rehabilitation plan suggestion. The difference merging processing includes three types of operations: suggestion item replacement, suggestion item deletion, and suggestion item order reordering. Suggestion item replacement is used to adjust the content field to cover the original suggestion content field. Suggestion item deletion is used to remove suggestion items marked as inapplicable by the doctor's terminal from the phased suggestion item set. Suggestion item order reordering is used to place suggestion items marked as priority execution by the doctor's terminal in the first position of display. After the difference merging processing is completed, an adjustment suggestion record is generated. The adjustment suggestion record includes an adjustment suggestion version number, an association index field, an adjustment content field, a review and signature field, and a write-back status field. The association index field inherits the association index field of the abnormal status and appends the suggestion version record number. The write-back status field is used to indicate whether the adjustment suggestion has been written into the home management system and synchronized to the mobile terminal. Furthermore, when writing adjustment suggestion records, the storage module increments the suggestion version record, establishes a parent-child association between the adjustment suggestion version number and the original suggestion version record number, and updates the processing status field of abnormal states to "processed," causing the online guidance module to remove the corresponding task from the task queue. Subsequently, the data transmission unit synchronizes the adjustment suggestion records to the home management system and the mobile terminal. After receiving the adjustment suggestions, the mobile terminal binds them to the training record entry, enabling subsequent training record submissions and daily record reports to reference the adjustment suggestion version number. This allows the S400's change collection and change analysis processing to read the new suggestion criteria and review signature fields. Understandably, the adjustment suggestion version number, parent-child association, and write-back status field constitute the minimum set essential for write-back closure, supporting the suggestion evolution and retrospective analysis of the same patient group's file index in subsequent cycles. The terminal display template and prompt text are extended fields used for unified interactive presentation but do not change the core operational chain of write-back closure.

[0099] Regarding automation and triggering conditions, the anomaly detection in this step is triggered by the data processing unit when the suggested version record is written, the acquisition window record is closed, or the review status changes. The triggering conditions correspond to the completion of suggestion generation, the end of the data acquisition cycle, and the storage of review results, respectively. The online guidance processing is triggered by the online guidance module when an anomaly status is written or when a pending anomaly status is polled. The triggering condition corresponds to the occurrence of an anomaly status. The adjustment suggestion generation and write-back are triggered by the algorithm module when the doctor submits the adjustment suggestion input on the terminal. The triggering condition corresponds to the completion of online guidance processing. The storage module jointly registers the anomaly detection condition version, the suggestion template library version, the suggestion version record number, and the adjustment suggestion version number, and writes the input version set and output version set of each run into the run record. This ensures that subsequent loops of S300's key monitoring related indicator establishment processing and S400's change acquisition and change analysis processing can read the suggestion caliber consistent with the adjustment suggestion version number, thereby forming a continuous closed-loop operation trajectory under the same patient group file index. The adjustment suggestions output in this step are not the final output, but are written back to the home management system to enter the next round of monitoring and analysis. The adjustment suggestions are recorded in the system as output field names and are called as the reference for the subsequent execution of the "key monitoring related indicators" in S400 and the review basis for the "phased rehabilitation plan suggestions" in S500, so that the anomaly judgment and online guidance processing form a cross-step connection with the previous steps.

[0100] In an engineering implementation, stroke patients receive phased rehabilitation plan suggestions from the home management system via a mobile terminal in a home environment and conduct rehabilitation training. After training, the mobile terminal submits training records and daily records at fixed times each day. During follow-up visits, the doctor's terminal submits follow-up records and processes review tasks pushed by the online guidance module. Once the data collection window is closed, the data processing unit matches and verifies the phased rehabilitation plan suggestions with the change analysis record summary based on anomaly detection criteria. If a change tag mismatch occurs and the review status is "not reviewed," the storage module registers the anomaly, and the online guidance module generates a suggestion review task and pushes it to the doctor's terminal. The doctor's terminal views the task attachments and enters adjustment suggestions in the remote access unit. The algorithm module performs suggestion item replacement and rearrangement on the original phased suggestion item set to generate adjustment suggestion records. The storage module registers the adjustment suggestion version number and updates the anomaly status processing status field. The data transmission unit synchronizes the adjustment suggestion records to the home management system and the mobile terminal. Subsequently, the mobile terminal continues to submit training records and daily records under the new recommendation criteria. In the next acquisition window, the S400 reads the adjustment recommendation version number and uses it as the associated index for recommendation generation, so that subsequent anomaly judgment and online guidance processing continue the record chain under the same patient group file index.

[0101] In summary, this step establishes a consistency verification link between the phased rehabilitation plan recommendations and their underlying data, constrained by anomaly detection conditions. It also accumulates correlation indexes and source summaries in abnormal states. The online guidance module generates doctor terminal tasks and adjustment suggestions based on these abnormal states, ensuring consistency between the adjustment suggestions and phase markers and patient group profile indexes. The adjustment suggestions complete the write-back closure through version numbers and write-back status fields, becoming the input for subsequent cyclical steps, thus providing a traceable closed-loop operational trajectory for the entire process.

[0102] Example 2: Figure 2 This diagram illustrates a structural block diagram of a stroke home rehabilitation guidance system based on multidimensional classification, according to an embodiment of the present invention. Figure 2 As shown, the structure may include:

[0103] Data acquisition module 01 is used to acquire the basic information of stroke patients and collect changes in key monitoring indicators. This data acquisition module is connected to the data processing module. Specifically, the data acquisition module receives registration and follow-up information of stroke patients in the home management system to form the basic information of the stroke patients. Under the constraints of key monitoring indicators output by the data processing module, it completes the implementation of data collection object binding, data collection period registration, and data collection channel switching. Data collection object binding establishes a one-to-one correspondence between the basic information of the stroke patients and the key monitoring indicators. Data collection period registration records the start and end times of each data collection session. The aforementioned acquisition channel switching is used to perform priority selection between different sources corresponding to the same key monitoring-related indicator. During the acquisition process, the data acquisition module performs integrity verification and missing data marking on the acquired values. The missing data marking is associated with the corresponding time period record and the changes of the acquired key monitoring-related indicator are organized into change acquisition records according to the acquisition time period. The change acquisition records serve as the carrier of the changes of the key monitoring-related indicator and are transmitted to the data processing module. The data processing module calls the change acquisition records accordingly to complete the change analysis and processing. The data acquisition module also retains the source identifier and time period record of the change acquisition records for storage and retrieval by the storage module.

[0104] Data processing module 02 is used to classify the basic information of the stroke patients according to disease characteristics, neurological deficit location, severity, and comorbidities to obtain classification results. Based on the classification results, it performs ten-category patient group mapping processing to generate ten patient groups. Based on the ten patient groups, it performs key monitoring related indicator establishment processing to generate key monitoring related indicators. Based on the key monitoring related indicators, it performs change collection and change analysis processing to generate phased rehabilitation plan suggestions. The data processing module is connected to the remote monitoring module. Specifically, the data processing module receives the basic information of the stroke patients from the data collection module and sequentially performs field extraction, standardization, and classification in the classification processing chain. The rule matching process involves several steps. First, field extraction extracts disease characteristics, neurological deficit locations, severity, and comorbidities from the stroke patient's basic information and forms structured entries. Second, standardization maps and merges different expressions of the same field to maintain a consistent set of values. Third, classification rule matching compares each structured entry with a preset classification rule and outputs the classification result. Fourth, the data processing module performs a ten-category patient group mapping process after obtaining the classification result. This process includes mapping rule selection, conflict arbitration, and mapping registration. Mapping rule selection determines the corresponding mapping rule item based on the field combination of the classification result. Conflict arbitration is used to... When multiple mapping rule items are simultaneously satisfied, a unique mapping item is output according to a preset priority. The mapping registration is used to associate the classification result with the output unique mapping item and generate the ten patient groups. After generating the ten patient groups, the data processing module performs key monitoring related indicator establishment processing. The key monitoring related indicator establishment processing includes indicator template loading, indicator binding with patient groups, and indicator collection constraint generation. The indicator template loading is used to load key monitoring related indicator templates corresponding to the ten patient groups. The indicator binding with patient groups is used to establish an index relationship between the indicator entries in the key monitoring related indicator templates and the ten patient groups. The indicator collection constraint generation is used to form the collection frequency. The data collection period and missing test marker processing rules are aggregated into the key monitoring-related indicators. After forming the key monitoring-related indicators, the data processing module calls the change collection records passed in by the data collection module to complete change collection and change analysis processing. The change analysis processing includes data cleaning, change calculation, and analysis result generation. The data cleaning is used to perform consistency processing on outliers and missing test markers in the change collection records and form an analyzable dataset. The change calculation is used to perform time series processing on the changes of the key monitoring-related indicators and calculate change segments. The analysis result generation is used to match the change segments with the phased rules corresponding to the ten patient groups and generate the phased rehabilitation plan suggestions.The proposed phased rehabilitation plan is transmitted as an output to the remote monitoring module and serves as input to the online guidance module. Simultaneously, the data processing module writes the classification results, the ten patient groups, the key monitoring indicators, and the proposed phased rehabilitation plan into the storage module. Upon receiving adjustment suggestions from the online guidance module, the module establishes a correlation between these suggestions and the proposed phased rehabilitation plan, triggering a subsequent call for change collection and analysis.

[0105] The remote monitoring module 03 is used to transmit the phased rehabilitation plan suggestions and access the doctor's terminal. The remote monitoring module is connected to the online guidance module. Specifically, the remote monitoring module receives the phased rehabilitation plan suggestions from the data processing module and performs transmission encapsulation, session management, and arrival confirmation processing. The transmission encapsulation is used to merge the phased rehabilitation plan suggestions with the stroke patient identifier, generation time, and corresponding ten patient group indexes to form a transmission message. The session management is used to maintain the remote access session with the doctor's terminal and register a session identifier for each transmission message. The arrival confirmation processing is used to send back confirmation information after the doctor's terminal receives the message and associate the confirmation information with the session identifier for archiving. During the access process to the doctor's terminal, the remote monitoring module verifies the login status and access permissions of the doctor's terminal. If the verification fails, a rejection record is output and written to the storage module. If the verification passes, the remote monitoring module sends the phased rehabilitation plan suggestion to the online guidance module and simultaneously transmits the session identifier, enabling the online guidance module to reuse the session identifier to complete the return path binding when generating adjustment suggestions. The remote monitoring module also receives the return message from the online guidance module and forwards it to the data processing module as the transmission carrier for the adjustment suggestions. The association between the return message, the session identifier, and the phased rehabilitation plan suggestion is written to the storage module.

[0106] The online guidance module 04 is used to perform anomaly detection and online guidance processing based on the phased rehabilitation plan recommendations, generate adjustment suggestions, and write the adjustment suggestions back to the data processing module. Specifically, the online guidance module receives the phased rehabilitation plan recommendations and the session identifier from the remote monitoring module, and calls the analysis results of the changes in the key monitoring-related indicators provided by the data processing module to complete the anomaly detection. The anomaly detection includes loading of detection conditions, condition matching, and anomaly record generation. The loading of detection conditions is used to load anomaly detection conditions corresponding to the ten patient groups. The condition matching is used to compare the analysis results of the changes in the key monitoring-related indicators with the anomaly detection conditions item by item and output anomaly markers. The anomaly record generation is used to associate the anomaly markers with the session identifier and the phased rehabilitation plan recommendations to form anomaly records. After outputting the anomaly markers, the online guidance module performs anomaly detection and online guidance processing to generate adjustment suggestions and write the adjustment suggestions back to the data processing module. The online guidance module enters the online guidance processing chain, which includes suggestion editing, suggestion verification, and suggestion encapsulation. Suggestion editing is used to adjust the suggested items of the phased rehabilitation plan in the doctor's terminal interface and form candidate adjustment suggestions. Suggestion verification is used to check the consistency of the candidate adjustment suggestions with the ten patient group indexes and output verification prompts for missing items. Suggestion encapsulation is used to merge the verified candidate adjustment suggestions with the abnormal records and the session identifier to form the adjustment suggestions. The online guidance module writes the adjustment suggestions back to the data processing module through the remote monitoring module. The data processing module registers the adjustment suggestions accordingly and triggers subsequent change collection and change analysis processing. Simultaneously, the online guidance module writes the abnormal records and the adjustment suggestions into the storage module and maintains their index association with the phased rehabilitation plan suggestions.

[0107] Storage module 05 is used to store the basic information of the stroke patients, the classification results, the ten patient groups, the key monitoring indicators, the phased rehabilitation plan suggestions, and the adjustment suggestions, and provides a read interface to the data processing module. Specifically, the storage module establishes a data writing channel with the data acquisition module, the data processing module, the remote monitoring module, and the online guidance module, and establishes an index structure according to the stroke patient identifier and the generation time, registering the basic information of the stroke patients, the classification results, the ten patient groups, the key monitoring indicators, the phased rehabilitation plan suggestions, and the adjustment suggestions as searchable records respectively; during the writing process, the storage module performs consistency checks on records under the same stroke patient identifier. The consistency check includes field integrity checks and association checks. The field integrity checks are used to determine null values ​​in the required fields of each record and generate missing records. The association checks are used to check the classification results and the ten patient groups, the ten patient groups and the key monitoring indicators, and the... The key monitoring indicators are checked against the index relationships between the phased rehabilitation plan recommendations and the adjustment recommendations, and a related check record is generated. The missing records and the related check records are written to the storage module and made available for reading by the data processing module. The reading interface provides the data processing module with access methods for searching by stroke patient identifier, generation time, and the ten patient group indexes. The data processing module calls the reading interface to obtain historical classification results, historical phased rehabilitation plan recommendations, and historical adjustment recommendations, and completes the write-back registration and subsequent change collection and change analysis processing linkage. The storage module generates an access record for each read request and archives it in association with the index structure of the corresponding record. The access records and the adjustment recommendations together form a closed-loop traceable data foundation.

Claims

1. A method for guiding home-based stroke rehabilitation based on multidimensional classification, characterized in that, include: S100. Obtain the basic information of stroke patients, classify them according to disease characteristics, location of neurological deficit, severity, and comorbidities, and obtain the classification results. S200. Based on the classification results, perform 10 patient group mapping processing to generate 10 patient groups; S300. Based on the aforementioned 10 patient groups, key monitoring-related indicators are established and processed to generate key monitoring-related indicators; S400. Based on the key monitoring indicators, collect and analyze changes to generate phased rehabilitation plan recommendations. S500: Based on the aforementioned phased rehabilitation plan recommendations, perform anomaly detection and online guidance processing, generate adjustment suggestions, and write them back.

2. The method according to claim 1, characterized in that, The process of classifying and managing disease characteristics, location of neurological deficits, severity, and comorbidities includes: The disease feature classification process includes enumerating and normalizing stroke types, normalizing the onset time on a time axis to generate a disease duration field, and outputting classification labels containing stroke type, onset time, recurrence information, and a description of the current rehabilitation stage. The classification and processing of neurological deficit sites includes structured analysis, key phrase extraction and conflict resolution for four types of neurological deficits: language, motor, cognitive and psychological, and finally mapping the final phrases to corresponding deficit site classification labels. The severity classification process includes determining the source priority and merging the severity descriptions of the doctor's terminal and the mobile terminal, and outputting the severity level or the pending confirmation mark. Comorbidity classification processing includes structurally splitting comorbidity descriptions to generate comorbidity entries containing entry sources and confirmation status, and verifying the association between past medication information and comorbidity entries.

3. The method according to claim 1, characterized in that, The process of mapping 10 patient groups includes: The classification results are subjected to field decomposition, standardization, and consistency checks, including source consistency checks and logical consistency checks. The preset mapping rule set is loaded from the storage module. The mapping rule set is maintained by version number. Each rule entry consists of a combination of four types of enumerated values: disease feature classification label, neurological deficit site classification label, severity level, and comorbidity classification label. Rule matching is performed using a two-level matching chain, which consists of: Level 1 matching, which uses the classification labels of the neurological deficit sites and severity levels in the classification results to form a combination key and selects a set of candidate rule entries from the mapping rule set; and Level 2 matching, which uses the disease feature classification labels and comorbidity classification labels in the classification results to perform cross-filtering on the set of candidate rule entries to determine the final rule entries. When the classification result has a pending confirmation marker, a mapping confidence status field is attached to the 10 patient groups output by the mapping process.

4. The method according to claim 1, characterized in that, The process of generating 10 patient groups includes: Based on the final rule entries, 10 patient groups are generated. The 10 patient groups include a patient group identifier, a mapping confidence state field, and an association index field. The association index field is used to record the version number of the classification result on which the generation is based and the version number of the mapping rule set. The generated 10 patient groups and the associated index field are written into the patient group file of the storage module to form a patient group file index.

5. The method according to claim 1, characterized in that, The process of establishing and processing key monitoring indicators includes: The group identifier, mapping confidence state field, and association index field of the 10 patient groups are parsed from the patient group files. Based on the group identifier, a set of rules for key monitoring indicators is loaded from the storage module for matching. The set of rules for key monitoring indicators is maintained in the form of a version number. When a match is successful, a key monitoring related indicator structure is generated based on the matched rule entries. The key monitoring related indicator structure includes an indicator field list and a collection frequency configuration. The indicator field list consists of a subset of health data and training data fields corresponding to the language, motor, cognitive, and psychoneurological function deficits. The collection frequency configuration includes a collection frequency switching rule based on the training time period marker to switch between rest and exercise or rehabilitation training states. The generated key monitoring indicator structure and the rule version number on which it is based are written into the monitoring configuration file of the storage module.

6. The method according to claim 1, characterized in that, The process of change acquisition and change analysis includes: The data transmission unit executes the collection scheduling according to the collection frequency configuration in the key monitoring indicators, switches the collection frequency according to the training time period mark, and performs field existence verification and field value consistency verification on the reported health data and training data, generating change collection records containing collection timestamp, data source, missing test mark and review mark, which are organized by the storage module in units of collection window; The data processing unit reads change collection records associated with the key monitoring indicators from the storage module, performs a combined processing chain of time sequence comparison, intra-window comparison, and inter-window comparison, generates change analysis records containing change direction labels, fluctuation labels, and mutation labels, and performs isolation processing on data with missing measurement marks or verification marks.

7. The method according to claim 1, characterized in that, The process of generating key monitoring indicators includes: Based on the change analysis records and the patient group file index, a stage switching determination is made according to a preset set of stage switching rule entries to obtain a stage marker; based on the stage marker and the patient group file index, rehabilitation nursing, rehabilitation training, and psychological nursing suggestion entries are matched and assembled from the suggestion template library maintained by the storage module to form a stage suggestion entry set; a suggestion version record containing the patient group file index, the version number of the key monitoring related indicators, the stage marker, and the stage switching basis field is registered for the stage suggestion entry set to generate a stage rehabilitation plan suggestion.

8. The method according to claim 1, characterized in that, The process of anomaly detection and online guidance includes: Anomaly Detection and Handling: The data processing unit, based on the proposed phased rehabilitation plan and its associated version records, change analysis records, and change collection records, and according to the anomaly detection conditions loaded from the storage module and bound to the patient group profile index, performs suggestion consistency verification and execution feedback verification. The suggestion consistency verification verifies the correspondence between the proposed phased rehabilitation plan and its generation basis, and the execution feedback verification verifies the consistency between the training records submitted by the mobile terminal and the rehabilitation training suggestion items in the proposed phased rehabilitation plan. When the verification fails, a structured anomaly status is generated, including an anomaly type field, an anomaly source field, an association index field, and a processing status field. The association index field includes at least the patient group profile index, the version number of the key monitoring related indicators, and the suggestion version record number. The anomaly status is then written to the storage module. Online guidance processing: The online guidance module generates an online guidance task based on the abnormal state and pushes it to the doctor's terminal; it receives structured adjustment suggestions returned by the doctor's terminal, which include fields for adjustment content, applicable stage, applicable patient group, and review and signature.

9. The method according to claim 1, characterized in that, The process of generating and writing back adjustment suggestions includes: The adjustment suggestion input is merged with the phased rehabilitation plan suggestion to generate an adjustment suggestion record. The adjustment suggestion record includes an adjustment suggestion version number that is associated with the suggestion version record number, a write-back status field, and the associated index field. The adjustment suggestion record is written to the storage module, the processing status field of the abnormal status is updated, and the adjustment suggestion record is synchronously written back to the home management system and the mobile terminal through the data transmission unit.

10. A stroke home rehabilitation guidance system based on multidimensional classification, characterized in that, include: Data acquisition module, data processing module, remote monitoring module, online guidance module, and storage module; The modules are connected in sequence to implement the method as described in any one of claims 1-9.

Images