A medical treatment accompanying intelligent scheduling method and system based on multi-dimensional image matching

Abstract

The application relates to the technical field of intelligent scheduling, and discloses a medical treatment accompanying diagnosis intelligent scheduling method and system based on multi-dimensional portrait matching, wherein the method comprises the following steps: calculating the opportunity exposure rate of unverified attributes of each accompanying doctor, and generating an attribute verification missing attribution label set; adopting a slow decay function for a structural missing attribute, adopting an exponential decay function for a competitive elimination attribute to calculate a timeliness coefficient, and generating an effective ability portrait vector; deducing an expected medical treatment stage timeline, dividing independent service sections according to idle time windows; establishing a mapping relationship between a low complexity service section and a structural missing attribute, respectively generating each service section distribution scheme, performing three-target Pareto sorting, and outputting a relay scheduling comprehensive scheme; and pushing a whole-process node state and an inspection report to a family member end in real time.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling technology, specifically to an intelligent scheduling method and system for accompanying patients to medical appointments based on multi-dimensional profile matching. Background Technology

[0002] Medical escort service platforms typically maintain a multi-dimensional capability profile for each escort, covering attributes such as professional skills, communication skills, and familiarity with departmental procedures, and also maintain a needs profile for patients, thereby achieving the matching and allocation of escorts and patients.

[0003] Existing scheduling methods apply a uniform exponential decay timeliness correction to each dimension of the companion's competency profile. When a certain attribute fails to receive positive validation for an extended period, regardless of whether the reason for the lack of validation is that the corresponding patient demand is extremely rare or that companions are consistently rejected due to insufficient competitiveness, the same profile decay result occurs. This leads to the systematic forgetting of scarce competency attributes such as sign language communication and rare disease companionship, making it impossible to find candidate companions with the corresponding competency when such scarce demand arises.

[0004] Furthermore, existing scheduling methods typically bind patients to a single escort throughout all stages of their medical treatment. The skill requirements for escorts vary significantly across different stages, making it impossible for a single escort to provide optimal matching at every stage. Simultaneously, family members on remote medical visits are unable to obtain real-time information about the patient's stage and examination results, remaining in an information blind spot for extended periods, thus impacting the service experience. Summary of the Invention

[0005] This invention provides a method and system for intelligent scheduling of medical companions based on multi-dimensional profile matching, which solves the technical problems in related technologies, such as the inaccuracy of the profile of the companion's ability due to lack of verification opportunities, resulting in a decline in the quality of scheduling and matching, and the lack of transparency of information about the medical process to family members.

[0006] This invention discloses an intelligent scheduling method for medical accompaniment based on multi-dimensional profile matching and real-time transparency throughout the entire process, comprising the following steps: Obtain the global service demand history, count the demand activity index of each attribute, calculate the opportunity exposure rate of each companion's unverified attributes, mark unverified attributes with an opportunity exposure rate below a preset threshold as structural missing type, and mark unverified attributes with an opportunity exposure rate above a preset threshold but not assigned as competitive elimination type. For attributes with structural missing types, a slow decay function is used, and for attributes with competitive elimination types, an exponential decay function is used to calculate the timeliness coefficient. These coefficients are then multiplied dimension by dimension with the original multidimensional competency profile vector of the companion physician to generate an effective competency profile vector. We analyze patient needs profiles and use Monte Carlo simulation based on process templates to extrapolate the expected timeline of medical treatment stages. Based on the idle time windows between stages, we divide the medical treatment process into multiple independent service segments and extract the stage-level capability requirement vector and complexity score of each independent service segment. Select low-complexity service segments and semantically match their auxiliary capability tags with structurally missing attributes to generate a mapping table. Based on the mapping table, generate a low-complexity service segment allocation scheme with verification objectives. Based on multi-dimensional matching calculation, generate a capability adaptation allocation scheme for high-complexity service segments. Jointly execute multi-objective optimization to output a relay scheduling comprehensive scheme. During the medical treatment process, the system pushes real-time status updates and estimated time consumption for each stage of the process to the family members.

[0007] Furthermore, the formula for calculating the chance exposure rate is as follows: ,in For opportunity exposure rate, This refers to the number of times the accompanying therapist is included in the candidate list in the scheduling of needs related to this attribute. This refers to the total frequency of needs related to the patient needs profile within the recent time window for this attribute; when When the value is zero, the corresponding unverified attribute is directly marked as a structural missing type.

[0008] Furthermore, the formula for calculating the gradual decay function is as follows: The formula for calculating the exponential decay function is as follows: ,in For slow decay coefficient, The original attenuation coefficient, The time interval since the most recent positive verification. The value is less than The slow decay function sets a lower threshold for the timeliness coefficient. ,when Below Time As output.

[0009] Furthermore, the process of analyzing the patient needs profile includes: extracting the visit type identifier, target hospital and department preferences, disease complexity features, and escort service preference vectors; if the patient is a follow-up visit or undergoing in-depth examination, the patient's medical history summary, index of previous examination reports, and historical diagnostic conclusions are read, vectorized, and concatenated with the basic needs profile vector to generate an enhanced patient needs profile vector; correspondingly, a medical history-related ability dimension is added to the escort's effective ability profile vector to ensure consistency between the dimensions of the two vectors.

[0010] Furthermore, the first stage of the capability requirement vector Dimensional values Calculated by the following formula: ,in This represents the total number of stage nodes contained in this independent service segment. For the first The estimated duration of each stage node, The total estimated duration of this independent service segment, For the first The stage node is at the The capability requirement intensity value across each capability dimension; the complexity score. The calculation formula is: ,in For the first The process complexity attribute value of each stage node.

[0011] Furthermore, the multidimensional matching calculation uses weighted cosine similarity, and the calculation formula is as follows: ,in To match quality scores, Let be the number of dimensions of the image vector. For the first Weight coefficients for each dimension The first vector of patient needs profile Each dimension value The first effective capability profile vector for companion therapists Each dimension value.

[0012] Furthermore, the multi-objective optimization performs Pareto ranking with three objectives: maximizing the quality of multi-dimensional profile matching in the global service segment, minimizing the idle rate of the escort, and maximizing the coverage of structural missing attribute verification. The solution with the highest comprehensive utility value is selected from the non-dominated solution set as the final output. The smoothness of the connection between independent service segments is introduced as an auxiliary constraint in the Pareto ranking, requiring that when adjacent independent service segments are assigned to different escorts, the difference between the escorts in the communication style dimension of the latter escort is lower than the switching tolerance threshold.

[0013] Furthermore, it also includes: when the actual completion time of each stage exceeds the expected range of the expected medical treatment stage timeline, causing the start time of subsequent independent service segments to shift, re-searching the available companion physician set within the time shift window, recalculating the multi-dimensional profile matching and allocation scheme for subsequent independent service segments based on the effective capability profile vector, outputting the updated relay scheduling instruction and synchronously updating and pushing the subsequent stage time estimate to the family end.

[0014] Furthermore, it also includes a closed-loop process for real-time synchronization of examination reports and service feedback: After the escort collects the images of the examination report, a quality check is performed on the clarity and completeness of the images. Once the check is passed, the images are uploaded to the patient's file and a summary of key indicators is pushed to the family member's end. After each independent service segment is completed, patient feedback evaluation data is obtained. For escorts assigned to low-complexity service segments, it is checked whether the structural missing attributes are actually triggered and applied. The attribution labels are updated based on the triggering situation and feedback evaluation data. At least two independent positive verifications are required before the attribution labels can be completely removed and restored to the verified state.

[0015] This invention discloses a medical accompaniment intelligent scheduling system based on multi-dimensional profile matching and real-time transparency throughout the entire process, used to execute the aforementioned multi-dimensional profile matching and real-time transparency throughout the entire process intelligent scheduling method for medical accompaniment, including: The attribute attribution analysis module is used to obtain the global service demand history, count the demand activity index of each attribute, calculate the opportunity exposure rate of each escort's unverified attributes, and generate attribute verification missing attribution tag set. The differentiated decay module is used to calculate the timeliness coefficient by applying a slow decay function to structural missing attributes and an exponential decay function to competitive elimination attributes, thereby generating an effective capability profile vector for each escort. The demand profile parsing module is used to parse the patient demand profile, extract the visit type identifier, preference information and escort service preference vector, and generate an enhanced patient demand profile vector for patients with follow-up visits or in-depth examinations. The timeline simulation module is used to simulate and extrapolate the patient's expected medical treatment timeline using Monte Carlo simulation based on process templates and departmental queuing data. The process segmentation module is used to identify idle time windows based on the expected medical treatment stage timeline and divide the medical treatment process into multiple independent service segments, extracting the stage-level capability requirement vector and complexity score of each independent service segment; The verification mapping module is used to filter low-complexity service segments and semantically match their auxiliary capability tags with structurally missing attributes to generate a mapping relationship table; The allocation scheme generation module is used to generate low-complexity service segment allocation schemes with verification targets based on the mapping relationship table, and to generate high-complexity service segment capability adaptation allocation schemes based on multi-dimensional matching calculations. The global optimization module is used to perform multi-objective Pareto sorting and output a comprehensive relay scheduling scheme for low-complexity service segment allocation schemes and high-complexity service segment capacity adaptation allocation schemes. The real-time push and dynamic scheduling module is used to obtain the geographical coordinates of the escorts in real time and push the real-time status updates of the entire process nodes and the stage time estimation to the family. When the actual time deviation exceeds the expected range, it will re-search for available escorts and output the updated relay scheduling instructions. The report synchronization and feedback closed-loop module is used to upload the examination report images to the patient's file after quality verification and push the summary of key indicators. It also obtains patient feedback evaluation data to detect the triggering of structural missing attributes and updates the attribution labels.

[0016] The present invention has the following beneficial effects.

[0017] By combining demand activity indicators and opportunity exposure rates, the reasons for unverified attributes are classified into two categories: structural missing and competitive elimination. Slow decay function protection is applied to structural missing attributes to ensure that scarce capability attributes remain discoverable in multidimensional matching, thus overcoming the problem of systematic forgetting of scarce capability attributes caused by uniform exponential decay.

[0018] By incorporating past medical history, examination reports, and diagnostic conclusions into the enhanced patient needs profile, and integrating the service preference vector of escorts, the matching calculation can cover department recommendations and personalized escort adaptation needs in follow-up visit scenarios.

[0019] By breaking down the medical treatment process into multiple independent service segments based on available time windows and implementing relay scheduling, each independent service segment is executed by the escort with the most suitable capabilities for that stage, overcoming the problem of uneven capabilities of a single escort across stages in the full-process binding model. At the same time, low-complexity service segments are transformed into verification scenarios with structurally missing attributes, continuously generating verification opportunities during normal operation.

[0020] With real-time status updates throughout the entire process and instant synchronization of examination reports, family members at remote locations can continuously stay informed about the patient's medical progress and examination results, eliminating information blind spots for family members. Attached Figure Description

[0021] Figure 1 This is a flowchart of the intelligent scheduling method for medical accompaniment based on multi-dimensional profile matching and real-time transparency throughout the entire process provided in this embodiment of the invention; Figure 2 This is a schematic diagram illustrating the comparison of attribution analysis of key competency attributes of escort A provided in an embodiment of the present invention; Figure 3 This is a schematic diagram comparing the numerical values ​​of each dimension of the effective competence profile of the escort A provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the expected duration range of each stage of medical treatment for patient Wang, provided in an embodiment of the present invention; Figure 5This is a schematic diagram comparing the idle time window and switching threshold after each stage of the medical treatment process provided in the embodiments of the present invention; Figure 6 This is a schematic diagram of the complexity score and estimated total duration of each service segment provided in the embodiments of the present invention; Figure 7 This is a schematic diagram of the cosine similarity between a low-complexity service segment and a structurally missing attribute provided in an embodiment of the present invention; Figure 8 This is a schematic diagram comparing the quality scores of candidate companions in each service segment according to an embodiment of the present invention; Figure 9 This is a schematic diagram of the global multi-objective optimization three-objective normalized score and comprehensive utility value provided in the embodiment of the present invention. Detailed Implementation

[0022] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0023] At least one embodiment of the present invention discloses an intelligent scheduling method for medical accompaniment based on multi-dimensional profile matching and real-time transparency throughout the entire process. Figure 1 This includes the following steps: Step 1: Attribute validation and attribution analysis.

[0024] The system retrieves the global service demand history and calculates the frequency of demand for each attribute within the recent time window, based on capability dimensions, within the patient demand profile. It then calculates the demand activity index for each attribute. For each companion's multi-dimensional capability profile, it retrieves the demand activity index for each unverified attribute and the number of candidate participations for each companion in the scheduling of related demands for that unverified attribute, calculating the opportunity exposure rate. Unverified attributes with an opportunity exposure rate below a preset threshold are marked as structurally missing, while those with an opportunity exposure rate above the preset threshold but still not assigned are marked as competitively eliminated. This generates an attribute verification missing attribution label set for each companion.

[0025] The formula for calculating opportunity exposure rate is: ; in, For opportunity exposure rate, This refers to the number of times the accompanying therapist is included in the candidate list in the scheduling of needs related to this attribute. This represents the total frequency of needs related to the patient needs profile within the recent time window for this attribute. and All are dimensionless quantities. When the value is zero, it indicates that there is no demand for this attribute in the near future, and the corresponding unverified attribute is directly marked as a structural missing type.

[0026] It should be noted that the length of the recent time window is configured differently based on the attribute type. For high-frequency, general attributes (such as basic communication skills and familiarity with common departmental procedures), the recent time window is set to 30 to 60 days; for low-frequency, scarce attributes (such as sign language communication skills and experience in accompanying rare diseases), the recent time window is set to 90 to 180 days to avoid distortion of the demand frequency statistics for scarce attributes due to an excessively short recent time window.

[0027] Step 2, Calculation of Differentiated Timeliness Decay.

[0028] For attributes marked as structurally missing, a slow decay function is used to calculate the timeliness coefficient, and for attributes marked as competitively eliminated, an exponential decay function is used to calculate the timeliness coefficient. The timeliness coefficients of each dimension after differential decay are multiplied dimension-by-dimensionally by the original multidimensional competency profile vector of the companion, generating an effective competency profile vector for each companion.

[0029] The original numerical units of each dimension of the original multidimensional competency profile vector may differ (e.g., rating score, percentage, years of experience, etc.). Before performing dimension-by-dimensional multiplication, the original numerical values ​​of each dimension are normalized using a mean normalization process based on the range, thus scaling the numerical values ​​of each dimension uniformly to a uniform mean. The interval eliminates the influence of dimensional differences on the time-related decay calculation results.

[0030] The formula for calculating the timeliness coefficient of structural missing attributes is: ; in, The timeliness coefficient of the structural missing attribute. It is a natural constant. For slow decay coefficient, This is the time interval (in days) since the most recent positive validation. The formula for calculating the timeliness coefficient of competitive elimination attributes is: ; in, The timeliness coefficient of the competitive elimination attribute. It is a natural constant. The original attenuation coefficient, The meaning is the same as above. The value is less than This makes the decay rate of structurally missing attributes significantly lower than that of competitively eliminated attributes.

[0031] In this embodiment of the application, in order to ensure that the slow decay function maintains a reasonable lower limit value even when it has not been verified for a long time, a lower limit threshold for the timeliness coefficient is also set for the slow decay function. When the calculation is obtained Below At that time, with This serves as the timeliness coefficient output for the attribute, preventing structurally missing attributes from becoming completely invalid after an extremely long period of unverification.

[0032] Step 3: Patient needs profile analysis and medical treatment timeline projection.

[0033] When a patient submits a request for companionship, the system analyzes the patient's needs profile and extracts the following: visit type identifier, target hospital and department preference, target doctor preference, disease complexity characteristics, and companion service preference vector. The visit type identifier includes initial visit, follow-up visit, in-depth examination, and surgical consultation. The companion service preference vector covers dimensions such as gender, years of experience, and communication style. Non-numerical fields within each dimension (such as gender and communication style category) are converted to numerical representations using a coded classification data method.

[0034] If the patient is a follow-up visit or undergoing in-depth examination, the patient's medical history summary, index of previous examination reports, and historical diagnostic conclusions associated with the patient's needs profile are further read. The medical history data is then vectorized and concatenated with the basic needs profile vector to generate an enhanced patient needs profile vector.

[0035] In a typical medical treatment process template library, the most similar process template is retrieved based on the treatment type identifier and target department preference. Based on the process template's stage node list and inter-node transition probability matrix, combined with current departmental waiting time data, Monte Carlo simulation is used to extrapolate the patient's expected medical treatment timeline, outputting the expected start time interval, expected duration interval, and expected idle time window between stages for each stage. The Monte Carlo simulation generates multiple medical treatment path samples by randomly sampling the inter-node transition probability matrix and the departmental waiting time distribution multiple times. The statistical interval of the time nodes for each path sample is then used as the output expected time interval.

[0036] It should be noted that the process template's stage node list includes standard nodes such as registration and waiting, outpatient visit, laboratory tests, waiting for reports, and medication pickup. The node transition probability matrix reflects the possibility and probability value of transitions between nodes under different types of medical visits.

[0037] It should be noted that in the enhanced patient needs profile vector, the vectorization of past medical history summaries, indexes of previous examination reports, and historical diagnostic conclusions uses text embedding to extract semantic features, generating a fixed-dimensional medical history feature vector. This medical history feature vector is dimensionally independent of the basic needs profile vector. The total dimension of the resulting enhanced patient needs profile vector is the sum of the basic needs profile dimension and the medical history feature dimension. Correspondingly, in subsequent matching calculations, a medical history-related capability dimension (such as experience in follow-up consultations and report interpretation skills) is added to the companion's effective capability profile vector to maintain consistency in the dimensions of both vectors. For patients seeking initial consultations or surgical consultations, the basic needs profile vector is directly used as the patient-side input for matching calculations, and the companion's effective capability profile vector does not add a medical history-related capability dimension; the dimensions of both vectors remain consistent.

[0038] Step 4: Extracting the segmented and phased capability requirements of the medical treatment process.

[0039] Based on the projected timeline of the expected medical treatment stages, idle time windows between stages are identified. When the projected duration of an idle time window exceeds the handover threshold, the patient's complete medical treatment process is divided into multiple independent service segments.

[0040] For each independent service segment, the capability requirement tags corresponding to each stage node within the segment are extracted. These tags are then weighted and merged according to the proportion of each stage node's expected duration to the total duration of the independent service segment, generating a stage-level capability requirement vector for that segment. The stage-level capability requirement vector is then... Dimensional values Calculated by the following formula: ; in, This represents the total number of stage nodes contained in this independent service segment. For the first The estimated duration of each stage node, The total estimated duration of this independent service segment (i.e. ), For the first The stage node is at the The intensity value of capability requirement in each capability dimension Indexed for capability dimensions. and The time units are the same, and the ratio is... For dimensionless weights, This is a dimensionless intensity value. The dimensions are consistent.

[0041] Simultaneously, the complexity score of this independent service segment is calculated by weighting the process complexity attributes of each stage node within this segment. Complexity Score The calculation formula is: ; in, For the first The process complexity attribute value of each stage node is obtained by mapping the pre-configured complexity level of each node in the process template library. The dimensionless ranking value and the dimensionless weight Multiply and then sum. The dimensions are consistent.

[0042] It should be noted that the relay switching threshold is determined based on the size of the target hospital and the response time for dispatching escorts, ranging from 15 to 45 minutes. When the expected duration of the idle time window does not exceed the relay switching threshold, the preceding and following stages are included in the same independent service segment and served continuously by the same escort.

[0043] In this embodiment of the application, in order to avoid the discomfort caused by patients frequently contacting different caregivers in a short period of time due to excessive segmentation, an upper limit constraint is set on the number of independent service segments after segmentation. When the number of independent service segments generated by segmentation exceeds the upper limit, adjacent independent service segments are merged in order of increasing idle time window duration until the number of independent service segments meets the upper limit constraint.

[0044] Step 5: Mapping low-complexity service segments to structurally missing attributes.

[0045] Independent service segments with complexity scores below the applicable threshold are selected from all independent service segments to form a low-complexity service segment set. A list of auxiliary capability tags that are non-core requirements but can trigger applications is extracted from each low-complexity service segment in this set. Semantic matching is performed between the auxiliary capability tag list and the structurally missing attributes in the attribute verification missing attribution tag set to generate a mapping table between low-complexity service segments and verifiable structurally missing attributes. Semantic matching is performed by calculating the cosine similarity between the text embedding vectors of the auxiliary capability tags and the structurally missing attributes. Both text embedding vectors involved in the cosine similarity calculation are generated by the same text embedding model, have the same dimension, and a mapping relationship is established when the similarity exceeds the matching threshold.

[0046] It should be noted that the assistive capability label refers to the capability dimension that is not a primary functional requirement in this low-complexity service segment but can be applied during execution. For example, the core requirement of the medication dispensing stage is familiarity with pharmacy procedures, but this stage can also trigger the practical application of assistive capabilities such as sign language communication (for hearing-impaired patients) and dialect communication (for elderly patients who do not speak Mandarin).

[0047] Step 6: Generation of segmented differential matching and allocation schemes.

[0048] Based on the mapping table, for each low-complexity service segment, a set of candidate companions with corresponding structurally missing attributes and satisfying spatiotemporal reachability constraints is retrieved. Using the satisfaction of basic capability requirements of the low-complexity service segment as a hard constraint and the priority of structurally missing attribute verification as the sorting criterion, a low-complexity service segment allocation scheme with verification objectives is generated.

[0049] For the remaining high-complexity service segments, a multi-dimensional matching calculation is performed between the patient demand profile vector (including the service preference dimension; for patients undergoing follow-up visits or in-depth examinations, an enhanced patient demand profile vector is used) and the effective capability profile vectors of candidate escorts. The matching quality score for each candidate escort is output, and capability adaptation allocation schemes for high-complexity service segments are generated by sorting them from high to low matching quality scores. The multi-dimensional matching calculation uses weighted cosine similarity, calculated using the following formula: ; in, To match quality scores, Let be the number of dimensions of the image vector. For the first Weight coefficients for each dimension The first vector of patient needs profile Each dimension value The first effective capability profile vector for companion therapists Each dimension value The dimension index is [not specified]. Both the patient needs profile vector and the companion / attendant effective competence profile vector have [not specified] dimensions. As described in step 3, for patients undergoing follow-up visits or in-depth examinations, the effective capability profile vector of the escort has added a medical history-related capability dimension, and the dimensions of both vectors remain consistent; for patients undergoing initial visits or surgical consultations, the dimensions of both vectors are also consistent. and All values ​​have been normalized in steps 2 and 3, and the values ​​of each dimension are within the same dimension range, so they can be directly used in weighted calculations.

[0050] It should be noted that the spatiotemporal accessibility constraint refers to whether the candidate escort can reach the corresponding area of ​​the target hospital before the expected start time of the low-complexity service segment. Specifically, it is determined by whether the expected travel time between the escort's current geographical location or the end location of the previous independent service segment and the target location is less than the available buffer time before the start of the low-complexity service segment.

[0051] It should be noted that the method for determining the satisfaction of basic capability requirements in low-complexity service segments is consistent with the multi-dimensional matching calculation in high-complexity service segments. That is, the stage-level capability requirement vector of the low-complexity service segment is used as the patient-side input, and the weighted cosine similarity is calculated with the effective capability profile vector of the candidate escort. When the obtained matching quality score exceeds the preset basic satisfaction threshold, the candidate escort is deemed to meet the hard constraint condition and can enter the subsequent candidate escort set sorted by verification priority.

[0052] Step 7: Solve the global multi-objective optimization problem.

[0053] The low-complexity service segment allocation scheme with verification objectives and the high-complexity service segment capability adaptation allocation scheme are jointly input into the global optimization solution. Pareto ranking is performed with three objectives: maximizing the quality of multi-dimensional profile matching for global service segments, minimizing the idle rate of caregivers, and maximizing the coverage of structurally missing attribute verification. The output is a relay scheduling comprehensive scheme that balances multi-dimensional profile matching quality and capability verification. Pareto ranking selects non-dominated solutions from the candidate solution set that are not completely dominated by other solutions in any of the three objective dimensions. The solution with the highest comprehensive utility value from the non-dominated solution set is selected as the final output. The three objectives are respectively processed by mean normalization based on range and then weighted and summed according to preset weights to obtain the comprehensive utility value.

[0054] It should be noted that the global service segment multi-dimensional profile matching quality is the weighted average of the matching quality scores of all independent service segments in the current candidate scheme, with the weight being the proportion of the expected duration of each independent service segment to the total service duration of the current scheduling; the companion idle rate is the average proportion of the time that all companions participating in the scheduling in the current candidate scheme are not covered by independent service segments within the scheduling time range to the available time of each companion; the structural missing attribute verification coverage rate is the proportion of the number of structural missing attributes in the current candidate scheme that are assigned to low complexity service segments and have corresponding mapping relationships to the total number of all structural missing attributes to be verified.

[0055] In this embodiment of the application, in order to take into account the continuity of patient experience in multi-objective optimization, the smoothness of the connection between independent service segments is introduced as an auxiliary constraint in Pareto ranking: when adjacent independent service segments are assigned to different escorts, the difference in communication style between the escort in the later independent service segment and the escort in the earlier independent service segment is required to be lower than the switching tolerance threshold, so as to avoid patients from feeling uncomfortable due to excessive differences in escort styles.

[0056] Step 8: Real-time status push and dynamic scheduling updates throughout the entire process.

[0057] During the patient's medical treatment process, the real-time geographical coordinates of the escort are obtained. The real-time distance between the escort and key points within the target hospital is calculated using geofencing. Real-time status updates covering all stages of the process are pushed to the family member's device, including order confirmation, en route to the hospital, arrival at the hospital, registration completion, waiting for treatment, outpatient visit, heading to examination, examination in progress, waiting for report, medication pickup, and service completion. While en route to the hospital, the estimated distance between the escort and the hospital entrance is simultaneously pushed to the family member's device.

[0058] When the status of each node changes, the estimated remaining time for the current stage and the estimated start time for each subsequent key step are pushed synchronously based on Monte Carlo simulation.

[0059] When the actual completion time of each stage exceeds the expected range of the expected medical treatment stage timeline, causing the start time of subsequent independent service segments to shift, the available companion physician set within the time shift window is retrieved again. Based on the effective capability profile vector, the multi-dimensional profile matching and allocation scheme of subsequent independent service segments is recalculated, the updated relay scheduling instruction is output, and the estimated time consumption of subsequent stages is simultaneously updated and pushed to the family member's end.

[0060] It should be noted that the geofence is a pre-configured polygonal area representing the boundary of the target hospital buildings. Key points include the hospital main gate, entrances to each department floor, and pharmacy windows. The real-time distance between the accompanying person and the key points is estimated by combining the straight-line distance between the geographic location coordinates and the key point coordinates with the internal navigation path.

[0061] In this embodiment of the application, in order to reduce the risk of information overload on the family members by status push, the push frequency is controlled in a hierarchical manner: during the high-change phase (such as on the way to the hospital or waiting for medical treatment), the location and time consumption updates are pushed at fixed time intervals; during the low-change phase (such as during the examination), the push is triggered only when the status actually changes, so as to avoid meaningless repeated notifications.

[0062] Step 9: The inspection report is synchronized in real time.

[0063] Once the accompanying physician completes the examination and obtains the report, an instant upload process is triggered. The accompanying physician uses a terminal to photograph or scan the report. The system performs quality checks on the captured images, including clarity and completeness checks. Upon successful verification, the report is immediately uploaded to the patient's file, and a notification is sent to the family member's device indicating that the examination results are ready. The family member can then view the report image and a summary of key indicators automatically generated by the system on their device. The report image is simultaneously added to the patient's past medical history file to continuously enrich the patient's profile data.

[0064] It should be noted that the sharpness detection determines whether the acquired image is blurry by calculating the variance of the Laplacian operator, while the integrity detection determines whether there are missing or cropped parts by identifying whether the boundaries of the report page are completely contained within the acquired image area. When the quality verification fails, a re-capture prompt is sent to the patient's terminal along with the specific problem type.

[0065] It should be noted that the key indicator summary extracts text content from the report image through optical character recognition. Based on the pre-configured departmental examination item indicator template, it matches and extracts the numerical results and reference ranges of each test item from the text content, and marks abnormal indicators that exceed the reference range to generate the key indicator summary.

[0066] Step 10: Service feedback and attribute verification closed loop.

[0067] After each independent service segment is completed, patient feedback evaluation data is obtained. For companions assigned to low-complexity service segments, it is checked whether the structured missing attribute was actually triggered and applied during this service. Based on the triggering status and feedback evaluation data, the verification timestamp of the structured missing attribute is updated and the attribution label is reset. If the structured missing attribute is actually triggered and the feedback evaluation data is positive, the attribution label of the structured missing attribute is removed from the structured missing type and restored to the verified state; if the structured missing attribute is not triggered, the original attribution label remains unchanged. The verification results are included in the historical statistics of subsequent demand activity indicators.

[0068] Meanwhile, real-time status viewing data and satisfaction scores from family members will be incorporated into the evaluation of the transparency of the service process, which will be used to continuously optimize the time distribution parameters of each stage in the Monte Carlo simulation.

[0069] In this embodiment of the application, to increase the credibility of the attribute verification results, for structurally missing attributes of the triggered application, at least two independent positive verifications are required before the attribution label of the structurally missing attribute can be completely removed and restored to the verified state. A single positive verification only affects the slow decay coefficient. Further reduce, but not completely reset, to avoid inaccurate judgment of attribute timeliness caused by a single, accidental positive feedback.

[0070] On the other hand, this invention also proposes an intelligent scheduling system for medical accompaniment based on multi-dimensional profile matching and real-time transparency throughout the entire process, including: The attribute attribution analysis module is used to obtain the global service demand history, count the demand activity index of each attribute, calculate the opportunity exposure rate of unverified attributes for each escort, and generate an attribute verification missing attribution tag set.

[0071] The differentiated decay module is used to calculate the timeliness coefficient by applying a slow decay function to structural missing attributes and an exponential decay function to competitive elimination attributes, thereby generating an effective capability profile vector for each escort.

[0072] The demand profile parsing module is used to parse the patient demand profile, extract the visit type identifier, preference information and escort service preference vector, and generate an enhanced patient demand profile vector for patients with follow-up visits or in-depth examinations.

[0073] The timeline simulation module is used to simulate and project the expected timeline of a patient's medical treatment based on process templates and departmental queuing data using Monte Carlo simulation.

[0074] The process segmentation module is used to identify idle time windows based on the expected medical treatment stage timeline and divide the medical treatment process into multiple independent service segments, extracting the stage-level capability requirement vector and complexity score of each independent service segment.

[0075] The verification mapping module is used to filter low-complexity service segments and perform semantic matching between the auxiliary capability tags of low-complexity service segments and structurally missing attributes to generate a mapping relationship table.

[0076] The allocation scheme generation module is used to generate low-complexity service segment allocation schemes with verification targets based on the mapping relationship table, and to generate high-complexity service segment capability adaptation allocation schemes based on multi-dimensional matching calculations.

[0077] The global optimization module performs a three-objective Pareto sort on the low-complexity service segment allocation scheme and the high-complexity service segment capability adaptation allocation scheme, and outputs a comprehensive relay scheduling scheme.

[0078] The real-time push module is used to obtain the geographical coordinates of the accompanying physician in real time and push real-time status updates and stage time estimates to the family members.

[0079] The dynamic scheduling update module is used to re-retrieve available companions and output updated relay scheduling instructions when the actual time offset exceeds the expected range.

[0080] The report synchronization module is used to verify the quality of the examination report images collected by the accompanying physician and then upload them to the patient's file, and push a report ready notification and a summary of key indicators to the family member's end.

[0081] The feedback loop module is used to acquire patient feedback evaluation data, detect the triggering of structural missing attributes and update the verification timestamp and attribution label, and incorporate family member behavior data into the service process transparency effect evaluation.

[0082] By way of example, the system proposed in this invention may also include various features and combinations thereof in the method embodiments, which will not be elaborated here.

[0083] This invention, through joint analysis of demand activity indicators and opportunity exposure rates, categorizes the reasons for unverified attributes in the companion's competency profile into two types: structural deficiencies and competitive elimination. A slow decay function is applied to structurally deficient attributes for protection, ensuring their continued discoverability in multi-dimensional matching. When a corresponding scarce demand arises, the system can retrieve candidate companions with that capability, overcoming the problem of systematically forgetting scarce capabilities due to uniform exponential decay. By incorporating past medical history, examination reports, and diagnostic conclusions into the enhanced patient demand profile and integrating companion service preference vectors, the matching calculation can cover precise department recommendations and personalized companion adaptation needs in follow-up visits. By decomposing the medical process into multiple independent service segments and implementing relay scheduling based on idle time windows, each independent service segment is executed by the companion with the most suitable capability for that stage. This overcomes the problem of uneven capability across stages for a single companion in the full-process binding model. Simultaneously, low-complexity service segments are transformed into verification scenarios for structurally deficient attributes, continuously creating verification opportunities during normal operation without relying on periods of low system load. With real-time status updates and instant synchronization of examination reports throughout the entire process, family members at remote locations can continuously stay informed about the patient's medical progress and examination results, eliminating the negative impact of information blind spots on service experience and family members' sense of security.

[0084] See Figures 2-9 The following application examples are proposed: A general hospital's (code name: Hospital A) escort service platform provided follow-up appointment escort services for an elderly patient with chronic kidney disease (code name: Patient Wang). Patient Wang has mobility issues and hearing impairment; his children work in another city and are unable to accompany him to the hospital in person. This follow-up appointment involved three parts: a nephrology outpatient visit, renal function blood tests, and an ultrasound examination. The platform currently has several registered escorts (code names: Escort A, Escort B, Escort C, and Escort D), and the multi-dimensional competency profiles of each escort have been maintained in the system. Among them, Escort A possesses sign language communication skills, but this ability has not been positively validated for a long time and is the key target of attribute attribution analysis in this scheduling.

[0085] Step 1: Attribute validation missing attribution analysis.

[0086] The platform reads the recent global service request history. Sign language communication is a low-frequency and scarce skill, with a corresponding recent time window set at 90 to 180 days. Within this time window, service requests involving sign language communication occurred a total of 3 times across the platform. Accompanying therapist A was not included in the candidate list in any of these three related needs scheduling processes. Opportunity Exposure The sign language communication ability was below the preset threshold, therefore it was marked as a structural deficiency. Within the same time window, the "complex medical record communication" ability was requested 28 times across the platform. The escort A was included in the candidates 21 times but was never finally assigned, indicating a high opportunity exposure rate, exceeding the preset threshold. This attribute was marked as a competitive elimination type.

[0087] Table 1. Results of attribution analysis for missing attribute verification of the patient companion (A).

[0088] Step 2: Calculation of differential timeliness decay.

[0089] The attribution labels obtained in Step 1 directly drive the selection of the decay coefficient. For the sign language communication ability of the caregiver A (structural deficiency), a slow decay coefficient is used. For complex medical case communication skills (competitive elimination), the original attenuation coefficient is used. ,and The time interval between each attribute and the most recent positive validation. The effective periods are 95 days and 40 days respectively. The timeliness coefficient of sign language communication ability is calculated as follows: The calculated value is higher than the lower threshold. Calculated values ​​are used directly; communication skills for complex medical records are based on... The calculation shows a greater attenuation. Before multiplying dimension by dimension, the original values ​​of each dimension of the companion A (including score, years of service, etc.) have all undergone mean normalization based on the range, and are uniformly scaled to... The interval is then multiplied by the corresponding timeliness coefficient to generate an effective capability profile vector.

[0090] Table 2. Calculation Process of Effective Competency Profile for Key Competency Dimensions of Accompanying Physician A

[0091] Step 3: Patient needs profile analysis and medical treatment timeline projection.

[0092] Patient Wang submitted a request for a follow-up visit companion service. The system analyzed his needs profile: the visit type was identified as a follow-up visit, the target department was nephrology, there was no preference for a target doctor, the complexity of his condition was moderate, and the companion service preference vector showed a gender preference code of 0 (no preference), a communication style preference code of 1 (patient and meticulous), and a work experience preference code of 2 (more than 3 years). Since the visit type was a follow-up visit, the system further read Patient Wang's past medical history summary, index of previous serum creatinine test reports, and historical diagnostic conclusions. After text embedding processing, a 64-dimensional medical history feature vector was generated, which was concatenated with the 32-dimensional basic needs profile vector to form an enhanced patient needs profile vector with a 96-dimensional dimension.

[0093] In the typical medical process template library, the system retrieves the process template that best matches "follow-up visit + nephrology department," obtaining a node list and a transition probability matrix between nodes for six stages: registration and waiting, outpatient visit, blood test, ultrasound examination, waiting for report, and medication pickup. Combining this with the daily queuing time data for each department at Hospital A, a Monte Carlo simulation is performed to extrapolate the expected timeline of the medical process stages, outputting the estimated start time interval and estimated duration interval for each stage.

[0094] Table 3. Expected Timeline Projection Results for Patient Wang's Medical Treatment Stages

[0095] Step 4: Segmentation of the medical treatment process and extraction of stage-level capability requirements.

[0096] Based on the simulation results, idle time windows between stages were identified. The relay switching threshold was set at 30 minutes based on the spatial scale of Hospital A. The idle time window between outpatient visits and blood tests was estimated to be 35 to 55 minutes; exceeding the 30-minute threshold formed the first cutoff point. The idle time window between ultrasound examinations and waiting for reports was estimated to be 50 to 70 minutes; exceeding the threshold formed the second cutoff point. The idle time windows between the remaining stages did not exceed the thresholds. After segmentation, three independent service segments were generated, which did not exceed the maximum number of independent service segments and did not require merging.

[0097] Service segment 1 includes two stages: registration and waiting for consultation, and outpatient consultation, with an estimated total duration of [duration missing]. Service segment 1 is a high-complexity service segment, lasting 1 minute; service segment 2 includes two stages: blood testing and ultrasound examination, with an estimated total duration of 1 minute. Service segment 3 is a low-complexity service segment, consisting of two stages: waiting for report and medication pickup. The estimated total time is [minutes]. Minutes represent low-complexity service segments. Values ​​of each dimension of the stage-level capacity requirement vector for each service segment. according to Weighted merging, complexity scoring according to Calculation, where For the first The estimated duration of each stage node, The estimated total duration of the service segment, For the first The node Dimensional capability requirement value, For the first The complexity attribute value of each node.

[0098] Table 4 Independent Service Segment Division and Complexity Scoring

[0099] Step 5: Mapping low-complexity service segments to structurally missing attributes.

[0100] The applicable threshold is set to 0.40 for service segment 2. ) and service section 3 ( All values ​​below the threshold form a low-complexity service segment set. An assistive capability tag list is extracted for service segment 3 (medication pickup stage), including assistive tags such as "sign language communication ability" and "dialect communication ability." The assistive capability tag list is semantically matched with the structurally missing attributes (sign language communication ability, rare disease care experience) in the attribute verification missing attribution tag set for escort A. The cosine similarity between the text embedding vectors is calculated: the cosine similarity between the "sign language communication ability" tag and the structurally missing attribute "sign language communication ability" is 0.96, exceeding the matching threshold of 0.80, thus establishing a mapping relationship; "Rare disease care experience" has no high-similarity corresponding item in the low-complexity service segment assistive tags, with the highest cosine similarity being 0.38, therefore no mapping relationship is established.

[0101] Table 5. Mapping relationship between low-complexity service segments and structurally missing attributes

[0102] Step 6: Generation of segmented differential matching and allocation schemes.

[0103] For service segment 3, a set of candidate escorts with sign language communication skills (structural missing) and meeting spatiotemporal accessibility constraints is retrieved. Escort A has the spatiotemporal conditions to reach the pharmacy area of ​​Hospital A before the expected start of service segment 3. The weighted cosine similarity between the stage-level capability requirement vector of Escort A in service segment 3 and the effective capability profile vector of Escort A is 0.83, which exceeds the basic satisfaction threshold of 0.70, satisfying the hard constraint. Therefore, Escort A is included in the candidate set and given the highest verification priority due to the mapping relationship with sign language communication skills. An allocation scheme with verification objectives is generated for service segment 3, prioritizing the allocation of Escort A.

[0104] For service segment 1 (high complexity), the matching quality score is calculated by combining the enhanced patient needs profile vector (96 dimensions) with the effective competency profile vectors of each candidate escort (with an added medical history-related competency dimension, also 96 dimensions) using the following formula: ; in For vector dimensions, For the first Dimensional weights, Vector for profiling patient needs dimensional components, Vector for effectively profiling the capabilities of companion therapists Dimensional components. Accompanying staff member B scored highest in the dimensions of familiarity with nephrology procedures and outpatient communication skills, with a matching quality score of 0.91, and was assigned to service segment 1.

[0105] Table 6. Matching Quality Scores and Allocation Results of Candidate Medical Companions for Each Service Segment

[0106] Step 7: Global multi-objective optimization solution.

[0107] The combination of assigning escort B to service segment 1, escort C to service segment 2, and escort A to service segment 3 is included in the candidate solution set and subjected to Pareto ranking along with other alternative combinations. The three objectives are: global service segment multi-dimensional profile matching quality (the matching quality score of each service segment is weighted and averaged according to the expected duration), escort idle rate (the average proportion of time when all participating escorts are not covered by any service segment), and structural missing attribute verification coverage rate (sign language communication ability is included in the service segment 3 mapping; there are 2 structural missing attributes to be verified, 1 of which is covered, with a coverage rate of 0.50). The average values ​​of the three objectives are normalized and then weighted and summed according to preset weights to obtain the comprehensive utility value. The current optimal solution has a comprehensive utility value of 0.87, the highest among the non-dominated solution sets, and is used as the final relay scheduling comprehensive solution output. When adjacent service segments are connected, the difference in communication style between companion B and companion C is 0.12, and the difference in communication style between companion C and companion A is 0.15. Both are lower than the handover tolerance threshold of 0.30, and the smoothness of the connection meets the auxiliary constraints.

[0108] Step 8: Real-time status push and dynamic scheduling updates throughout the entire process.

[0109] During the medical treatment process, the system obtains the real-time geographical coordinates of each escort. When escort B is en route to Hospital A, the system calculates the real-time distance between escort B and the key point at the hospital gate to be 1.2 kilometers, and pushes this information to the patient's family member's end, simultaneously pushing the "en route to hospital" status and the estimated remaining time until arrival. After escort B arrives at the hospital, the status of each node changes sequentially, and the family member's end receives real-time updates on statuses such as registration completion, waiting for treatment, and outpatient treatment, as well as the estimated start time of each subsequent stage. The actual completion time of the outpatient treatment exceeds the estimated range by 15 minutes, causing the start time of service segment 2 to be shifted later. The system re-searches for the availability of escort C within this time offset window, confirms availability, and then re-invokes the matching calculation in step 6 to regenerate the service segment 2 allocation scheme, outputting the updated relay scheduling instructions, and simultaneously updating the estimated time for subsequent stages on the family member's end.

[0110] During the low-change phase (ultrasound examination in progress), push notifications are triggered only when the status actually changes; during the high-change phase (waiting for appointment), location and time updates are pushed at fixed time intervals.

[0111] Step 9: Check the report and synchronize it instantly.

[0112] After accompanying patient Wang to complete blood tests and an ultrasound examination, caregiver C photographed the ultrasound report using a terminal. The system performed quality checks on the acquired images: clarity was checked using the Laplacian operator variance to ensure the image was not blurry; completeness was checked to confirm that the report page boundaries were fully contained within the acquired image area. Once both checks passed, the report image was immediately uploaded to patient Wang's file, and the family received a notification that the examination results were ready. The system performed optical character recognition on the report image, extracting numerical results and reference ranges for kidney length diameter, cortical thickness, etc., based on the nephrology ultrasound examination indicator template. Abnormal indicators exceeding the reference range (such as low cortical thickness) were marked, and a summary of key indicators was generated and pushed to the family. The report image was simultaneously added to patient Wang's past medical history file, enriching and enhancing the patient's needs profile data.

[0113] Step 10: Service feedback and attribute verification closed loop.

[0114] After each independent service segment was completed, patient Wang submitted feedback data. For service segment 3, the system checked whether the sign language communication ability of escort A was actually triggered: escort A communicated with hearing-impaired patient Wang using sign language at the medication window, and this attribute was confirmed to be triggered. The patient's children submitted a satisfaction rating of 4.8 out of 5 on the family side, indicating positive feedback data. Since escort A's sign language communication ability only had one positive verification record previously, according to the rule that attribution labels can only be completely removed after accumulating two independent positive verifications, only the attenuation coefficient was considered this time. Further reductions, incomplete resets, and attribution labels maintaining structural missing types while enhancing timeliness protection. Family member access behavior data and satisfaction scores are simultaneously incorporated into the transparency effect assessment; related data are used to update the time distribution parameters for each stage in the Monte Carlo simulation, optimizing the accuracy of subsequent medical treatment timeline projections.

[0115] The data flow path throughout the entire implementation process is clear and coherent. In Step 1, the attribute attribution tag set extracted from the global historical record directly drives the selection of the differential attenuation coefficient in Step 2, and the generated effective capability profile vector runs through all matching calculations in Step 6. The follow-up visit type identifier parsed in Step 3 triggers the generation of the enhanced patient demand profile vector, enabling the companion's effective capability profile vector in Step 6 to simultaneously add a medical history-related dimension, ensuring consistency between the two-sided vector dimensions. The expected medical treatment stage timeline deduced in Step 3 is directly input into the segmented calculation in Step 4. The segmentation results determine the set of low-complexity service segments in Step 5, and the mapping table further constrains the candidate retrieval range of low-complexity service segments in Step 6. The Pareto sort in Step 7 aggregates all allocation schemes from Step 6, and the output relay scheduling integrated scheme is monitored and executed in real time in Step 8. The dynamic scheduling update in Step 8 calls the matching calculation in Step 6 to regenerate instructions when the actual time offset occurs. The report data in Step 9 is written back to the patient file, and the verification results and feedback data in Step 10 flow back to the historical statistics and distribution parameters of Steps 1 and 2, forming a complete data loop.

[0116] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for intelligent scheduling of medical accompaniment services based on multi-dimensional profile matching and real-time transparency throughout the entire process, characterized in that, Includes the following steps: Obtain the global service demand history, count the demand activity index of each attribute, calculate the opportunity exposure rate of each companion's unverified attributes, mark unverified attributes with an opportunity exposure rate below a preset threshold as structural missing type, and mark unverified attributes with an opportunity exposure rate above a preset threshold but not assigned as competitive elimination type. For attributes with structural missing types, a slow decay function is used, and for attributes with competitive elimination types, an exponential decay function is used to calculate the timeliness coefficient. These coefficients are then multiplied dimension by dimension with the original multidimensional competency profile vector of the companion physician to generate an effective competency profile vector. We analyze patient needs profiles and use Monte Carlo simulation based on process templates to extrapolate the expected timeline of medical treatment stages. Based on the idle time windows between stages, we divide the medical treatment process into multiple independent service segments and extract the stage-level capability requirement vector and complexity score of each independent service segment. Select low-complexity service segments and semantically match their auxiliary capability tags with structurally missing attributes to generate a mapping table. Based on the mapping table, generate a low-complexity service segment allocation scheme with verification objectives. Based on multi-dimensional matching calculation, generate a capability adaptation allocation scheme for high-complexity service segments. Jointly execute multi-objective optimization to output a relay scheduling comprehensive scheme. During the medical treatment process, the system pushes real-time status updates and estimated time consumption for each stage of the process to the family members.

2. The intelligent scheduling method for multi-dimensional profile matching and real-time transparency throughout the entire medical escort process as described in claim 1, characterized in that, The formula for calculating the opportunity exposure rate is: ,in For opportunity exposure rate, This refers to the number of times the accompanying therapist is included in the candidate list in the scheduling of needs related to this attribute. This refers to the total frequency of needs related to the patient needs profile within the recent time window for this attribute; when When the value is zero, the corresponding unverified attribute is directly marked as a structural missing type.

3. The intelligent scheduling method for multi-dimensional profile matching and real-time transparency throughout the entire medical escort process as described in claim 1, characterized in that, The formula for calculating the slow decay function is as follows: The formula for calculating the exponential decay function is as follows: ,in For slow decay coefficient, The original attenuation coefficient, The time interval since the most recent positive verification. The value is less than The slow decay function sets a lower threshold for the timeliness coefficient. ,when Below Time As output.

4. The intelligent scheduling method for multi-dimensional profile matching and real-time transparency throughout the entire medical accompaniment process as described in claim 1, characterized in that, The process of analyzing the patient needs profile includes: extracting the visit type identifier, target hospital and department preferences, disease complexity features, and escort service preference vectors; if the patient is a follow-up visit or undergoing in-depth examination, the patient's medical history summary, index of previous examination reports, and historical diagnostic conclusions are read, vectorized, and concatenated with the basic needs profile vector to generate an enhanced patient needs profile vector; correspondingly, a medical history-related ability dimension is added to the escort's effective ability profile vector to ensure consistency between the two vector dimensions.

5. The intelligent scheduling method for multi-dimensional profile matching and real-time transparency throughout the entire medical escort process as described in claim 1, characterized in that, The first stage-level capability requirement vector Dimensional values Calculated by the following formula: ,in This represents the total number of stage nodes contained in this independent service segment. For the first The estimated duration of each stage node, The total estimated duration of this independent service segment, For the first The stage node is at the The capability requirement intensity value across each capability dimension; the complexity score. The calculation formula is: ,in For the first The process complexity attribute value of each stage node.

6. The intelligent scheduling method for multi-dimensional profile matching and real-time transparency throughout the entire medical escort process as described in claim 1, characterized in that, The multidimensional matching calculation uses weighted cosine similarity, and the calculation formula is as follows: ,in To match quality scores, Let be the number of dimensions of the image vector. For the first Weight coefficients for each dimension The first vector of patient needs profile Each dimension value The first effective capability profile vector for companion therapists Each dimension value.

7. The intelligent scheduling method for multi-dimensional profile matching and real-time transparency throughout the entire medical escort process as described in claim 1, characterized in that, The multi-objective optimization uses three objectives—maximizing the matching quality of multi-dimensional profiles of global service segments, minimizing the idle rate of companions, and maximizing the verification coverage of structurally missing attributes—to perform Pareto ranking. The solution with the highest comprehensive utility value is selected from the non-dominated solution set as the final output. The Pareto ranking introduces the smoothness of independent service segment connection as an auxiliary constraint, requiring that when adjacent independent service segments are assigned to different companions, the difference in communication style between the latter companion and the former companion is lower than the switching tolerance threshold.

8. The intelligent scheduling method for multi-dimensional profile matching and real-time transparency of the entire medical escort process as described in claim 1, characterized in that, Also includes: When the actual completion time of each stage exceeds the expected range of the expected medical treatment stage timeline, causing the start time of subsequent independent service segments to shift, the available companion physician set within the time shift window is re-searched. Based on the effective capability profile vector, the multi-dimensional profile matching and allocation scheme of subsequent independent service segments is recalculated, and the updated relay scheduling instruction is output and the estimated time consumption of subsequent stages is simultaneously updated and pushed to the family member's end.

9. The intelligent scheduling method for multi-dimensional profile matching and real-time transparency of the entire medical escort process as described in claim 1, characterized in that, It also includes a closed-loop process for real-time synchronization of examination reports and service feedback: After the escort collects the images of the examination report, a quality check is performed on the clarity and completeness of the images. Once the check is passed, the images are uploaded to the patient's file and a summary of key indicators is pushed to the family member's end. After each independent service segment is completed, patient feedback evaluation data is obtained. For escorts assigned to low-complexity service segments, it is checked whether the structural missing attributes are actually triggered and applied. The attribution labels are updated based on the triggering situation and feedback evaluation data. At least two independent positive verifications are required before the attribution labels can be completely removed and restored to the verified state.

10. A medical escort intelligent scheduling system based on multi-dimensional profile matching and real-time transparency throughout the entire process, used to execute the medical escort intelligent scheduling method based on multi-dimensional profile matching and real-time transparency throughout the entire process as described in any one of claims 1 to 9, characterized in that, include: The attribute attribution analysis module is used to obtain the global service demand history, count the demand activity index of each attribute, calculate the opportunity exposure rate of each escort's unverified attributes, and generate attribute verification missing attribution tag set. The differentiated decay module is used to calculate the timeliness coefficient by applying a slow decay function to structural missing attributes and an exponential decay function to competitive elimination attributes, thereby generating an effective capability profile vector for each escort. The demand profile parsing module is used to parse the patient demand profile, extract the visit type identifier, preference information and escort service preference vector, and generate an enhanced patient demand profile vector for patients with follow-up visits or in-depth examinations. The timeline simulation module is used to simulate and extrapolate the patient's expected medical treatment timeline using Monte Carlo simulation based on process templates and departmental queuing data. The process segmentation module is used to identify idle time windows based on the expected medical treatment stage timeline and divide the medical treatment process into multiple independent service segments, extracting the stage-level capability requirement vector and complexity score of each independent service segment; The verification mapping module is used to filter low-complexity service segments and semantically match their auxiliary capability tags with structurally missing attributes to generate a mapping relationship table; The allocation scheme generation module is used to generate low-complexity service segment allocation schemes with verification targets based on the mapping relationship table, and to generate high-complexity service segment capability adaptation allocation schemes based on multi-dimensional matching calculations. The global optimization module is used to perform multi-objective Pareto sorting and output a comprehensive relay scheduling scheme for low-complexity service segment allocation schemes and high-complexity service segment capacity adaptation allocation schemes. The real-time push and dynamic scheduling module is used to obtain the geographical coordinates of the escorts in real time and push the real-time status updates of the entire process nodes and the stage time estimation to the family. When the actual time deviation exceeds the expected range, it will re-search for available escorts and output the updated relay scheduling instructions. The report synchronization and feedback closed-loop module is used to upload the examination report images to the patient's file after quality verification and push key indicator summaries, obtain patient feedback evaluation data to detect the triggering of structural missing attributes and update the attribution labels.

Images