A patient image-based proton therapy intelligent guidance method and system

Abstract

The application provides a kind of intelligent guide diagnosis method and system based on patient portrait of proton therapy, it is related to medical data processing technical field, method includes: obtaining the historical proton therapy interactive text of patient operation terminal;According to historical proton therapy interactive text, the history portrait node similar graph of patient is constructed;Obtain the current intelligent question and answer dialogue text of current patient operation terminal;According to current intelligent question and answer dialogue text, select the unmarked feature node of current patient;The unmarked feature node of current patient is loaded into history portrait node similar graph, and the target guide diagnosis label in history portrait node similar graph is determined;According to target guide diagnosis label, from the candidate machine question and answer service path set that guide terminal server pre-establishes in machine question and answer service integration screening;The path sorting of candidate machine question and answer service path set is carried out, and the rendering sequence of proton therapy intelligent question and answer guide diagnosis interaction is generated.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and in particular to an intelligent triage method and system for proton therapy based on patient profiles. Background Technology

[0002] The field of medical data processing technology refers to the application of computer technology in the collection, processing, storage, analysis and management of medical information. It mainly involves patient information management, medical data mining, clinical decision support and information system integration, aiming to standardize the processing of medical-related data and support medical service processes.

[0003] Among them, proton therapy refers to a radiotherapy method that uses charged proton beams to irradiate tumor tissue. The intelligent proton therapy triage method refers to a method that guides and allocates the medical process based on the patient's basic information and clinical needs. It usually achieves the planning and guidance of the patient's medical path through technical means such as rule matching information query and human experience.

[0004] However, traditional intelligent triage methods for proton therapy can only guide patients through basic procedures based on their basic information and needs. When faced with complex consultation scenarios that require comprehensive assessment of the patient's condition, past treatments, examinations, and reports, they are mostly limited to simple rule matching, information retrieval, or rely on human experience for allocation. They cannot mine and reuse triage data related to current patients and similar patients in the past, nor can they combine past user interactions to intelligently filter and orderly display multiple service paths. Summary of the Invention

[0005] To address the limitations of traditional intelligent triage methods for proton therapy, which rely solely on basic patient information and needs for guidance, and are often limited to simple rule matching, information retrieval, or manual experience when faced with complex consultation scenarios requiring comprehensive assessment of the patient's condition, past treatments, examinations, and reports, these methods are unable to effectively mine and reuse triage data linking current patients with similar historical patients, nor can they combine past user interactions to intelligently filter and orderly display multiple service paths. This invention provides an intelligent triage method and system for proton therapy based on patient profiles.

[0006] The technical solutions provided by the embodiments of the present invention are as follows: The first aspect of this invention provides a proton therapy intelligent triage method based on patient profiles, comprising: S1: Obtain the historical proton therapy interaction text from the patient's operating terminal; S2: Construct a similarity graph of historical profile nodes for patients based on historical proton therapy interaction text; S3: Obtain the current intelligent question-and-answer dialogue text on the current patient's operating terminal; S4: Based on the current intelligent question-and-answer dialogue text, select the unmarked feature nodes of the current patient; S5: Load the unlabeled feature nodes of the current patient into the historical portrait node similarity map, and determine the target triage label in the historical portrait node similarity map; S6: Based on the target triage label, filter the candidate machine question-and-answer service path set from the machine question-and-answer service path set pre-established by the triage terminal server; S7: Sort the candidate machine question-answering service path set and generate a proton therapy intelligent question-answering guidance interactive rendering sequence.

[0007] A second aspect of this invention provides a proton therapy intelligent triage system based on patient profiles, comprising: processor; The memory stores computer-readable instructions, which, when executed by a processor, implement the proton therapy intelligent triage method based on patient profiles as described in the first aspect.

[0008] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent triage method for proton therapy based on patient profiles as described in the first aspect.

[0009] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this embodiment of the invention, the historical proton therapy interaction text of the patient's operating terminal is first obtained. Based on this text, a similarity graph of the patient's historical profile nodes is constructed, establishing an association system for historical patient triage data. Then, the current intelligent question-and-answer dialogue text of the current patient's operating terminal is obtained. Unmarked feature nodes of the current patient are selected and loaded into the historical profile node similarity graph. Target triage tags are determined based on the triage associations of historical patients, effectively reusing the triage associations between the current patient and similar historical patients, overcoming the limitations of traditional solutions that rely solely on rules and human experience. According to the determined target triage tags, a matching set of candidate machine question-and-answer service paths is selected from the machine question-and-answer service path set pre-established by the triage terminal server. The candidate machine question-and-answer service path set is then sorted to generate a proton therapy intelligent question-and-answer triage interaction rendering sequence, completing the accurate selection of multiple service paths and the orderly determination of their display order. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating an intelligent triage method for proton therapy based on patient profiles, provided as an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram of the structure of a proton therapy intelligent triage system based on patient profiles, provided in an embodiment of the present invention. Detailed Implementation

[0013] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0014] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0015] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0016] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0017] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0018] Reference manual attached Figure 1 The diagram illustrates a flowchart of an intelligent triage method for proton therapy based on patient profiles, provided by an embodiment of the present invention.

[0019] This invention provides a patient-profile-based intelligent triage method for proton therapy. This method can be implemented using a patient-profile-based intelligent triage device for proton therapy, which can be a terminal or a server. The processing flow of the patient-profile-based intelligent triage method for proton therapy may include the following steps: S1: Obtain the historical proton therapy interaction text on the patient's operating terminal.

[0020] S2: Construct a similarity graph of the patient's historical profile nodes based on the historical proton therapy interaction text.

[0021] Optionally, the historical profile node similarity graph specifically includes: the patient's historical feature vector and the actual triage destination category.

[0022] In one possible implementation, S2 specifically includes sub-steps S201 to S205: S201: Extract the patient's historical tumor type information, historical lesion location information, historical tumor stage information, historical treatment method information, historical pathology examination status identifier, and historical imaging report status identifier from the historical proton therapy interactive text, and map them into a preset multi-dimensional space to combine them into the patient's historical feature vector.

[0023] Specifically, firstly, the patient's operating terminal is the web-based or mobile Q&A page used by the patient. Historical proton therapy interaction text refers to text records generated from previous machine-based Q&A guidance sessions. Proton therapy refers to radiotherapy that uses proton beams to irradiate the lesion area. Tumor type information indicates the patient's disease name; lesion location information indicates the body region where the tumor is located; tumor stage information indicates the stage of disease progression; previous treatment information indicates the types of treatments the patient has received; pathology examination status indicates whether tissue or cell examinations have been completed; and imaging report status indicates whether imaging reports have been submitted or are readable. Next, historical conversation text for the same patient is retrieved from the Q&A record table stored on the guidance terminal server. The record source is limited to the guidance consultation input box, registration consultation entry, report query entry, and manually transcribed records. After retrieval, the records are arranged by patient ID, conversation time, text content, and entry source. Then, each text record is cleaned up, deleting blank entries, duplicates, garbled characters, and content containing only irrelevant punctuation. When multiple values ​​appear for the same field, the most recent valid text is registered; unrecognizable fields are registered as not provided. Next, the pre-defined multi-dimensional space is established by the triage personnel in the field configuration table. This is based on the six types of patient information that need to be extracted in this step. Each type of information occupies a fixed dimension position, and the dimension order is determined by the field number in the field configuration table. Tumor type, lesion location, tumor stage, previous treatment methods, pathology examination status identifier, and imaging report status identifier are all written in this order. Field values ​​are formed by a one-to-one correspondence between text items and numerical items in the configuration table. Next, the cleaned field table is read, and text items are converted into corresponding numerical items in the configuration table. Fields without provided values ​​are converted into preset missing values. The process of determining preset missing values ​​involves first listing all codes that can normally appear in each field, and then selecting a fixed value that does not overlap with any normal code as a unified marker for the missing field. For example, if a patient's previous consultation text mentions lung cancer, chest, stage III, chemotherapy, completed pathology, or uploaded imaging, these are written into the corresponding dimension positions and arranged according to the field number. Finally, the patient's historical feature vector is generated.

[0024] S202: Define the historical feature vector of each patient as a data node and obtain the pre-configured triage destination category.

[0025] Optionally, the specific categories for patient guidance include: patient registration, report inquiry, electronic invoice, and personalized Q&A with medical institutions.

[0026] Among them, a data node is a data record that carries a patient's historical feature vector.

[0027] It should be noted that the patient's historical feature vectors are registered separately according to the patient identifier, and each patient corresponds to only one independent data node. When the same patient has multiple historical sessions, they are first sorted by session time, and then the record that generated the most recent clear referral destination is read.

[0028] S203: Set the triage destination category as the preset proton therapy triage label of the data node, and generate a labeled node.

[0029] Among them, the tagged nodes are data records that have been written into the proton therapy triage destination.

[0030] Among them, the proton therapy referral label is the category of referral service that the patient with this history will eventually enter.

[0031] Specifically, the pre-configured triage destination categories are established by the triage terminal server's backend path maintenance page. The criteria are based on the four service entry points actually accessible on the current triage page. These entry point names are directly registered from the service menu on the backend page, corresponding to triage registration, report inquiry, e-invoices, and personalized Q&A with medical institutions, respectively. Next, the text of the last service entry point in the historical session is read and compared character-by-character with the four triage destination categories. If the number of characters matches and the position of each character is consistent, the category is registered as the proton therapy triage label for the data node. If the number of characters does not match or any character does not match, the comparison continues to the next category. If all categories do not match, the node is registered as a record pending verification and no valid triage label is written. For example, if a patient's historical session last entered the e-invoice download page, the proton therapy triage label registered for that node would be the e-invoice category, generating a tagged node.

[0032] S204: Calculate the Euclidean distance between the historical feature vectors of each patient.

[0033] Here, the Euclidean distance is the sum of the numerical differences between the historical feature vectors of two patients at the same dimensional position.

[0034] Specifically, the process reads labeled nodes and verifies the number of dimensions for each node. If the number of dimensions matches the field configuration table, it participates in distance calculation; otherwise, it is registered as an incalculable record. Next, the preset connection threshold range is determined by the distance distribution of similar labeled nodes in historical triage records. First, patient nodes that have clearly entered the same triage destination are selected from the historical records, and their distances are calculated pairwise. Then, these nodes are sorted by distance from smallest to largest. The upper limit is determined by the position where most similar labeled nodes can connect to each other and where different labeled nodes have minimal overlap. The lower limit is fixed at 0, indicating that the two vectors have no difference in position across dimensions. Then, the values ​​of any two labeled nodes are read dimension by dimension to obtain the difference between positions in the same dimension. Finally, all dimensional differences are summarized, and the Euclidean distance is calculated as the square root of the sum of the squares of the differences across dimensions, yielding the distance between the two nodes.

[0035] Furthermore, the distance value is compared with a preset connection threshold range. When the distance value is 0, it is registered as completely similar. When the distance value is greater than 0 and within the threshold range, an edge connection is established. When the distance value exceeds the upper limit of the threshold, no edge connection is established. Dimensions with missing values ​​are included in the calculation according to the missing value marker. When the number of missing dimensions exceeds half of the total number of fields, no edge connection is established between the nodes. For example, if two patients have similar tumor types, lesion locations, stages, previous treatment methods, pathological states, and imaging states, and their distance values ​​fall within the threshold range, then an edge connection is registered between the corresponding nodes of the two patients.

[0036] S205: Construct a similarity graph of historical portrait nodes based on Euclidean distance.

[0037] Among them, the historical portrait node similarity graph is a patient portrait association structure formed by labeled nodes and edge connections.

[0038] Among them, the edge connection relationship is the association record established between two labeled nodes due to the similarity of features.

[0039] In this embodiment of the invention, six types of core diagnostic and treatment information are extracted from historical interactive text and quantified into multi-dimensional feature vectors. Labeled nodes are generated by binding the actual triage destination of the patient. Then, node similarity associations are established based on Euclidean distance and a preset threshold to construct a similarity graph of historical profile nodes. This not only realizes the transformation of unstructured patient consultation text into a computable and quantifiable profile, fully covering the key dimensions required for proton therapy triage such as the patient's condition, previous treatment, and examination status, but also deeply binds patient feature similarity with historical triage results through graph structure. This provides a reliable computational carrier for the subsequent label propagation algorithm to match the current patient's triage label, effectively solving the problem that traditional triage cannot reuse the experience of triage for similar patients in the past. At the same time, the thresholded edge connection rules also take into account the accuracy and computational efficiency of similarity matching.

[0040] S3: Obtain the current intelligent question-and-answer dialogue text on the current patient's operating terminal.

[0041] S4: Based on the current intelligent question-and-answer dialogue text, select the unlabeled feature nodes of the current patient.

[0042] Specifically, the unlabeled feature nodes of the current patient include the current patient feature vector and unlabeled nodes.

[0043] In one possible implementation, S4 specifically includes sub-steps S401 and S402: S401: By comparing characters, extract the current patient's tumor type, lesion location, tumor stage, previous treatment methods, pathological examination status, and imaging report status from the current intelligent question-and-answer dialogue text, and map them to the same multidimensional space as the patient's historical feature vector for splicing to generate the current patient feature vector.

[0044] Specifically, the current intelligent question-and-answer dialogue text is the machine-generated question-and-answer text entered or selected by the current patient on the triage page, and the current patient feature vector is the numerical arrangement result of the current patient's six fields in the same multi-dimensional space. Next, the input text, option text, and recognizable text after the report upload are read from the current session on the patient's terminal. The read content is organized according to the patient's temporary identifier, session time, session round, and text source. Next, the current text is cleaned, deleting duplicate punctuation, meaningless whitespace, and characters that cannot form the meaning of the fields, retaining text directly corresponding to tumor type, lesion location, tumor stage, previous treatment methods, pathology examination status identifier, and imaging report status identifier. Next, the current text is first segmented into continuous word segments, and then compared character by character with the field vocabulary in the field configuration table. When a matching text item is found in the field vocabulary, that text item is registered in the corresponding field. When multiple candidate texts appear simultaneously, the last valid text appearing in the current session is used. Next, the current field is written to the established same-dimensional position. Fields that do not appear are written with the same missing value. Fields that conflict and cannot be distinguished by time are registered as fields to be verified.

[0045] For example, if the current patient enters nasopharyngeal carcinoma, head and neck, stage II, has undergone radiotherapy, pathology completed, and images not uploaded, these fields will be written to the corresponding dimension positions in sequence.

[0046] S402: Define the current patient feature vector as an unlabeled node, and use the unlabeled node as the current patient's unlabeled feature node.

[0047] Among them, unmarked nodes are current patient data records that have not yet been written into the triage destination category.

[0048] Specifically, the current patient's feature vector is read and recorded in the current session's temporary record table. The record includes the patient's temporary identifier, the current patient's feature vector, the field source, and the session time. Next, it checks if a manually designated referral destination already exists in the current session. If no manually designated destination exists, the record is recorded as an unmarked node; otherwise, it is not treated as an unmarked node. Then, if the current patient's feature vector contains missing fields, the missing values ​​are retained, and content not present in the current conversation is not added.

[0049] For example, if the current patient only provides the disease type and report status, but has not yet entered any triage path, then the current patient's feature vector is retained as unlabeled.

[0050] It should be noted that the unlabeled feature nodes of the current patient are the current patient records that need to be associated with the similarity map of historical profile nodes in this session. Next, the unlabeled nodes are read from the temporary record table, and the temporary patient identifier is checked against the session identifier in the current intelligent question-and-answer dialogue text. Then, if the identifiers are completely consistent, the unlabeled node is selected; if the identifiers are inconsistent, the corresponding record is searched again according to the session time and entry source; if a unique corresponding record is found, it is selected; if multiple corresponding records are found, the record with the latest time is read; if no corresponding record is found, it is registered as a record to be verified. Next, the current patient feature vector, field source, and missing field position in the selected record are read, and its unlabeled state is maintained. Finally, the unlabeled feature nodes of the current patient are generated.

[0051] For example, if a patient continuously enters multiple rounds of consultation text on the same triage page, and the temporary identifier remains consistent, then the unmarked node corresponding to that session will be selected.

[0052] In this embodiment of the invention, six types of diagnostic features that are completely consistent with the dimensions of historical profiles are extracted from the current intelligent question-and-answer dialogue text. These features are then mapped to a unified multidimensional space to generate a standardized current patient feature vector, which is then encapsulated as an unlabeled feature node. This ensures that the current patient data and the historical profile system are dimensionally identical and consistent in scope, providing a consistent computational carrier for subsequent loading of historical profile node similarity graphs for tag propagation and similarity matching. Furthermore, through mechanisms such as character word matching, missing field retention, manual triage destination verification, and conversation identifier verification, the invention adapts to real-time interaction scenarios and quickly completes the structuring transformation of unstructured dialogues while ensuring the authenticity of feature data and the accuracy of node attribution. This avoids repetitive calculations in scenarios requiring manual intervention and achieves efficient and accurate transformation of current patient real-time consultation information into computable triage matching units.

[0053] S5: Load the unlabeled feature nodes of the current patient into the historical portrait node similarity map, and determine the target triage label in the historical portrait node similarity map.

[0054] Among them, the target triage label specifically refers to the preset proton therapy triage label.

[0055] In one possible implementation, S5 specifically includes sub-steps S501 to S504: S501: Load the current patient's unlabeled feature nodes into the historical image node similarity graph, and calculate the current Euclidean distance between the current patient's unlabeled feature nodes and each labeled node in the multidimensional space using the label propagation algorithm.

[0056] The label propagation algorithm is a computational process that transmits the existing triage destinations of historical nodes to the current patient node according to the similarity between the nodes.

[0057] It should be noted that the process involves loading the similarity map of historical patient images and the unlabeled feature nodes of the current patient, reading the proton therapy referral labels on each historical node, and registering these labels as transferable soft labels. Next, the current Euclidean distance between the unlabeled feature nodes of the current patient and each labeled node is calculated. Before calculation, the dimensional positions of the two nodes are checked for consistency. If the dimensions are consistent, the numerical differences are read dimension by dimension. The current Euclidean distance is calculated as the square root of the sum of the squares of the numerical differences in each dimension. If the dimensions are inconsistent, the labeled node does not participate in this transfer.

[0058] S502: Convert the current Euclidean distance into similarity weights.

[0059] The similarity weight is a numerical value representing the degree of proximity between the current patient node and a certain historical node.

[0060] Specifically, the current Euclidean distance is converted into similarity weights. The conversion is based on the following principle: the smaller the distance value, the fewer the field differences, and the larger the weight; conversely, the larger the distance value, the more the field differences, and the smaller the weight. When the distance value exceeds the preset connection threshold, the weight is registered as 0. The upper limit of the similarity weight is determined by the case where the distance value is 0, because 0 indicates that the current patient feature vector and the historical feature vector are completely identical in all dimensions, and the similarity weight is the maximum registered value. When the distance value is between 0 and the upper threshold, the weight is registered in descending order of distance. The similarity weight is the difference between 1 and the proportion of the current Euclidean distance to the upper threshold of the preset connection threshold. When the distance value exceeds the upper threshold, no effective weight is registered.

[0061] For example, if a current patient and a historical patient are identical in tumor type, lesion location, and reporting status, but differ only in past treatment methods, then the weight of that historical node is higher than that of historical nodes with more significant field differences.

[0062] S503: Based on the similarity weight, the soft labels on each labeled node are transmitted along the edge connection relationship to the current patient's unlabeled feature node, and the guidance probability of each proton therapy guidance label being distributed to the current patient's unlabeled feature node is calculated.

[0063] Among them, soft tags are the categories of patient referral destinations that can be included in the referral process.

[0064] The triage probability is the proportion of a certain triage destination category that is assigned to the current patient's unlabeled feature node.

[0065] Among them, the edge connection relationship is the historical node association relationship that has been registered in the similarity graph of historical portrait nodes.

[0066] Specifically, the similarity weights and soft tags corresponding to each tagged node are read, and cumulative records are established for the categories of patient registration, report query, e-invoice, and personalized Q&A with medical institutions. Next, tagged nodes that can form effective weights with the current patient node are read along the existing edge connections in the historical profile node similarity graph, and the weight of each node is written into the cumulative record corresponding to its soft tag. Nodes with a weight of 0 are not written into the cumulative record. Then, the cumulative records of the four categories are proportionally converted: the patient referral probability = the ratio of the cumulative similarity weight of a certain category to the total cumulative similarity weight of all categories, thus obtaining the referral probability for that category. Next, if the cumulative weights of all categories are 0, the referral probabilities of all four categories are registered as indeterminate. If only one category has an effective cumulative weight, the probability of that category is registered as the highest, and the other categories are registered as 0. If multiple categories have effective cumulative weights, they are registered according to the proportion of each category's cumulative weight to the total cumulative weight. For example, if most of the effective historical nodes associated with the current patient belong to the report query category, and the corresponding cumulative weight value is the largest, then the report query category receives the highest referral probability. The end of the process generates the probability of each proton therapy referral label being distributed to the current patient's unlabeled feature nodes.

[0067] S504: Sort the triage probabilities in descending order, and use the proton therapy triage label bound to the highest triage probability as the target item to determine the target triage label.

[0068] The target item is the label with the highest probability selected from multiple triage probabilities.

[0069] Among them, the target triage label is the triage destination category used when matching the current patient with subsequent service paths.

[0070] Specifically, the system reads four categories of triage probabilities and sorts them from highest to lowest value. Next, if the first category contains only one category, the corresponding proton therapy triage tag is extracted. If the first category contains two or more categories with the same probability, the number of valid historical nodes corresponding to each category is read, and the category with the larger number is selected. If the number of valid historical nodes is still the same, the last occurrence of the service intent text in the current intelligent question-and-answer dialogue is read, and the category whose name is exactly the same is selected. If it is still impossible to distinguish, it is registered as pending confirmation and not forcibly written into a category. Then, the sorting is based on the triage probability itself; a higher probability indicates that historical nodes more closely related to the current patient's characteristics have a higher cumulative weight in that category, while a lower probability indicates that the category has a lower cumulative weight. Finally, the target triage tag is generated.

[0071] For example, if the report query category is ranked first and there are no other categories listed, then the report query category is extracted.

[0072] In this embodiment of the invention, relying on the constructed historical profile node similarity graph, the referral destinations of historically tagged nodes are used as soft labels through a label propagation algorithm. Similarity weights calculated using Euclidean distance are used as the basis for transmission, moving along node edge connections to the current patient's unlabeled node. Weighted cumulative calculations determine the distribution probability of various referral labels, and then, combined with probability descending sorting and multi-dimensional fallback rules for parallel scenarios, the target referral label is finally determined. This mechanism overcomes the limitations of traditional referral systems that rely on fixed rule matching and cannot reuse historical experience with similar patients. It can distinguish the reference value of different historical cases based on quantified similarity weights, making referral label determination more aligned with patients' actual treatment needs and historical treatment path patterns. Simultaneously, the accompanying boundary scenario processing rules cover special cases such as no effectively associated nodes and parallel probabilities, effectively improving the accuracy and robustness of referral classification. Furthermore, the entire process can be automated and executed in real-time, providing a reliable classification benchmark for the accurate selection of subsequent machine question-and-answer service paths.

[0073] S6: Based on the target triage label, select a set of candidate machine question-and-answer service paths from the set of machine question-and-answer service paths pre-established by the triage terminal server.

[0074] Among them, the candidate machine question-answering service path set is a list of paths that completely match the target triage label and can continue to participate in the ranking.

[0075] The candidate machine question-and-answer service path set specifically includes: the triage and registration path, the report query path, the electronic invoice path, and the personalized question-and-answer path for medical institutions.

[0076] Among them, the set of machine question-and-answer service paths is a list of service entry points that can be displayed and stored in the triage terminal server.

[0077] The triage and registration path is the machine-guided question-and-answer service path that leads to the registration-related pages.

[0078] The report query path is the machine Q&A service path that leads to the inspection report viewing page.

[0079] The electronic invoice path is the machine Q&A service path that leads to the invoice processing page.

[0080] Among them, the personalized Q&A path for medical institutions is the machine Q&A service path that leads to the institution's consultation and Q&A page.

[0081] In one possible implementation, S6 specifically includes sub-steps S601 to S603: S601: Based on the set of machine question-and-answer service paths, classify the categories.

[0082] The classification identifier is the triage category text to which each path belongs.

[0083] It should be noted that the system reads the established path list from the path maintenance page of the triage terminal server backend. This list consists of path name, path address, entry level, category identifier, activation status, and update time. Next, the pre-established path set is registered by backend maintenance personnel according to the actual accessible entry points on the patient's terminal. The criteria for this registration are the service entry points already open on the patient's terminal page; entry points not open on the patient's terminal page are not included in this set. Then, for each machine-guided question-and-answer service path, its name and the service type registered in the backend are read, and the path is categorized into the triage registration category, report query category, e-invoice category, or personalized medical institution question-and-answer category. The categorization text is consistent with the triage destination category. For example, the examination report viewing entry is categorized into the report query category, and the invoice download entry is categorized into the e-invoice category. Finally, a set of machine-guided question-and-answer service paths with category identifiers is generated.

[0084] S602: Compare the classifying identifier with the target triage label character by character to obtain the character comparison results.

[0085] Among them, the character comparison result is the registration result of whether the classification identifier and the target triage label are completely consistent.

[0086] Specifically, the system reads the target triage label and then reads the category identifier of each path in the machine question-and-answer service path set. Next, it compares the number of characters for two text entries; if the number of characters differs, it is recorded as inconsistent; if the number of characters is the same, it is compared character by character position; if all positions are the same, it is recorded as completely consistent; if any position differs, it is recorded as inconsistent. Then, if the target triage label is pending confirmation, all paths are recorded as unmatchable, and the character comparison results are generated at the end.

[0087] For example, if the target triage label is an e-invoice category and the classification identifier of a certain path is also an e-invoice category, then it is registered as completely consistent; if the classification identifier of a certain path is a report query category, then it is registered as inconsistent.

[0088] S603: Based on the character comparison results, extract the paths whose classification identifiers are completely consistent with the target triage label from the machine question-and-answer service path set, and combine all the extracted paths to generate a candidate machine question-and-answer service path set.

[0089] Specifically, the character comparison results are read, and path records in the machine question-and-answer service path set are read one by one. Next, paths whose category identifier and target triage label are completely identical are extracted; inconsistent paths and paths that do not match are not extracted. Then, the activation status of the extracted paths is read: paths that are enabled are retained, paths that are disabled are removed, and paths with an empty activation status field are registered as pending verification and not added to the candidate set. Finally, the retained paths are organized into a unified record according to path name, path address, category identifier, entry level, and update time, and a candidate machine question-and-answer service path set is generated at the end.

[0090] For example, when the target referral label is the report query category, the entry points for viewing imaging reports and pathology reports are retained, but the entry point for downloading electronic invoices is not retained.

[0091] In this embodiment of the invention, various triage service paths within the server are first uniformly assigned category identifiers. Then, through precise character comparison, the identifiers are matched with target triage tags. Simultaneously, a secondary screening is performed based on the path's activation status, ultimately yielding a set of compliant candidate service paths. This method strictly adheres to the triage classification results obtained in the previous step, achieving precise correspondence between tags and service paths, eliminating irrelevant or disabled invalid paths, and significantly reducing the scope of subsequent sorting calculations. This ensures that triage services match patient needs, simplifies subsequent processes, improves overall operational efficiency, and, at the same time, standardized path management guarantees the availability and stability of triage services.

[0092] S7: Sort the candidate machine question-answering service path set and generate a proton therapy intelligent question-answering guidance interactive rendering sequence.

[0093] The intelligent question-and-answer guidance and interactive rendering sequence for proton therapy specifically includes: the feature value and display priority score for each path.

[0094] In one possible implementation, S7 specifically includes sub-steps S701 to S707: S701: Obtain patient operation terminal information from the triage terminal server.

[0095] The patient's terminal information specifically includes: historical interface click status, historical page dwell time, and historical number of follow-up inquiries.

[0096] Among them, the historical interface click status is a record of whether the patient clicked and entered a certain service path.

[0097] Among them, the duration of time spent on a historical page is the length of time between when a patient opens a page and when they leave that page.

[0098] Among them, the number of times the patient asked follow-up questions in history refers to the number of times the patient continued to ask follow-up questions about the same issue after entering the pathway.

[0099] S702: Map the patient's terminal information to preset standard intervals to generate corresponding first feedback score, second feedback score and third feedback score.

[0100] The feedback score is an evaluation value obtained by converting the above behavioral records into the same numerical range.

[0101] It should be noted that the click entry point, click time, entry result, page open time, page close time, and follow-up question rounds are read from the behavior record table of the triage terminal server. The data source is the access record generated by the patient's terminal on the triage page. Next, the behavior records are cleaned up: records with empty click entry points are removed, records with page close time earlier than open time are removed, records with repeated clicks on the same entry point within 1 second are retained, and records with empty follow-up question rounds are recorded as 0. Next, the preset standard range is determined by the meaning of the behavior record values. The click status has only two basic results: entering and not entering. Entering the page indicates that the click is valid, and not entering the page indicates that the click is invalid. The dwell time after entering and immediately exiting is less than the minimum time required for normal page loading and reading, so it is recorded as an intermediate state. Next, the range of page dwell time is determined by the time required for the patient to read the machine question-and-answer service path page. Less than 3 seconds usually only completes the page opening and is insufficient to complete the reading of the page content. 3 to 30 seconds is enough to browse the title and main entry points, and more than 30 seconds is enough to view the page content and make operation selections. Next, the range of follow-up questions is determined by the number of follow-up questions themselves. 0 times indicates that the patient did not continue to ask follow-up questions after entering the path; 1 to 2 times indicates that there were still a few supplementary inquiries; and more than 2 times indicates that the path will generate multiple rounds of inquiries later. Then, the weights are determined by the directness of three types of records in the service path feedback: click status indicates whether the patient entered the path; page dwell time indicates the dwell time after entering; and the number of follow-up questions indicates whether the patient continued to ask about the same issue after entering. Therefore, weights are allocated in the order of click status, dwell time, and number of follow-up questions, with the sum of the three weights fixed at 1. Next, the first feedback score, second feedback score, and third feedback score are obtained respectively. The service path feedback score = the sum of the product of the weight value corresponding to the first feedback score and the first feedback score, the product of the weight value corresponding to the second feedback score and the second feedback score, and the product of the weight value corresponding to the third feedback score and the third feedback score. The service path feedback score is then generated.

[0102] For example, if a patient successfully enters the report query page after clicking the entry point, stays for 45 seconds, and asks 0 follow-up questions, then all three feedback items are in a relatively good registration position.

[0103] S703: The service path feedback score is obtained by weighted summation of the first feedback score, the second feedback score, and the third feedback score.

[0104] S704: Construct a target evaluation function based on the service path feedback score.

[0105] The target evaluation function includes establishing a function by using the service path feedback score as the evaluation benchmark, setting the feature value of each path in the candidate machine question-and-answer service path set and the current patient feature vector as the input variables of the function, and setting the display priority score as the output result of the function.

[0106] The objective evaluation function is a calculation relationship that transforms the service path feedback score, candidate path feature value, and current patient feature vector into a display priority score.

[0107] S705: Extract feature values ​​of each path from the set of candidate machine question answering service paths.

[0108] Among them, the characteristic values ​​of the path are the path category, entry level, historical click count, historical dwell time, historical follow-up question count, and the numerical record corresponding to the enabled status. The display priority score is a value used to arrange the display order of candidate paths.

[0109] S706: Input each feature value and the current patient feature vector into the target evaluation function, and calculate the priority of each path relative to the current patient feature vector to obtain the display priority score.

[0110] Specifically, the process reads the service path feedback score, candidate path records from the candidate machine question-and-answer service path set, and the current patient feature vector. Next, the target evaluation function is set based on the three types of input specified in the original steps. First, the service path feedback score is determined as the evaluation benchmark. Then, the feature value of each path and the current patient feature vector are used as input variables, outputting a unique display priority score. The input items in the function are written in a fixed order to avoid misalignment of fields between different paths. Next, the path feature values ​​undergo uniform scaling before input: the entry level is registered according to the number of page levels; the historical click count is registered according to the number of times the path appears in the historical behavior record; the historical dwell time is registered according to the second-level value in the historical record; the historical follow-up question count is registered as an integer; and the enabled status is registered as enabled or disabled. Next, the process of determining the preset weights is consistent with the weight setting of the service path feedback score. First, the service path feedback score is confirmed as the evaluation benchmark. Then, it is confirmed that the path category consistency status, entry level, historical clicks, dwell time, and follow-up questions are all fields related to the path display order, with a fixed total weight of 1. Each field is arranged according to its directness to the current patient's triage target, with the target triage tag consistency status taking priority. Next, the service path feedback score is read, followed by the entry level and historical behavior fields. Then, the above input items are read for each candidate path. The display priority score is the sum of the following four factors: the product of the weight value corresponding to the service path feedback score and the service path feedback score, the product of the weight value corresponding to the path category consistency status and the path category consistency status value, the product of the weight value corresponding to the entry level score and the entry level score, and the product of the weight value corresponding to the historical behavior score and the historical behavior score. This yields the priority of the path relative to the current patient's feature vector, and this priority is recorded as the display priority score. Finally, the display priority score of each path in the candidate machine question-and-answer service path set is generated.

[0111] For example, if the current target triage label is the report query category, and the candidate path also belongs to the report query category, with fewer entry levels, a longer historical dwell time, and fewer follow-up inquiries, then this path will receive a higher display priority score.

[0112] S707: Based on the display priority score, sort the candidate machine question-and-answer service path set in descending order and generate a proton therapy intelligent question-and-answer guidance interactive rendering sequence.

[0113] Among them, the intelligent question-and-answer guidance and interactive rendering sequence for proton therapy is the path arrangement result that the patient's operating terminal finally reads in the order of display.

[0114] It should be noted that the display priority score corresponding to each candidate machine Q&A service path is read and sorted from highest to lowest score. Next, if two paths have different scores, the path with the higher score is placed first, and the path with the lower score is placed later. If two paths have the same score, the path update time is read, and the path with the newer update time is placed first. If the update times are still the same, the entry level is read, and the path with the smaller entry level value is placed first. If the entry levels are still the same, they are sorted according to the original registration order in the machine Q&A service path set. Next, the enabled status is read again after sorting; enabled paths are retained, disabled paths are removed, and paths with an empty status are registered as pending verification and removed. Then, the retained paths are written to a rendering record readable by the patient's operating terminal. The record includes the path name, path address, target triage label, display priority score, and display order. Finally, a proton therapy intelligent Q&A triage interactive rendering sequence is generated.

[0115] For example, if the score for viewing the imaging report is the highest under the report query category, followed by the pathology report viewing path, and then the examination result query path, then the results will be written into the display record in this order.

[0116] In this embodiment of the invention, behavioral data such as patient historical clicks, page dwell time, and number of follow-up inquiries are comprehensively extracted. A service path feedback score is obtained through standardized mapping and weighted calculation. Then, an evaluation function is constructed by combining path feature values ​​and patient feature vectors to calculate the display priority score. Based on the score and multiple fallback rules, the paths are sorted, and finally, a directly displayable triage interaction rendering sequence is output. This method no longer relies solely on tag classification but incorporates users' actual usage habits and the path's inherent attributes, ensuring that the path display order aligns with patient preferences and user experience. Multi-level data cleaning, score calculation, and sorting rules also guarantee fair, reasonable, and logically rigorous results. Simultaneously, disabled paths are removed to ensure that all services displayed on the front end are available, comprehensively improving the practicality, user-friendliness, and overall service efficiency of intelligent triage.

[0117] Reference manual attached Figure 2 The diagram shows a schematic of the structure of a proton therapy intelligent triage system based on patient profiles provided by the present invention.

[0118] The present invention also provides a proton therapy intelligent triage system 20 based on patient profiles, applied to the above-mentioned proton therapy intelligent triage method based on patient profiles, comprising: Processor 201.

[0119] The memory 202 stores computer-readable instructions, which, when executed by the processor 201, implement the intelligent proton therapy triage method based on patient profiles as described in the method embodiment.

[0120] The intelligent proton therapy triage system 20 based on patient profiles provided by this invention can execute the above-mentioned intelligent proton therapy triage method based on patient profiles and achieve the same or similar technical effects. To avoid duplication, this invention will not elaborate further.

[0121] It should be understood that the processor in the embodiments of the present invention can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0122] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0123] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0124] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0125] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0126] It should be understood that, in various embodiments of the present invention, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0127] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0128] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0129] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0130] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0131] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0132] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0133] This invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the intelligent proton therapy triage method based on patient profiles as described in the method embodiments.

[0134] The present invention provides a computer-readable storage medium that can implement the steps and effects of the proton therapy intelligent triage method based on patient profiles in the above-described method embodiments. To avoid repetition, the present invention will not elaborate further.

[0135] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0136] The following points need to be explained: (1) The accompanying drawings of the embodiments of the present invention only involve the structures involved in the embodiments of the present invention. Other structures can refer to the general design.

[0137] (2) For clarity, the thickness of layers or regions is enlarged or reduced in the drawings used to describe embodiments of the invention, i.e., these drawings are not drawn to scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being “above” or “below” another element, the element may be “directly” located “above” or “below” the other element or there may be intermediate elements.

[0138] (3) Where there is no conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.

[0139] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A proton therapy intelligent triage method based on patient profiles, characterized in that, include: S1: Obtain the historical proton therapy interaction text from the patient's operating terminal; S2: Based on the historical proton therapy interaction text, construct a similarity graph of the patient's historical profile nodes; S3: Obtain the current intelligent question-and-answer dialogue text on the current patient's operating terminal; S4: Based on the current intelligent question-and-answer dialogue text, select the unmarked feature nodes of the current patient; S5: Load the unlabeled feature nodes of the current patient into the similarity map of the historical portrait nodes, and determine the target triage label in the similarity map of the historical portrait nodes; S6: Based on the target triage label, filter the set of candidate machine question-and-answer service paths from the set of machine question-and-answer service paths pre-established by the triage terminal server; S7: Sort the candidate machine question-answering service path set and generate a proton therapy intelligent question-answering guidance interactive rendering sequence.

2. The intelligent triage method for proton therapy based on patient profiles according to claim 1, characterized in that, The historical profile node similarity map specifically includes: the patient's historical feature vector and the actual triage destination category.

3. The intelligent triage method for proton therapy based on patient profiles according to claim 1, characterized in that, S2 specifically includes: S201: Extract the patient's historical tumor type information, historical lesion location information, historical tumor stage information, historical treatment method information, historical pathology examination status identifier, and historical imaging report status identifier from the historical proton therapy interactive text, and map them into a preset multi-dimensional space to form the patient's historical feature vector. S202: Define the historical feature vector of each patient as a data node and obtain the pre-configured triage destination category; S203: Set the triage destination category to the preset proton therapy triage label of the data node, and generate a labeled node; S204: Calculate the Euclidean distance between the historical feature vectors of each of the patients; S205: Construct a similarity graph of the historical portrait nodes based on the Euclidean distance.

4. The intelligent triage method for proton therapy based on patient profiles according to claim 3, characterized in that, The specific categories of patient guidance include: patient registration, report inquiry, electronic invoice, and personalized Q&A with medical institutions.

5. The intelligent triage method for proton therapy based on patient profiles according to claim 1, characterized in that, S4 specifically includes: S401: By comparing characters, extract the current patient's tumor type information, lesion location information, tumor stage information, previous treatment method information, pathological examination status identifier, and image report status identifier from the current intelligent question and answer dialogue text, and map them to the same multidimensional space as the patient's historical feature vector for splicing to generate the current patient feature vector; S402: Define the current patient feature vector as an unlabeled node, and use the unlabeled node as the current patient's unlabeled feature node.

6. The intelligent triage method for proton therapy based on patient profiles according to claim 1, characterized in that, S5 specifically includes: S501: Load the current patient's unlabeled feature nodes into the historical portrait node similarity graph, and calculate the current Euclidean distance between the current patient's unlabeled feature nodes and each labeled node in the multidimensional space using the label propagation algorithm; S502: Convert the current Euclidean distance into similarity weights; S503: Based on the similarity weight, the soft labels on each labeled node are transmitted along the edge connection relationship to the current patient's unlabeled feature node, and the guidance probability of each proton therapy guidance label being distributed to the current patient's unlabeled feature node is calculated. S504: Sort the referral probabilities in descending order, and take the proton therapy referral tag bound to the highest referral probability as the target item to determine the target referral tag.

7. The intelligent triage method for proton therapy based on patient profiles according to claim 1, characterized in that, S6 specifically includes: S601: Based on the set of machine question-and-answer service paths, assign a category identifier; S602: Compare the attribution classification identifier with the target triage label character by character to obtain the character comparison result; S603: Based on the character comparison result, extract the path whose belonging category identifier is completely consistent with the target triage label from the machine question-and-answer service path set, and combine all the extracted paths to generate the candidate machine question-and-answer service path set.

8. The intelligent triage method for proton therapy based on patient profiles according to claim 1, characterized in that, Specifically, S7 includes: S701: Obtain patient operation terminal information from the triage terminal server; S702: Map the patient's operating terminal information to preset standard intervals respectively to generate corresponding first feedback score, second feedback score and third feedback score; S703: The first feedback score, the second feedback score, and the third feedback score are weighted and summed to obtain the service path feedback score; S704: Construct a target evaluation function based on the service path feedback score; S705: Extract the feature values ​​of each path from the candidate machine question answering service path set; S706: Input each of the aforementioned feature values ​​and the current patient feature vector into the target evaluation function, and calculate the priority of each of the aforementioned paths relative to the current patient feature vector to obtain a display priority score; S707: Based on the display priority score, sort the candidate machine question-and-answer service path set in descending order and generate the proton therapy intelligent question-and-answer guidance interactive rendering sequence.

9. A proton therapy intelligent triage system based on patient profiles, characterized in that, include: processor; A memory storing computer-readable instructions, which, when executed by the processor, implement the intelligent proton therapy triage method based on patient profiles as described in any one of claims 1 to 8.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the intelligent triage method for proton therapy based on patient profiles as described in any one of claims 1 to 8.

Images