Knowledge retrieval enhancement method and device for endoscopic surgery navigation and surgery navigation system

Abstract

The application relates to the technical field of surgical navigation, in particular to a laparoscopic surgical navigation knowledge retrieval enhancement method and device and a surgical navigation system. The method comprises the following steps: in response to a query request, generating a structured initial query object; processing the initial query object based on preset query expansion rules to generate a query task; performing retrieval in the surgical navigation system based on the query task, generating structured surgical navigation knowledge data and outputting the same. The application can perform three-channel fusion retrieval based on a vector retrieval channel, a keyword retrieval channel and a graph retrieval channel in the surgical navigation system, output structured surgical navigation knowledge data, improve the accuracy and scene adaptability of knowledge retrieval, and the output surgical navigation knowledge data can be directly called by a navigation decision module of the surgical navigation system, thereby providing traceable knowledge support for the surgical navigation system. The application can effectively assist doctors in making surgical decisions and improve the safety and efficiency of surgery.

Description

Technical Field

[0001] This application relates to the field of surgical navigation technology, specifically to a knowledge retrieval enhancement method, device, and surgical navigation system for laparoscopic surgical navigation. Background Technology

[0002] Surgical robots (such as laparoscopic surgical robots) have become an important tool in minimally invasive surgery and are widely used in complex surgical scenarios such as urinary system tumor resection, gynecological disease treatment, and thoracic and abdominal organ surgery.

[0003] Surgical robots can integrate surgical navigation systems. These systems assist surgeons in making surgical decisions. Such systems typically focus on 3D reconstruction of preoperative images, intraoperative instrument tracking and positioning, or augmented reality displays, helping surgeons intuitively understand the patient's individualized anatomy. However, in complex and dynamic surgical navigation scenarios, surgical navigation systems still have significant limitations. For example, they cannot perceive and utilize real-time, multimodal dynamic information during surgery, and lack the ability to understand and query the dynamic intraoperative scene.

[0004] Therefore, surgical navigation systems suffer from the problem of not being able to provide comprehensive and accurate knowledge support. Summary of the Invention

[0005] This application provides a method, device, and surgical navigation system for enhancing knowledge retrieval in laparoscopic surgery. It enables fusion retrieval through three channels—vector retrieval, keyword retrieval, and graph retrieval—to output structured surgical navigation knowledge data. This improves the accuracy and scenario adaptability of knowledge retrieval. Furthermore, the output surgical navigation knowledge data can be directly accessed by the navigation decision module of the surgical navigation system, providing traceable knowledge support. This application can effectively assist surgeons in making surgical decisions, improving the safety and efficiency of surgery.

[0006] In a first aspect, embodiments of this application provide a knowledge retrieval enhancement method for laparoscopic surgical navigation, applied to a surgical navigation system. The method includes: generating a structured initial query object in response to a query request; wherein the query request includes at least one dimension of structured surgical query fields; processing the initial query object based on preset query expansion rules to generate a query task including the initial query object and a target query subtask; wherein the query expansion rules are used to characterize the mapping relationship between preset triggering conditions and query subtasks, and when the initial query object matches the preset triggering conditions, the target query subtask corresponding to the initial query object is obtained; performing a retrieval in the surgical navigation system based on the query task to generate structured surgical navigation knowledge data and output it.

[0007] Secondly, embodiments of this application provide a knowledge retrieval enhancement device for laparoscopic surgical navigation, applied to a surgical navigation system. The device includes: a first generation module, used to generate a structured initial query object in response to a query request; wherein the query request includes at least one dimension of structured surgical query field; a second generation module, used to process the initial query object based on preset query expansion rules to generate a query task including the initial query object and a target query subtask; wherein the query expansion rules are used to characterize the mapping relationship between preset triggering conditions and query subtasks, and when the initial query object matches the preset triggering conditions, the target query subtask corresponding to the initial query object is obtained; and a query module, used to perform a retrieval in the surgical navigation system based on the query task, generate structured surgical navigation knowledge data, and output it.

[0008] Thirdly, embodiments of this application provide a surgical navigation system configured to perform the knowledge retrieval enhancement method for laparoscopic surgical navigation as described in the first aspect and any implementation thereof.

[0009] The technical solution disclosed in this application involves a surgical navigation system that first responds to a query request by generating a structured initial query object. This transforms unstructured surgical query fields into a structured data format, laying the foundation for subsequent precise data queries. Furthermore, the surgical navigation system processes the initial query object using preset query expansion rules. When the initial query object meets preset triggering conditions (such as specific intraoperative risk events), it automatically expands to generate target query subtasks. This process breaks the limitations of relying on explicit user input, achieving a shift from "passive response" to "active prediction," enabling more comprehensive and accurate query tasks based on the potential risks or implicit needs of the current surgical scenario. Furthermore, the surgical navigation system performs retrieval based on a query task that includes an initial query object and query subtasks, generating and outputting structured surgical navigation knowledge data. It not only utilizes a three-channel fusion retrieval system (vector retrieval, keyword retrieval, and graph retrieval) to obtain surgical data directly matching the user's intent, but also simultaneously acquires association rules, risk warnings, or similar cases highly relevant to the current scenario. This effectively solves the problems of existing surgical navigation systems where the retrieved surgical data is limited in content and lacks scenario adaptability, improving the accuracy and scenario adaptability of knowledge retrieval. Moreover, the output surgical navigation knowledge data can be directly called by the navigation decision module of the surgical navigation system, providing traceable knowledge support. This embodiment can effectively assist doctors in making surgical decisions, improving the safety and efficiency of surgery. Attached Figure Description

[0010] Figure 1This is a schematic diagram of the structure of the surgical robot provided in the embodiments of this application; Figure 2 This is a schematic diagram of the surgical navigation system according to an embodiment of this application; Figure 3 This is a flowchart illustrating the knowledge retrieval enhancement method for laparoscopic surgical navigation provided in an embodiment of this application; Figure 4 This is a schematic diagram of the process for generating a query task provided in an embodiment of this application; Figure 5 This is a schematic diagram of the query task generation scenario provided in the embodiments of this application; Figure 6 This is a schematic diagram of the process of generating surgical navigation knowledge data based on query tasks, provided in an embodiment of this application. Figure 7 This is a schematic diagram of the three-channel retrieval process provided in the embodiments of this application; Figure 8 This is a flowchart illustrating the output of navigation prompts provided in an embodiment of this application; Figure 9 This is a schematic diagram of the process for generating surgical navigation knowledge data provided in an embodiment of this application; Figure 10 This is a schematic diagram illustrating the application scenario of the surgical navigation knowledge data provided in the embodiments of this application; Figure 11 This is a schematic diagram of the process for constructing a retrieval channel provided in an embodiment of this application; Figure 12 This is a schematic diagram of the knowledge retrieval enhancement device for laparoscopic surgical navigation provided in this application embodiment. Detailed Implementation

[0011] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of this application.

[0012] It should be noted that many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein. Therefore, the scope of protection of this application is not limited to the specific embodiments disclosed below.

[0013] In the description of this application, it should be understood that the terms "upper," "lower," "horizontal," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this application. In this application, unless otherwise expressly specified and limited, the first feature being "upper" or "lower" than the second feature can mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediate medium.

[0014] Surgical robots (such as laparoscopic surgical robots) have become an important tool in minimally invasive surgery and are widely used in complex surgical scenarios such as urinary system tumor resection, gynecological disease treatment, and thoracic and abdominal organ surgery.

[0015] Surgical robots can integrate surgical navigation systems. These systems assist surgeons in making surgical decisions. Such systems typically focus on 3D reconstruction of preoperative images, intraoperative instrument tracking and positioning, or augmented reality displays, helping surgeons intuitively understand the patient's individualized anatomy. However, in complex and dynamic surgical navigation scenarios, surgical navigation systems still have significant limitations. For example, they cannot perceive and utilize real-time, multimodal dynamic information during surgery, and lack the ability to understand and query the dynamic intraoperative scene.

[0016] Therefore, surgical navigation systems suffer from the problem of not being able to provide comprehensive and accurate knowledge support.

[0017] To address the aforementioned issues, this application provides a knowledge retrieval enhancement method for laparoscopic surgical navigation, which enables rapid querying of surgical data during the procedure, providing comprehensive and accurate knowledge support, thereby effectively assisting doctors in making surgical decisions and improving the safety and efficiency of the surgery.

[0018] Among them, the knowledge retrieval enhancement method for laparoscopic surgical navigation can be applied to surgical robots, which can integrate surgical navigation systems. Specifically, the knowledge retrieval enhancement method for laparoscopic surgical navigation can be configured in the surgical navigation system.

[0019] Figure 1 This is a schematic diagram of the structure of the surgical robot provided in the embodiments of this application.

[0020] like Figure 1 As shown, the surgical robot 100 can specifically be a laparoscopic surgical robot. This application embodiment only uses the laparoscopic surgical robot 100 as an example for illustration. The surgical robot 100 can also be other types of surgical robots, and this application embodiment does not limit this.

[0021] The surgical robot 100 may include a physician control platform 101, a patient surgical platform 102, and an imaging platform 103.

[0022] The patient surgical platform 102 may include a chassis, a column, a robotic arm connected to the column, and a lens arm and an instrument arm disposed at the end of a support assembly of the robotic arm.

[0023] The patient surgical platform 102 also includes an endoscope that is detachably mounted on a lens arm and is used to capture surgical video streams.

[0024] The patient surgical platform 102 also includes at least one surgical instrument detachably mounted on the instrument arm. The surgical instrument is used to perform surgical procedures, such as cutting, suturing, clamping, or electrocoagulation.

[0025] The physician control platform 101 may include a main console, a display, and an operating handle. An operator (e.g., a surgeon) can control the robotic arm on the patient surgical platform 102 via the operating handle to operate surgical instruments mounted at the end of the robotic arm.

[0026] The imaging platform 103 may include a display for displaying the surgical video stream acquired by the endoscope to present the surgical field of view to the operator and assist the operator in performing the surgical procedure.

[0027] The surgical robot 100 may include a controller, which may be set in the doctor control platform 101, the patient surgical platform 102 or the image platform 103, or it may be set independently.

[0028] The controller can be configured to input the surgical video stream acquired by the endoscope into the surgical navigation system, and execute a knowledge retrieval enhancement method for laparoscopic surgical navigation based on the surgical navigation system.

[0029] The following embodiments of this application will specifically describe the knowledge retrieval enhancement method for laparoscopic surgical navigation.

[0030] The display can be used to show navigation prompts output by the surgical navigation system, which guide the user in operating the lens arm and instrument arm.

[0031] It should be noted that the display in this embodiment can be a display on the image platform 103, or a display on the doctor control platform 101 or the patient surgical platform 102. This embodiment does not limit the display to any particular type.

[0032] The surgical robot 100 provided in this application embodiment integrates a surgical navigation system and executes a knowledge retrieval enhancement method for laparoscopic surgical navigation. It can acquire key data during the surgical process in real time, quickly query surgical data, and output corresponding navigation prompts. In this way, the surgical robot 100 provides comprehensive and accurate knowledge support, thereby effectively assisting doctors in making surgical decisions and improving the safety and efficiency of the surgery.

[0033] The surgical navigation system integrated into the surgical robot 100 is described in detail below.

[0034] Figure 2 This is a schematic diagram of the surgical navigation system according to an embodiment of this application.

[0035] like Figure 2 As shown, the surgical navigation system 200 in this embodiment may include a semantic map layer 201, an intraoperative perception layer 202, a registration and positioning layer 203, an intelligent decision-making layer 204, and an enhanced display layer 205.

[0036] Among them, the semantic map layer 201 is used to provide preoperative anatomical structure and spatial risk information.

[0037] The intraoperative sensing layer 202 is used to acquire information such as surgical stage, instruments, anatomical structures, and risk events based on intraoperative video.

[0038] The registration and positioning layer 203 is used to establish the correspondence between preoperative information and intraoperative space.

[0039] The intelligent decision layer 204 is used to generate task-level planning results by combining intraoperative status, surgical task graph, and knowledge retrieval results.

[0040] The enhanced display layer 205 is used to output corresponding navigation prompts.

[0041] Data can be exchanged between different layers using a standardized message format defined by JSON Schema, thus forming a closed-loop navigation process that combines map correction, perception decision-making, prompts and feedback, and knowledge enhancement.

[0042] The knowledge retrieval enhancement method for laparoscopic surgical navigation provided in this application embodiment can be specifically executed by the intelligent decision layer 204.

[0043] It should be noted that the following embodiments of this application all take the navigation scenario of partial nephrectomy as an example to illustrate the knowledge retrieval enhancement method for laparoscopic surgical navigation. It is also applicable to other types of surgery, and the embodiments of this application do not limit it.

[0044] Figure 3 This is a flowchart illustrating the knowledge retrieval enhancement method for laparoscopic surgical navigation provided in this application embodiment.

[0045] like Figure 3 As shown, in some embodiments, a knowledge retrieval enhancement method for laparoscopic surgical navigation can be applied to a surgical navigation system, and the method includes steps S100-S300.

[0046] Step S100: In response to the query request, a structured initial query object is generated, wherein the query request includes at least one dimension of surgical query fields.

[0047] Among them, a query request refers to a retrieval instruction issued by various functional layers of the surgical navigation system or by the user during the operation, which carries multiple dimensions of surgical information.

[0048] In one implementation, the surgical query fields include at least one of the following: query_id, current_phase, current_task, visible_structures, current_events, patient_features, and query_purpose.

[0049] For example, the content of the "Current Stage" field might include, for example, "Tumor Resection Stage." The content of the "Current Task" field might include, for example, "Tumor Boundary Confirmation." The content of the "Visible Structure" field might include, for example, "Tumor, Renal Parenchyma, Collecting System." The content of the "Current Event" field might include, for example, "Minor Bleeding." The content of the "Patient Characteristics" field might include, for example, "Tumor Located in the Lower Pole of the Left Kidney, 8.5cm in Size." 3 It is 2.1mm away from the collection system. The content of the target field for the query includes, for example, obtaining operation rules, risk explanations, and similar cases.

[0050] In this embodiment of the application, the content of each field in the surgical query field is not specifically limited.

[0051] In one implementation, the surgical navigation system generates a query request based on the query target field and the scene status field.

[0052] The query target field includes at least one of the query requirements input by the user and the retrieval requirements generated by the intelligent decision-making layer. For example, the query target field includes a query purpose field, which includes a first query purpose field input by the user via voice or text, such as "query current surgical risks and operational recommendations". The query purpose field also includes a second query purpose field automatically generated by the intelligent agent in the intelligent decision-making layer based on the current surgical stage and task, such as "automatically retrieve operational rules and similar cases related to the current anatomical structure".

[0053] The scene state field includes at least one of the stage field, structure field, and event field output by the intraoperative perception layer, as well as at least one of the personalized anatomical feature field output by the registration and positioning layer and the task field output by the intelligent decision-making layer.

[0054] For example, the scene state field includes at least one of the following: surgical stage field, visible structure field, surgical event field generated by the intraoperative perception layer, surgical object feature field generated by the registration and localization layer, and surgical task field generated by the intelligent decision-making layer.

[0055] The surgical object feature field can be generated based on the preoperative navigation semantic map in the registration and localization layer. The surgical task field can be generated based on the intelligent agent in the intelligent decision-making layer.

[0056] In this way, the structured initial query object can uniformly encode the multi-dimensional surgical query fields, which facilitates efficient querying of subsequent surgical data.

[0057] It should also be noted that in the structured initial query object, the content of fields other than the primary query objective field can be used as supplementary fields to the primary query objective field. This initially achieves contextual supplementation for the query request, ensuring that the query object not only contains a clear query objective but also carries multidimensional information about the current surgical status, providing a rich contextual foundation for subsequent query expansion and retrieval enhancement.

[0058] The following example uses a specific application scenario to illustrate the process of generating a query request.

[0059] For example, at time point A of a partial nephrectomy, currently in the tumor resection stage, with the tumor adjacent to the collecting system and accompanied by mild bleeding, the query request would be as follows: .

[0060] Step S200: The initial query object is processed based on the preset query expansion rules to generate a query task including the initial query object and the target query subtask; wherein, the query expansion rules are used to characterize the mapping relationship between the preset triggering conditions and the query subtasks, and the target query subtask corresponding to the initial query object is obtained when the initial query object matches the preset triggering conditions.

[0061] In one implementation, the query extension rule includes a rule identifier, preset trigger conditions, extension actions, and priority.

[0062] The rule identifier is a number used to uniquely identify each query extension rule, which facilitates system management and tracking.

[0063] The preset trigger conditions are used to match the field content of the surgical query field in the initial query object. In this way, if the field content of the surgical query field in the initial query object meets the preset trigger conditions, the corresponding extended action is triggered.

[0064] Extended actions include generating one or more target query subtasks.

[0065] Priority is used to determine the execution order when multiple query expansion rules are triggered simultaneously, ensuring that critical expansion actions are processed first.

[0066] For example, query expansion rules may include: The first query extension rule includes the first rule identifier R1, the first preset trigger condition, the first extension action, and the first priority.

[0067] The first preset trigger condition is, for example, "bleeding exists in the surgical event and its severity is greater than 0.3", the first extended action is, for example, generating the first target query subtask "bleeding handling rules in the surgical task stage", and the first priority is, for example, 1.

[0068] The second query extension rule includes the second rule identifier R2, the second preset trigger condition, the second extension action, and the second priority.

[0069] Among them, the second preset triggering condition is, for example, "the distance from the tumor to the ensemble system in the semantic map is less than a safety threshold (e.g., 3 mm)", the second extended action is, for example, generating the second target query subtask "precautions for resection and suturing when the tumor is adjacent to the ensemble system", and the second priority is, for example, 1.

[0070] The third query extension rule includes the third rule identifier R3, the third preset trigger condition, the third extension action, and the third priority.

[0071] Among them, the third preset triggering condition is, for example, "the tumor location in the semantic map is endophytic", the third extended action is, for example, generating the third target query subtask "endophytic renal tumor resection techniques and resection margin judgment", and the third priority is, for example, 2.

[0072] The fourth query extension rule includes the fourth rule identifier R4, the fourth preset trigger condition, the fourth extension action, and the fourth priority.

[0073] Among them, the fourth preset trigger condition is, for example, "surgical stage field = "vascular control" AND semantic map has accessory renal artery = True", the fourth extended action is, for example, generating the fourth target query subtask "accessory renal artery treatment rules and precautions", and the fourth priority is, for example, 2.

[0074] The fifth query extension rule includes the fifth rule identifier R5, the fifth preset trigger condition, the fifth extension action, and the fifth priority.

[0075] Among them, the fifth preset trigger condition is, for example, "the overall score of perceived quality is less than 0.6", the fifth extended action is, for example, generating the fifth target query subtask "standard operating procedures for the surgical stage", and the fifth priority is, for example, 3.

[0076] The sixth query extension rule includes the sixth rule identifier R6, the sixth preset trigger condition, the sixth extension action, and the sixth priority.

[0077] Among them, the sixth preset trigger condition is, for example, "default (no special conditions trigger)", the sixth extended action is, for example, generating the sixth target query subtask "surgical task related rules and next operation of surgical stage", and the sixth priority is, for example, 3.

[0078] The specific generation method of the query task will be described in detail in the subsequent embodiments of this application, and will not be repeated here.

[0079] In this way, after the query request is structured into an initial query object, the surgical navigation system does not directly perform retrieval based on the initial query object. Instead, it introduces query expansion rules, using the surgical query fields contained in the initial query object as input, and matching them one by one with the preset trigger conditions in the query expansion rules. This identifies the expanded retrieval subtasks (i.e., target query subtasks) that are highly relevant to the current surgical state, and the initial query object and the target query subtask together constitute the query task. Therefore, the surgical navigation system performs surgical data retrieval based on this query task, which not only covers the user's original intent but also automatically covers the derivative information needs that are highly related to the current surgical scenario. This makes the returned surgical data more comprehensive and closer to the actual intraoperative needs, effectively making up for the problem of insufficient information from a single query.

[0080] Step S300: Based on the query task, perform a retrieval in the surgical navigation system to generate structured surgical navigation knowledge data and output it.

[0081] In this way, the surgical navigation system performs data retrieval based on the expanded query task, realizing accurate and efficient querying of surgical data during the operation. This provides reliable knowledge support for the surgical navigation system to assist the surgeon in making intraoperative decisions, significantly improving surgical safety and efficiency.

[0082] Figure 4 This is a flowchart illustrating the process of generating a query task provided in an embodiment of this application.

[0083] like Figure 4As shown, in some embodiments, step S200 involves processing the initial query object based on preset query expansion rules to generate a query task that includes the initial query object and the target query subtask, including steps S210-S240.

[0084] Step S210: Validate the structured surgical query fields in the initial query object based on the preset trigger conditions in the query expansion rules, and obtain the validation results.

[0085] Further based on the aforementioned example, when the surgical event field in the initial query object is "there is bleeding in the surgical event and the severity is 0.4", the surgical navigation system verifies the surgical event field based on the first preset trigger condition in the first query expansion rule. The verification results include: the "severity is 0.4" in the surgical event field satisfies the "severity is greater than 0.3" in the first preset trigger condition, and the first preset trigger condition matches the surgical event field.

[0086] Step S220: If the verification result shows that the preset triggering condition matches the surgical query field, candidate subtasks are generated, where each candidate subtask has a corresponding priority.

[0087] Further based on the aforementioned example, the surgical navigation system may, for example, generate a first target query subtask to a sixth target query subtask based on the first query expansion rule to the sixth query expansion rule, as candidate subtasks.

[0088] Step S230: Select the top N candidate subtasks with the highest priority as the target query subtasks, where 0 < N ≤ 6.

[0089] For example, N=5, meaning the maximum number of tasks for the target query subtask is 5. It should be noted that this embodiment does not limit the specific value of N.

[0090] Further based on the aforementioned example, the surgical navigation system can, for instance, determine the target query subtask based on the top 5 candidate subtasks with the highest priority. It should be noted that, when priorities are the same, the surgical navigation system can sort candidate subtasks based on rule identifiers.

[0091] Step S240: Generate a query task based on the initial query object and the target query subtask.

[0092] Further based on the aforementioned example, when the initial query object triggers the first preset trigger condition and the second preset trigger condition, the surgical navigation system generates a query task based on the initial query object, the first target query subtask, and the second target query subtask.

[0093] In this way, the surgical navigation system introduces a priority-based candidate subtask selection mechanism to ensure that the query task focuses on the most important knowledge needs in the current surgical scenario, which facilitates improved retrieval efficiency and high relevance of retrieval results, providing targeted surgical data support for subsequent surgical decisions.

[0094] The implementation of steps S210-S240 above will be illustrated below based on specific application scenarios.

[0095] Figure 5 This is a schematic diagram of the query task generation scenario provided in the embodiments of this application.

[0096] like Figure 5 As shown, in one implementation, the initial query object obtained by the surgical navigation system includes surgical query fields such as a query purpose field, a patient characteristic field, and a current stage field. The query purpose field is used, for example, to query the current resection strategy; the patient characteristic field is used, for example, to indicate the characteristics of the tumor proximity set system; and the current stage field indicates the intraoperative status stage, specifically, for example, the tumor resection stage accompanied by mild bleeding.

[0097] After integrating the different surgical query fields in the initial query object, the surgical navigation system can load query extension rules.

[0098] After loading the query expansion rules, the surgical navigation system triggers the query expansion rules based on execution steps S210-S230 to generate the target query subtask.

[0099] For example, the target query subtask includes a first subtask, a second subtask, a third subtask, and a fourth subtask.

[0100] The process is divided into four subtasks: the first subtask queries for precautions regarding the resection and suturing of tumor-adjacent aggregates; the second subtask queries for rules governing bleeding during tumor resection; the third subtask queries for video clips of cases with similar tumor locations; and the fourth subtask queries for the next steps in the current task.

[0101] Further, the surgical navigation system executes step S240 to generate a query task based on the initial query object (e.g., including the query target field, patient characteristic field, and current stage field) and the target query subtask (e.g., including the first subtask, the second subtask, the third subtask, and the fourth subtask).

[0102] Figure 6 This is a schematic diagram of the process of generating surgical navigation knowledge data based on query task execution in an embodiment of this application.

[0103] like Figure 6As shown, in some embodiments, step S300 involves performing a retrieval in the surgical navigation system based on the query task, generating structured surgical navigation knowledge data, and outputting it, including steps S310-S330.

[0104] Step S310: Input the query task into the vector retrieval channel, keyword retrieval channel and map retrieval channel of the surgical navigation system respectively.

[0105] Step S320: Perform the search on the query task in the vector search channel, keyword search channel and graph search channel respectively, and obtain the corresponding vector search results, keyword search results and graph search results.

[0106] Step S330: Based on the vector retrieval results, keyword retrieval results, and graph retrieval results, generate structured surgical navigation knowledge data.

[0107] In this way, the surgical navigation system can obtain surgical data that matches the user's direct intent by integrating three channels: vector retrieval, keyword retrieval, and graph retrieval. It can also simultaneously obtain association rules, risk warnings, or similar cases that are highly relevant to the current scenario. This effectively solves the problem that the surgical data retrieved by the existing surgical navigation system is limited in content and lacks scenario adaptability, thus improving the accuracy and scenario adaptability of knowledge retrieval.

[0108] The following embodiments of this application will provide a detailed description of the above steps, which will not be repeated here.

[0109] Figure 7 This is a schematic diagram of the three-channel retrieval process provided in the embodiments of this application.

[0110] like Figure 7 As shown, in one implementation, step S320 involves performing a search on the query task in the vector search channel, keyword search channel, and graph search channel respectively, to obtain the corresponding vector search results, keyword search results, and graph search results, including steps S321-S323.

[0111] Step S321: In the vector retrieval channel, the query task is transformed into a query vector, and the query vector is matched with the preset vector index in the vector database to obtain the first preset surgical data that is matched as the vector retrieval result.

[0112] The vector database is used to store the mapping relationship between preset vector indexes and first preset surgical data.

[0113] In one implementation, the surgical navigation system vectorizes the query task into a query vector based on an embedding model. Further, based on the query vector, an approximate nearest neighbor (ANN) search is performed in a vector database (e.g., the Milvus vector database) to obtain the vector retrieval result.

[0114] For example, the surgical navigation system can obtain K_v semantically closest knowledge units and case descriptions as vector retrieval results. K_v is, for example, 3. This application does not limit the specific value of K_v in its embodiments.

[0115] Further, based on specific application scenarios in the aforementioned embodiments, the data content of the vector retrieval results will be explained.

[0116] For example, the vector search results include three search results, namely V1, V2 and V3.

[0117] V1 includes: "Operational techniques for protecting the collection system during tumor resection" (Similarity=0.92, Source=Expert experience).

[0118] V2 includes: "Management of wound bleeding during partial nephrectomy" (Similarity=0.87, Source=Guideline).

[0119] V3 includes: "Case of resection of tumor in the lower pole of the left kidney" (similarity=0.83, source=case report).

[0120] In this way, the surgical navigation system can quickly locate the search results that best match the semantics of the current surgical scenario by converting natural language queries into high-dimensional semantic vectors and performing efficient retrieval in the vector database, thus significantly improving retrieval efficiency and accuracy.

[0121] Step S322: In the keyword retrieval channel, extract keywords from the query task, and use the keyword search engine to match the keywords with the preset keyword index to obtain the matched second preset surgical data as the keyword retrieval result.

[0122] The keyword search engine is used to store the mapping relationship between the preset keyword index and the second preset surgical data.

[0123] In one implementation, the surgical navigation system extracts keywords (e.g., including "tumor", "collection system", "bleeding", "treatment rules") based on the query task, and performs a multi-field Boolean search on the keyword search engine based on the keywords to obtain keyword retrieval results.

[0124] For example, the keyword search results obtained by the surgical navigation system include: rule_name field, structure_name field, event_type field, and keyword_tags field.

[0125] Further, based on specific application scenarios in the aforementioned embodiments, the data content of keyword search results will be explained.

[0126] For example, the keyword search results include three results, namely K1, K2 and K3.

[0127] K1 includes: "RULE-089: The collecting system should be sutured when the tumor-collecting system distance is <3mm" (BM25=18.5, source=guideline).

[0128] K2 includes: “RULE-045: Management of bleeding during resection – electrocoagulation or suture hemostasis” (BM25=16.2, source=guideline).

[0129] K3 includes: "Collision system injury → postoperative urinary leakage risk" (BM25=14.8, source = expert consensus).

[0130] In this way, the surgical navigation system achieves precise matching through keyword retrieval channels, enabling it to quickly locate surgical data containing specific terms. This compensates for the shortcomings of vector retrieval in precise term matching and improves the comprehensiveness and reliability of retrieval results.

[0131] Step S323: In the graph retrieval channel, extract the target field from the query task, match the target field with the entities in the graph database, and obtain the matched third preset surgical data as the graph retrieval result based on the surgical processing rules corresponding to the relationship between the entities.

[0132] The target fields include at least one of the following: surgical task field, surgical event field, and anatomical structure field. The graph database is used to represent the relationships between different entities, and the relationships between different entities are used to represent the surgical processing rules. The surgical processing rules are used as the third preset surgical data.

[0133] In one implementation, the surgical navigation system starts with a node in the surgical task field or a node in the surgical event field, performs path matching in the graph database, obtains the surgical processing rules associated with that node, and uses the surgical processing rules as the graph retrieval results.

[0134] For example, the surgical navigation system starts with the node of the surgical task field, obtains the focus structure field along the task_focuses_on entity relationship, obtains the adjacent risk structure field along the anatomy_adjacent_to entity relationship, and obtains the surgical processing rule along the event_triggers_rule entity relationship.

[0135] For example, the surgical navigation system starts from the node of the surgical event field and retrieves the surgical processing rules along the edges of the event_triggers_rule entity relationship.

[0136] Further, based on specific application scenarios in the aforementioned embodiments, the data content of the map retrieval results will be explained.

[0137] For example, the map search results include a search result L.

[0138] L includes: from task node "T501 Tumor Boundary Confirmation" along task_focuses_on - "Tumor" - anatomy_adjacent_to - "Collection System" - event_triggers_rule - "RULE-089 Suture Collection System", path length: 3 hops, relationship chain: tumor boundary confirmation - focus on tumor - adjacent collection system - trigger suture rule.

[0139] In this way, the surgical navigation system can achieve semantic association reasoning through the graph retrieval channel, and can discover surgical processing rules that are indirectly related to the current surgical task or event. This makes up for the shortcomings of vector retrieval and keyword retrieval in understanding semantic relationships, and improves the depth and interpretability of retrieval results.

[0140] Figure 8 This is a flowchart illustrating the output navigation prompt information provided in an embodiment of this application.

[0141] like Figure 8 As shown, in some embodiments, step S300 involves performing a retrieval in the surgical navigation system based on the query task, generating structured surgical navigation knowledge data, and outputting it, including steps S340-360.

[0142] Step S340: Perform a search in the surgical navigation system based on the query task to obtain the search results.

[0143] Step S350: Based on the search results, generate surgical navigation knowledge data, wherein the surgical navigation knowledge data is used to represent structured surgical data fields, and the surgical data fields are used to describe the search results.

[0144] In one implementation, the surgical navigation system processes the retrieval results and encapsulates the processed results into a K1 structured output to generate surgical navigation knowledge data. The K1 structured output is, for example, in JSON format.

[0145] In one implementation, the surgical data fields include at least one of the following: query_id, timestamp, related_rules, related_structures, risk_explanations, recommended_actions, similar_cases, prompt_templates, confidence, and sources.

[0146] The query identifier (query_id) field is a string field and is used to describe a unique identifier for the query.

[0147] The timestamp field is of type float and is used to indicate the time stamp when the retrieval was completed.

[0148] The surgical operation rule field is a list (array) that describes the list of operation rules related to the current task. Each surgical operation rule field includes at least one of the following: rule identifier (rule_id), rule type (rule_type), rule content (rule_content), priority (priority), and source (source).

[0149] The surgical procedure structure field is a list field that describes a list of anatomical structures relevant to the current task. Each surgical procedure structure field includes at least one of the following: structure identifier (structure_id), name (name), surgical relevance (surgical_relevance), risk level (risk_level), and relationship to current task (relationship_to_current_task).

[0150] The surgical risk source and explanation fields are list types, used to describe the source and explanation of the current intraoperative risk. Each surgical risk source and explanation field includes at least one of the following: risk type (risk_type), trigger condition (trigger_condition), severity (severity), explanation (explanation), and recommended action (recommended_action).

[0151] The "Recommended Procedures and Precautions" field is a list type, used to describe a list of recommended procedures or precautions. This field is sorted by priority.

[0152] The surgical similarity case field is a list field used to describe similar cases or video clips. Each surgical similarity case field includes at least one of the following: case identifier (case_id), similarity score (similarity_score), key findings (key_findings), and video clip index (key_findings).

[0153] The prompt word template field is a list type, used to describe a structured template that the intelligent decision-making layer can directly call to generate navigation prompt information. The prompt word template field includes at least one of the following: prompt word identifier (template_id), prompt type (prompt_type), template content (template_content), and applicable conditions (applicable_conditions).

[0154] The field type for the surgical data confidence field is a list, used to describe the overall confidence level of the search results (e.g., 0-1).

[0155] The surgical data source field is a list field, used to describe the list of knowledge sources. The surgical data source field includes at least one of the following: data source type (source_type), data source name (source_name), retrieval channel (retrieval_channel), and credibility (credibility).

[0156] In this way, the surgical navigation knowledge data generated by the surgical navigation system carries the necessary information required by the intelligent decision-making layer, enabling the rapid provision of surgical navigation knowledge data during the operation, thereby effectively assisting doctors in making surgical decisions and improving the safety and efficiency of the operation.

[0157] Step S360: Based on the intelligent decision-making layer in the surgical navigation system, call up surgical navigation knowledge data and output navigation prompt information.

[0158] In one implementation, the intelligent decision-making layer in the surgical navigation system can access surgical navigation knowledge data via a standardized message interface. The worker module within the intelligent decision-making layer can directly parse the `related_rules`, `risk_explanations`, and `prompt_templates` fields from the K1 structured output surgical navigation knowledge data to generate navigation prompts. This process eliminates the need for additional natural language parsing, significantly reducing system latency and improving decision response speed.

[0159] The process of generating navigation prompts will be further explained using specific application scenarios from the foregoing embodiments as examples.

[0160] For example, after receiving the structured output of K1, the Worker module extracts related_rules[0] and prompt_templates[0], fills in the parameters, and generates navigation prompt information.

[0161] Navigational prompts include, for example, “Note: The tumor is only 2.1 mm from the collecting system. Be highly vigilant for damage to the collecting system during resection. Recommendation: Use a combination of blunt and sharp dissection. After resection, carefully check the integrity of the collecting system. If necessary, close with 4-0 absorbable sutures. (Based on: EAU Guidelines 2024, confidence level: 91%)”.

[0162] The embodiments of this application do not limit the specific content of the navigation prompt information.

[0163] In this way, the surgical navigation system generates standardized, clearly sourced, and confident navigation prompts without requiring additional natural language parsing. This makes the content of the navigation prompts traceable, interpretable, and auditable, significantly improving the reliability and usability of the surgical navigation system and ensuring that users can quickly obtain high-quality decision support during surgery.

[0164] Figure 9 This is a schematic diagram of the process for generating surgical navigation knowledge data provided in the embodiments of this application.

[0165] like Figure 9 As shown, in some embodiments, step S350, based on the search results, generates surgical navigation knowledge data, including steps S351-S356.

[0166] Step S351: Determine the target score for the retrieval results of different channels. The target score has a corresponding knowledge source credibility weight.

[0167] In one implementation, in each channel, the surgical navigation system determines the target score of the retrieval results for the current channel based on the product of the relevance score of the retrieval results, the knowledge source credibility weight of the retrieval results, and the channel weight of the retrieval results. The relevance score is used to characterize the relevance between the query task and the retrieval results.

[0168] In one implementation, the channel weights include the first channel weight corresponding to the vector retrieval result, the second channel weight corresponding to the keyword retrieval result, and the third channel weight corresponding to the graph retrieval result, wherein the third channel weight is greater than the second channel weight, and the second channel weight is greater than the first channel weight.

[0169] Based on the aforementioned example, the search results obtained by the surgical navigation system may include vector search results, keyword search results, and graph search results.

[0170] In one implementation, the surgical navigation system can calculate the target score of the retrieval results for each channel based on Formula 1.

[0171] Formula 1: S_base(r)=relevance_score(r)×source_credibility(r)×channel_weight.

[0172] Where r is the retrieval result returned by each channel, S_base(r) is the target score for each channel, relevance_score(r) is the relevance score of the retrieval result (used to evaluate the relevance between the retrieval result and the query task), source_credibility(r) is the credibility weight of the knowledge source, and channel_weight is the channel weight.

[0173] Specifically, when r is the vector retrieval result of the vector retrieval channel, relevance_score(r) is the vector similarity score between the vector retrieval result and the query task, source_credibility(r) is the first preset value (for example, if the knowledge source is a guideline reviewed by experts, the first preset value is 1.0; if the knowledge source is expert experience, the first preset value is 0.9; if the knowledge source is a case report, the first preset value is 0.7; if the knowledge source is an instructional video, the first preset value is 0.6), and channel_weight is the second preset value (for example, 0.8).

[0174] When r is the keyword search result of the keyword search channel, relevance_score(r) is the BM25 score normalization value between the keyword search result and the query task, source_credibility(r) is the third preset value (see the setting method of the first preset value), and channel_weight is the fourth preset value (for example, 0.9).

[0175] When r represents the graph retrieval result of the graph retrieval channel, relevance_score(r) is the inverse normalized value of the relationship path length between the graph retrieval result and the query task, source_credibility(r) is the fifth preset value (see the setting method of the first preset value), and channel_weight is the sixth preset value (e.g., 1.0). Among these, because the graph retrieval channel has accurate relationship paths and traceability, the weight of the graph retrieval channel is set to the highest.

[0176] In this way, by determining the target scores of the search results from different channels, the surgical navigation system can facilitate subsequent weighted fusion based on the scores of each channel, thereby selecting the surgical data with the highest overall quality to recommend to the user.

[0177] Step S352: If the channel includes a map retrieval channel, multiply the target score of the map retrieval result of the map retrieval channel by a preset enhancement coefficient to obtain the enhanced target score.

[0178] In one implementation, since the graph retrieval channel has the highest accuracy and traceability, if the relational path associated with the graph retrieval result contains a high-risk structure (such as the tail entity of the anatomy_adjacent_to entity relation being "collective system" or "major blood vessel"), the surgical navigation system can enhance the target score of the graph retrieval result of that channel by multiplying it by a preset enhancement coefficient β (e.g., β=1.2).

[0179] It should be noted that the preset enhancement coefficient β can be dynamically adjusted according to the type of surgery or user preferences. For example, β can be set to 1.3 in kidney surgery and 1.1 in liver surgery to highlight the importance of high-risk structures in surgical decisions. This application does not limit the specific value of the preset enhancement coefficient in its embodiments.

[0180] In this way, the surgical navigation system can prioritize and recommend precise surgical data closely related to high-risk anatomical structures, thereby improving surgical safety and decision reliability.

[0181] Step S353: In the case of identical knowledge units in the search results, retain the search result with the highest target score.

[0182] Further based on the aforementioned example, the surgical navigation system retains the search result with the highest target score (S_base) for multi-channel results pointing to the same knowledge unit (e.g., the same rule identifier (rule_id) or the same case identifier (case_id)), and merges its source and channel labels.

[0183] In this way, the surgical navigation system can effectively eliminate redundant data and avoid repeatedly recommending different versions of the same knowledge unit, thereby improving the simplicity and usability of the final results. At the same time, retaining the highest-scoring version ensures that users obtain the most reliable surgical data, further optimizing the accuracy of surgical decisions.

[0184] Step S354: Sort the final target scores of the search results of different channels in descending order. The final target scores include the enhanced target score corresponding to the graph search channel, the target score corresponding to the vector search channel, and the target score corresponding to the keyword search channel.

[0185] In one implementation, the surgical navigation system sorts the final target scores of all channels in descending order and prioritizes recommending the top Q surgical data points with the highest scores. Q is greater than or equal to 1.

[0186] For example, Q equals 10.

[0187] The specific value of Q is not limited in the embodiments of this application.

[0188] In this way, by sorting and truncating the search results, the surgical navigation system can present users with the most relevant and reliable surgical data, reducing the time cost for users to filter information. Step S355: Based on the top K final target scores, determine the candidate retrieval results, where 0 < K < 10.

[0189] In one implementation, to ensure the accuracy of the final output, the surgical navigation system needs to further filter the search results corresponding to the top K final target scores. Therefore, the surgical navigation system only uses the search results corresponding to the top K final target scores as candidate search results. Step S356: Based on the candidate search results, generate and output surgical navigation knowledge data.

[0190] In one implementation, step S356 includes steps S3561-S3562.

[0191] Step S3561: If there is conflicting data in the candidate search results, filter the conflicting data based on the knowledge source credibility weight corresponding to the final target score, and retain the candidate search results corresponding to the highest knowledge source credibility weight. Conflicting data is used to represent candidate search results from different sources that have the same knowledge unit and the same knowledge unit includes conflicting attribute information.

[0192] For example, when the query task includes "tumor safe resection margin distance", conflicting data may include, for example: Candidate search results from Source A (EAU Guidelines 2024): It is recommended that the resection margin be ≥1mm, emphasizing that "as long as the tumor is completely removed, the thickness of the resection margin does not affect the oncological prognosis".

[0193] Among them, the candidate search results from source A were obtained based on the keyword search channel, and their confidence level was 1.0.

[0194] Candidate search results from source B (a certain expert consensus): It is recommended that the cutting edge be ≥3mm, as it is considered safer.

[0195] Among them, the candidate search results for source B were obtained based on the vector search channel, and their confidence level was 0.9.

[0196] Candidate search results from source C (a case study): reported a local recurrence rate of 2.1% for cut margins <1 mm.

[0197] Among them, the candidate retrieval results for source C were obtained based on the vector retrieval channel, and their confidence level was 0.7.

[0198] In one implementation, if the surgical navigation system determines that there is conflicting data among the candidate search results (i.e., data from different knowledge sources that have contradictory conclusions about the same fact (e.g., different versions of guidelines have different recommendations on safe cutting margin distances)), then it retains the candidate search result with the highest credibility weight of the knowledge source among the conflicting data.

[0199] As can be seen from the previous example, the candidate search results from source A and source B conflict on the recommended value of "safe cutting edge distance", that is, the conflict between 1mm and 3mm.

[0200] Since the candidate search results from source A have the highest weight, the candidate search results from source A are retained.

[0201] It should be noted that when the credibility weight of knowledge sources is the same in conflicting data, all candidate search results are retained and marked as pending confirmation.

[0202] In this way, the surgical navigation system can ensure data accuracy while retaining the flexibility of clinical decision-making.

[0203] Step S3562: Based on the filtered candidate search results, generate and output surgical navigation knowledge data.

[0204] In one implementation, the surgical navigation system integrates the filtered candidate search results into structured surgical navigation knowledge data and outputs it.

[0205] In cases where candidate search results labeled as surgical navigation knowledge data are included in the surgical navigation knowledge data, the conflicting information must be noted in the `risk_explanations` field of the candidate search results. By clearly labeling conflicting information, the surgical navigation system helps clinicians fully understand the differences between various knowledge sources when making decisions.

[0206] For example, the conflicting information noted in the risk_explanations field may include: .

[0207] The calculation process of steps S351-S357 above will be further explained using a specific application scenario from the foregoing embodiments as an example.

[0208] Figure 10 This is a schematic diagram illustrating an application scenario for generating surgical navigation knowledge data provided in an embodiment of this application.

[0209] like Figure 10 As shown, in some embodiments, after generating a query task based on execution step S200, the surgical navigation system, based on execution steps S310-S320, inputs the query task into the vector retrieval channel, keyword retrieval channel, and graph retrieval channel of the surgical navigation system respectively to perform a three-channel parallel retrieval. Thus, the surgical navigation system can obtain the retrieval results of the three-channel parallel retrieval, namely the vector retrieval results, keyword retrieval results, and graph retrieval results involved in step S330. Since the vector retrieval results include vector retrieval results, the keyword retrieval results include keyword retrieval results, and the graph retrieval results include graph retrieval results, step 350 can be considered, for example, a sub-step of step S330 when the surgical data includes vector retrieval results, keyword retrieval results, and graph retrieval results.

[0210] In other words, the surgical navigation system can execute steps S351-S357 on the vector retrieval results output by the vector retrieval channel, the keyword retrieval results output by the keyword retrieval channel, and the graph retrieval results output by the graph retrieval channel.

[0211] For example, it can be known based on steps S351 and S352.

[0212] The target score S_base(r) of the map retrieval results corresponding to the map retrieval channel is... L =1.0×1.0×1.0=1.0.

[0213] Where 1 represents the relevance score of the map retrieval results, 1 represents the guide credibility weight, and 1 represents the map retrieval channel weight.

[0214] Since the map retrieval results are associated with high-risk structural "ensemble systems", then S_base(r L Multiply by the enhancement factor β=1.2 to get the enhanced target score of 1.2 (i.e. 1.0×1.2).

[0215] The target score S_base(r) of the vector retrieval results corresponding to the vector retrieval channel V1 =0.92×0.9×0.8=0.662.

[0216] Wherein, 0.92 is the relevance score of vector retrieval result V1, 0.9 is the credibility weight of expert experience, and 0.8 is the weight of vector retrieval channel.

[0217] The target score S_base(r) of the vector retrieval results corresponding to the vector retrieval channel V2 =0.87×1.0×0.8=0.696.

[0218] Here, 0.87 is the relevance score of vector retrieval result V2, 1.0 is the guide credibility weight, and 0.8 is the vector retrieval channel weight.

[0219] The target score S_base(r) of the keyword search results K1 corresponding to the keyword search channel K1 =0.88×1.0×0.9=0.792.

[0220] Here, 0.88 is the relevance score of keyword search result K1, 1.0 is the guide credibility weight, and 0.9 is the keyword search channel weight.

[0221] In other words, the surgical navigation system first performs multi-dimensional weighted scoring on the vector retrieval results, keyword retrieval results, and atlas retrieval results to obtain the target score. It then further performs high-risk result enhancement processing on the atlas retrieval results to obtain the enhanced target score corresponding to the atlas retrieval results.

[0222] Furthermore, based on steps S353-S355, it can be seen that...

[0223] The surgical navigation system first executes step S353, in the case of identical knowledge units in the search results, and then performs duplicate result deduplication and merging on the search results, retaining the search result with the highest target score.

[0224] Furthermore, by executing steps S354-S355, the surgical navigation system sorts the final target scores of the search results from different channels in descending order and obtains the top K final target scores, thereby achieving comprehensive score sorting and truncation processing of the final target scores and determining the candidate search results.

[0225] The surgical navigation system executes steps S353-S355, and the candidate retrieval results corresponding to the first 5 final target scores obtained include: The first candidate search result corresponding to the final target score of 1.2 includes: RULE-089 (set system stitching rule).

[0226] The second candidate search result corresponding to the final target score of 0.792 includes: RULE-045 (bleeding management rule).

[0227] The third candidate search result corresponding to the final target score of 0.696 includes: V2 (Guidelines for the Management of Wound Bleeding).

[0228] The fourth candidate search result corresponding to the final target score of 0.662 includes: V1 (protection set system trick).

[0229] The fifth candidate search result corresponding to the final target score of 0.598 includes: V3 (similar case).

[0230] Furthermore, based on step S356, it is known that...

[0231] By executing step S356, the surgical navigation system filters conflicting data based on the knowledge source credibility weight corresponding to the final target score when there is conflicting data in the candidate search results, and retains the candidate search results corresponding to the highest knowledge source credibility weight, thus realizing knowledge conflict verification processing.

[0232] The surgical navigation system executes step S356, and the obtained surgical navigation knowledge data includes:

[0233] .

[0234] In this way, by generating surgical navigation knowledge data, the surgical navigation system can provide clinicians with clear, reliable, and traceable decision support information, thereby improving the scientific nature and safety of surgical planning.

[0235] Figure 11 This is a schematic diagram of the process for constructing a retrieval channel provided in an embodiment of this application.

[0236] like Figure 11As shown, in one implementation, the knowledge retrieval enhancement method for laparoscopic surgical navigation further includes steps S010-S050.

[0237] Step S010: Construct a surgical knowledge resource base, which includes at least one knowledge unit.

[0238] Among them, the surgical knowledge resource base refers to the collection of surgical knowledge from different knowledge sources (such as surgical guidelines, expert consensus, expert-annotated surgical procedures, surgical teaching videos, preoperative images and navigation semantic maps, historical case data, etc.) after preprocessing.

[0239] A knowledge unit refers to the basic unit in a surgical knowledge resource base used to represent a single surgical knowledge point or operational element, such as a surgical procedure, a standard procedure for using an instrument, an anatomical structure identifier, a risk warning, or an operational parameter. Each knowledge unit can include attribute information, source identifier, and credibility weight to facilitate tracing and weight calculation during subsequent queries and scoring processes.

[0240] In this way, by building a unified surgical knowledge resource base, a standardized knowledge foundation can be provided for subsequent multi-channel retrieval and fusion scoring, thereby improving the accuracy and reliability of surgical data queries.

[0241] In one implementation, step S010 includes the following steps S011-S013.

[0242] Step S011: Obtain surgical knowledge data from different knowledge sources.

[0243] The knowledge sources may include at least one of the following: surgical guidelines, expert consensus, expert-annotated surgical procedures, surgical teaching videos, preoperative images and navigation semantic maps, and historical case data.

[0244] When the sources of knowledge include surgical guidelines and expert consensus, the content of surgical knowledge data includes, for example, the surgical procedure, risk management, and complication prevention for partial nephrectomy as outlined in the Urological Association guidelines.

[0245] When the knowledge source includes expert annotations of the surgical procedure, the content of the surgical knowledge data includes, for example, surgical stages, tasks, instruments, actions, and anatomical structure correspondences annotated by urology experts.

[0246] When the knowledge source includes surgical instruction videos, the content of the surgical knowledge data may include, for example, typical partial nephrectomy procedure steps and visualizations containing keyframes.

[0247] When the knowledge sources include preoperative images and navigation semantic maps, the content of surgical knowledge data includes spatial relationships such as the typical location of kidney tumors in different patient groups, their relationship with adjacent blood vessels, and the spatial distribution of risk areas.

[0248] When the knowledge source includes historical case data, the content of the surgical knowledge data includes, for example, the anonymized case characteristics, risk structure, surgical treatment results, and expert review records.

[0249] In this way, by acquiring surgical knowledge data from multiple sources, the surgical navigation system can comprehensively cover all kinds of information dimensions involved in the surgical process, thereby providing a rich and diverse data foundation for the subsequent construction of knowledge units.

[0250] Step S012 involves de-identifying, template organizing, version marking, and source registration of the surgical knowledge data, and generating knowledge units based on the processed surgical knowledge data.

[0251] De-identification is used to remove patient privacy information, such as name, ID number, and hospital number, from surgical knowledge data.

[0252] The template processing step is used to uniformly convert surgical knowledge data from different sources and in different formats into a preset structured template. For example, it transforms free text descriptions of surgical steps into standardized field combinations of "stage-task-action-instrument-anatomical structure". It should be noted that the structured template can have various structured settings, and this embodiment does not limit them.

[0253] Version tagging is used to tag knowledge units, such as tagging the version number, update date, and modification description of the knowledge unit, to facilitate subsequent tracking of knowledge evolution.

[0254] Among them, the source registration process is used to record the data source information of knowledge units, including source type, providing institution, acquisition time, etc.

[0255] It should be noted that when a knowledge unit contains content that requires manual review, the surgical navigation system can mark the review status of the knowledge unit. The review status includes, for example, pending review, approved review, and unapproved review.

[0256] In this way, the surgical navigation system can transform multi-source heterogeneous surgical knowledge data into unified and standardized knowledge units through processes such as de-identification, template organization, version marking, and source registration, thereby providing a standardized data foundation for subsequent knowledge retrieval, reasoning, and application.

[0257] Step S013: Construct a surgical resource library based on knowledge units.

[0258] In this way, the surgical navigation system integrates multiple knowledge units to generate a surgical resource library for structured processing.

[0259] Step S020: Construct a surgical knowledge graph, which includes entities obtained based on knowledge units, entity relationships between different entities, and surgical rules based on entities and entity relationships.

[0260] In one implementation, step S020 includes the following steps.

[0261] Step S021: Obtain entities and entity relationships between different entities based on knowledge units.

[0262] In one implementation, the surgical navigation system can extract or input entities and entity relationships from knowledge units.

[0263] In this context, an entity refers to a node in the surgical knowledge graph that has a specific meaning. Entity relations refer to the directed semantic relationships between different entities.

[0264] For example, in a navigation scenario for partial nephrectomy, the types of entities include at least one of the following: surgical phase, surgical task, anatomy, surgical instrument, action, risk event, rule, and case. This disclosure does not limit the types of entities.

[0265] The surgical stage entities are used to illustrate the various surgical stages of partial nephrectomy, such as the renal hilum dissection stage, vascular control stage, tumor resection stage, and wound suturing stage.

[0266] The attributes of a surgical phase entity include at least one of the following: phase identifier (phase_id), phase name (phase_name), phase order (phase_order), phase description (description), and typical duration range (typical_duration_range).

[0267] Surgical task entities are used to describe the fine-grained task nodes under each surgical stage, such as the task of freeing the renal artery, confirming the occlusion task, removing the tumor task, and suturing the collection system.

[0268] The attributes of a surgical task entity include at least one of the following: task identifier (task_id), task name (task_name), phase identifier (phase_id), preconditions (preconditions), completion conditions (completion_conditions), focus structures (focus_structures), risk structures (risk_structures), recommended instruments (recommended_instruments), and risk events (on_risk_event).

[0269] In this context, both the prerequisite and completion conditions are evaluateable Boolean expressions. For example, preconditions = "Renal hilum region visible == True AND dissection clamp in place == True".

[0270] Anatomical structures are used to describe the anatomical structures involved in the operation, such as the renal parenchyma, renal artery, renal vein, collecting system, tumor, etc.

[0271] The attributes of an anatomical entity include at least one of the following: structure_id, name, synonyms, function_type, surgical_relevance, and risk_level.

[0272] Among them, functional types include substantial functions, blood flow channels, urinary channels, pathological lesions, or benign lesions.

[0273] Among these, surgical relevance includes protection, attention, avoidance, or resection.

[0274] The risk levels range from 1 to 5, with each level indicating the severity of the accidental injury consequences. The higher the level, the more serious the consequences of the accidental injury.

[0275] The surgical instrument section describes the instruments used during the procedure, such as clamps, needle holders, suction devices, scissors, and electrocautery instruments.

[0276] The attributes of a surgical instrument entity include at least one of the following: instrument ID (instrument_id), instrument name (instrument_name), synonyms (synonyms), typical actions (typical_actions), and associated surgical phases (associated_phases).

[0277] The entity of the operation action is used to describe the operation action performed by the instrument, such as traction, separation, clamping, cutting, electrocoagulation, and suturing.

[0278] The attributes of an operation action entity include at least one of the following: action ID (action_id), action name (action_name), typical instruments (typical_instruments), and target structures (target_structures).

[0279] Risk event entities are used to describe adverse events that may occur during surgery, such as bleeding, urinary leakage, smoke obscuring, and unclear tumor boundaries.

[0280] The attributes of a risk event entity include at least one of the following: event ID (event_id), event type (event_type), typical causes (typical_causes), severity range (severity_range), and recommended rule IDs (recommended_rule_ids).

[0281] The processing rule entity is used to describe the rules for dealing with risks or guiding operations, such as risk avoidance rules, phase transition rules, and navigation prompt rules.

[0282] The attributes of a processing rule entity include at least one of the following: rule identifier (rule_id), rule type (rule_type), preconditions (preconditions), processing action (action), priority (priority), source (source), and review status (review_status).

[0283] The rule types include phase transition rules, risk avoidance rules, device selection rules, and prompt generation rules.

[0284] The prerequisite is a Boolean expression that can be evaluated.

[0285] Case examples are used to illustrate historical or teaching cases, such as similar tumor locations, similar vascular relationships, and similar treatment strategies.

[0286] The attributes of a case entity include case ID, patient profile, tumor characteristics, key findings, surgical outcome, and associated video segments.

[0287] The patient profile includes the patient's age range and gender after desensitization. Tumor characteristics include tumor location, size, and RENAL score. Key findings include intraoperative anatomical variations, risk events, and their management. Surgical outcomes include resection status, complications, and postoperative recovery.

[0288] For example, in the navigation scenario of partial nephrectomy, the types of entity relationships include at least one of the following: phase_contains_task, task_focuses_on, instrument_performs, action_on, anatomy_adjacent_to, event_triggers_rule, task_dependency, and case_related_to.

[0289] It should be noted that each entity relationship corresponds to a head entity and a tail entity at its two ends, and the relationship is directional. This disclosure does not limit the type of entity relationship.

[0290] In the context of a stage-includes-task relationship, the head entity represents the surgical stage entity, and the tail entity represents the surgical task entity. This relationship describes how a stage contains a task, such as the tumor resection stage containing tumor boundary confirmation.

[0291] In the task-focused structural relation, the head entity represents the surgical task entity, and the tail entity represents the anatomical structure entity. The task-focused structural relation is used to specify that the task focuses on a particular anatomical structure, such as the renal hilum anatomy focusing on the renal artery.

[0292] In the instrument action relationship, the head entity is the surgical instrument entity, and the tail entity is the action entity. The instrument action relationship is used to describe that the instrument can perform a certain action, such as the clamping action performed by a forceps.

[0293] In action relationships, the head entity represents the operational action entity, and the tail entity represents the anatomical structure entity. Action relationships are used to describe the action performed on a specific anatomical structure, such as suturing acting on an assembly system.

[0294] In the structural proximity risk relationship, the head entity and the tail entity are both anatomical structural entities. Anatomical structural proximity relationships are used to describe the spatial proximity relationships between structures, such as the tumor proximity aggregation system.

[0295] In the event triggering rule relationship, the head entity is the risk event entity, and the tail entity is the processing rule entity. The event triggering rule relationship is used to describe the corresponding processing rules triggered by a risk event, such as a bleeding event triggering a hemostasis rule.

[0296] In the task dependency relationship, the head entity is the surgical task entity, and the tail entity is the surgical stage entity. The task dependency relationship is used to describe the sequential execution order of tasks, such as proceeding to tumor resection after renal artery control is completed.

[0297] In case association, the head entity is the case entity, and the tail entities are entities of various types (e.g., anatomical structures and surgical tasks). Case association is used to describe the relationship between a case and various entities, such as a case being associated with a specific tumor location and vascular variations.

[0298] Step S022: Determine surgical rules based on entities and entity relationships.

[0299] Among them, surgical rules can be represented as a triple of pseudo-entity-entity relation-entity, which is used to describe the logical constraints in surgical knowledge.

[0300] Step S023: Store entities as nodes and entity relationships as edges in the graph database, and define surgical rules to generate a surgical knowledge graph.

[0301] Graph databases are a type of database that uses a graph structure to store and query data. They belong to the category of NoSQL databases and use vertices, edges, and properties to represent and store data, focusing on the relationships between entities.

[0302] The surgical navigation system generates a surgical knowledge graph by storing entities as nodes and entity relationships as edges in a graph database and defining surgical rules. The surgical knowledge graph can be defined as a metadata file in JSON format.

[0303] During the initialization of the surgical navigation system, it can load and verify metadata files for use in subsequent practical applications of the surgical knowledge graph.

[0304] For example, in the navigation scenario of partial nephrectomy, the initialized surgical knowledge graph contains approximately 1500 entity nodes (corresponding to 8 entity types), approximately 3200 relational edges, and approximately 200 surgical rules. After the surgical knowledge is structured and stored in the form of a graph database, the surgical navigation system can perform efficient path traversal and relational reasoning on the graph using the Cypher query language. For example, starting from the surgical task entity node "tumor boundary confirmation," it retrieves the anatomical structure entity node "tumor" along the task_focuses_on edge, then retrieves the high-risk anatomical structure entity node "set system" adjacent to "tumor" along the anatomy_adjacent_to edge, and further retrieves the surgical processing rule entity node "suture set system rule" triggered by the "set system" along the event_triggers_rule edge. In this way, the composite reasoning result that originally required multiple independent queries is obtained after a 3-hop path traversal, which greatly improves the knowledge retrieval efficiency and interpretability of the surgical navigation system in complex intraoperative scenarios.

[0305] In this way, the surgical navigation system transforms the knowledge that was originally scattered in guidelines, expert experience, surgical videos and case records into computable, searchable and traceable structured knowledge, providing clear task basis and risk interpretation for intelligent agents in the intelligent decision-making layer, supporting precise terminology query, relational reasoning and path tracing, and overcoming the shortcomings of flat text knowledge bases.

[0306] Step S030: Construct a visual case library based on surgical visual data, wherein the surgical visual data includes at least one of surgical videos, surgical keyframes, intraoperative images, and surgical teaching videos.

[0307] For example, in a partial nephrectomy navigation scenario, a visual case library can be built based on the partial nephrectomy surgical video, keyframes, intraoperative images, or instructional video clips.

[0308] In one implementation, step S030 includes the following steps.

[0309] Step S031: The surgical visual data is labeled based on feature information of at least one dimension to obtain labeled surgical visual data, wherein each feature information includes at least one labeling field.

[0310] The types of annotation fields include at least one of the following: surgical phase, surgical task, surgical instrument, anatomical structure, event, and risk level. This disclosure does not limit the specific types of annotation fields.

[0311] In one implementation, the annotation method may include at least one of manual annotation, semi-automatic annotation, or automatic annotation based on a pre-trained model. This disclosure does not limit the specific annotation method.

[0312] Step S032: Generate structured surgical visual data based on the labeled surgical visual data corresponding to the labeled fields.

[0313] The structured surgical visual data includes at least one structured field selected from case ID, video segment start and end frames, surgical phase, surgical task, visible structures, instruments, event, expert note, and keyframe IDs. This disclosure does not limit the types of structured fields in the visual case library.

[0314] The structured fields for Case Identifier, Video Segment Start and End Frames, Surgical Stage, Surgical Task, Visible Anatomical Structures, Instruments, and Events are as follows: Case Identifier (unique case number), Keyframe Identifier (list of keyframes), and Keyframe Identifier (list of keyframe numbers).

[0315] Step S033: Construct a visual case library based on structured surgical visual data.

[0316] The surgical navigation system employs multi-dimensional, systematic annotation and structured processing of surgical visual data. This ensures that each case in the visual case library possesses complete feature information, guaranteeing that surgical visual data is no longer isolated media files but rather searchable knowledge assets highly correlated with the surgical scenario context. Simultaneously, the structured visual case library lays the foundation for establishing cross-modal knowledge mapping relationships with the surgical knowledge graph. This allows textual rules and visual cases to jointly point to the same knowledge concept, enabling the surgical navigation system to simultaneously invoke rule-based knowledge and similar case experience when generating navigation prompts, significantly enhancing the clinical reference value of the navigation prompts.

[0317] Step S040: Establish knowledge mapping relationships based on the surgical knowledge graph and the visual case library, and generate a multimodal knowledge graph based on the knowledge mapping relationships.

[0318] In one implementation, step S040 includes the following steps S041-S047.

[0319] Step S041: Based on the anatomical structure field in the entity in the surgical knowledge graph and the visible structure annotation field in the annotation field of the visual case library, perform lowest-level name matching, where both the anatomical structure field and the visible structure annotation field include hierarchical names from high to low.

[0320] In one implementation, the anatomical structure field has a hierarchical naming system, which describes the structure from high to low, such as human body structure, urinary system, kidney, blood vessels, arteries, and renal artery.

[0321] It is evident that the structural annotation field also has a hierarchical naming system, which is described in a hierarchical structure from high to low. Its hierarchical naming system may be the same as or different from that of the anatomical structure field, for example, blood vessel, artery, and renal artery in that order.

[0322] The lowest level name refers to the most granular name in the hierarchy (e.g., the renal artery in the anatomical structure field, or the renal artery in the visible structure annotation field).

[0323] In one implementation, the surgical navigation system achieves the lowest-level name matching by matching the names and synonyms in the anatomical structure field with the visible structures annotation field in the annotation field.

[0324] Step S042: When the lowest-level names corresponding to the anatomical structure field and the visible structure annotation field match, a knowledge mapping relationship is established based on the lowest-level names corresponding to the anatomical structure field and the visible structure annotation field.

[0325] Further based on the aforementioned example, when the surgical navigation system determines that the lowest-level name of the anatomical structure field, "renal artery," matches the lowest-level name of the visible structure annotation field, "renal artery," it establishes a mapping relationship between the lowest-level name of the anatomical structure field, "renal artery," and the lowest-level name of the visible structure annotation field, "renal artery."

[0326] Step S043: If the lowest level names corresponding to the anatomical structure field and the visible structure annotation field do not match, traverse the level names of the previous level of the visible structure annotation field in turn and perform level matching with the anatomical structure field to obtain the target level name corresponding to the anatomical structure field. The target level name is the same as the current level name of the visible structure annotation field, or the semantic similarity between the target level name and the current level name of the visible structure annotation field is greater than a preset similarity threshold.

[0327] Further based on the aforementioned example, if the lowest-level name "renal artery" does not exist in the visible structure annotation field, the surgical navigation system will sequentially traverse the upper-level names of the visible structure annotation field and then perform hierarchical matching with the anatomical structure field to obtain the target hierarchical name corresponding to the anatomical structure field, such as "artery".

[0328] In other words, the surgical navigation system traversed the surgical knowledge graph and found entities that were similar in level or semantics to the visible structure annotation fields.

[0329] In this way, the surgical navigation system can align the visible structures labeled in the visual case library with the anatomical entities in the knowledge graph through hierarchical traversal and semantic matching, thereby establishing cross-modal knowledge associations. This hierarchical matching strategy effectively solves the alignment difficulties caused by differences in labeling granularity or inconsistencies in terminology, improving the accuracy and robustness of multimodal knowledge graph construction.

[0330] Step S044: If the target level name of the anatomical structure field matches the level name of the current level traversed by the visible structure annotation field, a knowledge mapping relationship is established based on the target level name of the anatomical structure field and the level name of the current level of the visible structure annotation field.

[0331] Further based on the aforementioned example, when the surgical navigation system determines that the target level name "artery" of the anatomical structure field matches the current level name "artery" traversed by the visible structure annotation field, it establishes a mapping relationship between the target level name "artery" of the anatomical structure field and the current level name "artery" of the visible structure annotation field.

[0332] Step S045: Entities with knowledge mapping relationships are identified as candidate entities.

[0333] Based on the aforementioned steps, the surgical navigation system can establish multiple mapping relationships and use the entities corresponding to these mapping relationships as candidate entities.

[0334] Step S046: Determine the target entity based on the surgical task annotation field and surgical stage annotation field in the annotation fields of the visual case library.

[0335] In one implementation, the surgical navigation system uses the surgical task (task) annotation field and the surgical phase (phase) annotation field, combined with the task_focuses_on relationship in the surgical knowledge graph, to filter out the candidate entities with the highest relevance to the current surgical phase or surgical task, which are then used as the final target entities.

[0336] Step S047: Generate a multimodal knowledge graph based on the target entity.

[0337] In one implementation, the surgical navigation system determines the target knowledge unit corresponding to the target entity.

[0338] It's important to note that the target knowledge unit's attributes include confidence level. If the confidence level of the target knowledge unit is lower than a preset threshold, the surgical navigation system marks the target knowledge unit as pending confirmation and triggers a manual review process. After the manual review is passed, the surgical navigation system writes the target knowledge unit into the multimodal knowledge graph and establishes a case-related-to relationship based on the target knowledge unit. In this way, semantic alignment is achieved between the surgical knowledge graph and the visual case library through the establishment of knowledge mapping relationships, thereby forming a unified multimodal knowledge representation.

[0339] For example, the terms "renal artery," "renal artery," and "anatomy-artery-renal" in the surgical knowledge graph and "renal artery" in the video case library are all mapped to the same anatomical structure entity.

[0340] In this way, the surgical navigation system generates a multimodal knowledge graph that associates knowledge units with surgical visual data, realizing cross-modal knowledge association. This makes it easier for the intelligent agent in the intelligent decision-making layer to simultaneously call upon rule knowledge and similar case experience when generating prompts.

[0341] Step S050: Based on the multimodal knowledge graph, knowledge units, and visual case library, at least one retrieval channel is generated. The retrieval channel is used to represent the mapping relationship between the retrieval index and the preset surgical data.

[0342] This application provides an example using a surgical navigation system that generates three retrieval channels. However, this application does not actually limit the specific number of retrieval channels.

[0343] In one implementation, the surgical navigation system generates vector retrieval channels, keyword retrieval channels, and graph retrieval channels based on a multimodal knowledge graph, knowledge units, and a visual case library. The vector retrieval channel includes a preset vector index for linking to a vector database, the keyword retrieval channel includes a preset keyword index for linking to a keyword search engine, and the graph retrieval channel includes a preset graph index for linking to a graph database.

[0344] In one implementation, the vector database corresponding to the vector retrieval channel includes the Milvus vector database. A preset vector index is used to retrieve semantically similar textual knowledge, case descriptions, and video clips. The index content of the preset vector index includes the semantic vectors of textual knowledge entries, the semantic vectors of case descriptions, and the visual feature vectors of video clips.

[0345] For example, the preset vector index is obtained by vectorizing knowledge units in 3072 dimensions based on the text-embedding-3-large embedding model.

[0346] It should be noted that this application does not limit the specific type of vector database in its embodiments. Furthermore, this application does not limit the embedding model type and dimension of the vector index; the vector index may include more or fewer index contents.

[0347] In one implementation, the keyword search engine corresponding to the keyword retrieval channel includes an Elasticsearch index. A pre-defined keyword index is used to retrieve specific terms, device names, structure names, or risk events. The index content of the pre-defined keyword index includes precise fields such as device names, anatomical terms, rule clauses, and phase names.

[0348] The default keyword index is configured as an inverted index, which supports Boolean queries and fuzzy matching, and can quickly locate knowledge units or surgical visual data containing specific terms.

[0349] It should be noted that this application does not limit the specific type of keyword search engine. Furthermore, this application does not limit the algorithms supported by the preset keyword index.

[0350] The graph database corresponding to the graph retrieval channel includes the Neo4j graph database. The preset graph index is used to traverse entity relationships, task dependencies, risk structures, and processing rules. The preset graph index content includes entities, relationships, attributes, and rule conditions, and supports path traversal and condition filtering based on the Cypher query language.

[0351] It should be noted that this application embodiment does not limit the specific type of graph database. Furthermore, this application embodiment does not limit the query language and traversal algorithm supported by the preset graph index.

[0352] In this way, the surgical navigation system provides a multi-channel retrieval index that can be queried and invoked. In actual surgical scenarios, it can quickly respond to query needs from different dimensions, providing a data foundation for subsequent intelligent decision-making and real-time assistance.

[0353] It's important to note that the surgical navigation system can be built upon three retrieval channels established using offline preprocessing tasks, a process that can be completed during system deployment. During online operation, the surgical navigation system only performs searches without modifying the index (except for periodic knowledge graph updates). By constructing three retrieval channels, the system can, upon receiving a query, drive multiple channels to perform searches in parallel. This allows for semantically broad similarity retrieval using the vector retrieval channel, precise keyword matching using the keyword retrieval channel, and logically rigorous relational path reasoning retrieval using the knowledge graph retrieval channel. Through the parallel operation of these three retrieval channels, the surgical navigation system can meet diverse surgical scenario query needs while maintaining low-latency online retrieval response capabilities, thus improving the real-time knowledge retrieval efficiency of the surgical navigation system.

[0354] In one implementation, the knowledge retrieval enhancement method for laparoscopic surgical navigation can be configured on a computing platform, such as an edge computing server.

[0355] The solutions disclosed herein are not limited to the embodiments mentioned above.

[0356] The knowledge retrieval enhancement method for laparoscopic surgical navigation provided in this disclosure has been described above. It is understood that the surgical data query device, in order to realize the various functions of this knowledge retrieval enhancement method for laparoscopic surgical navigation, may include corresponding hardware and software for implementing the hardware functions.

[0357] Those skilled in the art will readily recognize that the steps of the knowledge retrieval enhancement method for laparoscopic surgical navigation described in conjunction with the embodiments of this disclosure can be implemented in hardware, in a software-driven hardware manner, or in a combination of hardware and software. Whether a function is implemented in hardware or software-driven hardware 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 implementation should not be considered beyond the scope of this disclosure.

[0358] Figure 12 This is a schematic diagram of the knowledge retrieval enhancement device for laparoscopic surgical navigation provided in this application embodiment.

[0359] like Figure 12 As shown, in some embodiments, this application provides a knowledge retrieval enhancement device 1200 for laparoscopic surgical navigation, applied to a surgical navigation system. The device includes: The first generation module 1210 is used to generate a structured initial query object in response to a query request; wherein the query request includes at least one dimension of structured surgical query field; The second generation module 1220 is used to process the initial query object based on a preset query expansion rule to generate a query task including the initial query object and a target query subtask; wherein, the query expansion rule is used to characterize the mapping relationship between the preset triggering condition and the query subtask, and when the initial query object matches the preset triggering condition, the target query subtask corresponding to the initial query object is obtained. The query module 1230 is used to perform a retrieval in the surgical navigation system based on the query task, generate structured surgical navigation knowledge data, and output it.

[0360] The knowledge retrieval enhancement device for laparoscopic surgical navigation disclosed in this application first responds to a query request by generating a structured initial query object. In this way, the surgical navigation system transforms unstructured surgical query fields into a structured data format, laying the foundation for subsequent accurate data queries. Furthermore, the surgical navigation system processes the initial query object using preset query expansion rules. When the initial query object meets preset triggering conditions (such as specific intraoperative risk events), it automatically expands to generate target query subtasks. This process breaks the limitations of relying on explicit user input, realizing a shift from "passive response" to "active prediction," and enabling more comprehensive and accurate query tasks based on the potential risks or implicit needs of the current surgical scenario. Furthermore, the surgical navigation system performs retrieval based on a query task that includes an initial query object and query subtasks, generating and outputting structured surgical navigation knowledge data. It not only utilizes a three-channel fusion retrieval system (vector retrieval, keyword retrieval, and graph retrieval) to obtain surgical data directly matching the user's intent, but also simultaneously acquires association rules, risk warnings, or similar cases highly relevant to the current scenario. This effectively solves the problems of existing surgical navigation systems where the retrieved surgical data is limited in content and lacks scenario adaptability, improving the accuracy and scenario adaptability of knowledge retrieval. Moreover, the output surgical navigation knowledge data can be directly called by the navigation decision module of the surgical navigation system, providing traceable knowledge support. This embodiment can effectively assist doctors in making surgical decisions, improving the safety and efficiency of surgery.

[0361] It should be noted that in this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0362] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0363] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-included system or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0364] In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc. The above embodiments are merely specific embodiments of this application and are not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made based on the technical solutions of this application should be included within the scope of protection of this application.

Claims

1. A knowledge retrieval enhancement method for laparoscopic surgical navigation, applied to a surgical navigation system, characterized in that, The method includes: In response to a query request, a structured initial query object is generated; wherein the query request includes at least one dimension of structured surgical query fields; The initial query object is processed based on preset query expansion rules to generate a query task including the initial query object and a target query subtask; wherein, the query expansion rules are used to characterize the mapping relationship between preset triggering conditions and query subtasks, and when the initial query object matches the preset triggering conditions, the target query subtask corresponding to the initial query object is obtained. Based on the query task, a retrieval is performed in the surgical navigation system to generate structured surgical navigation knowledge data and output it.

2. The knowledge retrieval enhancement method for laparoscopic surgical navigation according to claim 1, characterized in that, The surgical query field includes a query target field and a scenario status field; wherein, the query target field includes at least one of the query requirements input by the user and the retrieval requirements generated by the intelligent decision layer; the scenario status field includes at least one of the stage field, structure field, and event field output by the intraoperative perception layer, and at least one of the personalized anatomical feature field output by the registration and positioning layer and the task field output by the intelligent decision layer.

3. The knowledge retrieval enhancement method for laparoscopic surgical navigation according to claim 1, characterized in that, The initial query object is processed based on preset query expansion rules to generate a query task that includes the initial query object and a target query subtask, including: Based on the preset triggering conditions in the query expansion rules, the structured surgical query fields in the initial query object are verified to obtain the verification results; If the verification result shows that the preset triggering condition matches the surgical query field, candidate subtasks are generated, wherein each candidate subtask has a corresponding priority. The top N candidate subtasks with the highest priority are determined as the target query subtasks, where 0 < N ≤ 6; The query task is generated based on the initial query object and the target query subtask.

4. The knowledge retrieval enhancement method for laparoscopic surgical navigation according to claim 1, characterized in that, The process of performing a retrieval in the surgical navigation system based on the query task, generating structured surgical navigation knowledge data, and outputting it includes: The query task is input into the vector retrieval channel, keyword retrieval channel, and map retrieval channel of the surgical navigation system, respectively. The query task is searched in the vector search channel, the keyword search channel, and the graph search channel respectively to obtain the corresponding vector search results, keyword search results, and graph search results. Based on the vector retrieval results, the keyword retrieval results, and the graph retrieval results, the structured surgical navigation knowledge data is generated.

5. The knowledge retrieval enhancement method for laparoscopic surgical navigation according to claim 4, characterized in that, The process of performing searches on the query task in the vector search channel, the keyword search channel, and the graph search channel respectively, to obtain corresponding vector search results, keyword search results, and graph search results, includes: In the vector retrieval channel, the query task is transformed into a query vector, and the query vector is matched with a preset vector index in the vector database to obtain the matched first preset surgical data as the vector retrieval result; wherein, the vector database is used to store the mapping relationship between the preset vector index and the first preset surgical data; In the keyword retrieval channel, keywords are extracted from the query task, and the keyword search engine is used to match the keywords with a preset keyword index to obtain the matched second preset surgical data as the keyword retrieval result; wherein, the keyword search engine is used to store the mapping relationship between the preset keyword index and the second preset surgical data; In the graph retrieval channel, target fields are extracted from the query task, and the target fields are matched with entities in the graph database. Based on the surgical processing rules corresponding to the relationships between the entities, the matched third preset surgical data is obtained as the graph retrieval result. The target fields include at least one of surgical task fields, surgical event fields, and anatomical structure fields. The graph database is used to represent the relationships between different entities, and the relationships between different entities are used to represent the surgical processing rules.

6. The knowledge retrieval enhancement method for laparoscopic surgical navigation according to claim 1, characterized in that, The process of performing a retrieval in the surgical navigation system based on the query task, generating structured surgical navigation knowledge data, and outputting it includes: Based on the query task, a search is performed in the surgical navigation system to obtain the search results; Based on the search results, structured surgical navigation knowledge data is generated, wherein the surgical navigation knowledge data is used to represent structured surgical data fields, and the surgical data fields are used to describe the search results; based on the intelligent decision layer of the surgical navigation system, the surgical navigation knowledge data is invoked to output navigation prompt information.

7. The knowledge retrieval enhancement method for laparoscopic surgical navigation according to claim 6, characterized in that, The surgical data fields include at least one of the following: query identifier field, timestamp field, surgical operation rule field, surgical treatment structure field, surgical risk source and explanation field, recommended surgical operation and precautions field, surgical similar case field, prompt word template field, surgical data confidence field, and surgical data source field.

8. The knowledge retrieval enhancement method for laparoscopic surgical navigation according to claim 5, characterized in that, The process of performing a retrieval in the surgical navigation system based on the query task, generating structured surgical navigation knowledge data, and outputting it includes: Determine the target score for the retrieval results from different channels, wherein the target score has a corresponding knowledge source credibility weight; When the channel includes the map retrieval channel, the target score of the map retrieval result is multiplied by a preset enhancement coefficient to obtain the enhanced target score; In cases where the search results contain the same knowledge unit, the search result with the highest target score is retained; The final target scores of the search results based on different channels are sorted in descending order, wherein the final target scores include the enhanced target score corresponding to the graph retrieval channel, the target score corresponding to the vector retrieval channel, and the target score corresponding to the keyword retrieval channel; Based on the top K final target scores of the sorted results, candidate retrieval results are determined, where 0 < K < 10; Based on the candidate search results, the surgical navigation knowledge data is generated and output.

9. The knowledge retrieval enhancement method for laparoscopic surgical navigation according to claim 8, characterized in that, The step of generating and outputting the surgical navigation knowledge data based on the candidate search results includes: In the case of conflicting data in the candidate search results, the conflicting data is filtered based on the knowledge source credibility weight corresponding to the final target score, and the candidate search results corresponding to the highest knowledge source credibility weight are retained. The conflicting data is used to characterize the candidate search results from different sources that have the same knowledge unit and the same knowledge unit includes conflicting attribute information. Based on the filtered candidate search results, the surgical navigation knowledge data is generated and output, wherein the surgical navigation knowledge data is used to characterize structured surgical data fields, and the surgical data fields are used to describe the surgical data.

10. The knowledge retrieval enhancement method for laparoscopic surgical navigation according to claim 8, characterized in that, The target score for determining the retrieval results of different channels includes: In each channel, the target score of the search result in the current channel is determined based on the product of the relevance score of the search result, the knowledge source credibility weight of the search result, and the channel weight of the search result, wherein the relevance score is used to characterize the relevance between the query task and the search result.

11. The knowledge retrieval enhancement method for laparoscopic surgical navigation according to claim 10, characterized in that, The channel weights include the first channel weight corresponding to the vector retrieval result, the second channel weight corresponding to the keyword retrieval result, and the third channel weight corresponding to the graph retrieval result, wherein the third channel weight is greater than the second channel weight, and the second channel weight is greater than the first channel weight.

12. A knowledge retrieval enhancement device for laparoscopic surgical navigation, applied to a surgical navigation system, characterized in that, The device includes: The first generation module is used to generate a structured initial query object in response to a query request; wherein the query request includes at least one dimension of structured surgical query field; The second generation module is used to process the initial query object based on preset query expansion rules to generate a query task including the initial query object and a target query subtask; wherein, the query expansion rules are used to characterize the mapping relationship between preset triggering conditions and query subtasks, and when the initial query object matches the preset triggering conditions, the target query subtask corresponding to the initial query object is obtained. The query module is used to perform a retrieval in the surgical navigation system based on the query task, generate structured surgical navigation knowledge data, and output it.

13. A surgical navigation system, characterized in that, Configured to perform the knowledge retrieval enhancement method for laparoscopic surgical navigation as described in any one of claims 1-11.

Images