Medical process flow intelligent diagnosis system and method based on knowledge graph

Abstract

The invention discloses a medical process flow intelligent diagnosis system and method based on a knowledge graph, and relates to the technical field of building design intelligence. The system comprises a drawing analysis module, a medical process knowledge graph, a knowledge retrieval module, an intelligent diagnosis module, a diagnosis report generation module and a manual auditing module. A medical process knowledge graph covering specification standards and hospital cases is constructed, an optical character recognition and target detection technology is adopted to analyze a drawing, a vector retrieval technology is utilized to retrieve related knowledge, intelligent diagnosis is performed based on a multi-modal artificial intelligence model, and the medical process knowledge graph is analyzed. And comprehensive inspection is carried out from five dimensions of general plane layout, department relationship, functional integrity, area ratio and fire evacuation, and a diagnosis report is automatically generated. According to the method, the diagnosis efficiency is remarkably improved, the diagnosis time of a single drawing is shortened to be within 5 minutes, the diagnosis accuracy rate reaches 85% or above, problem omission is effectively reduced, multi-standard collaboration is achieved, and continuous optimization is supported.

Description

Technical Field

[0001] This invention relates to the field of intelligent building design technology, specifically to an intelligent diagnostic system and method for medical processes based on knowledge graphs and artificial intelligence technologies. Background Technology

[0002] Medical process design is a core component of hospital architectural design, and its rationality and accuracy directly affect the hospital's operational efficiency, medical safety, and patient experience. The primary medical process flow mainly includes the hospital's overall site layout, medical function organization, departmental location relationships, and room size allocation, involving standards and regulations from multiple professional fields such as architecture, medicine, and fire protection.

[0003] Currently, the diagnosis and analysis of primary medical process flows mainly rely on manual labor, which presents the following technical problems: (1) Low efficiency of manual diagnosis. A complete first-level flow chart of hospital medical process requires consulting dozens of standards and specifications, such as the "Code for Design of General Hospital Buildings" GB51039, the "Code for Design of Clean Operating Room Buildings" GB50333, and the "Code for Fire Protection Design of Buildings" GB50016. Manual diagnosis of a single chart usually takes several hours or even days, which is difficult to meet the needs of rapid design iteration.

[0004] (2) High dependence on professional knowledge. Medical process design requires a mastery of knowledge from multiple disciplines such as architecture, medicine, and management. It places extremely high demands on the professional ability of designers. Non-medical process professionals find it difficult to accurately judge the rationality of the design scheme and are prone to making judgment errors.

[0005] (3) Problems that are easy to overlook. The regulations are numerous and complex, involving thousands of technical requirements. Manual inspection can easily overlook key issues, such as unreasonable layout of entrances and exits in the overall plan, lack of resource sharing among related departments, lack of necessary departments, and insufficient number of evacuation staircases. If these problems are not discovered in the planning stage, but are only discovered in the construction drawing stage or even the construction stage, it will lead to major design changes and cause huge time and economic losses.

[0006] (4) Difficulty in coordinating multiple standards. Medical process design involves standards from multiple fields such as architecture, fire protection, and medical and health care. There are overlaps and connections between different standards, and it is difficult for humans to fully grasp the coordination relationship between various standards. It is easy to meet the requirements of one standard but violate the requirements of another standard.

[0007] In existing technologies, some intelligent architectural drawing review systems have been applied to disciplines such as fire protection and structure. However, these systems are mainly based on fixed rule engines and use if-then rules for judgment, making it difficult to handle the complex semantic relationships and reasoning problems in medical process design. Furthermore, existing systems are mainly aimed at general architectural disciplines, and there is currently no intelligent diagnostic system specifically designed for the highly specialized and knowledge-intensive field of medical processes.

[0008] Therefore, there is an urgent need to develop a system and method that can efficiently, accurately, and comprehensively perform intelligent diagnosis of the first-level processes of medical technology in order to solve the above-mentioned technical problems. Summary of the Invention

[0009] The purpose of this invention is to provide an intelligent diagnostic system and method for medical processes based on knowledge graphs, in order to solve the technical problems of low efficiency of manual diagnosis, strong reliance on professional knowledge, easy omissions, and difficulty in coordinating multiple standards in the prior art.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: A knowledge graph-based intelligent diagnostic system for medical processes includes: The drawing parsing module is used to receive medical process flow drawings, extract text information from the drawings using optical character recognition technology, identify department areas, functional rooms, and process routes in the drawings using object detection algorithms, and convert the recognition results into structured data. The medical process knowledge graph includes an entity library, a relationship library, a standard library, and a case library. The entity library stores entity objects such as medical departments, functional rooms, and medical equipment. The relationship library stores the positional relationships, process relationships, standard constraint relationships, and area ratio relationships between entities. The standard library stores national and local standards related to medical building design. The case library stores medical process data of completed hospital projects. The knowledge retrieval module is used to retrieve relevant standard provisions and reference cases in the medical process knowledge graph based on the structured data output by the drawing parsing module. The knowledge retrieval module adopts vector retrieval technology to convert drawing information and knowledge base content into vector representations and realizes knowledge retrieval by calculating vector similarity. The intelligent diagnosis module is used to perform intelligent diagnosis on drawings based on retrieved knowledge. It includes a general layout diagnosis unit, a departmental relationship diagnosis unit, a functional integrity diagnosis unit, an area ratio diagnosis unit, and a fire evacuation diagnosis unit. The intelligent diagnosis module adopts a multimodal artificial intelligence model, takes the drawing information and retrieved knowledge as input, and generates diagnostic conclusions through semantic understanding and reasoning. The diagnostic report generation module is used to automatically generate a structured diagnostic report based on the diagnostic results of the intelligent diagnostic module. The diagnostic report includes a problem description, a location marker of the problem, the violated normative clauses, improvement suggestions, and a severity rating of the problem. The manual review module provides a visual interface that allows professionals to review, modify, and supplement diagnostic results, and feeds the reviewed results back to the medical process knowledge graph for continuous optimization.

[0011] Furthermore, the construction of the medical process knowledge graph includes the following steps: collecting relevant standard documents, standard drawings, and medical process flow data of completed hospital projects related to medical building design; automatically extracting medical departments, functional rooms, and medical equipment entities from the documents using a named entity recognition algorithm; identifying the positional relationships, process relationships, and standard constraint relationships between entities using a relation extraction algorithm; and storing the extracted entities and relationships in a graph database to form a medical process knowledge graph.

[0012] Furthermore, the overall layout diagnostic unit checks the following: whether the location of the outpatient entrance, emergency entrance, inpatient visitation entrance, and waste exit conforms to the principle of separating clean and contaminated areas in the medical process; whether the functional zoning of clean areas, semi-clean areas, and contaminated areas is clear; and whether the relationship between each entrance and exit and urban roads and surrounding buildings is reasonable.

[0013] Furthermore, the departmental relationship diagnosis unit uses an adjacency analysis algorithm to calculate the spatial distance and connectivity between departments and determine the rationality of the layout of the following related departments: the feasibility of resource sharing between the outpatient pharmacy and the emergency pharmacy; the convenient connection between the emergency department and the radiology and laboratory departments; the proximity of the operating room to the ICU and the pathology department; and the relevance between the delivery room and the neonatal department.

[0014] Furthermore, the fire evacuation diagnostic unit checks the following: whether the number and location of evacuation staircases meet the requirements of the regulations; whether the evacuation stairwells have natural ventilation and lighting conditions; whether the refuge rooms in ward buildings and operating departments are set up in accordance with the regulations; and whether the elevator lobby is set up in accordance with fire prevention requirements.

[0015] This invention also provides a knowledge graph-based intelligent diagnostic method for medical processes, comprising the following steps: S1: Drawing parsing steps: Receive medical process flow drawings, extract text information using optical character recognition technology, identify department areas and functional room locations using target detection algorithms, and convert the recognition results into structured data; S2: Knowledge retrieval step, converting the structured data of the drawings into query vectors, retrieving the knowledge entries with the highest similarity in the vector database of the medical process knowledge graph, and filtering out relevant standard provisions and reference cases; S3: Intelligent diagnostic steps, construct diagnostic prompts, combine drawing information and retrieved knowledge as input to a multimodal artificial intelligence model, conduct diagnosis according to five dimensions: overall layout, departmental relationships, functional integrity, area ratio, and fire evacuation, and evaluate the confidence level of the diagnostic results; S4: Report generation steps: Summarize the problem list for each diagnostic dimension, mark the location coordinates, standard clauses, and severity level of each problem, generate improvement suggestions, and output a structured diagnostic report; S5: Manual review and feedback steps. Professionals review diagnostic reports through a visual interface, correct misjudged issues, supplement missing issues, and use the review results as new training samples to update the knowledge graph.

[0016] Furthermore, the confidence assessment method in S3 is as follows: a baseline confidence threshold of 0.7 is set; when the confidence of the diagnostic result is lower than 0.7, it is marked as a suggestion for manual review; when the confidence of the diagnostic result is lower than 0.5, it is marked as a requirement for manual review; for diagnostic results involving fire safety, regardless of the confidence level, they are all marked as requiring manual review.

[0017] Furthermore, the improvement suggestion generation method in S4 is as follows: retrieve reference cases similar to the current project in terms of scale and level from the case library; extract solutions to similar problems in the reference cases; and generate targeted improvement suggestions in combination with the requirements of the specification clauses.

[0018] Furthermore, the continuous optimization method in S5 includes: using manually reviewed and confirmed correct diagnostic results as positive samples; using manually corrected misjudgment results as negative samples; periodically performing incremental training on the diagnostic model; and supplementing the knowledge graph with newly added normative clauses and case data.

[0019] Furthermore, the diagnostic report includes the following parts: basic project information, including hospital level, number of beds, and building area; diagnostic summary, including the total number of problems and statistics categorized by severity; detailed list of problems, categorized and displayed according to five dimensions: overall layout, departmental relationships, functional integrity, area allocation, and fire evacuation; drawing annotations, marking the location of problems on the original drawings and using different colors to distinguish the severity of problems; and improvement suggestions, providing specific improvement measures and reference cases for each problem.

[0020] Compared with the prior art, the present invention has the following beneficial effects: (1) Significantly improve diagnostic efficiency. Through technologies such as automatic drawing parsing, intelligent knowledge retrieval, and AI-assisted diagnosis, the diagnosis time for a single drawing is reduced from several hours to less than 5 minutes, improving diagnostic efficiency by more than 60%, greatly reducing labor costs, and meeting the needs of rapid design iteration.

[0021] (2) Lowering the professional knowledge threshold. The constructed medical process knowledge graph covers more than 20 standards and norms, more than 30 hospital cases, and more than 5,000 professional entities. It systematizes and digitizes expert experience and normative knowledge, and non-medical process professionals can also use the system to make preliminary diagnoses, reducing the reliance on professionals.

[0022] (3) Reduce omissions of problems. The system conducts a comprehensive inspection from five dimensions: overall layout, departmental relationships, functional integrity, area allocation, and fire evacuation. The diagnostic accuracy rate is over 85%, which significantly reduces omissions of problems compared to manual inspection and avoids major design defects in the scheme stage from continuing into the construction stage.

[0023] (4) Achieve multi-standard collaboration. The knowledge graph integrates standards and regulations from multiple fields such as architecture, fire protection, and healthcare. It establishes connections between different standard provisions through a relationship graph, automatically identifies standard conflicts and collaboration requirements, and solves the problem of multi-standard collaboration.

[0024] (5) Support for continuous optimization. Through the closed-loop mechanism of "intelligent diagnosis - manual review - feedback learning", the system can continuously optimize the knowledge graph and diagnostic model based on feedback data in actual applications, and the diagnostic accuracy will gradually improve, forming a virtuous cycle of becoming smarter the more it is used.

[0025] (6) Provide traceability. Each diagnostic conclusion can be traced back to specific regulatory provisions and reference cases, which enhances the credibility and persuasiveness of the diagnostic results and makes it easier for designers to understand the cause of the problem and make targeted improvements. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the construction of the medical process knowledge graph for this invention. Figure 2 This is a system architecture diagram of the present invention. Detailed Implementation

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

[0028] like Figure 2 As shown, the present invention provides a medical process intelligent diagnosis system based on knowledge graph, including a drawing parsing module, a medical process knowledge graph, a knowledge retrieval module, an intelligent diagnosis module, a diagnosis report generation module, and a manual review module.

[0029] I. Construction of Medical Process Knowledge Graph like Figure 1 As shown, the construction of a medical process knowledge graph includes the following steps: (1) Data Collection: Collect relevant standard documents, standard drawings, and hospital case data related to medical building design. Specifically, this includes: ① National Standards: More than 20 currently effective standards, including the "Code for Design of General Hospital Buildings" GB51039-2014, the "Technical Code for Clean Operating Room Buildings in Hospitals" GB50333-2013, the "Code for Fire Protection Design of Buildings" GB50016-2014 (2018 Edition), and the "Code for Design of Infectious Disease Hospitals" GB50849-2014.

[0030] ② Local standards: More than 10 provincial-level local hospital building design standards from Guangxi, Guangdong, Zhejiang and other provinces.

[0031] ③Standard drawing sets: 15 sets of national architectural standard design drawings related to medical buildings.

[0032] ④ Hospital Cases: More than 30 hospital projects of different levels, including Grade A tertiary hospitals, Grade B tertiary hospitals, and Grade II hospitals, completed in the past 5 years, including project master plans, primary flowcharts, department setup lists, area statistics tables, and other materials.

[0033] ⑤ Technical literature: More than 200 academic papers, technical reports, design guidelines and other documents related to medical process design.

[0034] (2) Knowledge extraction: Natural language processing technology is used to extract knowledge from the collected documents.

[0035] ① Named Entity Recognition (NER): Employs a BERT-based named entity recognition model to automatically identify medical department entities (such as outpatient department, emergency department, operating room, ICU, radiology department, laboratory department, etc.), functional room entities (such as consultation room, treatment room, procedure room, waiting area, pharmacy, registration and payment counter, etc.), medical equipment entities (such as CT scanner, MRI machine, DSA, linear accelerator, etc.), and process node entities (such as registration, consultation, examination, medication dispensing, hospitalization, surgery, etc.) from standardized documents.

[0036] ② Relation Extraction: A relation extraction algorithm based on dependency parsing is used to identify relationships between entities. The main relation types include: Locational relationships: such as "the emergency department should be located near the radiology department", "the operating room should be located near the ICU", "the waste disposal outlet should be far away from the main entrance", etc. Process relationships: such as "patient registration → consultation → examination → medication", "emergency patients → triage → resuscitation room / observation room → hospitalization / discharge", etc. Regulatory constraints include: "Tertiary hospitals must set up infectious disease clinics", "The area of ​​outpatient departments should account for 25%-30% of the total building area", and "Evacuation stairwells must have natural ventilation and lighting", etc. Area ratio: such as "outpatient clinic area ≥ 10㎡", "operating room net height ≥ 2.8m", "ward corridor net width ≥ 2.4m", etc.

[0037] (3) Knowledge Graph Construction: The extracted entities and relationships are stored in the Neo4j graph database to form a medical process knowledge graph. The knowledge graph includes: ① Entity Library: Stores entity objects such as medical departments, functional rooms, medical equipment, and process nodes, with a total of over 5,000 entities. Each entity contains attribute information; for example, the attributes of a department entity include department name, department category, applicable hospital level, standard area range, and main function description.

[0038] ② Relationship Database: Stores locational relationships, process relationships, normative constraints, area allocation relationships, etc., between entities, with a total of over 10,000 relationships. Each relationship includes attributes such as relationship type, relationship strength, and normative basis.

[0039] ③ Standards Library: Stores national and local standards related to medical building design, including standard name, standard number, clause content, clause explanation, and scope of application. The total number of standard clauses exceeds 3000.

[0040] ④ Case Library: Stores medical process data for completed hospital projects, including basic project information (hospital name, level, number of beds, building area, etc.), overall layout, department setup, area allocation, process organization, etc. There are more than 30 cases in total.

[0041] (4) Knowledge vectorization: Pre-trained language models (such as BERT, RoBERTa, etc.) are used to convert text information in the knowledge graph into vector representations and store them in the Milvus vector database to support efficient semantic retrieval.

[0042] II. Drawing Analysis Module The drawing parsing module receives medical process flow drawings and supports formats such as PDF, CAD (DWG / DXF), and images (JPG / PNG). Drawing parsing includes the following steps: (1) Text Information Extraction: Optical Character Recognition (OCR) technology is used to extract text information from the drawings. In this embodiment, the PaddleOCR open-source framework is used to perform text recognition on the drawings and extract text information such as department names, room names, area labels, and process labels. The recognition accuracy rate reaches over 90%.

[0043] (2) Graphic Element Recognition: A deep learning-based target detection algorithm is used to identify graphic elements in the drawings. In this embodiment, the YOLOv8 target detection model is used. By training on 50 annotated drawings, the automatic recognition of graphic elements such as department area boundaries, functional room locations, workflow routes, and entrance / exit locations is achieved. The recognition accuracy rate reaches over 85%.

[0044] (3) Structured Data Conversion: Convert the recognition results into a structured data format (JSON format) and establish coordinate mapping relationships for drawing elements. Structured data includes: ① Department List: Information such as the name, location coordinates, area, and floor of each department; ② Room list: Information such as the name, location coordinates, area, and department of each functional room; ③ List of entrances and exits: Information such as the type of each entrance and exit (outpatient entrance, emergency entrance, visitor entrance, waste disposal exit, etc.), location coordinates, and orientation; ④ Process route: Information such as the path of each medical process, the departments and rooms connected; ⑤ Evacuation facilities: Information such as the location and number of evacuation staircases, elevators, and refuge rooms.

[0045] III. Knowledge Retrieval Module The knowledge retrieval module searches for relevant knowledge in the medical process knowledge graph based on the structured data output by the drawing parsing module. The retrieval process is as follows: (1) Query vector generation: Key information (such as hospital level, number of beds, department list, entrance and exit settings, etc.) in the structured data of the drawings are extracted into natural language descriptions and converted into query vectors using a pre-trained language model.

[0046] (2) Vector retrieval: Similarity retrieval is performed in the Milvus vector database to retrieve the top-50 knowledge items with the highest similarity to the query vector. The retrieval uses the cosine similarity calculation method, and the retrieval time is less than 1 second.

[0047] (3) Knowledge screening: The search results are screened a second time. Inapplicable knowledge items are filtered out based on hospital level, project scale, and other conditions. Finally, 20-30 highly relevant normative articles and 3-5 reference cases are selected.

[0048] (4) Knowledge organization: The selected standard provisions are classified and organized according to five dimensions: overall layout, departmental relationship, functional integrity, area ratio, and fire evacuation, so as to facilitate the use of the intelligent diagnosis module.

[0049] IV. Intelligent Diagnostic Module The intelligent diagnostic module employs a multimodal artificial intelligence model for diagnostic analysis. This embodiment can utilize large language models with image understanding and text generation capabilities, such as commercial models like GPT-4V, Claude3, and GeminiProVision, or open-source visual language models like Qwen-VL and CogVLM.

[0050] The intelligent diagnostic module includes five diagnostic units: (1) General layout diagnostic unit Inspection content: ① Location of the outpatient entrance: Is it close to the main urban road for the convenience of patients? Is it located on the main facade of the building for easy identification? Is it conveniently connected to the parking lot?

[0051] ② Emergency entrance location: Is it set up independently, separate from the outpatient entrance? Is it easily accessible by ambulances? Is it accessible 24 hours a day? Is it the shortest distance from the emergency department?

[0052] ③ Location of the inpatient visitation entrance: Is it set up separately from the outpatient and emergency entrance to reduce cross-interference? Is it convenient for visitors to reach the ward building?

[0053] ④ Location of waste disposal outlet and morgue exit: Is it far from major pedestrian traffic to avoid cross-infection? Is it located in a concealed location? Is it convenient for waste transport vehicles to enter and exit?

[0054] ⑤ Functional zoning: Are the zoning of clean areas (operating rooms, ICUs, delivery rooms, etc.), semi-clean areas (wards, clinics, etc.), and contaminated areas (sewage treatment, morgues, etc.) clearly defined? Are there clear separations and flow organization between each zone?

[0055] Diagnostic method: The location of entrances and exits and functional zoning information in the drawings are compared with the specifications in the knowledge graph. A large language model is used for semantic understanding and reasoning to determine whether it meets the specifications. For example, the model will analyze the "spatial distance between the emergency room entrance and the outpatient entrance" and the "relative positional relationship between the waste disposal outlet and the main entrance," and provide a judgment and reasoning on whether it is reasonable.

[0056] (2) Departmental Relationship Diagnosis Unit Inspection content: ①Outpatient pharmacy and emergency pharmacy: Can resources be shared, with one pharmacy serving two departments to improve resource utilization?

[0057] ② Emergency Department and Radiology / Laboratory Departments: Are they easily accessible to each other? Can emergency patients quickly reach the examination departments? Do the examination departments have dedicated emergency examination channels or dedicated emergency examination rooms?

[0058] ③ Are the operating room, ICU, and pathology department located close to each other? Can postoperative patients be quickly transferred to the ICU? Can surgical specimens be quickly sent to the pathology department?

[0059] ④ Delivery room and neonatal ward: Are they located close to each other? Can newborns be transferred quickly?

[0060] ⑤ Endoscopy Center and Sterilization Supply Center: Whether it facilitates rapid delivery of endoscopes after cleaning and sterilization.

[0061] Diagnostic method: An adjacency analysis algorithm is used to calculate the spatial distance (straight-line distance and walking distance) and connectivity (whether they are on the same floor, whether there is a convenient passage, etc.) between departments. Combined with the association requirements in the knowledge graph, the system determines whether the departmental location relationships are reasonable. For example, if the "Emergency Department" and "Radiology Department" are not on the same floor and are not connected by a dedicated elevator, the system will determine that the "departmental relationship is unreasonable" and provide improvement suggestions.

[0062] (3) Functional integrity diagnostic unit Inspection content: ① Completeness of Departmental Setup: Based on the hospital's level (Grade III Class A, Grade III Class B, Grade II, etc.), check whether all required medical departments are included. For example, a Grade III hospital must have: outpatient department, emergency department, inpatient department, operating room, ICU, radiology department, laboratory department, pathology department, pharmacy department, sterilization supply center, blood bank, nutrition kitchen, infectious disease clinic, etc.

[0063] ② Supporting facilities: Check whether the auxiliary rooms of the medical departments are fully equipped, such as whether the consultation room is equipped with a waiting area, toilet, doctor's office, nurse station, etc.

[0064] ③ Special function configuration: Check whether the special functions required by the standard are configured, such as refuge rooms, barrier-free facilities, and temporary medical waste storage rooms.

[0065] Diagnostic method: Compare the list of departments in the drawing with the department setting standards in the knowledge graph, and use set difference operation to find the missing departments. The system will list the "missing departments" and indicate the relevant standard basis.

[0066] (4) Area ratio diagnostic unit Inspection content: ① Area Proportion of Each Medical Functional Module: Check whether the area allocation of each functional module, such as outpatient department, inpatient department, medical technology department, and administrative and logistical departments, meets the standards. For example, the outpatient department area should generally account for 25%-30% of the total building area, the inpatient department area should account for 40%-50%, and the medical technology department area should account for 20%-25%.

[0067] ② Area of ​​individual departments: Check whether the area of ​​each department meets the standard requirements. For example, the area of ​​the inpatient department calculated based on the number of beds, and the area of ​​the outpatient department calculated based on the daily outpatient volume, etc.

[0068] ③ Functional room area: Check whether the area of ​​a single functional room meets the standard requirements. For example, the area of ​​a consultation room should be ≥10㎡, the area of ​​a ward should be ≥6㎡ / bed, and the area of ​​an operating room should be 25-50㎡ depending on the grade.

[0069] Diagnostic method: Based on the area data in the drawings, calculate the area and proportion of each functional module, department, and room, and compare it with the standard area indicators in the knowledge graph. For cases where the area exceeds or falls short of the standard, the system will provide the percentage deviation and improvement suggestions.

[0070] (5) Fire evacuation diagnostic unit Inspection content: ① Number of evacuation staircases: Based on the building area, floor height, and number of people, check whether the number of evacuation staircases meets the requirements of the "Code for Fire Protection Design of Buildings".

[0071] ② Location of evacuation staircases: Check whether the layout of the evacuation staircases is reasonable and whether the evacuation distance meets the requirements (usually ≤30m or ≤40m). Check whether the first-floor evacuation staircases can directly evacuate to the outside.

[0072] ③Natural ventilation and lighting in evacuation stairwells: Check whether the evacuation stairwells have natural ventilation and lighting conditions and whether they have external windows.

[0073] ④ Refuge room setup: Check whether the refuge room setup in ward buildings, operating rooms and other places meets the standards.

[0074] ⑤ Elevator lobby: Check whether the elevator (especially elevators that open into the corridor or hospital street) has a lobby and whether the lobby meets the fire separation requirements.

[0075] Diagnostic method: Based on the evacuation facility information in the drawings, combined with parameters such as building area, floors, and functions, the relevant clauses of the "Code for Fire Protection Design of Buildings" are checked. Because fire safety is of paramount importance, all fire evacuation diagnostic results, regardless of confidence level, are marked as "must be manually reviewed".

[0076] V. Diagnostic Report Generation Module The diagnostic report generation module automatically generates a structured diagnostic report based on the diagnostic results from the intelligent diagnostic module. The report includes the following parts: (1) Basic information of the project: hospital name, hospital level (Grade III Class A / Grade III Class B / Grade II Class A), number of beds, total building area, land area, number of building floors, etc.

[0077] (2) Diagnosis summary: total number of problems, number of serious problems (such as missing necessary departments, fire evacuation not in compliance with regulations, etc.), number of general problems (such as departmental relationships not being optimized, etc.), number of suggestions (such as optional optimization suggestions).

[0078] (3) Detailed list of issues: Issues are categorized and displayed according to five dimensions: overall layout, departmental relationships, functional integrity, area allocation, and fire evacuation. Each issue includes: ① Problem description: Describe the problem you found in natural language, such as "The emergency room entrance is too far from the main road in the city, which makes it inconvenient for patients to seek medical treatment".

[0079] ② Problem location marking: Mark the coordinates of the problem's location on the drawing to help designers quickly locate it.

[0080] ③ Violated regulations: List the specific regulations that are violated, such as "Article 4.1.3 of the Code for Design of General Hospital Buildings GB51039-2014: The emergency department shall have an independent entrance and exit and shall be convenient for emergency vehicles to park."

[0081] ④ Problem Severity Rating: Divided into three levels: "Serious" (red), "Moderate" (yellow), and "Recommended" (blue). Serious problems refer to those that violate mandatory regulations, affect the hospital's functionality or safety; moderate problems refer to those that are not optimized enough and affect ease of use; and recommended items refer to optional optimization measures.

[0082] ⑤ Improvement suggestions: Provide specific and feasible improvement measures for each problem. For example, for the problem of "the distance between the emergency department and the radiology department is too far", the improvement suggestion may be "to suggest that the radiology department be moved to the same floor or adjacent floor as the emergency department and a dedicated elevator be set up to connect them" or "to suggest that a dedicated emergency examination channel and dedicated emergency CT / DR equipment be set up in the radiology department".

[0083] (4) Drawing annotation: Mark the location of the problem on the original drawing, and use different colors to distinguish the severity of the problem (red for serious problems, yellow for general problems, and blue for suggestions). Generate annotated drawings for designers to view intuitively.

[0084] (5) Reference Cases: Provide excellent case studies similar in scale and level to the current project for reference. For example, for a tertiary hospital project with 800 beds, the system will recommend 3 excellent design cases of tertiary hospitals of similar scale, showing their overall layout, departmental organization, process design, etc.

[0085] VI. Manual Review Module The manual review module provides a visual web interface, allowing professionals to review, modify, and supplement diagnostic reports. Key functions include: (1) Problem Confirmation: Professional staff will review each diagnosed problem to confirm its accuracy. For misdiagnosed problems (false positives), they can be marked as "misdiagnosed" and the reason can be explained; for correct problems, they can be marked as "confirmed".

[0086] (2) Problem Supplement: If the system misses some problems (false negatives), professionals can manually add problems and fill in the problem description, location, standard basis, improvement suggestions and other information.

[0087] (3) Modification suggestions: Professionals can modify or supplement the improvement suggestions given by the system to make the suggestions more in line with the actual situation of the project.

[0088] (4) Feedback Submission: After the review is completed, the professionals submit the review results. The system uses the review data as new training samples to optimize the knowledge graph and diagnostic model.

[0089] (5) Continuous optimization: ① Correctly confirmed diagnoses are used as positive samples and added to the case library of the knowledge graph; ② The manually corrected misjudgments are used as negative samples to adjust the parameters and prompt word strategy of the diagnostic model, so as to avoid similar misjudgments from happening again; ③ Regularly (monthly) incrementally train the diagnostic model to improve its accuracy and generalization ability; ④ Add newly released standards and norms and newly completed hospital cases to the knowledge graph in a timely manner to keep the knowledge base up-to-date.

[0090] VII. System Deployment and Operation This system can be deployed in the following hardware and software environments: Hardware environment: Server: Configured with an Intel Xeon or AMD EPYC quad-core or higher CPU, 16GB or more of RAM, and 200GB or more of SSD storage space; Optional configuration with NVIDIA GPU accelerator cards (such as RTX4090, A100, etc.) to accelerate image recognition and model inference; Network environment: Supports intranet deployment (high data security) or cloud deployment (easy access for multiple users).

[0091] Software environment: Operating system: Linux (Ubuntu 20.04 or CentOS 8 recommended) or Windows Server 2019 / 2022; Databases: Neo4j 4.4+ (graph database), Milvus 2.2+ (vector database), PostgreSQL 13+ (relational database used to store project and user information); Development language: Python 3.8+; Main dependencies: PyTorch 2.0+, Transformers 4.30+, LangChain 0.1+, PaddleOCR 2.6+, OpenCV 4.7+, FastAPI 0.100+ (Web framework); Frontend: Vue.js 3+, ElementPlus (UI component library).

[0092] System operation process: (1) Users upload medical process flow diagrams (PDF / CAD / image format) through the Web interface; (2) The system calls the drawing parsing module to extract the text and graphic information in the drawing and convert it into structured data; (3) The system calls the knowledge retrieval module to search for relevant standards and cases in the medical process knowledge graph; (4) The system calls the intelligent diagnosis module to perform diagnostic analysis in five dimensions based on the large language model; (5) The system calls the diagnostic report generation module to generate a structured diagnostic report, including a text report and annotated drawings; (6) Users can view the diagnostic report through the Web interface and review, modify and supplement the report; (7) When users submit the review results, the system will use the feedback data for continuous optimization of the knowledge graph and model.

[0093] Example 2: Case Study on Medical Process Diagnosis in the Emergency and Outpatient Building of a Tertiary Hospital This embodiment uses the first-level flow chart of the medical process of the outpatient and emergency building of a planned tertiary general hospital (800 beds) in a certain city as an example to illustrate the application effect of the system of the present invention.

[0094] (1) Basic Project Information: Hospital Name: The Second People's Hospital of a Certain City (Grade III Class A General Hospital) Construction scale: 800 beds in total, with a building area of ​​28,000 square meters for outpatient and emergency services. Daily outpatient volume: Designed to accommodate 3,000 outpatient visits per day. Number of floors: 2 underground floors, 6 floors above ground (2) Drawing analysis results: The system automatically identified: 35 medical departments (18 outpatient departments, emergency department, radiology department, laboratory department, endoscopy center, etc.) 120 functional rooms (consultation rooms, treatment rooms, examination rooms, waiting areas, pharmacies, etc.) There are four main entrances / exits (outpatient main entrance, emergency entrance, inpatient visitation entrance, and logistics entrance / exit). 6 evacuation staircases, 4 passenger elevators, and 2 freight elevators Drawing analysis time: 2 minutes and 15 seconds.

[0095] (3) Knowledge retrieval results: The system retrieved: There are 58 relevant regulations (35 from the "Code for Design of General Hospital Buildings", 15 from the "Code for Fire Protection Design of Buildings", and 8 from other regulations). Three case studies were provided (800 beds in a municipal first people's hospital, 900 beds in a provincial people's hospital, and 750 beds in a university-affiliated hospital). Knowledge retrieval time: 0.8 seconds.

[0096] (4) Intelligent diagnostic results: The system detected the following issues: ① Site layout issues (2): Serious Problem: The emergency room entrance is too far from the main urban road (approximately 150 meters), which does not comply with Article 4.1.3 of the "Code for Design of General Hospital Buildings," which states that "the emergency department should have an independent entrance and exit, and should be convenient for ambulances to park." Recommendation for Improvement: Move the emergency room entrance to the east facade, closer to the main urban road.

[0097] General Issue: The waste disposal outlet is too close to the main entrance of the outpatient department (approximately 30 meters), which may pose a risk of cross-infection. Recommendation for Improvement: Move the waste disposal outlet to a concealed location on the north side of the building.

[0098] ② Departmental relationship issues (3): Serious Problem: The emergency pharmacy and outpatient pharmacy do not share resources, resulting in resource waste. Furthermore, the two pharmacies are far apart (approximately 80 meters), hindering unified management. Improvement Recommendation: Merge them into one pharmacy, with separate dispensing windows on both sides to serve both outpatient and emergency patients.

[0099] Common Problem: The emergency department is located on the first floor, and the radiology department is on the second floor, with no dedicated elevator connecting them. Emergency patients experience significant time delays in reaching the radiology department. Recommendation for Improvement: Move the radiology department to the first floor, adjacent to the emergency department, or install a dedicated emergency medical elevator.

[0100] General Issue: The operating room is located on the fifth floor, and the ICU is on the fourth floor. This misalignment between the two floors makes postoperative patient transfers inconvenient. Recommendation for Improvement: Move the ICU to the fifth floor, placing it on the same floor as the operating room.

[0101] ③ Functional integrity issue (1): Serious Problem: Lack of an infectious disease clinic. According to Article 4.1.6 of the "Code for Design of General Hospital Buildings," tertiary hospitals must have an infectious disease clinic. Recommendation for Improvement: Add an infectious disease clinic to a separate area on the first floor, including examination rooms, treatment rooms, observation rooms, restrooms, and other functional rooms, with a separate entrance and exit.

[0102] ④ Area ratio issues (2): General Issue: The outpatient department has an area of ​​6,500 square meters, accounting for 23.2% of the total building area, which is below the standard range (25%-30%). Recommendation for Improvement: Appropriately increase the area of ​​the outpatient consultation rooms and waiting area.

[0103] General Issue: The medical technology department occupies 7800㎡, accounting for 27.9% of the total building area, slightly exceeding the standard range (20%-25%). Improvement Suggestion: Optimize the layout of the medical technology departments and reduce the area of ​​auxiliary rooms.

[0104] ⑤ Fire evacuation issue (1 problem): Serious Problem: The evacuation staircase on the east side of the first floor cannot directly lead to the outside, with an evacuation distance of 8 meters, exceeding the standard requirement of ≤5 meters. Improvement Suggestion: Adjust the location of the evacuation staircase, or add an evacuation door between the evacuation staircase and the outside.

[0105] Diagnosis time: 2 minutes and 10 seconds.

[0106] (5) Generation of diagnostic reports: The system generates a diagnostic report, including: Issue Summary: A total of 9 issues were identified, including 4 serious issues and 5 minor issues; Detailed list of issues: Displayed categorized by five dimensions, with each issue including description, location, standard reference, and improvement suggestions; Mark the drawings: Mark the locations of 4 serious problems in red and the locations of 5 minor problems in yellow on the original drawings; Reference Cases: Showcasing three outstanding design examples from hospitals of similar size.

[0107] Report generation time: 0 minutes and 25 seconds.

[0108] Total time: 4 minutes and 50 seconds.

[0109] (6) Manual review: The project was reviewed by a senior medical process engineer from Hualu Group. The review results are as follows: Eight questions were confirmed to be accurate (accuracy rate 88.9%). One issue was judged as a misjudgment: "The area of ​​the medical technology department is slightly high." The review experts believed that the project had a lot of medical technology equipment and the area ratio was reasonable. One missing issue to add: "Insufficient number of seats in the waiting area". Based on a daily outpatient volume of 3,000 people, there should be about 150 seats in the waiting area, but only 100 seats were actually provided.

[0110] Accuracy rate after review: 8 / 9 = 88.9%, omission rate: 1.

[0111] (7) Application effect: Diagnostic efficiency: The diagnosis is completed in 4 minutes and 50 seconds, which is about 60 times more efficient than manual diagnosis (which usually takes 4-6 hours). Diagnostic accuracy: 88.9%, significantly reducing the number of problems missed; Design Optimization: Based on the diagnostic report, the design team made targeted optimizations to the solution, avoiding major changes in the subsequent construction drawing stage, saving about 2 weeks of design time and avoiding potential economic losses of about 500,000 yuan.

[0112] Example 3: Continuous System Optimization After six months of practical application, the system has diagnosed 25 hospital projects and collected 225 pieces of feedback data for manual review. Through continuous optimization: (1) Knowledge graph expansion: 150 new regulations have been added (including the newly released "Architectural Design Standard for Infectious Disease Hospitals" in 2024, etc.). Eight new hospital cases were added; 300 new entities and 500 new relationships were added.

[0113] (2) Diagnostic model optimization: Adjust the diagnostic prompt strategy based on feedback data to reduce false positives; Optimize the confidence assessment algorithm to improve the accuracy of identifying low-confidence problems.

[0114] (3) Performance improvement: The diagnostic accuracy rate increased from 85% initially to 92%; The false positive rate decreased from 15% to 8%; The omission rate decreased from an average of 1.5 issues per project to 0.8 issues.

[0115] The above embodiments demonstrate that the system of the present invention can significantly improve the diagnostic efficiency and accuracy of medical processes, and its performance can be continuously improved through optimization, demonstrating good practical value. Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A knowledge graph-based intelligent diagnostic system for medical processes, characterized in that, include: The drawing parsing module is used to receive medical process flow drawings, extract text information from the drawings using optical character recognition technology, identify department areas, functional rooms, and process routes in the drawings using object detection algorithms, and convert the recognition results into structured data. The medical process knowledge graph includes an entity library, a relationship library, a standard library, and a case library. The entity library stores, but is not limited to, the following entity objects: medical departments, functional rooms, and medical equipment. The relationship library stores the positional relationships, process relationships, standard constraint relationships, and area ratio relationships between entities. The standard library stores national and local standards related to medical building design. The case library stores medical process data of completed hospital projects. The knowledge retrieval module is used to retrieve relevant standard provisions and reference cases in the medical process knowledge graph based on the structured data output by the drawing parsing module. The knowledge retrieval module adopts vector retrieval technology to convert drawing information and knowledge base content into vector representations and realizes knowledge retrieval by calculating vector similarity. The intelligent diagnosis module is used to perform intelligent diagnosis on drawings based on retrieved knowledge. It includes a general layout diagnosis unit, a departmental relationship diagnosis unit, a functional integrity diagnosis unit, an area ratio diagnosis unit, and a fire evacuation diagnosis unit. The intelligent diagnosis module adopts a multimodal artificial intelligence model, takes the drawing information and retrieved knowledge as input, and generates diagnostic conclusions through semantic understanding and reasoning. The diagnostic report generation module is used to automatically generate a structured diagnostic report based on the diagnostic results of the intelligent diagnostic module. The diagnostic report includes a problem description, a location marker of the problem, the violated normative clauses, improvement suggestions, and a severity rating of the problem. The manual review module provides a visual interface that allows professionals to review, modify, and supplement diagnostic results, and feeds the reviewed results back to the medical process knowledge graph for continuous optimization.

2. The medical process intelligent diagnostic system based on knowledge graphs according to claim 1, characterized in that: The construction of the medical process knowledge graph includes the following steps: Collect relevant specifications, standard drawings, and medical process data from completed hospital projects related to medical building design. Named entity recognition algorithms are used to automatically extract entities such as medical departments, functional rooms, and medical equipment from documents; A relation extraction algorithm is used to identify positional relationships, process relationships, and normative constraint relationships between entities; The extracted entities and relationships are stored in a graph database to form a medical process knowledge graph.

3. The intelligent diagnostic system for medical processes based on knowledge graphs according to claim 1, characterized in that: The site layout diagnostic unit checks the following: Whether the locations of the outpatient entrance, emergency entrance, inpatient visitation entrance, and waste disposal exit comply with the principle of separating clean and dirty medical processes; Are the functional zones for clean areas, semi-clean areas, and contaminated areas clearly defined? Are the relationships between each entrance / exit and the city roads and surrounding buildings reasonable? 4. The intelligent diagnostic system for medical processes based on knowledge graphs according to claim 1, characterized in that: The departmental relationship diagnosis unit uses an adjacency analysis algorithm to calculate the spatial distance and connectivity between departments and determine the rationality of the layout of the following related departments: Feasibility of resource sharing between outpatient and emergency pharmacies; Convenient connections between the emergency department and the radiology and laboratory departments; The operating room is located adjacent to the ICU and pathology department; The connection between the delivery room and the neonatal department.

5. The intelligent diagnostic system for medical processes based on knowledge graphs according to claim 1, characterized in that: The fire evacuation diagnostic unit checks the following: Do the number and location of evacuation staircases meet the requirements of the regulations? Does the evacuation stairwell have natural ventilation and lighting? Whether the refuge rooms in the ward building and operating room are set up in accordance with regulations; Does the elevator lobby meet fire safety requirements? 6. The intelligent diagnostic system for medical processes based on knowledge graphs according to claim 1, characterized in that, Its diagnostic method includes the following steps: S1: Drawing parsing steps: Receive medical process flow drawings, extract text information using optical character recognition technology, identify department areas and functional room locations using target detection algorithms, and convert the recognition results into structured data; S2: Knowledge retrieval step, converting the structured data of the drawings into query vectors, retrieving the knowledge entries with the highest similarity in the vector database of the medical process knowledge graph, and filtering out relevant standard provisions and reference cases; S3: Intelligent diagnostic steps, construct diagnostic prompts, combine drawing information and retrieved knowledge as input to a multimodal artificial intelligence model, conduct diagnosis according to five dimensions: overall layout, departmental relationships, functional integrity, area ratio, and fire evacuation, and evaluate the confidence level of the diagnostic results; S4: Report generation steps: Summarize the problem list for each diagnostic dimension, mark the location coordinates, standard clauses, and severity level of each problem, generate improvement suggestions, and output a structured diagnostic report; S5: Manual review and feedback steps. Professionals review diagnostic reports through a visual interface, correct misjudged issues, supplement missing issues, and use the review results as new training samples to update the knowledge graph.

7. The medical process intelligent diagnostic system based on knowledge graphs according to claim 6, characterized in that: The confidence assessment method in S3 is as follows: Set the baseline confidence threshold to 0.7; When the confidence level of a diagnostic result is below 0.7, it is marked as requiring manual review. When the confidence level of a diagnostic result is below 0.5, it is marked as requiring manual review. All diagnostic results related to fire safety, regardless of their confidence level, are marked as requiring manual review.

8. The intelligent diagnostic system for medical processes based on knowledge graphs according to claim 6, characterized in that: The method for generating improvement suggestions in S4 is as follows: Search the case library for reference cases that are similar in size and level to the current project; Extract solutions to similar problems from reference cases; Based on the requirements of the regulations, generate targeted improvement suggestions.

9. The intelligent diagnostic system for medical processes based on knowledge graphs according to claim 6, characterized in that: The continuous optimization method in S5 includes: Correct diagnostic results confirmed by manual review will be used as positive samples; Manually corrected misjudgments are used as negative samples; Regularly perform incremental training on the diagnostic model; The newly added regulatory provisions and case data will be added to the knowledge graph.

10. The intelligent diagnostic system for medical processes based on knowledge graphs according to claim 6, characterized in that: The diagnostic report includes the following parts: Basic project information, including hospital level, number of beds, and building area; Diagnostic summary, including the total number of problems and statistics categorized by severity; The detailed list of issues is categorized and displayed according to five dimensions: overall layout, departmental relationships, functional integrity, area allocation, and fire evacuation. Mark the location of the problem on the original drawing, and use different colors to distinguish the severity of the problem; Improvement suggestions include specific improvement measures and reference cases for each problem.

Images