Doctor work task intelligent allocation method and device

Abstract

The invention relates to a doctor work task intelligent allocation method and device, and belongs to the technical field of medical informatization. The method comprises the following steps: extracting key medical terms in a physical examination report to be distributed, and constructing a report feature vector; calculating a standard workload integral for each physical examination report; calculating the similarity between the report feature vector and a professional skilled label set in the doctor portrait model, and generating a doctor-report matching degree score; calculating the current load value of each doctor in the candidate doctor set, and performing weighted calculation on the doctor-report matching degree score and the current load value to generate a comprehensive score; and selecting the doctor with the highest comprehensive score to execute task allocation. The problems of low professional matching degree and uneven workload in traditional physical examination report distribution are solved, and the medical resource utilization efficiency and the physical examination report processing quality are improved.

Description

Technical Field

[0002] This invention relates to the field of medical information technology, and in particular to a method and device for intelligent allocation of doctors' work tasks, which is applicable to the intelligent allocation and management of medical resources in large physical examination centers or hospital health management departments. Background Technology

[0004] Healthcare informatics is a crucial area of ​​modern hospital management, especially in large health checkup centers or hospital health management departments, where the level of intelligence in information systems directly impacts the efficiency and fairness of medical resource allocation. Intelligent task allocation, as an important component of healthcare informatics, is of great significance for improving the efficiency of medical resource utilization.

[0005] Currently, common task allocation techniques mainly include simple round-robin and queue-length-based load balancing. Simple round-robin allocates tasks in a fixed order, without considering task complexity or differences in executor capabilities; queue-length-based load balancing only focuses on the number of tasks, ignoring task quality and attribute differences, leading to uneven resource allocation.

[0006] Existing medical task allocation systems typically employ a single-dimensional mechanism, such as distributing tasks equally based solely on the number of examinees or reports. This mechanism fails to adequately consider the varying complexity of different examination types, individual differences among examinees, and differences in physicians' professional expertise. Consequently, some physicians may be overburdened with handling a large number of complex reports, while others are relegated to relatively simpler reports, resulting in wasted resources and inefficiency. Furthermore, these systems lack real-time monitoring and dynamic adjustment capabilities, making them unable to cope with sudden changes in traffic or personnel adjustments.

[0007] The main problems with existing technologies include: a crude allocation method that fails to differentiate task complexity; uneven workload distribution leading to inefficient use of medical resources; low professional matching, failing to assign specific types of tasks to the most skilled professionals; and a lack of dynamic adjustment mechanisms, making it unable to adapt to real-time changes in the work environment. These problems severely restrict the improvement of work efficiency and service quality in medical institutions. Summary of the Invention

[0009] The purpose of this invention is to provide a method and device for intelligent allocation of doctors' work tasks, aiming to solve the problems of low professional matching, uneven workload, and lack of dynamic adjustment mechanism in the existing medical task allocation process.

[0010] To achieve the above objectives, the present invention provides a method for intelligent allocation of physician work tasks, comprising the following steps:

[0011] Collect doctors' personal information and historical review report data, perform text mining and natural language processing on the historical review report data, extract doctors' professional expertise tags and calculate tag weight coefficients to generate a doctor profile model, wherein the doctor profile model includes doctors' basic information, a set of professional expertise tags, ability weight coefficients and a real-time work queue;

[0012] Obtain the physical examination report data to be assigned, identify the physical examination type, analyze the examinee's personal attributes, and extract key indicators of the report data. Perform comprehensive calculation according to the preset workload calculation formula to generate the standard workload score of the physical examination report.

[0013] Keyword extraction and standardization are performed on the physical examination report to generate a report feature vector. The similarity between the report feature vector and the set of professional expertise tags in the doctor profile model is calculated to generate a doctor-report matching score.

[0014] Based on the doctor-report matching score, a candidate doctor set is selected, the current workload value of each doctor in the candidate doctor set is calculated, the doctor-report matching score and the current workload value are weighted and calculated to generate a comprehensive score, and the doctor with the highest comprehensive score is selected to perform task assignment.

[0015] The system collects and processes actual time and quality evaluation data of physical examination reports, compares the standard workload integral predicted by the system with the actual processing time, performs deviation analysis and machine learning optimization, and updates the parameters of the workload calculation formula and the professional expertise label weights in the doctor profile model.

[0016] Optionally, text mining and natural language processing are performed on the historical review report data to extract doctors' professional expertise tags and calculate tag weight coefficients, including:

[0017] Based on the historical review report data, text mining technology is used to extract high-frequency disease keywords and physical examination types, and combined with the doctor's personal professional background information, an initial set of professional expertise tags is generated.

[0018] For each tag in the initial set of professional expertise tags, the doctor's historical accuracy, processing efficiency, and number of reports handled are statistically analyzed, and a weighted calculation is performed to generate the weight coefficient for each tag;

[0019] Based on the weight coefficients of each tag, tags with weight coefficients exceeding a preset threshold are selected to generate a set of professional expertise tags in the doctor profile model.

[0020] Optionally, a comprehensive calculation is performed according to a preset workload calculation formula to generate the standard workload integral of the physical examination report, including:

[0021] Based on the physical examination type in the physical examination report, query the physical examination type points mapping table to obtain the basic score for the physical examination type;

[0022] Based on the age group, gender, and marital status of the subjects, a correction factor calculation model is established to calculate the personnel attribute correction factor;

[0023] Natural language processing is performed on the departmental summaries and test results in the physical examination report to extract abnormal indicators and key medical terms, query the medical keyword-integration mapping table, and calculate the keyword integration value.

[0024] The standard workload score of the physical examination report is generated by summing the base score of the physical examination type, the personnel attribute correction factor, and the keyword score bonus.

[0025] Optionally, the similarity between the report feature vector and the set of professional expertise tags in the doctor profile model is calculated to generate a doctor-report matching score, including:

[0026] Obtain each keyword and its weight in the report feature vector, obtain each tag and its weight coefficient in the professional expertise tag set, and construct a feature-tag matching matrix;

[0027] The semantic similarity between the report keywords and the doctor's expertise tags in the feature-tag matching matrix is ​​calculated, and a weighted sum is generated by combining their respective weights to produce the doctor-report matching score.

[0028] Optionally, calculating the current workload value of each doctor in the candidate doctor set includes:

[0029] Based on the candidate doctor set, extract the real-time work queue data of each doctor, count the number of reports currently pending for each doctor and the standard workload points of each report, perform cumulative calculation, and generate the cumulative workload points of each doctor.

[0030] Obtain the capability weight coefficient of each doctor, divide the cumulative workload score by the capability weight coefficient, perform standardization processing, and generate the current workload value of each doctor in the candidate doctor set.

[0031] Optionally, text mining techniques are used to extract high-frequency disease keywords and physical examination types, and combined with the doctor's personal professional background information, an initial set of professional expertise tags is generated, including:

[0032] Based on the historical review report data, a doctor's personal corpus is constructed, a medical terminology dictionary and a common disease terminology database are established, and a basic knowledge base is generated.

[0033] Define a context window for medical keywords in the doctor's personal corpus, collect words within a preset range before and after the medical keywords, and generate contextual environment data for each medical keyword;

[0034] Frequent pattern mining is performed on the contextual environment data to identify combinations and co-occurrence patterns of medical terms, and the frequency of occurrence of each medical term in different contextual environments is statistically analyzed to generate term contextual frequency distribution data.

[0035] Based on the term context frequency distribution data, high-frequency, low-information terms that appear due to routine descriptions are removed, and the net frequency of each medical term in different context environments is calculated to generate term net frequency data.

[0036] Based on the net frequency data of the terms, professional specificity, and medical importance, the contribution weight of each medical term to the doctor's professional expertise is calculated, and terms with high contribution weights and their contextual pattern combinations are selected to generate the initial set of professional expertise tags.

[0037] Optionally, natural language processing is performed on the departmental summaries and laboratory test results in the physical examination report to extract abnormal indicators and key medical terms, including:

[0038] The physical examination report is segmented according to department and examination type, a structured text representation is established, and segmented report data is generated;

[0039] The sliding window technique is applied to the segmented data of the report to scan each medical term and its context, extract key medical terms and descriptive words within a preset range before and after them, and generate term context data.

[0040] Based on a pre-trained medical terminology pattern library, pattern matching is performed on the terminology context data to identify medical terminology combination patterns, extract abnormal indicators and their descriptive contexts, and generate an abnormal indicator context set.

[0041] For each indicator in the abnormal indicator context set, the influence of modifiers on the clinical meaning of the terminology is analyzed, mild, moderate and severe abnormalities are distinguished, and indicator severity classification data is generated.

[0042] Based on the medical knowledge base and the severity grading data of the indicators, the clinical importance score of each abnormal indicator is calculated, and common minor abnormalities and serious abnormalities requiring professional attention are distinguished, and the key medical terms and their importance scores are generated.

[0043] Optionally, the current workload value of each doctor in the candidate doctor set is calculated, and a comprehensive score is generated by weighting the doctor-report matching score and the current workload value, including:

[0044] Based on the candidate doctor set and the physical examination reports to be assigned, a doctor-report bipartite graph model is constructed, with doctors as the first type of nodes and physical examination reports as the second type of nodes. The doctor-report matching score is used as the edge weight to generate a workload graph.

[0045] For each doctor node in the workload graph, the number of assigned reports, the total report complexity, and the connectivity with the physical examination report node are counted. The load centrality of each doctor node is calculated, and load centrality data is generated.

[0046] Based on the load centrality data, the doctor node with the highest load centrality is identified, and the edge weights of the doctor node with the highest load centrality are adjusted to reduce its connection strength with non-critical reporting nodes, thereby generating an optimized workload graph.

[0047] The optimized workload graph is iteratively optimized by repeatedly identifying nodes with high load centrality and adjusting edge weights until the load centrality of each doctor node tends to be balanced, thus generating a balanced workload graph.

[0048] Based on the balanced workload map, the standardized workload index of each doctor is recalculated as the current workload value. The doctor-report matching score and the current workload value are weighted and calculated to generate the comprehensive score.

[0049] Optionally, based on the balanced workload map, the standardized workload index of each doctor is recalculated as the current workload value. The doctor-report matching score and the current workload value are weighted and calculated to generate the comprehensive score. The doctor with the highest comprehensive score is selected to perform task assignments, including:

[0050] Based on the doctor-report matching score, a candidate doctor set is established for each physical examination report to be assigned. According to the structural characteristics of the balanced workload diagram, a report assignment sequence is generated, and an initial assignment scheme is generated.

[0051] Constraints are identified in the initial allocation scheme, and the hard requirements that specific diseases must be handled by doctors with specific specialties are extracted to generate a set of edge constraints.

[0052] Based on the report allocation sequence and the set of edge constraints, the most suitable doctor is assigned to each physical examination report in sequence. After each allocation, the workload distribution is reassessed to generate allocation process data.

[0053] Based on the allocation process data, monitor the current load distribution, dynamically adjust the professional matching degree threshold and load balancing weight coefficient, optimize subsequent allocation decisions, and generate a dynamically adjusted allocation scheme.

[0054] The dynamically adjusted allocation scheme is verified to confirm that all constraints are met and the optimal load balancing state is achieved. The actual report allocation operation is then executed, and the work queues of each doctor are updated.

[0055] The present invention also provides a smart task allocation device for doctors, comprising:

[0056] The doctor profile building module is used to collect doctors' personal information, educational background, professional title and historical review report data, perform text mining and natural language processing on the historical review report data, extract doctors' professional expertise tags and calculate tag weight coefficients to generate doctor profile models;

[0057] The report workload assessment module is used to acquire the physical examination report data to be assigned, identify the physical examination type, analyze the examinee's personal attributes, and extract key report indicators from the physical examination report data, and perform comprehensive calculations according to the preset workload calculation formula to generate the standard workload score of the physical examination report.

[0058] The professional matching calculation module is used to extract and standardize keywords from the physical examination report, generate a report feature vector, calculate the similarity between the report feature vector and the set of professional expertise tags in the doctor profile model, and generate a doctor-report matching score.

[0059] The task allocation decision module is used to filter a set of candidate doctors based on the doctor-report matching score, calculate the current workload value of each doctor in the candidate doctor set, perform a weighted calculation on the doctor-report matching score and the current workload value to generate a comprehensive score, and select the doctor with the highest comprehensive score to perform task allocation;

[0060] The dynamic optimization module is used to collect actual time and quality evaluation data for processing physical examination reports, compare the standard workload integral predicted by the system with the actual processing time, perform deviation analysis and machine learning optimization, and update the parameters of the workload calculation formula and the professional expertise label weights in the doctor profile model.

[0061] The present invention has the following beneficial effects:

[0062] (1) A multi-dimensional doctor profile model was constructed, including basic information, professional expertise tags, ability weights and real-time work queues, which breaks through the limitations of traditional simple polling or random allocation.

[0063] (2) A refined report workload quantification mechanism was proposed. Through multi-dimensional evaluation of physical examination type, personnel attributes and key report indicators, standard workload points were calculated for each physical examination report, which solved the problem of uneven workload caused by the large difference in difficulty of different reports.

[0064] (3) An intelligent allocation algorithm was designed to simultaneously take into account both professional matching degree and load balancing, and to achieve dynamic balance between the two through adjustable weight coefficients.

[0065] (4) An adaptive optimization closed-loop system was established. Based on actual review efficiency data, the workload calculation model and doctor professional labels were continuously optimized to form a self-learning and continuous optimization closed-loop system. Attached Figure Description

[0067] Figure 1 Flowchart of the assignment method for the intelligent director system;

[0068] Figure 2 A flowchart of a method for intelligently assigning tasks to doctors;

[0069] Figure 3 A schematic diagram illustrating the construction of a doctor's profile model;

[0070] Figure 4 This is a block diagram of the overall architecture of the system of the present invention. Detailed Implementation

[0072] 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. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0073] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0074] like Figure 1 and 2 As shown, this invention provides a method for intelligent allocation of doctor's work tasks, including the following steps:

[0075] S1: Collect doctors' personal information and historical review report data, perform text mining and natural language processing on the historical review report data, extract doctors' professional expertise tags and calculate tag weight coefficients to generate a doctor profile model, wherein the doctor profile model includes doctors' basic information, a set of professional expertise tags, ability weight coefficients and a real-time work queue.

[0076] Doctor portrait model such as Figure 3 As shown.

[0077] S2: Obtain the physical examination report data to be assigned, identify the physical examination type, analyze the examinee's personal attributes, and extract key indicators of the report data, and perform comprehensive calculations according to the preset workload calculation formula to generate the standard workload score of the physical examination report;

[0078] S3: Extract keywords from the physical examination report and standardize them to generate a report feature vector. Calculate the similarity between the report feature vector and the set of professional expertise tags in the doctor profile model to generate a doctor-report matching score.

[0079] S4: Based on the doctor-report matching score, filter the candidate doctor set, calculate the current workload value of each doctor in the candidate doctor set, perform a weighted calculation on the doctor-report matching score and the current workload value, generate a comprehensive score, and select the doctor with the highest comprehensive score to perform task assignment;

[0080] S5: Collect actual time and quality evaluation data for processing physical examination reports, compare the system's predicted standard workload integral with the actual processing time, perform deviation analysis and machine learning optimization, and update the parameters of the workload calculation formula and the professional expertise label weights in the doctor profile model.

[0081] Step S1 involves the construction of the doctor profile model, which is the foundation of the entire task allocation system. The system first collects doctors' personal information, including basic data such as name, title, professional background, and work experience, as well as historical medical examination reports reviewed by the doctors. Using advanced text mining and natural language processing techniques, the system extracts keywords and terms reflecting the doctors' professional expertise from these historical report data. The system not only focuses on the frequency of individual keywords but also analyzes their contextual environment and semantic relationships to ensure that the extracted professional tags have practical clinical significance. For example, the system can identify that a doctor demonstrates high efficiency and accuracy in processing reports related to "pulmonary nodules," and thus designates "pulmonary nodules" as that doctor's professional expertise tag. Simultaneously, the system calculates a weight coefficient for each expertise tag, comprehensively evaluating the doctor's historical accuracy, efficiency, and quantity in handling related reports. Finally, the system generates a comprehensive doctor profile model, including the doctor's basic information, a set of professional expertise tags, ability weight coefficients, and real-time work queue status, providing data support for subsequent task matching.

[0082] Step S2 is responsible for quantifying the workload of the medical examination reports to be assigned. After acquiring the data of the medical examination reports to be assigned, the system first identifies the type of medical examination, such as pre-employment medical examination, general health check-up, or cadre medical examination, etc. Different types of medical examinations have different basic workloads. Next, the system analyzes the personnel attributes of the examinees, including demographic characteristics such as age group, gender, and marital status. These factors affect the complexity of the report and the focus of attention. Subsequently, the system performs natural language processing on the departmental summaries and test results in the medical examination report to extract abnormal indicators and key medical terms. For example, findings such as "pulmonary nodules," "elevated blood sugar," or "abnormal electrocardiogram" will increase the complexity of the report and the difficulty of review. According to the preset workload calculation formula, the system comprehensively calculates the basic score of the medical examination type, the personnel attribute correction factor, and the keyword score bonus, and finally generates the standard workload score for the medical examination report. This multi-dimensional workload assessment mechanism ensures that the system can accurately reflect the actual workload of different reports, laying the foundation for fair task allocation.

[0083] Step S3 intelligently matches medical examination reports with doctors' professional expertise. The system performs in-depth analysis of the medical examination reports to be assigned, extracting key medical terms and abnormal indicators, and standardizing them to transform them into structured report feature vectors. These feature vectors contain the core medical content and clinical significance of the reports. Subsequently, the system calculates the semantic similarity between these report feature vectors and the set of professional expertise tags in each doctor's profile model. This calculation process considers not only precise keyword matching but also semantic relevance; for example, "myocardial ischemia" and "coronary heart disease," although different in wording, are highly related in medicine. The system considers the weight coefficients of the tags, focusing on the degree of matching between the doctor's area of ​​expertise and the report content. Through this multi-dimensional similarity calculation, the system ultimately generates a matching score for each doctor and the reports to be assigned, quantifying the doctor's professional suitability for handling such reports. This step ensures that reports are assigned to doctors with the most relevant professional knowledge and experience, improving the quality and efficiency of the review process.

[0084] Step S4 is responsible for the final task allocation decision, seeking an optimal balance between professional matching and load balancing. The system first filters doctors with matching scores exceeding a preset threshold based on the doctor-report matching score calculated in step S3, forming a candidate doctor set. For each doctor in this candidate set, the system calculates their current workload, i.e., the total number and complexity of reports being processed. The system considers not only the number of reports but also the standard workload score for each report, ensuring the accuracy of workload assessment. Simultaneously, the system considers the doctor's ability weighting coefficient, standardizing the workload value to allow for fair comparison among doctors with different qualifications and abilities. After obtaining the doctor-report matching score and the doctor's current workload value, the system generates a final comprehensive score using an adjustable weighted calculation model. This comprehensive score reflects both professional matching and workload balance, and the system selects the doctor with the highest comprehensive score as the final task allocation recipient. This decision-making mechanism ensures professional matching while avoiding excessive workloads for some doctors, achieving optimal resource allocation.

[0085] Step S5 establishes a closed-loop system for dynamic monitoring and adaptive optimization. After task allocation and execution, the system continuously collects data on the actual time spent by doctors processing physical examination reports, their operational trajectories, and quality evaluation results. The system compares and analyzes this actual data with the predicted standard workload integral from step S2 to identify deviations and inaccuracies in the workload calculation model. For example, the system may find that the actual processing time for a certain type of abnormal indicator is significantly longer than expected, or that a doctor's processing efficiency for a specific type of report is significantly higher than average. Based on these deviation analysis results, the system applies machine learning algorithms to continuously optimize and adjust the parameters of the workload calculation formula, making the predictions more accurate. Simultaneously, the system updates the professional expertise labels and their weights in the doctor profile model, ensuring that the model dynamically reflects changes and developments in doctors' professional abilities. This adaptive optimization mechanism creates a closed loop of data collection, analysis, adjustment, and verification. As usage time increases, the system's intelligence level continuously improves, and the accuracy and fairness of task allocation are also continuously enhanced.

[0086] In a preferred embodiment of the present invention, text mining and natural language processing are performed on the historical review report data to extract doctors' professional expertise tags and calculate tag weight coefficients, including:

[0087] S1.1: Based on the historical review report data, text mining technology is used to extract high-frequency disease keywords and physical examination types, and combined with the doctor's personal professional background information, an initial set of professional expertise tags is generated;

[0088] S1.2: For each tag in the initial set of professional expertise tags, calculate the doctor's historical accuracy, processing efficiency, and number of reports handled, perform weighted calculations, and generate weight coefficients for each tag;

[0089] S1.3: Based on the weight coefficients of each label, select labels whose weight coefficients exceed a preset threshold to generate a set of professional expertise labels in the doctor profile model.

[0090] Steps S1.1 to S1.3 describe in detail how the system extracts professional expertise tags from doctors' historical review report data and quantifies their weights.

[0091] In S1.1, the system first analyzes a database of medical examination reports reviewed by doctors, using text mining technology to identify frequently occurring disease keywords and examination types. This process not only focuses on word frequency but also considers the clinical value and importance of the terms. The system then combines this with the doctor's professional background information, such as education, professional training, and professional title, matching the mined keywords with the doctor's basic professional area to generate a preliminary candidate set of professional expertise tags. This approach, which incorporates professional background information, ensures the clinical relevance of the tags and avoids potential biases that may arise from relying solely on statistical data.

[0092] Step S1.2 further refines the evaluation of each candidate tag. The system not only focuses on the frequency of tag occurrence but, more importantly, analyzes the performance indicators of physicians when handling reports related to that tag. These indicators mainly include three aspects: historical accuracy, i.e., the accuracy of physicians' diagnoses of this type of report, which can be obtained through quality control feedback or peer review; processing efficiency, i.e., the average time required for physicians to complete this type of report; and the number of reports processed, reflecting the physician's experience in handling such reports. The system performs a weighted calculation on these three indicators to generate a comprehensive weight coefficient for each tag. This step ensures that the tag weight not only reflects the frequency of occurrence but also embodies the physician's actual professional level in the field.

[0093] Step S1.3 is the final label selection stage. Based on the weight coefficients calculated in the previous step, the system sets an appropriate threshold and retains only labels with weights exceeding this threshold as the doctor's core professional expertise. This selection mechanism ensures that the doctor profile model only includes the professional fields that the doctor is truly proficient in, improving the accuracy of subsequent task matching. Through these three steps, the system ultimately generates a precise set of doctor professional expertise labels based on empirical data, providing core professional competence data for the doctor profile model.

[0094] In one specific embodiment of the present invention, text mining technology is used to extract high-frequency disease keywords and physical examination types, and combined with the doctor's personal professional background information, an initial set of professional expertise tags is generated, including:

[0095] S1.1.1: Based on the aforementioned historical review report data, construct a doctor's personal corpus, establish a medical terminology dictionary and a common disease terminology database, and generate a basic knowledge base;

[0096] S1.1.2: Define a context window for medical keywords in the doctor's personal corpus, collect words within a preset range before and after the medical keywords, and generate contextual environment data for each medical keyword;

[0097] S1.1.3: Perform frequent pattern mining on the contextual environment data to identify combinations and co-occurrence patterns of medical terms, count the frequency of occurrence of each medical term in different contextual environments, and generate term contextual frequency distribution data.

[0098] S1.1.4: Based on the term context frequency distribution data, high-frequency, low-information terms that appear due to routine descriptions are removed, and the net frequency of each medical term in different context environments is calculated to generate term net frequency data.

[0099] S1.1.5: Based on the net frequency data of the terms, professional specificity and medical importance, calculate the contribution weight of each medical term to the doctor's professional expertise, screen the terms with high contribution weight and their contextual pattern combinations, and generate the initial set of professional expertise labels.

[0100] Steps S1.1.1 to S1.1.5 are a deeper expansion of S1.1, detailing the technical implementation of the text mining process. In S1.1.1, the system first constructs a doctor's personal corpus, which is a collection of all report texts that the doctor has reviewed throughout history. Simultaneously, the system establishes a medical terminology dictionary and a common disease terminology database as a foundational knowledge base. These knowledge bases contain standard medical terminology, descriptions of common diseases, and professional abbreviations, providing reference standards and semantic support for subsequent text analysis.

[0101] Step S1.1.2 introduces the concept of a context window, a crucial technique in natural language processing for understanding word semantics. The system defines a context window for each medical keyword, typically consisting of a fixed number of words preceding and following the keyword, or a complete sentence. By collecting these contextual words, the system can capture the keyword's usage environment and semantic information. For example, the word "nodule" may appear the same in "small lung nodules" and "thyroid nodules," but their clinical meanings and corresponding specialties differ; contextual information helps the system correctly distinguish these differences.

[0102] Step S1.1.3 performs frequent pattern mining on the collected contextual data to identify combinations and co-occurrence patterns of medical terms. The system not only counts the frequency of individual terms but also focuses on the co-occurrence relationships of term combinations. For example, the combination of "pulmonary nodule" and "CT enhancement" may indicate a physician's expertise in pulmonary imaging diagnosis. The term contextual frequency distribution data generated by the system reflects the physician's language patterns and professional preferences when dealing with different types of medical problems.

[0103] Step S1.1.4 is a crucial data cleansing step. Medical reports contain a large number of high-frequency but low-information terms, such as "within the normal range" and "no obvious abnormalities." Although these terms appear frequently, they do not reflect the doctor's professional expertise. The system identifies and removes these low-information terms by analyzing the distribution patterns of terms in different contexts, and calculates the "net frequency" of each medical term, that is, the actual frequency of influence after excluding routine descriptions, thereby generating more clinically meaningful net frequency data for terms.

[0104] Step S1.1.5 is the final step in tag generation. The system comprehensively considers three dimensions: net frequency of terms, specialty specificity, and medical importance. Specialty specificity refers to the distinguishing power of a term for a specific medical specialty, such as "ST segment elevation" being highly specific to cardiovascular diseases. Medical importance considers the weight of a term in clinical decision-making; for example, terms related to malignant tumors are usually more important than those related to minor abnormalities. Based on these three dimensions, the system calculates the contribution weight of each term to the physician's specialty, and finally selects terms with high contribution weights and their contextual pattern combinations to form an initial set of specialty expertise tags. This set includes not only single terms but also meaningful combinations of terms, providing a rich and multidimensional feature representation for the physician's professional profile.

[0105] like Figure 4 As shown, in another preferred embodiment of the present invention, a comprehensive calculation is performed according to a preset workload calculation formula to generate the standard workload integral of the physical examination report, including:

[0106] S2.1: Based on the physical examination type in the physical examination report, query the physical examination type points mapping table to obtain the basic score for the physical examination type;

[0107] S2.2: Based on the age group, gender, and marital status of the examinee, establish a correction factor calculation model to calculate the personnel attribute correction factor;

[0108] S2.3: Perform natural language processing on the departmental summaries and test results in the physical examination report, extract abnormal indicators and key medical terms, query the medical keyword-integration mapping table, and calculate the keyword integration value.

[0109] S2.4: Sum the base score of the physical examination type, the personnel attribute correction factor, and the keyword score bonus to generate the standard workload score of the physical examination report.

[0110] Steps S2.1 to S2.4 detail how the system calculates the standard workload points for each medical examination report, which is a crucial foundation for achieving fair task allocation. In S2.1, the system first identifies the type of medical examination report, such as routine medical examination, pre-employment medical examination, cadre medical examination, or special medical examination. Different types of medical examination reports differ significantly in content complexity and review requirements. For example, cadre medical examinations typically include more examination items and require more rigorous risk assessments; while special medical examinations conduct in-depth examinations for specific health issues, requiring professional judgment in the relevant field. The system assigns a basic score to each type of medical examination by querying a pre-established medical examination type point mapping table. This score reflects the basic workload level of that type of report.

[0111] Step S2.2 considers the impact of the examinee's personal attributes on the report complexity. The system establishes a correction factor calculation model based on demographic characteristics such as the examinee's age group, gender, and marital status. Age is a particularly important factor because different age groups have different health risk characteristics and examination priorities. For example, older adults often have more chronic diseases and abnormal indicators that need to be assessed; while women of childbearing age need to pay special attention to reproductive health-related indicators. The system calculates a correction factor based on these attributes to adjust the base workload score, making it more accurately reflect the actual workload required to process the report.

[0112] Step S2.3 is the most complex and crucial part of the workload calculation. The system performs in-depth natural language processing on the departmental summaries and test results in the physical examination report, identifying and extracting abnormal indicators and key medical terms. These abnormal indicators and key terms are the core factors determining the complexity of the report, as they usually require professional judgment, risk assessment, and health guidance from doctors. For example, findings such as "pulmonary nodules," "abnormal liver function," or "ST segment changes on electrocardiogram" will increase the complexity of report review. The system queries a pre-set medical keyword-integration mapping table, assigning corresponding integral values ​​to each identified key medical term. These integral values ​​comprehensively consider the clinical importance, degree of abnormality, and processing difficulty of the term, ultimately calculating the keyword integral value.

[0113] Step S2.4 is the final calculation step. The system comprehensively sums the basic score for the physical examination type, the personnel attribute correction factor, and the keyword score bonus obtained from the first three steps to generate the standard workload score for the physical examination report. This score is a quantitative indicator that integrates multiple factors, accurately reflects the actual workload required to process the report, and provides an objective basis for subsequent fair task allocation.

[0114] In one specific embodiment of the present invention, natural language processing is performed on the departmental summaries and laboratory test results in the physical examination report to extract abnormal indicators and key medical terms, including:

[0115] S2.3.1: The physical examination report is segmented according to department and examination type, a structured text representation is established, and the report segment data is generated;

[0116] S2.3.2: Apply sliding window technology to the segmented data of the report, scan each medical term and its context, extract key medical terms and descriptive words within a preset range before and after them, and generate term context data;

[0117] S2.3.3: Based on a pre-trained medical terminology pattern library, perform pattern matching on the terminology context data, identify medical terminology combination patterns, extract abnormal indicators and their descriptive contexts, and generate an abnormal indicator context set.

[0118] S2.3.4: For each indicator in the abnormal indicator context set, analyze the influence of modifiers on the clinical meaning of the terminology, distinguish between mild, moderate and severe abnormalities, and generate indicator severity grading data;

[0119] S2.3.5: Based on the medical knowledge base and the severity grading data of the indicators, calculate the clinical importance score of each abnormal indicator, distinguish between common minor abnormalities and serious abnormalities that require professional attention, and generate the key medical terms and their importance scores.

[0120] Steps S2.3.1 to S2.3.5 demonstrate in detail how the system extracts abnormal indicators and key medical terms from physical examination reports using natural language processing technology. In S2.3.1, the system first segments the entire physical examination report by department and examination type, such as imaging examinations, laboratory tests, and internal medicine examinations, and establishes a structured text representation. This segmentation process facilitates subsequent analysis using appropriate professional knowledge for different types of medical information, improving the accuracy of terminology extraction.

[0121] Step S2.3.2 introduces the sliding window technique, a commonly used text analysis method in natural language processing. The system applies this technique to segmented report data, using a fixed-size "window" that moves progressively across the text to scan each medical term and its surrounding context. This not only extracts individual medical terms but also captures descriptive words before and after them, such as modifiers like "mild," "obvious," and "requires further follow-up," which are crucial for understanding the clinical meaning of the terms. Through this process, the system generates detailed terminology context data, laying the foundation for subsequent semantic analysis.

[0122] Step S2.3.3 utilizes a pre-trained medical terminology pattern library to perform pattern matching on the terminology context data. This pattern library contains common medical terminology combination patterns and expressions, which can help the system accurately identify professional terms and abnormal descriptions in physical examination reports. Through this pattern matching, the system can effectively extract abnormal indicators and their descriptive contexts, such as "mild fatty infiltration of the liver" and "scattered small nodular shadows seen in both lungs, approximately 5 mm in diameter." These extracted abnormal indicators and their contexts form an abnormal indicator context set, providing detailed information for subsequent severity assessment.

[0123] Step S2.3.4 classifies the severity of the extracted abnormal indicators. The system analyzes the modifiers in the context of each abnormal indicator, assessing the impact of these modifiers on the clinical meaning of the terminology. For example, modifiers such as "mild" and "slightly" usually indicate lower severity, while "obvious," "significant," and "severe" indicate higher severity. The system also considers the degree of deviation of numerical indicators from the normal range, such as blood pressure, blood glucose, and liver function indicators. Through this analysis, the system can classify abnormal indicators into three severity levels: mild, moderate, and severe, forming indicator severity classification data.

[0124] Step S2.3.5 is the final critical medical terminology assessment stage. Based on a professional medical knowledge base and the severity grading data obtained in the previous step, the system calculates a clinical importance score for each abnormal indicator. This score comprehensively considers the severity, clinical significance, and potential health risks of the indicator. For example, even "mild" pulmonary nodules may have high clinical importance and require careful evaluation and follow-up; while some "significant" mild abnormalities in laboratory indicators may have limited clinical significance. Through this assessment, the system can distinguish between common minor abnormalities and serious abnormalities requiring professional attention, ultimately generating critical medical terms and their importance scores, providing an accurate basis for workload point calculation.

[0125] In a preferred embodiment of the present invention, the similarity between the report feature vector and the set of professional expertise tags in the doctor profile model is calculated to generate a doctor-report matching score, including:

[0126] S3.1: Obtain each keyword and its weight in the feature vector of the report, obtain each tag and its weight coefficient in the set of professional expertise tags, and construct a feature-tag matching matrix;

[0127] S3.2: Calculate the semantic similarity between the report keywords and the doctor's expertise tags in the feature-tag matching matrix, combine their respective weights, and perform a weighted sum to generate the doctor-report matching score.

[0128] Steps S3.1 and S3.2 detail how the system intelligently matches medical examination reports with doctors' professional expertise, which is the core mechanism for ensuring that reports are assigned to relevant specialties. In S3.1, the system first acquires two sets of key data: one is the report feature vector extracted from the medical examination report, containing various key medical terms and their weights; the other is the set of professional expertise tags obtained from the doctor's profile model, containing various areas of expertise and their weight coefficients. Based on these two sets of data, the system constructs a feature-tag matching matrix, a two-dimensional table structure. The horizontal axis represents keywords in the report, and the vertical axis represents the doctor's professional expertise tags. Each cell in the matrix records the semantic similarity between the corresponding keyword and tag. This matrix structure enables the system to systematically evaluate the multi-dimensional matching relationship between the report content and the doctor's professional expertise, rather than simply keyword matching.

[0129] Step S3.2 calculates the similarity of each cell in the matrix and generates the final matching score. The system calculates the semantic similarity of each pair of report keywords and doctor's expertise tags in the feature-tag matching matrix. This calculation is not limited to exact matching of completely identical words, but can identify semantic associations between medical terms. For example, "coronary heart disease" and "myocardial ischemia" are different terms, but are highly related in medicine; similarly, "type 2 diabetes" and "poor blood sugar control" also have a significant semantic association. The system can accurately calculate the semantic similarity between these terms through a pre-trained medical domain semantic model. After obtaining the similarity of each pair of keywords and tags, the system performs a weighted summation based on their respective weights. This means that more important anomalies in the report and the doctor's most proficient areas of expertise have a higher influence in the matching calculation. Finally, the system generates a comprehensive doctor-report matching score, which quantifies the doctor's professional suitability in handling this report. This intelligent matching mechanism based on semantic similarity far surpasses traditional keyword matching methods, and can more accurately assign reports to doctors with the most relevant professional knowledge and experience.

[0130] In a preferred embodiment of the present invention, calculating the current workload value of each doctor in the candidate doctor set includes:

[0131] S4.1: Based on the candidate doctor set, extract the real-time work queue data of each doctor, count the number of reports to be processed by each doctor and the standard workload points of each report, perform cumulative calculation, and generate the cumulative workload points of each doctor.

[0132] S4.2: Obtain the capability weight coefficient of each doctor, divide the cumulative workload integral by the capability weight coefficient, perform standardization processing, and generate the current workload value of each doctor in the candidate doctor set.

[0133] Steps S4.1 and S4.2 detail how the system calculates a doctor's current workload, a crucial step in achieving fair task allocation. In S4.1, the system first acquires real-time work queue data for each doctor in the candidate doctor set. This data reflects all medical examination reports that each doctor is currently processing or awaiting processing. The system not only counts the number of reports in each doctor's queue but, more importantly, considers the actual workload of each report. The system extracts the standard workload integral for each report (these integrals are calculated in step S2) and then sums the standard workload integrals of all reports in each doctor's queue to obtain the cumulative workload integral for each doctor. This calculation method, based on integrals rather than simply the number of reports, more accurately reflects the doctor's actual workload because the processing time and effort required for reports of different complexities vary greatly.

[0134] Step S4.2 further considers the impact of individual physician ability differences on workload assessment. The system obtains the ability weight coefficient for each physician, which reflects the physician's work efficiency and ability to handle complex reports. For example, senior physicians typically have higher efficiency and processing capabilities than junior physicians. The system divides the cumulative workload score calculated in the previous step by the physician's ability weight coefficient, performs standardization processing, and generates the final current workload value. This standardization process ensures that the workload can be fairly compared among physicians with different qualifications and abilities. For example, for the same workload of 100 points, the standardized workload value of a senior physician with an ability weight of 1.5 will be 66.7, while the standardized workload value of a junior physician with an ability weight of 0.8 will be 125, reflecting that the junior physician has a heavier actual burden when facing the same workload. By comprehensively considering workload scores and individual abilities, the system can accurately calculate the actual workload of each physician, providing an objective basis for subsequent task allocation decisions and ensuring that workload is fairly allocated while taking into account ability differences.

[0135] In a preferred embodiment of the present invention, the current workload value of each doctor in the candidate doctor set is calculated, and a weighted calculation is performed on the doctor-report matching score and the current workload value to generate a comprehensive score, including:

[0136] S4.A: Based on the candidate doctor set and the physical examination reports to be assigned, construct a doctor-report bipartite graph model, with doctors as the first type of nodes and physical examination reports as the second type of nodes, and use the doctor-report matching score as the edge weight to generate a workload graph;

[0137] S4.B: For each doctor node in the workload graph, count the number of assigned reports, the total report complexity, and the connectivity with the physical examination report node; calculate the load centrality of each doctor node; and generate load centrality data.

[0138] S4.C: Based on the load centrality data, identify the doctor node with the highest load centrality, adjust the edge weights of the doctor node with the highest load centrality to reduce its connection strength with non-critical reporting nodes, and generate an optimized workload graph.

[0139] S4.D: Iteratively optimize the optimized workload graph by repeatedly identifying nodes with high load centrality and adjusting edge weights until the load centrality of each doctor node tends to be balanced, thus generating a balanced workload graph.

[0140] S4.E: Based on the balanced workload map, recalculate the standardized workload index of each doctor as the current workload value, and perform a weighted calculation on the doctor-report matching score and the current workload value to generate the comprehensive score.

[0141] Steps S4.A to S4.E detail how the system achieves dynamic balancing and optimal task allocation of doctors' workload through a graph theory model. In S4.A, the system introduces the advanced concept of a bipartite graph model, a special graph structure consisting of two disjoint sets of nodes, with edges connecting only nodes of different types. The system treats doctors as the first type of node and medical examination reports as the second type of node, using the previously calculated doctor-report matching score as the weight of the edges connecting them, thus constructing a complete workload graph. This graph structure can intuitively represent all possible doctor-report allocation relationships and their suitability, providing a mathematical foundation for subsequent load balancing calculations.

[0142] Step S4.B introduces the concept of centrality from network analysis to assess the workload distribution of physicians. The system performs multi-dimensional analysis on each physician node in the workload graph: it counts the number of reports assigned to each physician, calculates the total complexity of these reports, and analyzes the connectivity between physician nodes and physical examination report nodes (i.e., how many reports they have potential assignment relationships with). Based on these indicators, the system calculates the "load centrality" of each physician node, a comprehensive indicator reflecting the degree of load concentration of physicians in the entire workload distribution network. Physicians with high load centrality are often highly skilled and suitable for handling multiple types of reports, but this also means they face a greater risk of workload stress.

[0143] Step S4.C implements load adjustment for doctors with high workload centrality. The system first identifies the doctor nodes with the highest workload centrality; these doctors are the "bottleneck" in the current workload allocation. Then, the system selectively adjusts the edge weights between these high-load doctors and physical examination reports, specifically reducing the connection strength between them and "non-critical report" nodes. "Non-critical reports" refer to reports that do not require specific medical expertise to process. Through this edge weight adjustment, the system guides subsequent task allocation algorithms to reduce the allocation of non-core tasks to these high-load doctors, conserving their energy for reports that truly require their professional strengths, thereby generating an optimized workload graph.

[0144] Step S4.D extends the optimization process of S4.C into an iterative algorithm. The system repeatedly identifies the doctor node with the highest load centrality in the current graph, adjusts the edge weights accordingly, and then recalculates the load centrality of each node. This process is repeated until the load centrality of all doctor nodes tends to be balanced. This iterative optimization method simulates a "fluid equilibrium" process, continuously guiding the workload from high-pressure areas to low-pressure areas, eventually reaching a relatively balanced state and generating a balanced workload graph. This method is particularly suitable for handling dynamically changing work environments and can adapt to constantly changing report inflows and doctor availability.

[0145] Step S4.E is the final comprehensive score calculation stage. Based on the balanced workload map, the system recalculates the standardized workload index for each doctor as their current workload value. This updated workload value not only considers the doctor's current actual workload but also reflects the relative workload distribution within the entire system. The system then weights the doctor-reporter matching score with this updated current workload value to generate the final comprehensive score. In this weighted calculation, the system can adjust the weighting coefficients to balance the priorities of the two objectives of professional matching and workload balance, enabling a flexible task allocation strategy.

[0146] In one specific embodiment of the present invention, based on the balanced workload map, the standardized workload index of each doctor is recalculated as the current workload value. The doctor-report matching score and the current workload value are weighted and calculated to generate the comprehensive score. The doctor with the highest comprehensive score is selected to perform task assignment, including:

[0147] Step A: Based on the doctor-report matching score, establish a candidate doctor set for each physical examination report to be assigned, and generate a report assignment sequence and an initial assignment scheme according to the structural characteristics of the balanced workload diagram;

[0148] Step B: Identify the constraints of the initial allocation scheme, extract the hard requirements that specific diseases must be handled by doctors with specific specialties, and generate a set of edge constraints.

[0149] Step C: Based on the report allocation sequence and the set of edge constraints, assign the most suitable doctor to each physical examination report in sequence, re-evaluate the workload distribution after each allocation, and generate allocation process data;

[0150] Step D: Based on the allocation process data, monitor the current load distribution, dynamically adjust the professional matching threshold and load balancing weight coefficient, optimize subsequent allocation decisions, and generate a dynamically adjusted allocation scheme.

[0151] Step E: Verify the dynamically adjusted allocation scheme to confirm that all constraints are met and the optimal load balancing state is achieved. Then, perform the actual report allocation operation and update the work queues of each doctor.

[0152] Steps A through E further detail the specific implementation process of selecting the best doctor for task allocation based on the balanced workload map. In step A, the system first establishes a candidate doctor set for each medical examination report to be assigned. This set includes all doctors professionally suitable to handle the report. Then, based on the structural characteristics of the balanced workload map, the system generates a report allocation sequence. This sequence determines the processing priority of the medical examination reports, typically prioritizing reports with high complexity or requiring specific professional skills to ensure that these critical reports are assigned to the most suitable doctor. Based on this sequence and the candidate doctor set, the system generates an initial allocation scheme.

[0153] Step B introduces the concept of constraints, which is particularly important for task allocation in the medical field. The system identifies constraints on the initial allocation scheme, paying special attention to the hard requirement that certain specific diseases must be handled by physicians with specific expertise. For example, certain high-risk cardiovascular abnormalities may require review by a cardiologist, or certain complex tumor markers may require assessment by an oncologist. The system extracts these hard requirements, generating a set of edge constraints to ensure that the final allocation scheme meets these necessary professional requirements.

[0154] Step C implements the actual report allocation process. Following the pre-determined report allocation sequence, the system assigns the most suitable doctor to each medical examination report sequentially. In each allocation decision, the system comprehensively considers the doctor's professional suitability, current workload, and all relevant constraints. After each allocation, the system immediately updates the workload status of the relevant doctors and reassesses the overall workload distribution to ensure that subsequent allocation decisions are based on the latest system state. This step-by-step allocation and real-time updating method can adapt to dynamic changes during the allocation process, avoiding allocation results that are locally optimal but globally unbalanced.

[0155] Step D introduces a dynamic adjustment mechanism, enabling the system to adaptively adjust its allocation strategy based on real-time load distribution. The system continuously monitors the current load distribution. If it detects excessive workloads for certain doctors or resource shortages in certain specialties, it dynamically adjusts two key parameters: the specialty matching threshold and the load balancing weight coefficient. For example, when the system load is light, the specialty matching threshold can be increased to prioritize assigning reports to the most relevant doctor; conversely, when the system load is heavy, the specialty matching threshold can be appropriately lowered to increase the weight of load balancing, ensuring a more even distribution of workload. This dynamic adjustment mechanism allows the system to find the optimal balance between specialty matching and load balancing, generating a dynamically adjusted allocation scheme.

[0156] Step E is the final verification and execution phase. The system comprehensively verifies the dynamically adjusted allocation scheme, confirming that it meets all previously defined constraints and achieves optimal load balancing. After verification, the system executes the actual report allocation operation, formally assigning each medical examination report to the selected doctors and updating each doctor's work queue. This step ensures that the theoretically optimal allocation scheme is accurately translated into practical operation, while also providing the system with the latest work status data, laying the foundation for the next round of task allocation.

[0157] like Figure 4 As shown, the present invention also provides a doctor's work task intelligent allocation device, comprising:

[0158] The doctor profile building module is used to collect doctors' personal information, educational background, professional title and historical review report data, perform text mining and natural language processing on the historical review report data, extract doctors' professional expertise tags and calculate tag weight coefficients to generate doctor profile models;

[0159] The report workload assessment module is used to acquire the physical examination report data to be assigned, identify the physical examination type, analyze the examinee's personal attributes, and extract key report indicators from the physical examination report data, and perform comprehensive calculations according to the preset workload calculation formula to generate the standard workload score of the physical examination report.

[0160] The professional matching calculation module is used to extract and standardize keywords from the physical examination report, generate a report feature vector, calculate the similarity between the report feature vector and the set of professional expertise tags in the doctor profile model, and generate a doctor-report matching score.

[0161] The task allocation decision module is used to filter a set of candidate doctors based on the doctor-report matching score, calculate the current workload value of each doctor in the candidate doctor set, perform a weighted calculation on the doctor-report matching score and the current workload value to generate a comprehensive score, and select the doctor with the highest comprehensive score to perform task allocation;

[0162] The dynamic optimization module is used to collect actual time and quality evaluation data for processing physical examination reports, compare the standard workload integral predicted by the system with the actual processing time, perform deviation analysis and machine learning optimization, and update the parameters of the workload calculation formula and the professional expertise label weights in the doctor profile model.

[0163] In a preferred embodiment of the present invention, the task allocation decision module includes:

[0164] The workload graph construction unit is used to construct a doctor-report bipartite graph model based on the candidate doctor set and the physical examination reports to be assigned, and generate a workload graph.

[0165] The load centrality calculation unit is used to calculate the load centrality of each doctor node in the workload diagram and generate load centrality data.

[0166] The graph structure optimization unit is used to identify the doctor node with the highest load centrality, adjust the edge weights and perform iterative optimization to generate a balanced workload graph.

[0167] The comprehensive scoring calculation unit is used to calculate the standardized workload index of each doctor based on the balanced workload map, and to perform a weighted calculation on the doctor-report matching score and the standardized workload index to generate a comprehensive score.

[0168] The assignment execution unit is used to select the doctor with the highest overall score, perform task assignment, and update the doctor's work queue.

[0169] The working process of the present invention is further illustrated below through specific embodiments:

[0170] Example: A health checkup center has three doctors:

[0171] Doctor A: Specializes in cardiovascular diseases (tags: hypertension, arrhythmia) and health management.

[0172] Doctor B: Specializes in tumor screening (tags: lung nodules, thyroid nodules, tumor markers).

[0173] Doctor C: Specializes in endocrine and digestive system diseases (tags: diabetes, fatty liver, Helicobacter pylori).

[0174] A medical examination report from a 50-year-old male cadre showed preliminary screening results indicating "pulmonary nodules" and "elevated blood sugar".

[0175] Workload Calculation: Cadre physical examination (3.0 points) + 50-year-old male (+0.5 points) + keyword "pulmonary nodules" (+1.5 points) + keyword "elevated blood sugar" (+0.8 points) = Total 5.8 standard workload points.

[0176] Specialty Matching: The report keywords were matched with the expertise tags of Doctor B (Oncology) and Doctor C (Endocrinology). By calculating the feature-tag matching matrix, the matching score for Doctor B was 0.85, and the matching score for Doctor C was 0.62.

[0177] Load balancing: Assume that doctor B's total load score is 45 points and doctor C's is 38 points. By constructing a workload graph model, the standardized load values ​​for each doctor are calculated: doctor B's is 1.2 and doctor C's is 0.9.

[0178] Allocation Decision: Although the report is a better match for Doctor B's expertise, for the sake of immediate load balancing, the system performs a weighted calculation of the match and load value (weights set to 0.6:0.4), resulting in a comprehensive score of 0.29 for Doctor B and 0.32 for Doctor C. Ultimately, the system allocates the report to Doctor C, who currently has a lower total load and a higher comprehensive score.

[0179] If the report contains more urgent or specific indicators (such as "suspected malignant tumor"), the weight ratio of professional matching degree and burden value can be adjusted through rule settings (such as adjusting it to 0.8:0.2) to prioritize the doctor with the highest professional matching degree to handle such reports.

[0180] This embodiment illustrates how the system makes trade-off decisions between professional matching and load balancing, achieving intelligent allocation that takes into account both professional matching and workload balancing.

[0181] It should be noted that the terms "an embodiment," "embodiment," and "exemplary embodiment" used herein do not necessarily refer to the same embodiment or example. Furthermore, these embodiments and examples do not limit the scope of the invention unless explicitly stated otherwise. In addition, the embodiments and features described herein can be combined with each other unless otherwise specified.

[0182] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0183] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. The present invention is not limited to the above embodiments. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical solutions of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for intelligent allocation of doctor's work tasks, characterized in that, Includes the following steps: Collect doctors' personal information and historical review report data, perform text mining and natural language processing on the historical review report data, extract doctors' professional expertise tags and calculate tag weight coefficients to generate a doctor profile model, wherein the doctor profile model includes doctors' basic information, a set of professional expertise tags, ability weight coefficients and a real-time work queue; Obtain the physical examination report data to be assigned, identify the physical examination type, analyze the examinee's personal attributes, and extract key indicators of the report data. Perform comprehensive calculation according to the preset workload calculation formula to generate the standard workload score of the physical examination report. Keyword extraction and standardization are performed on the physical examination report to generate a report feature vector. The similarity between the report feature vector and the set of professional expertise tags in the doctor profile model is calculated to generate a doctor-report matching score. Based on the doctor-report matching score, a candidate doctor set is selected, the current workload value of each doctor in the candidate doctor set is calculated, the doctor-report matching score and the current workload value are weighted and calculated to generate a comprehensive score, and the doctor with the highest comprehensive score is selected to perform task assignment. The system collects and processes actual time and quality evaluation data of physical examination reports, compares the standard workload integral predicted by the system with the actual processing time, performs deviation analysis and machine learning optimization, and updates the parameters of the workload calculation formula and the professional expertise label weights in the doctor profile model.

2. The method according to claim 1, characterized in that, Text mining and natural language processing were performed on the historical review report data to extract doctors' professional expertise tags and calculate tag weight coefficients, including: Based on the historical review report data, text mining technology is used to extract high-frequency disease keywords and physical examination types, and combined with the doctor's personal professional background information, an initial set of professional expertise tags is generated. For each tag in the initial set of professional expertise tags, the doctor's historical accuracy, processing efficiency, and number of reports handled are statistically analyzed, and a weighted calculation is performed to generate the weight coefficient for each tag; Based on the weight coefficients of each tag, tags with weight coefficients exceeding a preset threshold are selected to generate a set of professional expertise tags in the doctor profile model.

3. The method according to claim 1, characterized in that, The standard workload score for the medical examination report is generated by comprehensively calculating the workload according to the preset workload calculation formula, including: Based on the physical examination type in the physical examination report, query the physical examination type points mapping table to obtain the basic score for the physical examination type; Based on the age group, gender, and marital status of the subjects, a correction factor calculation model is established to calculate the personnel attribute correction factor; Natural language processing is performed on the departmental summaries and test results in the physical examination report to extract abnormal indicators and key medical terms, query the medical keyword-integration mapping table, and calculate the keyword integration value. The standard workload score of the physical examination report is generated by summing the base score of the physical examination type, the personnel attribute correction factor, and the keyword score bonus.

4. The method according to claim 1, characterized in that, Calculate the similarity between the report feature vector and the set of professional expertise tags in the doctor profile model, and generate a doctor-report matching score, including: Obtain each keyword and its weight in the report feature vector, obtain each tag and its weight coefficient in the professional expertise tag set, and construct a feature-tag matching matrix; The semantic similarity between the report keywords and the doctor's expertise tags in the feature-tag matching matrix is ​​calculated, and a weighted sum is generated by combining their respective weights to produce the doctor-report matching score.

5. The method according to claim 1, characterized in that, Calculating the current workload value of each doctor in the candidate doctor set includes: Based on the candidate doctor set, extract the real-time work queue data of each doctor, count the number of reports currently pending for each doctor and the standard workload points of each report, perform cumulative calculation, and generate the cumulative workload points of each doctor. Obtain the capability weight coefficient of each doctor, divide the cumulative workload score by the capability weight coefficient, perform standardization processing, and generate the current workload value of each doctor in the candidate doctor set.

6. The method according to claim 2, characterized in that, Text mining techniques were used to extract high-frequency disease keywords and physical examination types. Combined with doctors' personal professional background information, an initial set of professional expertise tags was generated, including: Based on the historical review report data, a doctor's personal corpus is constructed, a medical terminology dictionary and a common disease terminology database are established, and a basic knowledge base is generated. Define a context window for medical keywords in the doctor's personal corpus, collect words within a preset range before and after the medical keywords, and generate contextual environment data for each medical keyword; Frequent pattern mining is performed on the contextual environment data to identify combinations and co-occurrence patterns of medical terms, and the frequency of occurrence of each medical term in different contextual environments is statistically analyzed to generate term contextual frequency distribution data. Based on the term context frequency distribution data, high-frequency, low-information terms that appear due to routine descriptions are removed, and the net frequency of each medical term in different context environments is calculated to generate term net frequency data. Based on the net frequency data of the terms, professional specificity, and medical importance, the contribution weight of each medical term to the doctor's professional expertise is calculated, and terms with high contribution weights and their contextual pattern combinations are selected to generate the initial set of professional expertise tags.

7. The method according to claim 3, characterized in that, Natural language processing was performed on the departmental summaries and laboratory test results in the physical examination report to extract abnormal indicators and key medical terms, including: The physical examination report is segmented according to department and examination type, a structured text representation is established, and segmented report data is generated; The sliding window technique is applied to the segmented data of the report to scan each medical term and its context, extract key medical terms and descriptive words within a preset range before and after them, and generate term context data. Based on a pre-trained medical terminology pattern library, pattern matching is performed on the terminology context data to identify medical terminology combination patterns, extract abnormal indicators and their descriptive contexts, and generate an abnormal indicator context set. For each indicator in the abnormal indicator context set, the influence of modifiers on the clinical meaning of the terminology is analyzed, mild, moderate and severe abnormalities are distinguished, and indicator severity classification data is generated. Based on the medical knowledge base and the severity grading data of the indicators, the clinical importance score of each abnormal indicator is calculated, and common minor abnormalities and serious abnormalities requiring professional attention are distinguished, and the key medical terms and their importance scores are generated.

8. The method according to claim 1, characterized in that, Calculate the current workload value of each doctor in the candidate doctor set, and perform a weighted calculation on the doctor-report matching score and the current workload value to generate a comprehensive score, including: Based on the candidate doctor set and the physical examination reports to be assigned, a doctor-report bipartite graph model is constructed, with doctors as the first type of nodes and physical examination reports as the second type of nodes. The doctor-report matching score is used as the edge weight to generate a workload graph. For each doctor node in the workload graph, the number of assigned reports, the total report complexity, and the connectivity with the physical examination report node are counted. The load centrality of each doctor node is calculated, and load centrality data is generated. Based on the load centrality data, the doctor node with the highest load centrality is identified, and the edge weights of the doctor node with the highest load centrality are adjusted to reduce its connection strength with non-critical reporting nodes, thereby generating an optimized workload graph. The optimized workload graph is iteratively optimized by repeatedly identifying nodes with high load centrality and adjusting edge weights until the load centrality of each doctor node tends to be balanced, thus generating a balanced workload graph. Based on the balanced workload map, the standardized workload index of each doctor is recalculated as the current workload value. The doctor-report matching score and the current workload value are weighted and calculated to generate the comprehensive score.

9. The method according to claim 8, characterized in that, Based on the balanced workload map, the standardized workload index of each doctor is recalculated as the current workload value. The doctor-report matching score and the current workload value are weighted and calculated to generate the comprehensive score. The doctor with the highest comprehensive score is selected to perform task assignments, including: Based on the doctor-report matching score, a candidate doctor set is established for each physical examination report to be assigned. According to the structural characteristics of the balanced workload diagram, a report assignment sequence is generated, and an initial assignment scheme is generated. Constraints are identified in the initial allocation scheme, and the hard requirements that specific diseases must be handled by doctors with specific specialties are extracted to generate a set of edge constraints. Based on the report allocation sequence and the set of edge constraints, the most suitable doctor is assigned to each physical examination report in sequence. After each allocation, the workload distribution is reassessed to generate allocation process data. Based on the allocation process data, monitor the current load distribution, dynamically adjust the professional matching degree threshold and load balancing weight coefficient, optimize subsequent allocation decisions, and generate a dynamically adjusted allocation scheme. The dynamically adjusted allocation scheme is verified to confirm that all constraints are met and the optimal load balancing state is achieved. The actual report allocation operation is then executed, and the work queues of each doctor are updated.

10. A smart task allocation device for doctors, characterized in that, include: The doctor profile building module is used to collect doctors' personal information, educational background, professional title and historical review report data, perform text mining and natural language processing on the historical review report data, extract doctors' professional expertise tags and calculate tag weight coefficients to generate doctor profile models; The report workload assessment module is used to acquire the physical examination report data to be assigned, identify the physical examination type, analyze the examinee's personal attributes, and extract key report indicators from the physical examination report data, and perform comprehensive calculations according to the preset workload calculation formula to generate the standard workload score of the physical examination report. The professional matching calculation module is used to extract and standardize keywords from the physical examination report, generate a report feature vector, calculate the similarity between the report feature vector and the set of professional expertise tags in the doctor profile model, and generate a doctor-report matching score. The task allocation decision module is used to filter a set of candidate doctors based on the doctor-report matching score, calculate the current workload value of each doctor in the candidate doctor set, perform a weighted calculation on the doctor-report matching score and the current workload value to generate a comprehensive score, and select the doctor with the highest comprehensive score to perform task allocation; The dynamic optimization module is used to collect actual time and quality evaluation data for processing physical examination reports, compare the standard workload integral predicted by the system with the actual processing time, perform deviation analysis and machine learning optimization, and update the parameters of the workload calculation formula and the professional expertise label weights in the doctor profile model.

11. The apparatus according to claim 10, characterized in that, The task allocation decision module includes: The workload graph construction unit is used to construct a doctor-report bipartite graph model based on the candidate doctor set and the physical examination reports to be assigned, and generate a workload graph. The load centrality calculation unit is used to calculate the load centrality of each doctor node in the workload diagram and generate load centrality data. The graph structure optimization unit is used to identify the doctor node with the highest load centrality, adjust the edge weights and perform iterative optimization to generate a balanced workload graph. The comprehensive scoring calculation unit is used to calculate the standardized workload index of each doctor based on the balanced workload map, and to perform a weighted calculation on the doctor-report matching score and the standardized workload index to generate a comprehensive score. The assignment execution unit is used to select the doctor with the highest overall score, perform task assignment, and update the doctor's work queue.

Images