Data processing method based on inspection results

Abstract

The embodiment of the invention relates to a data processing method based on inspection results. The method comprises the following steps: receiving initial inspection report data of a hospital laboratory information system (LIS), an image archiving and communication system (PACS) and an electronic medical record system (EMR); the initial inspection report data comprises at least one of texts, numerical values and images; the image comprises at least one of an electrocardiogram and a pathological section; analyzing texts, numerical values and images in the initial inspection report data, extracting key medical entities, mapping the key medical entities into standardized fields, and generating target inspection report data according to the standardized fields, the target inspection report data being structured inspection report data; matching the target inspection report with a preset rule set, and determining a first number of rules from the rule set; determining a target rule according to the priority of the first number of rules; and sending result information corresponding to the target rule to the doctor terminal.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a data processing method based on inspection results. Background Art

[0002] Current clinical testing systems, such as hospital laboratory information systems (LIS), picture archiving and communication systems (PACS), and electronic medical records (EMR), only display test results. Doctors must manually analyze the data for diagnosis, which presents the following problems:

[0003] 1. Inefficiency: The testing and diagnosis processes are separated, and doctors need to switch between multiple platforms;

[0004] 2. Risk of misdiagnosis: Manual analysis relies on experience, and complex cases are prone to omissions;

[0005] 3. Insufficient data utilization: Unstructured inspection reports, such as text and images, are difficult to be parsed by automated systems.

[0006] Although existing CDSS systems can provide diagnostic recommendations, their rules are fixed and rely on structured input, making them unable to adapt to the needs of multi-format test data and dynamic medical scenarios. Summary of the Invention

[0007] The purpose of the present invention is to provide a data processing method based on inspection results to solve the problems existing in the prior art.

[0008] To achieve the above object, the present invention provides a data processing method based on inspection results, comprising:

[0009] Receiving initial test report data from a hospital's laboratory information system (LIS), picture archiving and communication system (PACS), or electronic medical record system (EMR); the initial test report data includes at least one of text, numerical values, and images; and the images include at least one of an electrocardiogram and a pathology slide;

[0010] Parsing the text, numerical values, and images in the initial medical inspection report data to extract key medical entities, mapping the key medical entities to standardized fields, and generating target medical inspection report data based on the standardized fields, wherein the target medical inspection report data is structured medical inspection report data;

[0011] Matching the target inspection report with a preset rule set, and determining a first number of rules from the rule set;

[0012] determining a target rule according to the priorities of the first number of rules;

[0013] The result information corresponding to the target rule is sent to the doctor terminal.

[0014] In one possible implementation, parsing the text, numerical values, and images in the initial test report data, extracting key medical entities, mapping the key medical entities to standardized fields, and generating target test report data based on the standardized fields specifically includes:

[0015] Clustering the initial inspection report data to extract common structural features; the common structural features include inspection items;

[0016] When new inspection report data is detected, the text is preprocessed to remove redundant characters and segment semantic units;

[0017] Determine whether the parsed semantic unit exists in the parsing template library. If not, perform entity recognition on the semantic unit based on a model in the medical field to identify key entities. The key entities include inspection items, numerical values ​​and units, and descriptive text.

[0018] Perform regular expression matching and standard conversion on inspection items, values ​​and units, and descriptive texts to generate structured first inspection report data;

[0019] Extracting text descriptions from inspection images using optical character recognition (OCR) technology and processing image features using a convolutional neural network to generate structured numerical indicators. The image feature processing includes identifying abnormal cell morphology from pathological sections and converting the abnormal cell morphology into structured numerical indicators based on predefined standardized fields.

[0020] Target inspection report data are obtained based on the structured first inspection report data and the structured numerical indicators.

[0021] In a possible implementation, calculating the intersection-over-union ratio of each rule in the first number of rules specifically includes:

[0022] The intersection-over-union ratio of each rule is determined according to the formula IoU=number of matching rules / total number of rules associated with the disease.

[0023] In a possible implementation, the method further includes:

[0024] The weighted intersection-over-union (IoU) of each rule is calculated using the formula IoU1=Σ(matching rule weight) / Σ(all association rule weights).

[0025] In a possible implementation, the target inspection report is matched with a first rule set, and the first number of rules matched from the first rule set specifically include:

[0026] Calculating the overlap ratio of any two rules, and when the overlap ratio is greater than an overlap ratio threshold, classifying the rules into the first rule set;

[0027] When any two rules point to the same disease, a higher similarity weight is assigned to the two rules, and the rules are classified into the second rule set;

[0028] Counting the probability of the rules being triggered simultaneously, determining the rules whose probability is greater than a preset probability, and classifying the rules into a third rule set;

[0029] According to a clustering algorithm, the first rule set, the second rule set, and the third rule set are represented as feature vectors, and clustered to obtain a plurality of clusters;

[0030] When the target inspection report triggers any cluster, the intersection-and-union ratio of multiple rules in the cluster is calculated;

[0031] When the target inspection report triggers multiple clusters, the intersection-and-union ratio of each cluster is calculated according to the cluster priority, and the target cluster is determined. Then, in the target cluster, the intersection-and-union ratio of multiple rules within the target cluster is determined;

[0032] The rule whose intersection-over-union ratio is greater than the intersection-over-union ratio threshold is determined to be one of the first number of rules.

[0033] In one possible implementation, the clusters include anemia cluster, infection cluster, metabolic disease cluster, and respiratory disease cluster; the conditions of the anemia cluster include HGB and MCV; the conditions of the infection cluster include white blood cells, CRP, and body temperature; the metabolic disease cluster includes blood glucose, lactate, and pH; and the respiratory disease cluster includes white blood cells, body temperature, and pH.

[0034] In a possible implementation, the method further includes:

[0035] Receive the diagnosis results sent by the doctor's terminal;

[0036] The preset rule set is updated according to the diagnosis result.

[0037] In a possible implementation, the method further includes:

[0038] According to the time information, the target inspection report data before the current time is obtained; the target inspection report data includes the name of the abnormal indicator, the value of the abnormal indicator and the diagnosis result of the abnormal indicator;

[0039] Determining corresponding rules based on the target inspection report data;

[0040] According to the rules, the weights of the clusters are updated.

[0041] In a possible implementation, the method further includes:

[0042] According to the current geographical location information, obtaining target inspection and examination report data within the geographical location information range;

[0043] Determining corresponding rules based on the target inspection report data;

[0044] According to the rules, the weights of the clusters are updated.

[0045] In a possible implementation, the method further includes:

[0046] Get patient ID;

[0047] According to the patient ID, retrieve the past history corresponding to the patient ID; the past history includes major diseases, chronic diseases, infectious diseases, surgical history and allergy history;

[0048] Matching is performed based on the major diseases, chronic diseases, infectious diseases, surgical history, and allergy history, and a preset rule set;

[0049] When a rule is matched, the target test report data of the patient ID is matched again with the matched rule;

[0050] When no rule is matched, the target test report data of the patient ID is matched with the rule set.

[0051] By applying the data processing method based on test results provided by an embodiment of the present invention, initial test report data from a hospital laboratory information system (LIS), an image archiving and communication system (PACS), and an electronic medical record system (EMR) is received; the initial test report data includes at least one of text, numerical values, and images; the images include at least one of electrocardiograms and pathological sections; the text, numerical values, and images in the initial test report data are parsed to extract key medical entities, and the key medical entities are mapped to standardized fields, and target test report data is generated based on the standardized fields, the target test report data being structured test report data; the target test report is matched with a preset rule set, and a first number of rules are determined from the rule set; the target rule is determined based on the priority of the first number of rules; and the result information corresponding to the target rule is sent to a doctor terminal. Thus, after the test report data is structured, the data is unified, and according to a clustering algorithm, a diagnosis result is automatically given based on the test results, reducing dependence on manual analysis, improving user experience, and avoiding omissions in complex cases. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A schematic flow chart of a data processing method based on inspection results provided by an embodiment of the present invention;

[0053] Figure 2 for Figure 1 The specific content of step 120;

[0054] Figure 3 for Figure 1 The specific content of step 130 in FIG. DETAILED DESCRIPTION

[0055] To make the objectives, technical solutions, and advantages of the present invention more apparent, the present invention will be further described in detail below with reference to the accompanying drawings. It is apparent that the embodiments described are only some, not all, of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are intended to fall within the scope of protection of the present invention.

[0056] Figure 1 The following is a flow chart of a data processing method based on inspection results provided by an embodiment of the present invention. Figure 1 , the technical solution of the present invention is described with specific embodiments. Figure 1 As shown, this application includes the following steps:

[0057] Step 110, receiving initial test report data from the hospital's laboratory information system LIS, picture archiving and communication system PACS, and electronic medical record system EMR; the initial test report data includes at least one of text, numerical values, and images; and the images include at least one of electrocardiograms and pathological sections.

[0058] Specifically, the present application can be directly connected to the LIS, PACS, and EMR systems to automatically obtain the test report data from each system. The text in the report data can be various test reports, and the numerical values ​​are the numerical values ​​corresponding to the test reports or images. The images can be electrocardiograms, electroencephalograms, MRI results, etc., and the present application does not limit this.

[0059] Step 120: Parse the text, numerical values, and images in the initial medical inspection report data to extract key medical entities, map the key medical entities to standardized fields, and generate target medical inspection report data based on the standardized fields. The target medical inspection report data is structured medical inspection report data.

[0060] Specifically, step 120 includes the following steps:

[0061] Step 1201, pre-processing the text, removing redundant characters and segmenting semantic units;

[0062] Specifically, we use regular expressions to remove irrelevant symbols and extra spaces, and segment the text by paragraphs or punctuation marks to obtain semantic units.

[0063] Step 1202: Determine whether the parsed semantic unit exists in the parsing template library. If not, perform entity recognition on the semantic unit based on a model in the medical field to identify key entities. Key entities include inspection items, numerical values ​​and units, and descriptive text.

[0064] Match the semantic units with the regular rules in the template library. If there is no match, use the NLP model to identify unmapped inspection items.

[0065] Among them, the test items include but are not limited to blood routine, biochemical indicators, etc. The value is the value of the test item, the unit is the unit of the value, and the descriptive text is, for example, increased, decreased, etc.

[0066] Step 1203 , performing regular expression matching and standard conversion on the inspection items, values ​​and units, and descriptive text to generate structured first inspection report data;

[0067] Specifically, the present application can process a variety of initial test and inspection reports in advance, including converting each report into a feature vector, such as detecting the existence of each item through coding, 1 if it exists, and 0 if it does not exist, extracting the item position feature, for example, if the report has a fixed paragraph structure, such as "blood routine" and "biochemical indicators", record the location of the item, and use a clustering algorithm to divide the report into K categories according to the feature vector, such as K = 5, corresponding to blood routine, blood gas analysis, imaging reports, etc., and analyze the structural similarity between reports through a tree diagram. For each type of report, count the high-frequency test and inspection items, such as "HGB, WBC, PLT" in the blood routine category, and then generate a public structure template. The template structure includes template type, test and inspection items, numerical format, paragraph position, etc.

[0068] Step 1204: extracting text descriptions from the inspection image using optical character recognition (OCR) technology, and performing image feature processing using a convolutional neural network to generate structured numerical indicators. Image feature processing includes identifying abnormal cell morphology from pathological sections and converting abnormal cell morphology into structured numerical indicators based on predefined standardized fields.

[0069] Specifically, a custom model is trained to adapt the medical report font, a pre-trained model is used for transfer learning, medical record slice images are labeled, such as normal, cancerous, inflammatory, etc., and the model output is mapped to standardized indicators, such as the proportion of abnormal cells at 30%.

[0070] Step 1205: Obtain target inspection report data based on the structured first inspection report data and the structured numerical indicators.

[0071] Furthermore, after step 120 and before step 130, the present application may further include: obtaining a patient ID;

[0072] According to the patient ID, retrieve the patient's corresponding medical history; medical history includes major diseases, chronic diseases, infectious diseases, surgical history and allergy history;

[0073] Matching is performed based on major diseases, chronic diseases, infectious diseases, surgical history, and allergy history, as well as preset rule sets;

[0074] When a rule is matched, the target test report data of the patient ID is matched again with the matched rule;

[0075] When no rule is matched, the target test report data of the patient ID is matched with the rule set.

[0076] Specifically, the patient ID is obtained from the target test report, and then the patient's medical history is called to match the medical history with the rule library. When the match is successful, the successfully matched rule is used as one of the first number of rules in step 130.

[0077] Step 130 , matching the target inspection report with a preset rule set, and determining a first number of rules from the rule set;

[0078] Step 140, determining a target rule based on the priorities of the first number of rules;

[0079] Specifically, step 130 includes the following steps:

[0080] Step 1301, calculating the overlap ratio of any two rules, and when the overlap ratio is greater than the overlap ratio threshold, classifying the rule into the first rule set;

[0081] Specifically, calculate the overlap ratio of the two rules in the condition part (such as inspection indicators). For example:

[0082] The conditions of Rule A are “HGB < 12 and MCV < 80”;

[0083] The conditions of Rule B are “HGB < 12 and Ferritin < 15”;

[0084] The overlap index between the two is HGB, and the overlap degree is 1 / 2.

[0085] The overlap can be calculated according to the following formula:

[0086] Assume that the condition fields of rule A and rule B are C A and C B , then the overlap ratio is:

[0087]

[0088] If Overlap(A, B) is greater than the overlap ratio threshold, for example, the overlap ratio threshold is 1 / 2, then rules A and B can be classified into the same rule set, which can be marked as the first rule set. Step 1302: When any two rules point to the same disease, a higher similarity weight is assigned to the two rules, and the rules are classified into the second rule set;

[0089] Specifically, if two rules point to the same type of disease, such as the target disease fields of the rules are the same, such as anemia-related, they are given a higher similarity weight and are included in the same rule set.

[0090] The rules may be merged by weighted overlap ratio, such as by merging the rules by Overlap(A, B)*weight.

[0091] Step 1303: Count the probabilities of the rules being triggered simultaneously, determine the rules whose probabilities are greater than a preset probability, and classify the rules into a third rule set;

[0092] Specifically, the probability of statistical rules being triggered simultaneously in actual applications is calculated, such as the frequency of simultaneous triggering of Rule A and Rule B in historical data. Rules with high co-occurrence frequencies are more similar, and rules with probabilities greater than the probability threshold are grouped into the same rule set.

[0093] The trigger probability can be calculated using the following formula: P(A∩B) = number of reports that trigger A and B simultaneously / total number of reports.

[0094] Step 1304: According to a clustering algorithm, the first rule set, the second rule set, and the third rule set are represented as feature vectors and clustered to obtain a plurality of clusters.

[0095] Specifically, K-means or hierarchical clustering algorithms are applied to group the rules. The rules are represented as feature vectors that may include the following features: the encoding of each field, such as the inspection items and threshold range; the target disease label, which is one-hot encoded; condition indicators, disease categories, and historical trigger times;

[0096] Using clustering algorithms such as K-means and hierarchical clustering, rule sets are grouped into multiple clusters, each representing a class of related rules. For example, Cluster 1 represents anemia-related rules, with conditions involving HGB, MCV, and Ferritin; Cluster 2 represents infection-related rules, with conditions involving white blood cells, CRP, and body temperature; and Cluster 3 represents metabolic disease rules, with conditions involving blood glucose, lactate, and pH. Cluster distance metrics can be weighted Euclidean distance or cosine similarity, combined with overlap ratio and disease label weighting. The number of clusters can be optimized using the elbow method or silhouette coefficient.

[0097] Step 1305 , when the target inspection report triggers any cluster, the intersection-over-union ratio of multiple rules in the cluster is calculated;

[0098] When the target inspection report triggers a cluster, it is necessary to calculate the intersection-and-union (IoU) of the rules within the cluster. The specific steps include trigger rule extraction, IoU calculation, and IoU threshold screening.

[0099] Trigger rule extraction is based on the target test report content, such as the test item results, and matches all triggered rules in the cluster. For example, Rule A: Hemoglobin <12g / dL → triggers anemia; Rule B: Mean corpuscular volume <80fL → triggers iron deficiency anemia. Calculate the intersection and union ratio of the rule condition fields. If the condition fields of rules A and B are CA and CB respectively, the intersection and union ratio is For example, if both rule A and rule B check hemoglobin and red blood cell volume, the intersection is 2, the union is 2, and IoU = 1.

[0100] If the IoU of the rules within a cluster is greater than a threshold, for example, the threshold is 0.8, then these rules are considered to be highly overlapping, and only one of them is retained, such as the rule with the highest priority.

[0101] Step 1306: When the target inspection report triggers multiple clusters, the intersection-and-union ratio of each cluster is calculated based on the cluster priority, and the target cluster is determined. Then, within the target cluster, the intersection-and-union ratio of multiple rules within the target cluster is determined.

[0102] Specifically, when the patient's target test report data triggers multiple clusters, it is necessary to sort them according to the priority of the clusters, such as disease severity, rule triggering frequency, etc., and calculate the IoU of each cluster in turn.

[0103] Clusters can have priorities, represented by their weights. A larger weight indicates a higher priority. Clusters can have both preset and dynamic priorities. The preset priority is acute disease clusters > chronic disease clusters. Dynamic priorities are determined based on the trigger frequency of vinegar in historical reports. The highest-priority cluster is selected and its intersection-of-union (IoU) ratio is calculated. If the IoU ratio does not meet the threshold, clusters with lower priorities are examined. Within the target cluster, the rule with the highest IoU ratio is selected for output. This rule may be a single rule or a combination of overlapping rules.

[0104] For example, in one example, cluster 1 is an anemia-related cluster, and cluster 1 includes rule A hemoglobin <12 g / dL and rule B hemoglobin <12 g / dL and ferritin <30 ng / mL.

[0105] Since the condition fields completely overlap, the intersection-over-union (IoU) of rules A and B is 1. If the threshold is 0.9, only rule B is retained because it is stricter.

[0106] Cluster 2 is a diabetes-related cluster. Cluster 2 includes rule C: fasting blood glucose > 126 mg / dL; rule D: glycosylated hemoglobin > 6.5%. Rules C and D have no overlapping conditions, and the intersection-over-union ratio IoU(C, D) = 0, so both are retained.

[0107] If the report triggers cluster 1 and cluster 2, and cluster 1 has a higher priority, rule B will be output eventually because the IoU meets the threshold.

[0108] Because rules within the same cluster have high overlap in conditions, their intersection-to-union ratio (IRU) better reflects the core characteristics of the disease. Therefore, rules within each cluster are prioritized. For example, rules are divided into multiple levels based on priority (e.g., urgent > high > medium > low). The priorities of different rules within the same cluster can also be represented by weights: higher weights indicate higher priority. Rules with higher priorities are processed first.

[0109] For example, a cluster includes rules A and B. Rule A has an urgent priority: "PaO2 < 60 mmHg → hypoxemia"; Rule B has a medium priority: "Blood sugar > 7.0 mmol / L → diabetes risk." The processing order is to check rule A first, then rule B.

[0110] Each rule is assigned a weight, such as based on the strength of medical evidence or historical accuracy, and rules with higher weights take precedence in case of conflict.

[0111] In another example, rule 1, weight 0.9: "WBC>12×10 9 / L AND body temperature>38℃→bacterial infection"; Rule 2 (weight 0.7): "CRP>10mg / L→inflammatory response".

[0112] If both are matched, the diagnostic suggestion of Rule 1 is adopted first.

[0113] When there is a logical conflict, the rules pointing to different diseases are logically verified and the conflict is adjudicated based on the context or additional data.

[0114] For example, Rule A: "Cough + fever → influenza", Rule B: "Cough + swollen lymph nodes → tuberculosis", solution: If the patient has "fever" and "swollen lymph nodes" at the same time, further examination of chest X-ray or laboratory indicators, such as PPD test, is required for judgment.

[0115] When calculating the intersection over union (IoU), the intersection over union (IoU) of each rule can be determined according to the formula IoU = number of matching rules / total number of rules associated with the disease.

[0116] Alternatively, the intersection over union (IoU) of each rule is determined according to the formula IoU = number of matching rules / total number of rules associated with the disease.

[0117] In another example, the intra-cluster rule intersection ratio is determined as follows:

[0118] Define a set of rules within a cluster: Cluster 0: Contains reports 1 and 4, involving the project {HGB, WBC, PLT}. Cluster 1: Contains reports 2 and 5, involving the project {WBC, PLT}. Cluster 2: Contains report 3, involving the project {HGB, PLT}.

[0119] Calculate the intra-cluster rule intersection-union ratio: cluster 0, report 1 and report 4:

[0120] Rule 1: {HGB, WBC}, Rule 4: {HGB, WBC, PLT}, Intersection: {HGB, WBC}, Union: {HGB, WBC, PLT}.

[0121] IoU**=2(number of intersection items) / 3(number of union items)≈0.67;

[0122] Cluster 1 (Report 2 and Report 5):

[0123] Rule 2: {WBC, PLT}, Rule 5: {PLT}, Intersection: {PLT}, Union: {WBC, PLT}, IoU** = 1 / 2 = 0.5.

[0124] Cluster 2, only report 3: single rule, no intersection-union calculation.

[0125] Step 1307 : Determine that the rule whose IoU ratio is greater than the IoU ratio threshold is one of the first number of rules.

[0126] Specifically, if the rules of multiple clusters are triggered, the global intersection-union ratio is calculated comprehensively according to the priority or weight of the clusters;

[0127] For example, the urgent disease cluster has a higher weight than the general disease cluster.

[0128] In one example, the input data is "HGB":11.0,"MCV":75,"Ferritin":10,"WBC":15×10 9 / L;

[0129] After the rules are matched, the rules for anemia in cluster 1 are triggered: HGB < 12, MCV < 80, and Ferritin < 15; at the same time, the rules for infection in cluster 2 are triggered: WBC > 12.

[0130] Perform an IoU calculation: Intra-cluster calculation: The IoU of the anemia cluster is 3 / 3 = 1.0; Cross-cluster weighting: The IoU of the infection cluster is 1 / 1 = 1.0, but due to its low priority, the overall weight is 0.7. The recommended results are "Primary Diagnosis": "Iron Deficiency Anemia", "Secondary Diagnosis": "Bacterial Infection", and "Confidence": [0.95, 0.85].

[0131] In the present application, before performing rule matching through a clustering algorithm, the following may also be included: filtering invalid rules.

[0132] Specifically, invalid rules are excluded based on contraindication conditions. First, the contraindication conditions in each rule (such as age restrictions, gender restrictions, and specific medical history) are checked. If the patient's test data meets the contraindication conditions of a rule, the rule is directly excluded.

[0133] For example, the rule: "If age < 18 years, exclude prostate cancer-related diagnoses" → If the patient is 15 years old, all prostate cancer rules will be invalid.

[0134] Step 150: Send the result information corresponding to the target rule to the doctor terminal.

[0135] Specifically, the result information corresponding to the rule, including the diagnosis result and the confidence level of each diagnosis result, may be sent to the doctor terminal.

[0136] Furthermore, the present application further comprises the following steps:

[0137] Receive the diagnosis results sent by the doctor's terminal;

[0138] Based on the diagnosis results, the preset rule set is updated.

[0139] Furthermore, the application may also include the following steps:

[0140] According to the time information, the target inspection report data before the current time is obtained; the target inspection report data includes the name of the abnormal indicator, the value of the abnormal indicator and the diagnosis result of the abnormal indicator;

[0141] Determine the corresponding rules based on the target inspection report data;

[0142] According to the rules, update the weight of the cluster.

[0143] Specifically, the priority or weight of rules is dynamically adjusted based on epidemiological data or the latest medical guidelines.

[0144] For example, during flu season, rules related to respiratory symptoms will be prioritized. Temporary rules are in effect: New temporary rules are added to respond to public health emergencies.

[0145] Furthermore, this application may also include:

[0146] According to the current geographical location information, obtain the target inspection and examination report data within the geographical location information range;

[0147] Determine the corresponding rules based on the target inspection report data;

[0148] According to the rules, update the weight of the cluster.

[0149] Specifically, for example, a certain disease breaks out in a certain place, and the current geographic location information happens to be within this area. Then, within this area, the rule corresponding to the disease may be a body temperature greater than 37 degrees. If this rule is not in the rule base, a temporary rule can be added to update the rule base. If this rule is in the rule base, the weight of the cluster where the rule "body temperature greater than 37 degrees" is located can be increased, so that the cluster where this rule is located can be matched more quickly. The weight of the cluster and the weight of the rule within the cluster can also be increased, so that the rule within the cluster can be matched more quickly.

[0150] By applying the data processing method based on test results provided by an embodiment of the present invention, initial test report data from LIS, PACS, and EMR is received; the initial test report data includes at least one of text, numerical values, and images; the images include at least one of electrocardiograms and pathological sections; the text, numerical values, and images in the initial test report data are parsed to extract key medical entities, and the key medical entities are mapped to standardized fields, and target test report data is generated based on the standardized fields, and the target test report data is structured test report data; the target test report is matched with a preset rule set, and a first number of rules are determined from the rule set; the target rule is determined based on the priority of the first number of rules; and the result information corresponding to the target rule is sent to the doctor terminal. Thus, after the test report data is structured, the data is unified, and according to the clustering algorithm, a diagnosis result is automatically given based on the test results, which reduces the dependence on manual analysis, improves the user experience, and also avoids omissions in complex cases.

[0151] Professionals should also be further aware that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0152] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0153] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A data processing method based on inspection results, characterized in that: The method comprises: Receiving initial test report data from a hospital's laboratory information system (LIS), picture archiving and communication system (PACS), or electronic medical record system (EMR); the initial test report data includes at least one of text, numerical values, and images; and the images include at least one of an electrocardiogram and a pathology slide; Parsing the text, numerical values, and images in the initial medical inspection report data to extract key medical entities, mapping the key medical entities to standardized fields, and generating target medical inspection report data based on the standardized fields, wherein the target medical inspection report data is structured medical inspection report data; Matching the target inspection report with a preset rule set, and determining a first number of rules from the rule set; determining a target rule according to the priorities of the first number of rules; The result information corresponding to the target rule is sent to the doctor terminal.

2. The method according to claim 1, characterized in that The parsing of the text, numerical values, and images in the initial test report data, extracting key medical entities, mapping the key medical entities to standardized fields, and generating target test report data based on the standardized fields specifically includes: Clustering the initial inspection report data to extract common structural features; the common structural features include inspection items; When new inspection report data is detected, the text is preprocessed to remove redundant characters and segment semantic units; Determine whether the parsed semantic unit exists in the parsing template library. If not, perform entity recognition on the semantic unit based on a model in the medical field to identify key entities. The key entities include inspection items, numerical values ​​and units, and descriptive text. Perform regular expression matching and standard conversion on inspection items, values ​​and units, and descriptive texts to generate structured first inspection report data; Extracting text descriptions from inspection images using optical character recognition (OCR) technology and processing image features using a convolutional neural network to generate structured numerical indicators. The image feature processing includes identifying abnormal cell morphology from pathological sections and converting the abnormal cell morphology into structured numerical indicators based on predefined standardized fields. Target inspection report data are obtained based on the structured first inspection report data and the structured numerical indicators.

3. The method according to claim 1, characterized in that The target inspection report is matched with the first rule set, wherein the first number of rules matched from the first rule set specifically include: Calculating the overlap ratio of any two rules, and when the overlap ratio is greater than an overlap ratio threshold, classifying the rules into the first rule set; When any two rules point to the same disease, a higher similarity weight is assigned to the two rules, and the rules are classified into the second rule set; Counting the probability of the rules being triggered simultaneously, determining the rules whose probability is greater than a preset probability, and classifying the rules into a third rule set; According to a clustering algorithm, the first rule set, the second rule set, and the third rule set are represented as feature vectors, and clustered to obtain a plurality of clusters; When the target inspection report triggers any cluster, the intersection-and-union ratio of multiple rules in the cluster is calculated; When the target inspection report triggers multiple clusters, the intersection-and-union ratio of each cluster is calculated according to the cluster priority, and the target cluster is determined. Then, in the target cluster, the intersection-and-union ratio of multiple rules within the target cluster is determined; The rule whose intersection-over-union ratio is greater than the intersection-over-union ratio threshold is determined to be one of the first number of rules.

4. The method according to claim 3, characterized in that The calculating of the intersection-over-union ratio of the multiple rules in the cluster specifically includes: According to the formula IoU=number of matching rules / total number of rules associated with diseases, the intersection-over-union ratio of each rule is determined.

5. The method according to claim 3, characterized in that Calculating the intersection-over-union ratio of the multiple rules in the cluster includes: The weighted intersection-over-union (IoU) of each rule is calculated using the formula IoU1=Σ(matching rule weight) / Σ(all association rule weights).

6. The method according to claim 1, characterized in that The clusters include anemia cluster, infection cluster, metabolic disease cluster, and respiratory disease cluster; the conditions of the anemia cluster include HGB and MCV; the conditions of the infection cluster include white blood cells, CRP, and body temperature; the metabolic disease cluster includes blood glucose, lactic acid, and pH, and the respiratory disease cluster includes white blood cells, body temperature, and pH.

7. The method according to claim 1, characterized in that The method then further comprises: Receive the diagnosis results sent by the doctor's terminal; The preset rule set is updated according to the diagnosis result.

8. The method according to claim 1, characterized in that The method further comprises: According to the time information, the target inspection report data before the current time is obtained; the target inspection report data includes the name of the abnormal indicator, the value of the abnormal indicator and the diagnosis result of the abnormal indicator; Determining corresponding rules based on the target inspection report data; According to the rules, the weights of the clusters are updated.

9. The method according to claim 1, characterized in that The method further comprises: According to the current geographical location information, obtaining target inspection and examination report data within the geographical location information range; Determining corresponding rules based on the target inspection report data; According to the rules, the weights of the clusters are updated.

10. The method according to claim 1, characterized in that The method further comprises: Get patient ID; According to the patient ID, retrieve the past history corresponding to the patient ID; the past history includes major diseases, chronic diseases, infectious diseases, surgical history and allergy history; Matching is performed based on the major diseases, chronic diseases, infectious diseases, surgical history, and allergy history, and a preset rule set; When a rule is matched, the target test report data of the patient ID is matched again with the matched rule; When no rule is matched, the target test report data of the patient ID is matched with the rule set.

Images