Medical image report diagnosis modification type judgment method and system

Abstract

The invention provides a medical image report diagnosis modification type judgment method which comprises the following steps: preprocessing a medical image report before modification and a medical image report after modification, and presenting report data before modification and report data after modification in the form of sentence pairs by the preprocessed data; identifying named entities and named entity types in the sentence pairs; pairing the named entities in each sentence pair, and outputting the paired named entities and named entity types; taking a comma or full stop separator as a unit, and judging the modification type of each unit according to a preset logic rule based on whether the paired named entities are the same or not and the named entity types corresponding to the named entities. The invention further discloses a medical image report diagnosis modification type judgment system. According to the method, multiple algorithms are combined and utilized, the image report diagnosis modification types are accurately counted, the rule is clear, the individuation degree in the specific field is higher, computing resource consumption is low, and the speed is higher.

Description

Technical Field

[0001] The present invention relates to the field of medical information, and more specifically, to a method and system for determining a diagnosis modification type of a medical imaging report. Background Art

[0002] In the process of writing imaging reports, the same report will be initially written by junior doctors and reviewed and revised by senior doctors. In order to evaluate the quality of junior doctors' report writing and to check for gaps in their knowledge, it is necessary to first classify the modifications to the diagnostic impression in medical imaging reports. In the prior art, the main text classification technologies are as follows:

[0003] The rule-based text classification method classifies text according to some manually formulated rules. The rules are different due to different scenarios.

[0004] Statistical text classification methods, such as machine learning and deep learning, perform text classification through supervised learning with large amounts of labeled data.

[0005] A text classification method based on the prompt words of a large language model. It mainly refers to directly inputting the prompt words into the large language model to obtain the classification result without fine-tuning or training.

[0006] The rule-based text classification method is not universal and is designed for application scenarios. In addition, as the number of rules increases, the maintainability of the system rules will deteriorate and some conflicts will arise.

[0007] The upper limit of accuracy of statistical text classification methods is relatively higher than that of rule-based methods, and they are more favored by the academic community. However, in the industrial world, they usually rely on a large amount of labeled high-quality data, and require a large amount of data annotation for different tasks, resulting in high application costs. In addition, in the medical field, the long-tail distribution of data is very common, resulting in a serious imbalance in the proportion of different types of data, which makes model training more difficult.

[0008] The text classification method based on the large language model prompt words can use the trained language model for classification, basically does not require manual labeling, and has a relatively low cost. Although the results of answering general questions are closer to those of humans, for professional questions and certain personalized questions, after testing, the accuracy (about 70%) is significantly different from that of the rule-based method (about 95%). In addition, the large language model has high hardware requirements and slow speed. Summary of the invention

[0009] In view of this, the main purpose of the present invention is to provide a method and system for determining the diagnostic modification type of a medical imaging report. The method can automatically analyze the modification type of an imaging report, and can be used as a branch method for evaluating the writing quality of an imaging report, thereby greatly reducing the difficulty of evaluating the quality of the report. It can provide a reference for primary reporting doctors to provide targeted suggestions for checking for omissions and filling gaps, and solves the problems existing in the prior art of poor rule maintainability, high consumption of computing resources, the need to annotate a large amount of data with classification labels, low accuracy, and high maintenance costs.

[0010] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0011] On the one hand, the present invention provides a method for determining the modification type of a medical imaging report diagnosis, comprising: preprocessing a medical imaging report before modification and a medical imaging report after modification, presenting the preprocessed data as sentence pairs to represent the report data before modification and after modification; identifying named entities and named entity types in the sentence pairs; pairing the named entities in each sentence pair, and outputting the paired named entities and named entity types; wherein named entities that fail to match are inserted into a sequence in the original order; taking a comma or period separator as a unit, determining the modification type of each unit according to preset logical rules based on whether the paired named entities are the same and the named entity types corresponding to the named entities; wherein the modification types are missed diagnosis, misdiagnosis, positioning error, unrecognized features, feature recognition errors, diagnostic logic errors, text grammar and format errors, inconsistent diagnostic confidence, inconsistent language habits, degree judgment errors, and clinical communication problems.

[0012] Preferably, the steps of identifying the named entities and named entity types in the sentence pairs include: constructing a named entity dictionary, the named entity dictionary including named entities, named entity types and synonyms corresponding to the named entities; wherein the categories of the named entity dictionary are disease type, imaging feature, location, diagnostic confidence, degree, and clinical communication; based on the named entity dictionary, identifying the named entities and named entity types from the sentence pairs by longest character matching.

[0013] Preferably, the preprocessing of the medical imaging report before and after modification comprises the following steps: identifying sentence boundaries: using a period or a colon as a sentence boundary, breaking the medical imaging report text into short sentences that express complete meanings; adjusting word order: comparing each short sentence before and after modification, if when the short sentence is moved to a corresponding position, the increase in the length of the longest common subsequence before and after diagnosis compared to the length of the short sentence before and after diagnosis reaches a preset value, then it is determined that the short sentence needs to be moved to the corresponding position; sentence pairing: based on the longest common subsequence, using a packaged algorithm to obtain the corresponding positions and corresponding relationships of the short sentences before and after modification, and generating four moving pointers according to the corresponding positions and corresponding relationships, which respectively correspond to the starting position and the end position of the short sentence before modification, and the starting position and the end position of the short sentence after modification, and combining the positions of the short sentence boundaries, continuously updating the four pointers, recording the matched short sentences, and forming the sentence pair.

[0014] Preferably, the step of pairing the named entities in each sentence pair includes: defining an array of named entities corresponding to the short sentence before modification as X, whose number is M, X[i].type represents the entity type of i named entities in X, and X[i].mention represents the entity mention content of i named entities in X; defining an array of named entities corresponding to the modified sentence as Y, whose number is N, Y[i].type represents the entity type of i named entities in Y, and Y[i].mention represents the entity mention content of i named entities in Y; defining a two-dimensional array dp of (M+1)*(N+1) to record the state, dp[i][j] represents the longest common subsequence number of the first i named entities before modification and the first j named entities after modification when the named entities are of the same type to be paired; lcs{X[i].mention,Y[j].mention} represents the longest common subsequence number of the i-th entity and the j-th entity; and defining the state transition equation as: dp[0][j]=0; (0<=j<=N+1)

[0015] dp[i][0]=0; (0<=i<=M+1)

[0016] dp[i][j]=dp[i-1][j-1]+lcs{X[i].mention,Y[j].mention}; (X[i].type==Y[i].type anddp[i][j]>=max{dp[i-1][j],dp[i][j-1]})

[0017] dp[i][j]=max{dp[i][j-1],dp[i-1][j]}; (X[i].type≠Y[i].type or dp[i][j]<

[0018] max{dp[i-1][j],dp[i][j-1]})

[0019] According to the state array, backtrack the named entities that can be matched and choose any combination.

[0020] Preferably, the method also includes: when each of the units includes multiple modification types, the modification type with the highest priority is used as the modification type of the unit; wherein the priority levels are arranged as follows: misdiagnosis> missed diagnosis> feature recognition error> feature not recognized> positioning error> diagnostic logic error> clinical communication problem> degree judgment error> text grammar and format errors> inconsistent diagnostic confidence> inconsistent language habits.

[0021] On the other hand, the present invention also provides a medical imaging report diagnosis modification type judgment system, the system comprising: a data preprocessing module, a data recognition module, a named entity pairing module and a modification type judgment module, wherein the data preprocessing module is connected to the data recognition module, and is used to preprocess the medical imaging report before modification and the medical imaging report after modification, and the preprocessed data presents the report data before modification and the modified report data in the form of sentence pairs; the data recognition module is respectively connected to the data preprocessing module and the named entity pairing module, and is used to identify the named entities and named entity types in the sentence pairs; the named entity pairing module is respectively connected to the data recognition module and the modification type judgment module The modules are connected to pair the named entities in each sentence pair and output the paired named entities and named entity types; wherein the named entities that fail to match are inserted into the sequence in the original order; the modification type judgment module is connected to the named entity pairing module and is used to take a comma or period separator as a unit, and based on whether the paired named entities are the same and the named entity types corresponding to the named entities, judge the modification type of each unit according to preset logical rules; wherein the modification types are missed diagnosis, misdiagnosis, positioning error, unrecognized features, feature recognition errors, diagnostic logic errors, text grammar and format errors, inconsistent diagnostic confidence, inconsistent language habits, degree judgment errors, and clinical communication problems.

[0022] Preferably, the data identification module includes: a dictionary construction unit and an extraction unit, wherein the dictionary construction unit is connected to the extraction unit and is used to construct a named entity dictionary, the named entity dictionary includes named entities, named entity types and synonyms corresponding to named entities; wherein the categories of the named entity dictionary are disease type, imaging features, positioning, diagnostic confidence, degree, and clinical communication; the extraction unit is connected to the dictionary construction unit and is used to identify named entities and named entity types from the sentence pairs by the longest character matching method based on the named entity dictionary.

[0023] Preferably, the data preprocessing module comprises: a sentence boundary recognition unit, a word order adjustment unit and a sentence pairing unit, wherein the sentence boundary recognition unit is connected to the word order adjustment unit, and is used to split the medical imaging report text into short sentences expressing complete meanings with a period or a colon as a sentence boundary; the word order adjustment unit is respectively connected to the sentence boundary recognition unit and the sentence pairing unit, and is used to compare each short sentence before and after the modification, and if when the short sentence is moved to the corresponding position, the increase in the length of the longest common subsequence before and after the diagnosis exceeds the ratio of the length of the longest common subsequence before and after the diagnosis to the length of the moved short sentence reaches a preset value, then it is determined that the short sentence needs to be moved to the corresponding position; the sentence pairing unit is connected to the word order adjustment unit, and is used to obtain the corresponding positions and corresponding relationships of the short sentences before and after the modification based on the longest common subsequence using a packaged algorithm, and generate four moving pointers according to the corresponding positions and corresponding relationships, which respectively correspond to the starting position and the end position of the short sentence before the modification, and the starting position and the end position of the short sentence after the modification, and the four pointers are continuously updated in combination with the short sentence boundary positions to record the matched short sentences to form the sentence pair.

[0024] Preferably, pairing the named entities in each sentence pair includes: defining an array of named entities corresponding to the short sentence before modification as X, whose number is M, X[i].type represents the entity type of i named entities in X, and X[i].mention represents the entity mention content of i named entities in X; defining an array of named entities corresponding to the modified sentence as Y, whose number is N, Y[i].type represents the entity type of i named entities in Y, and Y[i].mention represents the entity mention content of i named entities in Y; defining a two-dimensional array dp of (M+1)*(N+1) to record the state, dp[i][j] represents the longest common subsequence number of the first i named entities before modification and the first j named entities after modification when the named entities are of the same type to be paired; lcs{X[i].mention,Y[j].mention} represents the longest common subsequence number of the i-th entity and the j-th entity; and defining the state transition equation as: dp[0][j]=0; (0<=j<=N+1)

[0025] dp[i][0]=0; (0<=i<=M+1)

[0026] dp[i][j]=dp[i-1][j-1]+lcs{X[i].mention,Y[j].mention}; (X[i].type==Y[i].type anddp[i][j]>=max{dp[i-1][j],dp[i][j-1]})

[0027] dp[i][j]=max{dp[i][j-1],dp[i-1][j]}; (X[i].type≠Y[i].type or dp[i][j]<

[0028] max{dp[i-1][j],dp[i][j-1]})

[0029] According to the state array, backtrack the named entities that can be matched and choose any combination.

[0030] Preferably, when each of the units includes multiple modification types, the system also includes a modification type determination module, which is connected to the modification type judgment module, and is used to use the modification type with the highest priority as the modification type of the unit; wherein the priority levels are ranked as follows: misdiagnosis> missed diagnosis> feature recognition error> feature not recognized> positioning error> diagnostic logic error> clinical communication problem> degree judgment error> text grammar and format errors> inconsistent diagnostic confidence> inconsistent language habits.

[0031] Technical effects of the present invention:

[0032] 1. The method of the present invention preprocesses the medical imaging report before and after modification. The preprocessed data presents the report data before and after modification in the form of sentence pairs, and pairs the named entities in each sentence pair. Based on whether the paired named entities are the same and the named entity types corresponding to the named entities, the modification type of each unit is judged according to preset logical rules. The present invention uses a variety of algorithms to accurately count the types of diagnostic modifications in imaging reports. Compared with the statistical-based text classification method, the rules are clear, the interpretability is stronger, the labor cost is lower, and a large amount of data with classification labels does not need to be annotated. Only the corresponding classification dictionary (including synonyms) needs to be maintained. The data long tail problem is relatively small, and the update and maintenance cost is low. On the classification data of the modification type of the lumbar report that has been analyzed, the comprehensive accuracy is above 95%, making quality evaluation and teaching applications feasible.

[0033] 2. The method of the present invention pre-processes the image report before and after modification, identifies sentence boundaries, breaks the report text into short sentences, adjusts the word order, and processes sentence pairing. By simplifying the complex, a more fine-grained text analysis is performed to reduce the complexity of the task; among them, the sentence pairing processing does not rely on accuracy, but can be completed only through the longest common subsequence at the character level, thereby improving the overall robustness;

[0034] 3. Compared with the text classification method of large language model prompt words, the method of the present invention has a high degree of personalization in specific fields, a high accuracy rate, very little computing resource consumption, and a faster speed; this method can be used as a branch of report writing quality evaluation, which greatly reduces the difficulty of report quality evaluation, and can provide junior reporting doctors with targeted suggestions for finding omissions and filling gaps, and push learning materials in a targeted manner for areas where errors are often made, making the evaluation of the diagnostic level of junior doctors more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0036] Figure 1 A flow chart of a method for determining a medical imaging report diagnosis modification type according to Embodiment 1 of the present invention is shown;

[0037] Figure 2 A schematic diagram showing an example of named entity types - image features in a named entity dictionary in a method for determining a modification type of a medical image report diagnosis according to Embodiment 1 of the present invention is shown;

[0038] Figure 3 A schematic diagram of the structure of a medical imaging report diagnosis modification type determination system according to Embodiment 2 of the present invention is shown;

[0039] Figure 4 A schematic diagram of the structure of a medical imaging report diagnosis modification type determination system according to Embodiment 3 of the present invention is shown;

[0040] Figure 5 A schematic diagram showing an example of named entity types - image features in a named entity dictionary in a medical image report diagnosis modification type determination system according to Embodiment 3 of the present invention is shown;

[0041] Figure 6 A schematic diagram of the structure of a medical imaging report diagnosis modification type determination system according to a fourth embodiment of the present invention is shown;

[0042] Figure 7 A schematic diagram of the structure of a medical imaging report diagnosis modification type determination system according to Embodiment 5 of the present invention is shown. DETAILED DESCRIPTION

[0043] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments.

[0044] Embodiment 1

[0045] The present invention extracts named entities and entity types in sentences by modifying the correspondence between sentences before and after modification, identifies the correspondence between named entities before and after modification, and obtains the final report modification type based on the matched named entities and preset logical rules. The content before and after the image diagnosis modification is compared to classify the modification types. The types mainly include: missed diagnosis, misdiagnosis, diagnostic logic error, location error, inconsistent degree, inconsistent diagnostic confidence, inconsistent language habits, grammatical or spelling errors, unrecognized features, feature recognition errors, etc.

[0046] For example, see the following table:

[0047]

[0048] Figure 1 FIG. 4 is a flow chart of a method for determining a type of modification of a medical imaging report diagnosis according to a first embodiment of the present invention; Figure 1 As shown, the method comprises the following steps:

[0049] Step S101, preprocessing the medical imaging report before and after modification, and presenting the report data before and after modification in the form of sentence pairs;

[0050] The original imaging report is extracted from the hospital's RIS system or PACS system, and the imaging report before and after modification is used as input information of the program.

[0051] The preprocessing of the medical imaging report before and after modification comprises the following steps:

[0052] Identify sentence boundaries: Use periods or colons as sentence boundaries to break the medical imaging report text into short sentences that express complete meanings.

[0053] The technical idea adopted is: since most imaging reports are relatively standardized, in most cases it is more accurate to use periods as the basis for sentence recognition. Currently, we mainly use periods and colons as sentence boundaries.

[0054] There are many methods for sentence boundary identification tasks, such as through other rules or through statistical-based methods, typically deep learning methods.

[0055] Adjust word order: Compare each short sentence before and after the modification. If, when the short sentence is moved to the corresponding position, the increase in the longest common subsequence length before and after the diagnosis reaches a preset value compared with the short sentence before the move, it is determined that the short sentence needs to be moved to the corresponding position;

[0056] Sometimes the modification of the imaging report may be to cut and paste the text. For the accuracy of the subsequent analysis, especially the text matching before and after the modification is greatly affected, it is necessary to restore the cut and paste operation, take the modified diagnosis as the target, and adjust the sentence of the diagnosis before the modification.

[0057] The technical idea is to move the position of each sentence in the diagnosis before modification, taking into account that some content may be modified after operations such as cutting and pasting. If the increase in the longest common subsequence length of the modified diagnosis content compared to the length before the move accounts for a certain proportion of the moved sentences (for example, it is set to 40%), then it is considered that the sentence should be moved. Record the best way to move the sentence last time, move the next sentence, and find the best way to move the sentence while maintaining the original sentence order as much as possible.

[0058] Sentence matching: Based on the longest common subsequence, the packaged algorithm is used to obtain the corresponding positions and corresponding relationships (insertion, deletion, replacement, and sameness) of the short sentences before and after modification. This type of algorithm has specific implementations in many programming languages ​​(such as SequenceMatcher in Python's standard library difflib);

[0059] According to the corresponding positions and the corresponding relationships, four moving pointers are generated, which correspond to the starting position and the ending position of the short sentence before modification, and the starting position and the ending position of the short sentence after modification, respectively. Combined with the boundary positions of the short sentences, the four pointers are continuously updated to record the matched paired short sentences to form the sentence pairs.

[0060] The technical idea adopted is: using dynamic programming algorithm and double pointer algorithm, while ensuring that the current sentence order remains unchanged, to find the matching combination that can make the longest common subsequence of two sentences the longest, so as to find the best correspondence. The sentence pairing step is not a mandatory step, because there is also a named entity pairing step later, but the named entity pairing has a certain dependence on the accuracy of named entity recognition, while the sentence pairing does not have this dependence and can be completed only through the longest common subsequence at the character level, which helps to improve the overall robustness.

[0061] Step S102, identifying the named entities and named entity types in the sentence pair:

[0062] The step of identifying the named entities and the named entity types in the sentence pairs comprises:

[0063] Build a named entity dictionary; the named entity dictionary needs to be continuously updated and maintained.

[0064] The named entity dictionary includes named entities, named entity types, and synonyms corresponding to the named entities;

[0065] The categories of the named entity dictionary are disease type, imaging feature, location, diagnostic confidence, degree, and clinical communication; based on the named entity dictionary, named entities and named entity types are identified from the sentence pairs by longest character matching.

[0066] According to different classification standards, the corresponding named entity type can be adjusted. There are many other methods for named entity recognition, many of which are based on language models.

[0067] Figure 2 FIG. 1 shows an example schematic diagram of named entity types in a named entity dictionary, namely, image features, in a method for determining a modification type of a medical image report diagnosis according to Embodiment 1 of the present invention; Figure 2 As shown, the named entity categories are imaging features, including abnormal spinal cord signal, abnormal disc signal, abnormal vertebral signal, vertebral compression, spinal canal abnormality, vascular variation, intraspinal epidural lipomatosis, etc.

[0068] Step S103, pairing the named entities in each sentence pair, and outputting the paired named entities and named entity types; wherein the unmatched named entities are inserted into the sequence in the original order;

[0069] The technical idea adopted is: through the dynamic programming algorithm, the optimal correspondence between named entities is calculated while ensuring that the named entity types are the same and the character-level common subsequences are the longest.

[0070] The step of pairing the named entities in each sentence pair includes:

[0071] Define the named entity array corresponding to the short sentence before modification as X, whose number is M, X[i].type represents the entity type of the i-named entity in X, and X[i].mention represents the entity mention content of the i-named entity in X; define the named entity array corresponding to the modified sentence as Y, whose number is N, Y[i].type represents the entity type of the i-named entity in Y, and Y[i].mention represents the entity mention content of the i-named entity in Y;

[0072] Define a (M+1)*(N+1) two-dimensional array dp to record the state. dp[i][j] represents the longest common subsequence number of the first i named entities before modification and the first j named entities after modification, when the named entities are of the same type and can be paired. lcs{X[i].mention,Y[j].mention} represents the longest common subsequence number of the i-th entity and the j-th entity. The state transition equation is defined as: dp[0][j] = 0; (0 <= j <= N+1)

[0073] dp[i][0]=0; (0<=i<=M+1)

[0074] dp[i][j]=dp[i-1][j-1]+lcs{X[i].mention,Y[j].mention}; (X[i].type==Y[i].type anddp[i][j]>=max{dp[i-1][j],dp[i][j-1]}0029

[0075] dp[i][j]=max{dp[i][j-1],dp[i-1][j]}; (X[i].type≠Y[i].type or dp[i][j]<

[0076] max{dp[i-1][j],dp[i][j-1]})

[0077] According to the state array, we can trace back the named entities that can be matched. Sometimes we can get more than one matching method. We can choose any combination. We can insert the unmatched entities into the sequence in their original order.

[0078] Step S104, taking a comma or a period separator as a unit, based on whether the paired named entities are the same and the named entity types corresponding to the named entities, the modification type of each unit is determined according to preset logical rules, and finally the modification type of the entire sentence is summarized and output; wherein the modification types are missed diagnosis, misdiagnosis, positioning error, unrecognized features, feature recognition errors, diagnostic logic errors, text grammar and format errors, inconsistent diagnostic confidence, inconsistent language habits, degree judgment errors, and clinical communication problems.

[0079] This classification method is formulated based on medical-related needs. The specific classification method and judgment logic are as follows:

[0080] -Missed diagnosis: Generally, there is no description of the disease type before the modification, but there is a description of the disease type after the modification.

[0081] - Misdiagnosis: The disease types at the same location are different before and after modification, or there are disease types before modification but they are normal after modification.

[0082] - Positioning error: The disease type or characteristics are consistent before and after modification, but the anatomical position or spatial relationship is inconsistent.

[0083] -Feature not recognized: There is no image feature before the corresponding position is modified, but there is an image feature after the modification.

[0084] - Feature recognition error: There is an image feature before the corresponding position is modified but there is no image feature after the modification, or there are affected features before and after the modification but the features are inconsistent.

[0085] -Diagnostic logic error: the importance of disease types is not ranked correctly, and the order of sentences has been adjusted.

[0086] -Text grammar and format errors: Text and format errors, such as typos and punctuation.

[0087] -Inconsistent diagnostic confidence: The words "possible, highly likely, not excluded" or the expression of possibility before and after the modification are inconsistent.

[0088] - Inconsistent language habits: The writing habits before and after the modification are inconsistent, and synonyms are used.

[0089] - Error in judgment of severity: The description of the severity of the disease before and after the modification is inconsistent. The relevant severity descriptions include: mild, slight, slightly, slightly, severe, serious, significant, etc.

[0090] -Clinical communication issues: For example, clinical questions are not answered, the text is normal before modification but denied to be normal or the disease after modification; or there are hints for clinical work, such as the words "contrast", "comparison", "suggestion", etc.

[0091] In order to achieve the above classification, we constructed six types of word dictionaries, including disease type, imaging characteristics, location, diagnostic confidence, degree, and clinical communication, and recorded the named entities in each word dictionary.

[0092] The judgment of missed diagnosis and misdiagnosis depends on the disease category dictionary; positioning error depends on the positioning dictionary; feature failure and feature recognition error depend on the image feature dictionary; inconsistent language habits depend on the synonyms defined in each dictionary; degree judgment errors depend on the degree dictionary; clinical communication problems depend on the clinical communication dictionary; diagnostic logic errors are judged based on whether the word order adjustment in the previous steps occurs.

[0093] In order to deal with the relative duplication of multiple modification categories in some cases and make the classification results closer to people's intuitive feelings, we set certain priority rules according to the importance of categories. In the same minimum classification unit, if a category with a higher priority appears, the category with a lower priority will be overwritten. Specifically:

[0094] When each of the units includes multiple modification types, the modification type with the highest priority will be used as the modification type of the unit; wherein the priority levels are arranged as follows: misdiagnosis> missed diagnosis> feature recognition error> feature not recognized> positioning error> diagnostic logic error> clinical communication problem> degree judgment error> text grammar and format errors> inconsistent diagnostic confidence> inconsistent language habits.

[0095] The following takes the modification type determination of a lumbar vertebra report as an example to illustrate the steps implemented by the method of the present invention:

[0096] -Original Report

[0097]

[0098] - Sentence boundary recognition

[0099] In this example, the period and colon are used as sentence boundaries for splitting. After splitting, each line can basically express a relatively complete meaning.

[0100]

[0101] - Word order adjustment

[0102] According to the results of the previous step, it was found that "L4-5 horizontal spinal canal stenosis" should have been cut to the back, and the lower position was simply modified to "L3-5". This is indeed the case through program calculation. If the sentence is moved to the corresponding position, the length of the longest common subsequence before and after the diagnosis increases by 8 compared to the original. The length of this sentence is 10, accounting for more than 40%, so it is determined that the sentence should be moved; as for moving other sentences, there is no gain in the length of the longest common subsequence, so it is not moved, and the final result is as follows.

[0103]

[0104]

[0105] - Sentence pairing

[0106]

[0107] -Named Entity Recognition

[0108] From the aligned report above, we take a typical sentence for the following analysis.

[0109] In the dictionary of named entity recognition, "L5", "S1", and "right side" are in the positioning dictionary, "intervertebral disc herniation" is in the disease type dictionary, and "nerve root compression" is in the image feature dictionary, so the corresponding named entities can be identified from the text.

[0110]

[0111]

[0112] - Named Entity Pairing

[0113] By calculating the defined algorithm, we can get the following named entity matching results:

[0114]

[0115] -Modify type judgment

[0116] According to the named entity matching results, before and after the modification, there are only two additional entities after the modification, namely positioning and image features. Therefore, the error types are positioning error and feature unrecognition. Since the priority of feature unrecognition is higher than positioning error, the final modification type is feature unrecognition.

[0117]

[0118]

[0119] More examples below:

[0120]

[0121]

[0122] The embodiment of the present invention can preprocess the medical imaging report before modification and the medical imaging report after modification. The preprocessed data presents the report data before modification and the modified report data in the form of sentence pairs, and pairs the named entities in each sentence pair. Based on whether the paired named entities are the same and the named entity types corresponding to the named entities, the modification type of each unit is judged according to preset logical rules. The present invention uses a variety of algorithms to accurately count the types of diagnostic modifications of imaging reports. Compared with the statistical-based text classification method, the rules are clear, the interpretability is stronger, the labor cost is lower, and there is no need to annotate a large amount of data with classification labels. Only the corresponding classification dictionary (including synonyms) needs to be maintained. The data is less troubled by the long tail problem and the update and maintenance cost is low. On the classification data of the modification type of the lumbar report that has been analyzed, the comprehensive accuracy is above 95%, which makes quality evaluation and teaching applications It becomes feasible; the embodiment of the present invention pre-processes the imaging report before and after the modification, identifies the sentence boundaries, breaks the report text into short sentences, adjusts the word order, and processes the sentence pairing. By simplifying the complex, a more fine-grained text analysis is performed to reduce the complexity of the task; wherein, the sentence pairing processing does not rely on accuracy, but can be completed through the longest common subsequence at the character level, thereby improving the overall robustness; compared with the text classification method of the large language model prompt word, the embodiment of the present invention has a high degree of personalization in specific fields, a high accuracy rate, very little computing resource consumption, and a faster speed; this method can be used as a branch of report writing quality evaluation, which greatly reduces the difficulty of report quality evaluation, can provide targeted suggestions for primary reporting doctors to check for omissions and fill in the gaps, and push learning materials in a targeted manner for places where errors are often made, so as to make a more accurate evaluation of the diagnostic level of primary doctors.

[0123] Embodiment 2

[0124] Figure 3 FIG. 4 shows a schematic diagram of the structure of a medical imaging report diagnosis modification type determination system according to Embodiment 2 of the present invention; Figure 3 As shown, the system includes: a data preprocessing module 10, a data identification module 20, a named entity pairing module 30 and a modification type judgment module 40, wherein:

[0125] The data preprocessing module 10 is connected to the data identification module 20 and is used to preprocess the medical imaging report before and after modification, and the preprocessed data presents the report data before and after modification in the form of sentence pairs;

[0126] The original imaging report is extracted from the hospital's RIS system or PACS system, and the imaging report before and after modification is used as input information of the program.

[0127] The data identification module 20 is connected to the data preprocessing module 10 and the named entity pairing module 30 respectively, and is used to identify the named entities and named entity types in the sentence pairs;

[0128] The named entity pairing module 30 is connected to the data identification module 20 and the modification type determination module 40 respectively, and is used to pair the named entities in each sentence pair and output the paired named entities and named entity types; wherein the unmatched named entities are inserted into the sequence according to the original order;

[0129] The pairing of named entities in each sentence pair includes: defining an array of named entities corresponding to the short sentence before modification as X, whose number is M, X[i].type represents the entity type of i named entities in X, and X[i].mention represents the entity mention content of i named entities in X; defining an array of named entities corresponding to the modified sentence as Y, whose number is N, Y[i].type represents the entity type of i named entities in Y, and Y[i].mention represents the entity mention content of i named entities in Y; defining a two-dimensional array dp of (M+1)*(N+1) to record the state, dp[i][j] represents the longest common subsequence number of the first i named entities before modification and the first j named entities after modification when the named entities are of the same type to be paired; lcs{X[i].mention,Y[j].mention} represents the longest common subsequence number of the i-th entity and the j-th entity; and defining the state transition equation as: dp[0][j]=0; (0<=j<=N+1)

[0130] dp[i][0]=0; (0<=i<=M+1)

[0131] dp[i][j]=dp[i-1][j-1]+lcs{X[i].mention,Y[j].mention}; (X[i].type==Y[i].type anddp[i][j]>=max{dp[i-1][j],dp[i][j-1]})

[0132] dp[i][j]=max{dp[i][j-1],dp[i-1][j]}; (X[i].type≠Y[i].type or dp[i][j]<

[0133] max{dp[i-1][j],dp[i][j-1]})

[0134] According to the state array, backtrack the named entities that can be matched and choose any combination.

[0135] The modification type determination module 40 is connected to the named entity pairing module 30 and is used to determine the modification type of each unit according to a preset logical rule based on whether the paired named entities are the same and the named entity types corresponding to the named entities, with a comma or a period separator as a unit;

[0136] Among them, the modification types are missed diagnosis, misdiagnosis, positioning error, unrecognized features, feature recognition error, diagnostic logic error, text grammar and format error, inconsistent diagnostic confidence, inconsistent language habits, degree judgment error, and clinical communication problems.

[0137] This classification method is formulated based on medical-related needs. The specific classification method and judgment logic are as follows:

[0138] -Missed diagnosis: Generally, there is no description of the disease type before the modification, but there is a description of the disease type after the modification.

[0139] - Misdiagnosis: The disease types at the same location are different before and after modification, or there are disease types before modification but they are normal after modification.

[0140] - Positioning error: The disease type or characteristics are consistent before and after modification, but the anatomical position or spatial relationship is inconsistent.

[0141] -Feature not recognized: There is no image feature before the corresponding position is modified, but there is an image feature after the modification.

[0142] - Feature recognition error: There is an image feature before the corresponding position is modified but there is no image feature after the modification, or there are affected features before and after the modification but the features are inconsistent.

[0143] -Diagnostic logic error: the importance of disease types is not ranked correctly, and the order of sentences has been adjusted.

[0144] -Text grammar and format errors: Text and format errors, such as typos and punctuation.

[0145] -Inconsistent diagnostic confidence: The words "possible, highly likely, not excluded" or the expression of possibility before and after the modification are inconsistent.

[0146] - Inconsistent language habits: The writing habits before and after the modification are inconsistent, and synonyms are used.

[0147] - Error in judgment of severity: The description of the severity of the disease before and after the modification is inconsistent. The relevant severity descriptions include: mild, slight, slightly, slightly, severe, serious, significant, etc.

[0148] -Clinical communication issues: For example, clinical questions are not answered, the text is normal before modification but denied to be normal or the disease after modification; or there are hints for clinical work, such as the words "contrast", "comparison", "suggestion", etc.

[0149] In order to achieve the above classification, we constructed six types of word dictionaries, including disease type, imaging characteristics, location, diagnostic confidence, degree, and clinical communication, and recorded the named entities in each word dictionary.

[0150] The judgment of missed diagnosis and misdiagnosis depends on the disease category dictionary; positioning error depends on the positioning dictionary; feature failure and feature recognition error depend on the image feature dictionary; inconsistent language habits depend on the synonyms defined in each dictionary; degree judgment errors depend on the degree dictionary; clinical communication problems depend on the clinical communication dictionary; diagnostic logic errors are judged based on whether the word order adjustment in the previous steps occurs.

[0151] Embodiment 3

[0152] Figure 4 FIG. 4 shows a schematic diagram of the structure of a medical imaging report diagnosis modification type determination system according to Embodiment 3 of the present invention; Figure 4 As shown, the data identification module 20 includes: a dictionary construction unit 202 and an extraction unit 204, wherein:

[0153] The dictionary construction unit 202 is connected to the extraction unit 204 and is used to construct a named entity dictionary, wherein the named entity dictionary includes named entities, named entity types, and synonyms corresponding to the named entities; wherein the categories of the named entity dictionary are disease type, imaging feature, location, diagnostic confidence, degree, and clinical communication;

[0154] The extraction unit 204 is connected to the dictionary construction unit 202, and is used to identify named entities and named entity types from the sentence pairs by longest character matching based on the named entity dictionary.

[0155] According to different classification standards, the corresponding named entity type can be adjusted. There are many other methods for named entity recognition, many of which are based on language models.

[0156] Figure 5 FIG. 1 shows an example schematic diagram of named entity types in a named entity dictionary—image features in a method for determining a modification type of a medical image report diagnosis according to Embodiment 3 of the present invention; Figure 5 As shown, the named entity categories are imaging features, including abnormal spinal cord signal, abnormal disc signal, abnormal vertebral signal, vertebral compression, spinal canal abnormality, vascular variation, intraspinal epidural lipomatosis, etc.

[0157] Embodiment 4

[0158] Figure 6 FIG. 4 shows a schematic diagram of the structure of a medical imaging report diagnosis modification type determination system according to Embodiment 4 of the present invention; Figure 6 As shown, the data preprocessing module 10 includes: a sentence boundary recognition unit 102, a word order adjustment unit 104 and a sentence pairing unit 106, wherein:

[0159] The sentence boundary recognition unit 102 is connected to the word order adjustment unit 104 and is used to separate the medical imaging report text into short sentences each expressing a complete meaning, using a period or a colon as a sentence boundary;

[0160] The technical idea adopted is: since most imaging reports are relatively standardized, in most cases it is more accurate to use periods as the basis for sentence recognition. Currently, we mainly use periods and colons as sentence boundaries.

[0161] There are many methods for sentence boundary identification tasks, such as through other rules or through statistical-based methods, typically deep learning methods.

[0162] The word order adjustment unit 104 is connected to the sentence boundary identification unit 102 and the sentence pairing unit 106 respectively, and is used to compare each short sentence before and after the modification. If when the short sentence is moved to the corresponding position, the increase in the longest common subsequence length before and after the diagnosis accounts for the proportion of the moved short sentence to the moved short sentence reaches a preset value, then it is determined that the short sentence needs to be moved to the corresponding position;

[0163] Sometimes the modification of the imaging report may be to cut and paste the text. For the accuracy of the subsequent analysis, especially the text matching before and after the modification is greatly affected, it is necessary to restore the cut and paste operation, take the modified diagnosis as the target, and adjust the sentence of the diagnosis before the modification.

[0164] The technical idea is to move the position of each sentence in the diagnosis before modification, taking into account that some content may be modified after operations such as cutting and pasting. If the increase in the longest common subsequence length of the modified diagnosis content compared to the length before the move accounts for a certain proportion of the moved sentences (for example, it is set to 40%), then it is considered that the sentence should be moved. Record the best way to move the sentence last time, move the next sentence, and find the best way to move the sentence while maintaining the original sentence order as much as possible.

[0165] The sentence pairing unit 106 is connected to the word order adjustment unit 104 and is used to obtain the corresponding positions and corresponding relationships (insertion, deletion, replacement, sameness) of the short sentences before and after the modification based on the longest common subsequence using a packaged algorithm. Such algorithms are specifically implemented in many programming languages ​​(such as SequenceMatcher in Python's standard library difflib);

[0166] According to the corresponding positions and the corresponding relationships, four moving pointers are generated, which correspond to the starting position and the ending position of the short sentence before modification, and the starting position and the ending position of the short sentence after modification, respectively. Combined with the boundary positions of the short sentences, the four pointers are continuously updated to record the matched paired short sentences to form the sentence pairs.

[0167] The technical idea adopted is: using dynamic programming algorithm and double pointer algorithm, while ensuring that the current sentence order remains unchanged, to find the matching combination that can make the longest common subsequence of two sentences the longest, so as to find the best correspondence. The sentence pairing step is not a mandatory step, because there is also a named entity pairing step later, but the named entity pairing has a certain dependence on the accuracy of named entity recognition, while the sentence pairing does not have this dependence and can be completed only through the longest common subsequence at the character level, which helps to improve the overall robustness.

[0168] Embodiment 5

[0169] In order to deal with the relative duplication of multiple modification categories in some cases and make the classification results closer to people's intuitive feelings, we set certain priority rules according to the importance of categories. In the same minimum classification unit, if a category with a higher priority appears, the category with a lower priority will be overwritten. Specifically:

[0170] Figure 7 FIG. 4 shows a schematic diagram of the structure of a medical imaging report diagnosis modification type determination system according to Embodiment 5 of the present invention. Figure 5 As shown, when each of the units includes multiple modification types, the system also includes a modification type determination module 50, which is connected to the modification type judgment module 40, and is used to use the modification type with the highest priority as the modification type of the unit; wherein the priority levels are arranged as follows: misdiagnosis> missed diagnosis> feature recognition error> feature not recognized> positioning error> diagnostic logic error> clinical communication problem> degree judgment error> text grammar and format errors> inconsistent diagnostic confidence> inconsistent language habits.

[0171] The following takes the modification type judgment of a lumbar spine report as an example to illustrate the implementation of this system:

[0172] -Original Report

[0173]

[0174] - Sentence boundary recognition

[0175] In this example, the period and colon are used as sentence boundaries for splitting. After splitting, each line can basically express a relatively complete meaning.

[0176]

[0177] - Word order adjustment

[0178] According to the results of the previous step, it was found that "L4-5 horizontal spinal canal stenosis" should have been cut to the back, and the lower position was simply modified to "L3-5". This is indeed the case through program calculation. If the sentence is moved to the corresponding position, the length of the longest common subsequence before and after the diagnosis increases by 8 compared to the original. The length of this sentence is 10, accounting for more than 40%, so it is determined that the sentence should be moved; as for moving other sentences, there is no gain in the length of the longest common subsequence, so it is not moved, and the final result is as follows.

[0179]

[0180]

[0181] - Sentence pairing

[0182]

[0183] -Named Entity Recognition

[0184] From the aligned report above, we take a typical sentence for the following analysis.

[0185] In the dictionary of named entity recognition, "L5", "S1", and "right side" are in the positioning dictionary, "intervertebral disc herniation" is in the disease type dictionary, and "nerve root compression" is in the image feature dictionary, so the corresponding named entities can be identified from the text.

[0186]

[0187]

[0188] - Named Entity Pairing

[0189] By calculating the defined algorithm, we can get the following named entity matching results:

[0190]

[0191] -Modify type judgment

[0192] According to the named entity matching results, before and after the modification, there are only two additional entities after the modification, namely positioning and image features. Therefore, the error types are positioning error and feature unrecognition. Since the priority of feature unrecognition is higher than positioning error, the final modification type is feature unrecognition.

[0193]

[0194] More examples below:

[0195]

[0196]

[0197] The embodiment of the present invention can preprocess the medical imaging report before modification and the medical imaging report after modification. The preprocessed data presents the report data before modification and the modified report data in the form of sentence pairs, and pairs the named entities in each sentence pair. Based on whether the paired named entities are the same and the named entity types corresponding to the named entities, the modification type of each unit is judged according to preset logical rules. The present invention uses a variety of algorithms to accurately count the types of diagnostic modifications of imaging reports. Compared with the statistical-based text classification method, the rules are clear, the interpretability is stronger, the labor cost is lower, and there is no need to annotate a large amount of data with classification labels. Only the corresponding classification dictionary (including synonyms) needs to be maintained. The data is less troubled by the long tail problem and the update and maintenance cost is low. On the classification data of the modification type of the lumbar report that has been analyzed, the comprehensive accuracy is above 95%, which makes quality evaluation and teaching applications It becomes feasible; the embodiment of the present invention pre-processes the imaging report before and after the modification, identifies the sentence boundaries, breaks the report text into short sentences, adjusts the word order, and processes the sentence pairing. By simplifying the complex, a more fine-grained text analysis is performed to reduce the complexity of the task; wherein, the sentence pairing processing does not rely on accuracy, but can be completed through the longest common subsequence at the character level, thereby improving the overall robustness; compared with the text classification method of the large language model prompt word, the embodiment of the present invention has a high degree of personalization in specific fields, a high accuracy rate, very little computing resource consumption, and a faster speed; this method can be used as a branch of report writing quality evaluation, which greatly reduces the difficulty of report quality evaluation, can provide targeted suggestions for primary reporting doctors to check for omissions and fill in the gaps, and push learning materials in a targeted manner for places where errors are often made, so as to make a more accurate evaluation of the diagnostic level of primary doctors.

[0198] From the above description, it can be seen that the above-mentioned embodiments of the present invention achieve the following technical effects: the embodiments of the present invention can pre-process the medical imaging report before modification and the medical imaging report after modification, and the pre-processed data presents the report data before modification and after modification in the form of sentence pairs, and pairs the named entities in each sentence pair, and judge the modification type of each unit according to the preset logical rules based on whether the paired named entities are the same and the named entity types corresponding to the named entities; the present invention uses a variety of algorithms to accurately count the types of diagnostic modifications of imaging reports. Compared with the statistical-based text classification method, the rules are clear, the interpretability is stronger, the labor cost is lower, and there is no need to annotate a large amount of data with classification labels. Only the corresponding classification dictionary (including synonyms) needs to be maintained; it is less troubled by the long-tail problem of data, and the update and maintenance cost is low; on the lumbar report modification type classification data that has been analyzed, the comprehensive accuracy is 95 % or more, making quality evaluation and teaching applications feasible; the embodiment of the present invention pre-processes the imaging report before and after modification, identifies the sentence boundaries, splits the report text into short sentences, adjusts the word order, and processes the sentence pairing. By simplifying the complex, a more fine-grained text analysis is performed to reduce the complexity of the task; wherein, the sentence pairing processing does not rely on accuracy, but can be completed by the longest common subsequence at the character level, thereby improving the overall robustness; compared with the text classification method of the large language model prompt word, the embodiment of the present invention has a high degree of personalization in specific fields, a high accuracy rate, very little computing resource consumption, and a faster speed; this method can be used as a branch of report writing quality evaluation, which greatly reduces the difficulty of report quality evaluation, can provide targeted suggestions for primary reporting doctors to check for omissions and fill in the gaps, and push learning materials in a targeted manner for places where mistakes are often made, so as to make a more accurate evaluation of the diagnostic level of primary doctors.

[0199] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0200] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for determining the type of modification of a medical imaging report diagnosis, characterized in that: include: Preprocessing the medical imaging report before and after the modification, and presenting the report data before and after the modification in the form of sentence pairs; identifying named entities and named entity types in the sentence pairs; Pair the named entities in each sentence pair and output the paired named entities and named entity types; among them, the unmatched named entities are inserted into the sequence in the original order; Taking a comma or period separator as a unit, the modification type of each unit is determined according to preset logical rules based on whether the paired named entities are the same and the named entity types corresponding to the named entities; wherein the modification types are missed diagnosis, misdiagnosis, positioning error, unrecognized features, feature recognition errors, diagnostic logic errors, text grammar and format errors, inconsistent diagnostic confidence, inconsistent language habits, degree judgment errors, and clinical communication problems.

2. The method for determining the type of modification of a medical imaging report diagnosis according to claim 1, characterized in that: The step of identifying the named entities and the named entity types in the sentence pairs comprises: Constructing a named entity dictionary, wherein the named entity dictionary includes named entities, named entity types, and synonyms corresponding to the named entities; wherein the categories of the named entity dictionary are disease type, imaging feature, location, diagnostic confidence, degree, and clinical communication; Based on the named entity dictionary, named entities and named entity types are identified from the sentence pairs by longest character matching.

3. The method for determining the type of modification of a medical imaging report diagnosis according to claim 1, characterized in that: The steps of preprocessing the medical imaging report before and after modification include: Identify sentence boundaries: Use periods or colons as sentence boundaries to break the medical imaging report text into short sentences that express complete meanings. Adjust word order: Compare each short sentence before and after the modification. If, when the short sentence is moved to the corresponding position, the increase in the longest common subsequence length before and after the diagnosis reaches a preset value compared with the short sentence before the move, it is determined that the short sentence needs to be moved to the corresponding position; Sentence pairing: Based on the longest common subsequence, the corresponding positions and corresponding relationships of the short sentences before and after the modification are obtained by using the packaged algorithm. According to the corresponding positions and corresponding relationships, four moving pointers are generated, corresponding to the starting position and the ending position of the short sentence before the modification, and the starting position and the ending position of the short sentence after the modification, respectively. Combined with the boundary positions of the short sentences, the four pointers are continuously updated to record the matched paired short sentences to form the sentence pairs.

4. The method for determining the type of modification of a medical imaging report diagnosis according to claim 3, characterized in that: The steps of pairing the named entities in each sentence pair include: Define the named entity array corresponding to the short sentence before modification as X, whose number is M, X[i].type represents the entity type of the i-named entity in X, and X[i].mention represents the entity mention content of the i-named entity in X; define the named entity array corresponding to the modified sentence as Y, whose number is N, Y[i].type represents the entity type of the i-named entity in Y, and Y[i].mention represents the entity mention content of the i-named entity in Y; Define a (M+1)*(N+1) two-bit array dp to record the state, dp[i][j] represents the longest common subsequence number of the first i named entities before modification and the first j named entities after modification, when the named entities are of the same type and can be paired; lcs{X[i].mention,Y[j].mention} represents the longest common subsequence number of the i-th entity and the j-th entity; The state transition equation is defined as: dp[0][j] = 0; (0 <= j <= N+1) dp[i][0]=0; (0<=i<=M+1) dp[i][j]=dp[i-1][j-1]+lcs{X[i].mention,Y[j].mention}; (X[i].type==Y[i].type and dp[i][j]>=max{dp[i-1][j],dp[i][j-1]}) dp[i][j]=max{dp[i][j-1],dp[i-1][j]}; (X[i].type≠Y[i].type or dp[i][j] <max{dp[i-1][j],dp[i][j-1]}) According to the state array, backtrack the named entities that can be matched and choose any combination.

5. The method for determining the type of modification of a medical imaging report diagnosis according to claim 1, characterized in that: The method also includes: when each of the units includes multiple modification types, the modification type with the highest priority is used as the modification type of the unit; wherein the priority levels are arranged as follows: misdiagnosis> missed diagnosis> feature recognition error> feature not recognized> positioning error> diagnostic logic error> clinical communication problem> degree judgment error> text grammar and format errors> inconsistent diagnostic confidence> inconsistent language habits.

6. A medical imaging report diagnosis modification type judgment system, characterized in that: The system includes: a data preprocessing module, a data identification module, a named entity pairing module and a modification type judgment module, wherein: The data preprocessing module is connected to the data identification module and is used to preprocess the medical imaging report before and after modification, and the preprocessed data presents the report data before and after modification in the form of sentence pairs; The data identification module is connected to the data preprocessing module and the named entity pairing module respectively, and is used to identify the named entities and named entity types in the sentence pairs; The named entity pairing module is connected to the data identification module and the modification type judgment module respectively, and is used to pair the named entities in each sentence pair and output the paired named entities and named entity types; wherein the unmatched named entities are inserted into the sequence according to the original order; The modification type judgment module is connected to the named entity pairing module, and is used to use a comma or period separator as a unit, and based on whether the paired named entities are the same and the named entity types corresponding to the named entities, judge the modification type of each unit according to preset logical rules; wherein the modification types are missed diagnosis, misdiagnosis, positioning error, unrecognized features, feature recognition errors, diagnostic logic errors, text grammar and format errors, inconsistent diagnostic confidence, inconsistent language habits, degree judgment errors, and clinical communication problems.

7. The medical imaging report diagnosis modification type determination system according to claim 6, characterized in that: The data identification module includes: a dictionary construction unit and an extraction unit, wherein: The dictionary construction unit is connected to the extraction unit and is used to construct a named entity dictionary, wherein the named entity dictionary includes named entities, named entity types, and synonyms corresponding to the named entities; wherein the categories of the named entity dictionary are disease type, imaging feature, location, diagnostic confidence, degree, and clinical communication; The extraction unit is connected to the dictionary construction unit, and is used to identify named entities and named entity types from the sentence pairs by longest character matching based on the named entity dictionary.

8. The medical imaging report diagnosis modification type determination system according to claim 6, characterized in that: The data preprocessing module includes: a sentence boundary recognition unit, a word order adjustment unit and a sentence pairing unit, wherein: The sentence boundary recognition unit is connected to the word order adjustment unit and is used to separate the medical imaging report text into short sentences each expressing a complete meaning, using a period or a colon as a sentence boundary; The word order adjustment unit is connected to the sentence boundary recognition unit and the sentence pairing unit respectively, and is used to compare each short sentence before and after the modification. If when the short sentence is moved to the corresponding position, the increase in the longest common subsequence length before and after the diagnosis accounts for the proportion of the moved short sentence to the moved short sentence reaches a preset value, then it is determined that the short sentence needs to be moved to the corresponding position; The sentence pairing unit is connected to the word order adjustment unit and is used to obtain the corresponding positions and corresponding relationships of the short sentences before and after the modification based on the longest common subsequence using a packaged algorithm. According to the corresponding positions and the corresponding relationships, four moving pointers are generated, which respectively correspond to the starting position and the ending position of the short sentence before the modification, and the starting position and the ending position of the short sentence after the modification. In combination with the boundary positions of the short sentences, the four pointers are continuously updated to record the matched paired short sentences to form the sentence pairs.

9. The medical imaging report diagnosis modification type determination system according to claim 8, characterized in that: Pairing the named entities in each sentence pair includes: Define the named entity array corresponding to the short sentence before modification as X, whose number is M, X[i].type represents the entity type of the i-named entity in X, and X[i].mention represents the entity mention content of the i-named entity in X; define the named entity array corresponding to the modified sentence as Y, whose number is N, Y[i].type represents the entity type of the i-named entity in Y, and Y[i].mention represents the entity mention content of the i-named entity in Y; Define a (M+1)*(N+1) two-bit array dp to record the state, dp[i][j] represents the longest common subsequence number of the first i named entities before modification and the first j named entities after modification, when the named entities are of the same type and can be paired; lcs{X[i].mention,Y[j].mention} represents the longest common subsequence number of the i-th entity and the j-th entity; The state transition equation is defined as: dp[0][j] = 0; (0 <= j <= N+1) dp[i][0]=0; (0<=i<=M+1) dp[i][j]=dp[i-1][j-1]+lcs{X[i].mention,Y[j].mention}; (X[i].type==Y[i].type and dp[i][j]>=max{dp[i-1][j],dp[i][j-1]}) dp[i][j]=max{dp[i][j-1],dp[i-1][j]}; (X[i].type≠Y[i].type or dp[i][j] <max{dp[i-1][j],dp[i][j-1]}) According to the state array, backtrack the named entities that can be matched and choose any combination.

10. The medical imaging report diagnosis modification type determination system according to claim 6, characterized in that: When each of the units includes multiple modification types, the system also includes a modification type determination module, which is connected to the modification type judgment module, and is used to use the modification type with the highest priority as the modification type of the unit; wherein the priority levels are arranged as follows: misdiagnosis> missed diagnosis> feature recognition error> feature not recognized> positioning error> diagnostic logic error> clinical communication problem> degree judgment error> text grammar and format errors> inconsistent diagnostic confidence> inconsistent language habits.

Images