Ultrasonic examination report auxiliary generation method, electronic device and storage medium

Abstract

The present application relates to the field of ultrasound examination report generation, and specifically to an ultrasound examination report auxiliary generation method, electronic device and storage medium, the method comprising: collecting ultrasound examination reports, obtaining structural data and diagnostic data; determining the data universality of ultrasound examination reports, obtaining the information contribution of ultrasound examination reports; dividing ultrasound examination reports into examination report clusters, obtaining the data quality excellence of examination report clusters; determining a training data set based on the data quality excellence of all examination report clusters to train a neural network model, obtaining an ultrasound examination report auxiliary generation model, and obtaining an ultrasound examination report auxiliary generation result based on the ultrasound examination report auxiliary generation model. The present application solves the problem of information redundancy corresponding to ultrasound examination reports, which leads to unreasonable auxiliary generation diagnostic suggestions of ultrasound examination report auxiliary generation models.

Description

Technical Field

[0001] The present application relates to the field of ultrasonic examination report generation, and in particular to an ultrasonic examination report auxiliary generation method, electronic equipment and storage medium. Background Art

[0002] Ultrasound examination is a non-invasive examination and an important means of diagnosing hepatobiliary surgical diseases. Hepatobiliary surgical ultrasound examination forms images through the physical characteristics of high-frequency sound waves such as reflection and scattering in human tissues. The structure, morphology and functional status of the hepatobiliary system can be observed through the formed images, and the ultrasound examination report can be output in combination with hospital database technology. The popularization of hospital database technology provides a training basis for neural network models for the auxiliary generation of ultrasound examination reports, and neural network models can be trained to obtain ultrasound examination report auxiliary generation models.

[0003] In order to obtain the ultrasound examination report auxiliary generation model, the ultrasound examination report needs to be used to train the neural network model. However, in the process of training the neural network model, the generalization ability of the neural network model is affected by the training data set composed of ultrasound examination reports. When the data in the training data set is redundant, the neural network model training process will not be able to accurately learn the diverse characteristics of pathology, which will limit the generalization ability of the ultrasound examination report auxiliary generation model and lead to unreasonable diagnostic recommendations generated by the ultrasound examination report auxiliary generation model. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of this application is to provide an ultrasonic examination report auxiliary generation method, electronic equipment and storage medium. The technical solutions adopted are as follows:

[0005] In a first aspect, an embodiment of the present application provides a method for assisting in generating an ultrasonic examination report, the method comprising the following steps:

[0006] Collecting a first preset number of ultrasound examination reports and extracting text data, and obtaining structural data and diagnostic data of each ultrasound examination report;

[0007] Any ultrasound examination report is recorded as a target ultrasound examination report. The data universality of the target ultrasound examination report is determined based on the difference between the structural data of the target ultrasound examination report and all other ultrasound examination reports. The information contribution of the target ultrasound examination report is obtained by combining the similarity between the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports and the difference in the number of data included in the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports.

[0008] The ultrasound examination reports are divided into examination report clusters according to their information contribution, and the cluster centers of the examination report clusters are obtained. The data quality of the examination report clusters is obtained according to the differences between all information contribution degrees contained in the examination report clusters and the distribution of all information contribution degrees contained in the examination report clusters.

[0009] According to the data quality of all inspection report clusters, a training data set is determined to train the neural network model, an ultrasound inspection report auxiliary generation model is obtained, and an ultrasound inspection report auxiliary generation result is obtained according to the ultrasound inspection report auxiliary generation model.

[0010] Furthermore, the method of determining the data universality of the target ultrasound examination report includes:

[0011] The Levenshtein distance between the structural data of the target ultrasound examination report and any other ultrasound examination report is recorded as the text data difference between the target ultrasound examination report and any other ultrasound examination report;

[0012] The cumulative sum of the text data differences between the target ultrasound examination report and all other ultrasound examination reports is recorded as the comprehensive text data difference of the target ultrasound examination report;

[0013] The reciprocal of the difference in comprehensive text data of the target ultrasound examination report was recorded as the data prevalence of the target ultrasound examination report.

[0014] Furthermore, the information contribution of the target ultrasound examination report is obtained by combining the similarity between the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports, and the difference in the number of data included in the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports, including the specific method of:

[0015] The Simhash value between the diagnostic data of the target ultrasound examination report and any other ultrasound examination report is recorded as the text data duplication degree between the target ultrasound examination report and any other ultrasound examination report; the cumulative sum of the text data duplication degrees of the target ultrasound examination report and all other ultrasound examination reports is recorded as the comprehensive text data duplication degree of the target ultrasound examination report;

[0016] The difference between the amount of data included in the diagnostic data of the target ultrasound examination report and any other ultrasound examination report is recorded as the text data volume difference between the target ultrasound examination report and any other ultrasound examination report; the cumulative sum of the text data volume differences between the target ultrasound examination report and all other ultrasound examination reports is recorded as the comprehensive text data volume difference of the target ultrasound examination report;

[0017] According to the repetition degree of the comprehensive text data and the difference in the amount of the comprehensive text data of the target ultrasound examination report, the data information diversity of the target ultrasound examination report is obtained, and the data information diversity of the target ultrasound examination report is positively correlated with the repetition degree of the comprehensive text data and the difference in the amount of the comprehensive text data of the target ultrasound examination report;

[0018] According to the data universality and data information diversity of the target ultrasound examination report, the information contribution of the target ultrasound examination report is obtained.

[0019] Furthermore, the information contribution of the target ultrasound examination report is obtained according to the data universality and data information diversity of the target ultrasound examination report. The specific process of obtaining the information contribution is as follows:

[0020] The linear normalized value of the product of the data universality and the data information diversity of the target ultrasound examination report is recorded as the information contribution of the target ultrasound examination report.

[0021] Furthermore, the method of dividing the ultrasound examination reports into examination report clusters according to the information contribution of the ultrasound examination reports and obtaining the cluster centers of the examination report clusters includes the following specific methods:

[0022] The absolute value of the difference between the information contributions of two ultrasound examination reports is used as the distance between the two information contributions, and the information contributions of all ultrasound examination reports are clustered to obtain the examination report clustering clusters and the cluster centers of the examination report clustering clusters.

[0023] Furthermore, the specific method of obtaining the data quality of the inspection report clusters includes:

[0024] The mean value of all information contributions contained in the inspection report cluster is recorded as the information contribution mean of the inspection report cluster;

[0025] The cumulative sum of the Euclidean distances between all information contributions contained in the inspection report clustering cluster and the cluster center of the inspection report clustering cluster is recorded as the first difference of the information contribution of the inspection report clustering cluster, the information entropy of all information contributions contained in the inspection report clustering cluster is recorded as the second difference of the information contribution of the inspection report clustering cluster, and the product of the first difference and the second difference of the information contribution of the inspection report clustering cluster is recorded as the comprehensive difference of the information contribution of the inspection report clustering cluster;

[0026] According to the information contribution mean and comprehensive difference of information contribution of inspection report clusters, the data quality of inspection report clusters is obtained.

[0027] Furthermore, the data quality of the inspection report clusters is obtained according to the information contribution mean and the comprehensive difference of the information contribution of the inspection report clusters, including the specific method of:

[0028] The ratio of the mean information contribution of the inspection report cluster to the comprehensive difference in information contribution is recorded as the data quality of the inspection report cluster.

[0029] Further, the training data set is determined according to the data quality of all inspection report clusters to train the neural network model, the ultrasound inspection report auxiliary generation model is obtained, and the ultrasound inspection report auxiliary generation result is obtained according to the ultrasound inspection report auxiliary generation model, including the specific method of:

[0030] The ultrasound examination report corresponding to the examination report cluster with the highest data quality is used as a training data set, and the VisionLLM multimodal large language model is trained using the training data set to obtain the trained VisionLLM multimodal large language model, and the trained VisionLLM multimodal large language model is recorded as an ultrasound examination report auxiliary generation model;

[0031] The ultrasound examination report is input into the ultrasound examination report auxiliary generation model to obtain the ultrasound examination report auxiliary generation result.

[0032] In a second aspect, an embodiment of the present application further provides an electronic device for assisting in the generation of ultrasound examination reports, the electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of any one of the above-mentioned methods when executing the computer program.

[0033] In a third aspect, an embodiment of the present application provides an auxiliary generation storage medium for ultrasound examination reports, wherein the auxiliary generation storage medium includes: a text data extraction module, a text data quality evaluation module, a text data set quality evaluation module, and an auxiliary generation module.

[0034] A text data extraction module, used to collect a first preset number of ultrasound examination reports and extract text data, and obtain structural data and diagnostic data of each ultrasound examination report;

[0035] A text data quality evaluation module is used to record any ultrasound examination report as a target ultrasound examination report, determine the data universality of the target ultrasound examination report based on the difference between the structural data of the target ultrasound examination report and all other ultrasound examination reports, and obtain the information contribution of the target ultrasound examination report by combining the similarity between the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports, and the difference in the number of data included in the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports;

[0036] The text data set quality evaluation module is used to divide the ultrasound examination reports into examination report clusters according to the information contribution of the ultrasound examination reports, and obtain the cluster center of the examination report cluster, and obtain the data quality of the examination report cluster according to the difference between all the information contribution contained in the examination report cluster and the distribution of all the information contribution contained in the examination report cluster;

[0037] The auxiliary generation module is used to determine the training data set to train the neural network model according to the data quality of all inspection report clusters, obtain the ultrasound inspection report auxiliary generation model, and obtain the ultrasound inspection report auxiliary generation result according to the ultrasound inspection report auxiliary generation model.

[0038] It can be seen from the above embodiments that the ultrasonic examination report auxiliary generation method, electronic device and storage medium provided in the embodiments of the present application have at least the following beneficial effects:

[0039] This application analyzes the text data extracted from each ultrasound examination report. First, considering that the stronger the consistency of the structural description text of the ultrasound image, the higher the similarity of the organ structure and function examination results between patients, and the more significant the universality of the patient's ultrasound examination report, based on the ultrasound examination reports with more significant universality as the training data set, the ultrasound examination report auxiliary generation model has stronger robustness, the universality of the ultrasound examination report is evaluated, and the data universality of the ultrasound examination report is obtained. Then, combining the ultrasound examination reports with obvious differences between the diagnostic data as the training data set, the ultrasound examination report auxiliary generation model has stronger learning ability and adaptability to pathological features, and the information contribution of the target ultrasound examination report is obtained. The information contribution of the target ultrasound examination report is an evaluation of the quality of the target ultrasound examination report, which improves the learning ability of the acquired ultrasound examination report auxiliary generation model on pathological features. learning ability and adaptability; further, the ultrasound examination reports with consistent robustness, learning ability and adaptability of the ultrasound examination report auxiliary generation model, pathological characteristics are divided into the same examination report cluster, and the quality of the examination report cluster is evaluated according to the stability of the quality of the ultrasound examination reports contained in the examination report cluster and the quality evaluation of the ultrasound examination reports contained in the examination report cluster, so as to obtain the data quality quality of the examination report cluster; finally, according to the data quality quality of all examination report clusters, the training data set is determined to train the neural network model, and the ultrasound examination report auxiliary generation model is obtained, and then the ultrasound examination report auxiliary generation result is obtained, and the influence of the ultrasound examination report with redundant information as the training data set on the ultrasound examination report auxiliary generation model is eliminated, so as to solve the problem that the information redundancy corresponding to the ultrasound examination report leads to unreasonable auxiliary generation diagnosis suggestions of the ultrasound examination report auxiliary generation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0041] Figure 1 A flowchart of a method for assisting in generating an ultrasonic examination report provided in one embodiment of the present application;

[0042] Figure 2 A flowchart for obtaining information contribution provided for an embodiment of the present application;

[0043] Figure 3 A flowchart for obtaining data quality provided by an embodiment of the present application;

[0044] Figure 4 A schematic diagram of the structure of a storage medium for assisting in the generation of an ultrasonic inspection report provided in one embodiment of the present application. DETAILED DESCRIPTION

[0045] In order to further explain the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the following is a detailed description of the ultrasonic inspection report auxiliary generation method, electronic device and storage medium proposed in the present application, its specific implementation method, structure, features and effects in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0046] Unless otherwise specified and limited, terms such as "comprises", "comprising" or any other variants thereof are intended to cover non-exclusive inclusion, so that a circuit structure, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such articles or devices. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the article or device including the element. In addition, the term "and\or" used herein includes any and all combinations of one or more related listed items. All technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of this application.

[0047] The following is a detailed description of a method for assisting in generating an ultrasonic inspection report, an electronic device, and a specific solution of a storage medium provided by the present application in conjunction with the accompanying drawings.

[0048] See also Figure 1 , which shows a flowchart of a method for assisting in generating an ultrasonic examination report provided by an embodiment of the present application, the method comprising the following steps:

[0049] S001: Collect a first preset number of ultrasound examination reports and extract text data to obtain structural data and diagnostic data of each ultrasound examination report.

[0050] The first preset value ultrasound examination report of the hepatobiliary surgery department is retrieved from the hospital database. The ultrasound examination report includes the ultrasound image of the liver, the ultrasound image of the gallbladder, the text describing the structure of the ultrasound image, and the text of the doctor's diagnosis opinion.

[0051] Preferably, in an embodiment of the present application, when retrieving ultrasound examination reports, the number of ultrasound examination reports retrieved in this embodiment is 10,000, that is, the value of the first preset value in this embodiment is 10,000. In actual application, as other implementation methods, the implementer can decide the value of the first preset value according to actual conditions, and this application does not impose any special restrictions.

[0052] It should be noted that for each patient, if the patient has multiple ultrasound examination reports, only the latest ultrasound examination report corresponding to the same patient will be retrieved.

[0053] Extract text data from each retrieved ultrasound examination report, including the structural description text of the ultrasound image and the doctor's diagnosis text. Perform data cleaning and entity recognition on the extracted text data to obtain structural data and diagnosis data.

[0054] Among them, the structural data is a long entity corresponding to the structural description text of the ultrasound image, and the content included includes the size of the liver and gallbladder, the shape of the liver and gallbladder, the visual features of the boundaries of the liver and gallbladder, and the inner diameter of the blood vessels, etc. The diagnostic data is a long entity corresponding to the doctor's diagnostic opinion text on the ultrasound image, and the content included includes the preliminary diagnostic opinion proposed by the doctor based on the ultrasound image.

[0055] Preferably, in one embodiment of the present application, data cleaning of the extracted text data includes cleaning the extracted text data of punctuation marks, invalid characters and stop words. This embodiment uses named entity recognition (NER) technology to identify entity words of the text data after data cleaning. Named entity recognition (NER) technology is a well-known technology and will not be described in detail. In actual application, as other implementation methods, implementers can choose other entity recognition algorithms according to actual conditions, and this application does not impose any special restrictions.

[0056] At this point, the structural data and diagnostic data of each ultrasound examination report were obtained.

[0057] S002: Record any ultrasound examination report as a target ultrasound examination report, determine the data universality of the target ultrasound examination report based on the difference between the structural data of the target ultrasound examination report and all other ultrasound examination reports, and obtain the information contribution of the target ultrasound examination report by combining the similarity between the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports, as well as the difference in the number of data contained in the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports.

[0058] The more consistent the structural description texts of ultrasound images in ultrasound examination reports of different patients are, the higher the similarity of the organ structures and functional examination results between patients is. At this time, the greater the similarity of the pathological conditions corresponding to the two patients is, the more significant the universality of the patients' ultrasound examination reports is. At this time, the patient's ultrasound examination report is used as part of the training data set to train the neural network model to obtain the ultrasound examination report auxiliary generation model, and the obtained ultrasound examination report auxiliary generation model is more robust.

[0059] Based on the differences between the structural data of different ultrasound examination reports, the data universality of each ultrasound examination report was determined.

[0060] Any ultrasound examination report is recorded as a target ultrasound examination report. This embodiment is described by taking any ultrasound examination report as an example.

[0061] Preferably, as an embodiment of the present application, the Levenshtein distance between the structural data of the target ultrasound examination report and any other ultrasound examination report is recorded as the text data difference between the target ultrasound examination report and any other ultrasound examination report; the cumulative sum of the text data differences between the target ultrasound examination report and all other ultrasound examination reports is recorded as the comprehensive text data difference of the target ultrasound examination report; and the inverse of the comprehensive text data difference of the target ultrasound examination report is recorded as the data universality of the target ultrasound examination report.

[0062] In the process of calculating the ratio, in order to avoid the situation where the denominator is zero, a preset value needs to be added to the denominator, and an embodiment of the preset value is 1.

[0063] Calculating the Levenshtein distance between text data is a well-known technique and will not be described in detail.

[0064] The data universality of any ultrasound examination report can be obtained in the same way, that is, for any ultrasound examination report retrieved, there is a corresponding data universality.

[0065] Specifically, the calculation formula for the data prevalence of ultrasound examination reports is:

[0066]

[0067] in, Indicates the prevalence of data from ultrasound examination reports; Indicates Structural data of an ultrasound examination report; Indicates Structural data of an ultrasound examination report; Indicates Structural data of the first ultrasound examination report and the The Levenshtein distance between the structural data of the two ultrasound examination reports; Indicates The number of ultrasound reports that were different; It represents the second preset value, which is used to prevent the denominator from being 0. In this embodiment, the second preset value is 1.

[0068] The smaller the difference between the structural data of the target ultrasound examination report and all other ultrasound examination reports, the more similar the hepatobiliary pathology of the patient corresponding to the target ultrasound examination report is to the patients corresponding to all other ultrasound examination reports, the more significant the universality of the target ultrasound examination report, and the more robust the ultrasound examination report-assisted generation model obtained by using the target ultrasound examination report as part of the training data set to train the neural network model. At this time, the data universality of the target ultrasound examination report is greater.

[0069] When the structural data of ultrasound examination reports are similar, the diagnostic data in ultrasound examination reports with similar structural data may be quite different due to factors such as the patient's pathology, the patient's personal physical fitness, and the cause of the disease. Therefore, while paying attention to the similarity of the patient's pathological conditions corresponding to the ultrasound examination reports, it is also necessary to consider the differences between the diagnostic data of the ultrasound examination reports. At the same time, considering that the neural network model is trained using ultrasound examination reports with differences as part of the training data set, the obtained ultrasound examination report auxiliary generation model has a stronger learning ability and adaptability to pathological characteristics, and the differences between the diagnostic data of ultrasound examination reports are evaluated to improve the learning ability and adaptability of the obtained ultrasound examination report auxiliary generation model to pathological characteristics.

[0070] According to the similarities between the diagnostic data of different ultrasound examination reports and the differences in the number of data contained in the diagnostic data of different ultrasound examination reports, the data information diversity of each ultrasound examination report is determined.

[0071] The Simhash text duplication checking algorithm is used to process the diagnostic data of all ultrasound examination reports to obtain the Simhash value between the diagnostic data of every two different ultrasound examination reports. It is a known technology to obtain the Simhash value by using the Simhash text duplication checking algorithm, and it will not be repeated here. The Simhash value is an evaluation of the similarity between text data.

[0072] The Simhash value between the diagnostic data of the target ultrasound examination report and any other ultrasound examination report is recorded as the text data duplication degree between the target ultrasound examination report and any other ultrasound examination report; the cumulative sum of the text data duplication degrees between the target ultrasound examination report and all other ultrasound examination reports is recorded as the comprehensive text data duplication degree of the target ultrasound examination report. The difference in the number of data contained in the diagnostic data of the target ultrasound examination report and any other ultrasound examination report is recorded as the text data volume difference between the target ultrasound examination report and any other ultrasound examination report; the cumulative sum of the text data volume difference between the target ultrasound examination report and all other ultrasound examination reports is recorded as the comprehensive text data volume difference of the target ultrasound examination report.

[0073] The data information diversity of the target ultrasound examination report is determined according to the comprehensive text data repetition and the comprehensive text data volume difference of the target ultrasound examination report. The data information diversity of the target ultrasound examination report is positively correlated with the comprehensive text data repetition and the comprehensive text data volume difference of the target ultrasound examination report.

[0074] Preferably, as an embodiment of the present application, a power with a natural constant as the base and the difference in the amount of comprehensive text data of the target ultrasound examination report as the exponent is recorded as the first exponential power of the target ultrasound examination report, and the product of the first exponential power of the target ultrasound examination report and the repetition degree of the comprehensive text data is recorded as the data information diversity of the target ultrasound examination report.

[0075] Specifically, the calculation formula for the diversity of data information in ultrasound examination reports is:

[0076]

[0077] In the formula, Indicates The diversity of data information in each ultrasound examination report; represents a natural constant; Indicates The difference in the amount of comprehensive text data in ultrasound examination reports; Indicates The duplication of comprehensive text data of ultrasound examination reports; Indicates The first exponential power of an ultrasound examination report.

[0078] When the difference between the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports is greater, the diagnostic data contained in the target ultrasound examination report is richer in information. When the neural network model is trained using the target ultrasound examination report as part of the training data set, the diversity of the information features contained in the training data set is more significant, and the acquired ultrasound examination report auxiliary generation model has stronger learning ability and adaptability to pathological characteristics. At this time, the data information diversity of the target ultrasound examination report is greater.

[0079] The data information diversity of any ultrasound examination report can be obtained in the same way, that is, for any ultrasound examination report retrieved, there is a corresponding data information diversity.

[0080] In actual application, as another implementation method, the product of the repetition of the comprehensive text data of the target ultrasound examination report and the difference in the amount of comprehensive text data is recorded as the data information diversity of the target ultrasound examination report.

[0081] According to the data universality and data information diversity of the target ultrasound examination report, the information contribution of the target ultrasound examination report is obtained.

[0082] Preferably, as an embodiment of the present application, the linear normalized value of the product of the data universality and the data information diversity of the target ultrasound examination report is recorded as the information contribution of the target ultrasound examination report.

[0083] Specifically, the calculation formula for the information contribution of the ultrasound examination report is:

[0084]

[0085] in, Indicates Information contribution of each ultrasound examination report; Indicates the prevalence of data from ultrasound examination reports; Indicates The diversity of data information in each ultrasound examination report;

[0086] When the data universality and data information diversity of the target ultrasound examination report are greater, the universality of the target ultrasound examination report is more significant, the diagnostic data information contained in the target ultrasound examination report is richer, and the neural network model is trained using the target ultrasound examination report as part of the training data set. The obtained ultrasound examination report auxiliary generation model is more robust, and has stronger learning ability and adaptability to pathological characteristics. Therefore, the target ultrasound examination report should be selected as part of the training data set. At this time, the information contribution of the target ultrasound examination report is greater.

[0087] The information contribution of any ultrasound examination report can be obtained in the same way. That is to say, for any ultrasound examination report retrieved, there is a corresponding information contribution.

[0088] At this point, the information contribution of all ultrasound examination reports is obtained. The information contribution acquisition flow chart is as follows: Figure 2 shown.

[0089] S003: Divide the ultrasound examination reports into examination report clusters according to their information contribution, and obtain the cluster center of the examination report cluster. According to the difference between all information contribution contained in the examination report cluster and the distribution of all information contribution contained in the examination report cluster, obtain the data quality of the examination report cluster.

[0090] The ultrasound examination reports are divided into examination report clusters according to their information contribution, and the cluster centers of the examination report clusters are obtained.

[0091] Specifically, the absolute value of the difference between the information contribution of two ultrasound examination reports is used as the distance between the two information contributions, and the information contribution of all ultrasound examination reports is clustered using the K-mediods clustering algorithm to obtain the examination report clustering clusters, and obtain the cluster center of the examination report clustering clusters. In this embodiment, the number of examination report clustering clusters is set to 10, and clustering using the K-mediods clustering algorithm is a well-known technology and will not be described in detail.

[0092] The ultrasound examination reports of the same examination report cluster were used as training data sets to train the neural network model. The robustness of the ultrasound examination report-assisted generation model obtained, as well as the learning ability and adaptability of pathological features were consistent.

[0093] Any inspection report cluster is recorded as a target inspection report cluster. This embodiment takes any inspection report cluster as an example for description.

[0094] According to the differences between the contributions of all information contained in the target inspection report clusters and the numerical distribution of the contributions of all information contained in the target inspection report clusters relative to the cluster centers of the target inspection report clusters, the data quality of the target inspection report clusters is obtained.

[0095] The mean of all information contributions contained in the target inspection report cluster is recorded as the mean information contribution of the target inspection report cluster; the sum of the Euclidean distances between all information contributions contained in the target inspection report cluster and the cluster center of the target inspection report cluster is recorded as the first difference of the information contribution of the target inspection report cluster; the information entropy of all information contributions contained in the target inspection report cluster is recorded as the second difference of the information contribution of the target inspection report cluster; the product of the first difference and the second difference of the information contribution of the target inspection report cluster is recorded as the comprehensive difference of the information contribution of the target inspection report cluster; the ratio of the mean information contribution of the target inspection report cluster to the comprehensive difference of the information contribution is recorded as the data quality excellence of the target inspection report cluster.

[0096] When the ultrasound examination reports of the target examination report clustering cluster are used as the training data set to train the neural network model, the stronger the robustness of the ultrasound examination report auxiliary generation model obtained, the stronger the learning ability and adaptability of the pathological characteristics, the greater the mean value of the information contribution of the target examination report clustering cluster, the higher the quality of the ultrasound examination reports contained in the target examination report clustering cluster, and at this time, the data quality of the target examination report clustering cluster is higher. When the difference between the information contribution of the ultrasound examination reports contained in the target examination report clustering cluster is smaller, the first difference, the second difference and the comprehensive difference of information contribution of the target examination report clustering cluster are smaller, the quality of the ultrasound examination reports contained in the target examination report clustering cluster is more stable, and the quality of the ultrasound examination reports contained in the target examination report clustering cluster is higher. At this time, the data quality of the target examination report clustering cluster is higher.

[0097] The data quality acquisition flow chart is as follows Figure 3 shown.

[0098] The data quality excellence of any inspection report cluster can be obtained in the same way. That is, for any inspection report cluster obtained, there is a corresponding data quality excellence.

[0099] At this point, the data quality of each inspection report cluster is obtained.

[0100] S004: Determine a training data set to train a neural network model based on the data quality of all inspection report clusters, obtain an ultrasound inspection report auxiliary generation model, and obtain an ultrasound inspection report auxiliary generation result based on the ultrasound inspection report auxiliary generation model.

[0101] The ultrasound examination reports corresponding to the inspection report cluster with the highest data quality were taken as the training data set, and the VisionLLM multimodal large language model was trained using the training data set. All evaluation indicators in all ultrasound examination reports in the training data set were labeled using the LabelImg target detection and annotation tool. The low-rank adaptation (LoRA) method was used for optimization training, and the LoRA level was set to 64. The loss function consisted of the focalloss loss function, the center point regression loss function, and the cross entropy loss function. The trained VisionLLM multimodal large language model was obtained, and the trained VisionLLM multimodal large language model was recorded as the ultrasound examination report auxiliary generation model.

[0102] Among them, the evaluation indicators include the size of the liver, the regularity of its morphology, the uniformity of the echo in the liver area, the clarity of the blood vessel direction, the inner diameter of the portal vein, the size of the gallbladder, the roughness of the gallbladder wall and the sound transmission effect; the annotated labels include no abnormality, slight abnormality and severe abnormality; the VisionLLM multimodal large language model consists of two model components: a language-guided image segmentor and an open task decoder based on LLM.

[0103] The VisionLLM multimodal large language model and the training process of the VisionLLM multimodal large language model are well-known technologies and will not be described in detail.

[0104] By inputting the ultrasound examination report into the ultrasound examination report auxiliary generation model, auxiliary diagnosis suggestions of the ultrasound examination report can be obtained. The auxiliary diagnosis suggestions of the ultrasound examination report are the auxiliary generation results of the ultrasound examination report.

[0105] At this point, the auxiliary generation of the ultrasound examination report is completed.

[0106] An embodiment of the present application also provides an electronic device for assisting in generating an ultrasonic inspection report, the electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, and when the processor executes the computer program, the steps of any one of the above-mentioned methods for assisting in generating an ultrasonic inspection report are implemented.

[0107] Based on the same inventive concept as the above method, the present application embodiment also provides an ultrasonic inspection report auxiliary generation storage medium, see Figure 4 , Figure 4 1 is a schematic diagram of the structure of an ultrasonic inspection report auxiliary generation storage medium provided by an embodiment of the present application. In this embodiment, the auxiliary generation storage medium includes various units for executing various steps in an embodiment corresponding to an ultrasonic inspection report auxiliary generation method. Figure 4 ,The auxiliary generation storage medium includes: a text data extraction module, a text data quality evaluation module, a text data set quality evaluation module, and an auxiliary generation module.

[0108] A text data extraction module, used to collect a first preset number of ultrasound examination reports and extract text data, and obtain structural data and diagnostic data of each ultrasound examination report;

[0109] A text data quality evaluation module is used to record any ultrasound examination report as a target ultrasound examination report, determine the data universality of the target ultrasound examination report based on the difference between the structural data of the target ultrasound examination report and all other ultrasound examination reports, and obtain the information contribution of the target ultrasound examination report by combining the similarity between the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports, and the difference in the number of data included in the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports;

[0110] The text data set quality evaluation module is used to divide the ultrasound examination reports into examination report clusters according to the information contribution of the ultrasound examination reports, and obtain the cluster center of the examination report cluster, and obtain the data quality of the examination report cluster according to the difference between all the information contribution contained in the examination report cluster and the distribution of all the information contribution contained in the examination report cluster;

[0111] The auxiliary generation module is used to determine the training data set to train the neural network model according to the data quality of all inspection report clusters, obtain the ultrasound inspection report auxiliary generation model, and obtain the ultrasound inspection report auxiliary generation result according to the ultrasound inspection report auxiliary generation model.

[0112] It is to be understood that the sequence of the embodiments of the present application described above is for description only and does not represent the advantages and disadvantages of the embodiments. The above describes specific embodiments of the present specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0113] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0114] The above content is only an implementation method of the present application and is not intended to limit the scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the protection scope of the present application.

Claims

1. A method for assisting in generating an ultrasonic examination report, characterized in that: The method comprises the following steps: Collecting a first preset number of ultrasound examination reports and extracting text data, and obtaining structural data and diagnostic data of each ultrasound examination report; Any ultrasound examination report is recorded as a target ultrasound examination report. The data universality of the target ultrasound examination report is determined based on the difference between the structural data of the target ultrasound examination report and all other ultrasound examination reports. The information contribution of the target ultrasound examination report is obtained by combining the similarity between the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports and the difference in the number of data included in the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports. The ultrasound examination reports are divided into examination report clusters according to their information contribution, and the cluster centers of the examination report clusters are obtained. The data quality of the examination report clusters is obtained according to the differences between all information contribution degrees contained in the examination report clusters and the distribution of all information contribution degrees contained in the examination report clusters. Determine a training data set to train a neural network model according to the data quality of all inspection report clusters, obtain an ultrasound inspection report auxiliary generation model, and obtain an ultrasound inspection report auxiliary generation result according to the ultrasound inspection report auxiliary generation model; The specific method for determining the prevalence of data reported by the target ultrasound examination includes: The Levenshtein distance between the structural data of the target ultrasound examination report and any other ultrasound examination report is recorded as the text data difference between the target ultrasound examination report and any other ultrasound examination report; The cumulative sum of the text data differences between the target ultrasound examination report and all other ultrasound examination reports is recorded as the comprehensive text data difference of the target ultrasound examination report; The inverse of the difference in comprehensive text data of the target ultrasound examination report was recorded as the data prevalence of the target ultrasound examination report; The specific method of combining the similarity between the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports, and the difference in the number of data included in the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports to obtain the information contribution of the target ultrasound examination report includes: The Simhash value between the diagnostic data of the target ultrasound examination report and any other ultrasound examination report is recorded as the text data duplication degree between the target ultrasound examination report and any other ultrasound examination report; the cumulative sum of the text data duplication degrees of the target ultrasound examination report and all other ultrasound examination reports is recorded as the comprehensive text data duplication degree of the target ultrasound examination report; The difference between the amount of data included in the diagnostic data of the target ultrasound examination report and any other ultrasound examination report is recorded as the text data volume difference between the target ultrasound examination report and any other ultrasound examination report; the cumulative sum of the text data volume differences between the target ultrasound examination report and all other ultrasound examination reports is recorded as the comprehensive text data volume difference of the target ultrasound examination report; According to the repetition degree of the comprehensive text data and the difference in the amount of the comprehensive text data of the target ultrasound examination report, the data information diversity of the target ultrasound examination report is obtained, and the data information diversity of the target ultrasound examination report is positively correlated with the repetition degree of the comprehensive text data and the difference in the amount of the comprehensive text data of the target ultrasound examination report; According to the data universality and data information diversity of the target ultrasound examination report, the information contribution of the target ultrasound examination report is obtained; The information contribution of the target ultrasound examination report is obtained according to the data universality and data information diversity of the target ultrasound examination report. The specific process of obtaining the information contribution of the target ultrasound examination report is as follows: The linear normalized value of the product of the data universality and the data information diversity of the target ultrasound examination report is recorded as the information contribution of the target ultrasound examination report; The specific method of obtaining the data quality of the inspection report cluster is as follows: The mean value of all information contributions contained in the inspection report cluster is recorded as the information contribution mean of the inspection report cluster; The cumulative sum of the Euclidean distances between all information contributions contained in the inspection report clustering cluster and the cluster center of the inspection report clustering cluster is recorded as the first difference of the information contribution of the inspection report clustering cluster, the information entropy of all information contributions contained in the inspection report clustering cluster is recorded as the second difference of the information contribution of the inspection report clustering cluster, and the product of the first difference and the second difference of the information contribution of the inspection report clustering cluster is recorded as the comprehensive difference of the information contribution of the inspection report clustering cluster; According to the information contribution mean and comprehensive difference of the inspection report clusters, the data quality of the inspection report clusters is obtained; The specific method for obtaining the data quality of the inspection report clusters according to the information contribution mean and information contribution comprehensive difference of the inspection report clusters is as follows: The ratio of the mean information contribution of the inspection report cluster to the comprehensive difference in information contribution is recorded as the data quality of the inspection report cluster.

2. The method for assisting in generating an ultrasonic examination report according to claim 1, characterized in that: The specific method of dividing the ultrasound examination reports into examination report clusters according to the information contribution of the ultrasound examination reports and obtaining the cluster centers of the examination report clusters is as follows: The absolute value of the difference between the information contributions of two ultrasound examination reports is used as the distance between the two information contributions, and the information contributions of all ultrasound examination reports are clustered to obtain the examination report clustering clusters and the cluster centers of the examination report clustering clusters.

3. The method for auxiliary generation of ultrasonic examination report according to claim 1, characterized in that: The method of determining the training data set to train the neural network model according to the data quality of all inspection report clusters, obtaining the ultrasound inspection report auxiliary generation model, and obtaining the ultrasound inspection report auxiliary generation result according to the ultrasound inspection report auxiliary generation model includes the following specific methods: The ultrasound examination report corresponding to the examination report cluster with the highest data quality is used as a training data set, and the VisionLLM multimodal large language model is trained using the training data set to obtain the trained VisionLLM multimodal large language model, and the trained VisionLLM multimodal large language model is recorded as an ultrasound examination report auxiliary generation model; The ultrasound examination report is input into the ultrasound examination report auxiliary generation model to obtain the ultrasound examination report auxiliary generation result.

4. An electronic device for assisting in generating ultrasonic inspection reports, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 3 are implemented.

5. An auxiliary generation device for ultrasonic examination report, implementing the method as claimed in claim 1, characterized in that: The auxiliary generation device comprises: A text data extraction module, used to collect a first preset number of ultrasound examination reports and extract text data, and obtain structural data and diagnostic data of each ultrasound examination report; A text data quality evaluation module is used to record any ultrasound examination report as a target ultrasound examination report, determine the data universality of the target ultrasound examination report based on the difference between the structural data of the target ultrasound examination report and all other ultrasound examination reports, and obtain the information contribution of the target ultrasound examination report by combining the similarity between the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports, and the difference in the number of data included in the diagnostic data of the target ultrasound examination report and all other ultrasound examination reports; The text data set quality evaluation module is used to divide the ultrasound examination reports into examination report clusters according to the information contribution of the ultrasound examination reports, and obtain the cluster center of the examination report cluster, and obtain the data quality of the examination report cluster according to the difference between all the information contribution contained in the examination report cluster and the distribution of all the information contribution contained in the examination report cluster; The auxiliary generation module is used to determine the training data set to train the neural network model according to the data quality of all inspection report clusters, obtain the ultrasound inspection report auxiliary generation model, and obtain the ultrasound inspection report auxiliary generation result according to the ultrasound inspection report auxiliary generation model.

Images