Statistical data processing method and system based on intelligent medicine
By extracting lesion characteristics and medical keywords, constructing tag sequences and classifying them, the problem of inefficient information retrieval in the existing technology is solved, and more efficient and accurate medical information retrieval is achieved.
Patent Information
- Application Number
- CN202410732053.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-06
- Publication Date
- 2025-01-03
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing medical statistical data processing methods lack a clear classification system, which leads to inefficient information retrieval and may lead to information omission or incorrect classification.
By collecting medical record data and medical images, lesion characteristics and medical keywords are extracted, and the disease label sequence is constructed, and similarity and distribution differences are used to classify, so as to achieve effective classification of medical images and medical record data.
It improves the efficiency and accuracy of medical information retrieval and provides better support and services to medical workers.
Smart Images
Figure CN118571498B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of medical statistical data processing, and in particular to a statistical data processing method and system based on intelligent medicine. Background Art
[0002] In the medical field, a large amount of medical data is generated in the hospital database every day and needs to be collected and stored. These medical data contain rich medical information, such as patients' medical records, medical imaging data, etc. The processing and analysis of these medical data are of great significance for medical research and assisting doctors in diagnosing and treating diseases. However, the data processing process of medical statistics often faces challenges such as large amount of image data, high complexity and difficulty in pattern extraction. Therefore, it is particularly important to use high-level methods based on intelligent medical engineering to process and analyze medical statistical data. Currently, data sources are often extracted by analyzing medical record statistical data, and comparative retrieval is achieved by matching medical images. However, this method does not have a clear classification system when performing comparative retrieval of data sources or medical images, which may cause some information to be omitted or misclassified, reducing the efficiency of information retrieval. Summary of the invention
[0003] In order to solve the technical problem that there is no clear classification system when using existing methods for information comparison and retrieval, resulting in low efficiency of information retrieval, the purpose of the present invention is to provide a statistical data processing method and system based on intelligent medicine. The technical solutions adopted are as follows:
[0004] In a first aspect, the present invention provides a statistical data processing method based on intelligent medicine, comprising:
[0005] Collect medical records and medical images of different patients as well as basic information of patients and symptom information of medical images;
[0006] Extracting lesion features according to the feature distribution of the lesion part in each medical image to obtain the pathological feature value of each medical image; constructing the symptom label sequence of each medical image according to the pathological feature value of each medical image, symptom information and basic information of the patient;
[0007] Extracting medical keywords from each medical record data to form a keyword set for each medical record data; obtaining the importance of each keyword in the keyword set according to the difference between the distribution of each keyword in the keyword set in the medical record data and the distribution of other keywords in the medical record data; and screening all keywords in the keyword set using the importance to obtain a medical tag sequence for each medical record data;
[0008] Medical images are classified according to the similarities between the disease label sequences of different medical images, and medical record data are classified according to the distribution differences between the medical label sequences of different medical record data to obtain the classification results of medical images and medical record data.
[0009] Preferably, obtaining the importance of each keyword in the keyword set according to the difference between the distribution of each keyword in the keyword set in the medical record data and the distribution of other keywords in the medical record data specifically includes:
[0010] For any keyword set, the correlation parameter between any two keywords is obtained according to the repetition of the distribution of any two keywords in the medical record data;
[0011] Record any one of the keywords in the keyword set as a target keyword, and record all the keywords in the keyword set except the target keyword as reference keywords; record any one of the reference keywords as a first reference keyword, and record the next reference keyword adjacent to the first reference keyword as a second reference keyword;
[0012] Normalizing the correlation parameters between the target keyword and each reference keyword to obtain a weight coefficient corresponding to each reference keyword;
[0013] Calculating the absolute value of the difference between the correlation parameter between the target keyword and the first reference keyword and the correlation parameter between the target keyword and the second reference keyword, and performing negative correlation mapping on the absolute value of the difference to obtain the correlation characteristic index of the first reference keyword;
[0014] The product of the weight coefficient corresponding to each reference keyword and the associated characteristic index is taken as the characteristic degree of each reference keyword, and the average value of the characteristic degrees of all reference keywords is calculated to obtain the importance degree of the target keyword.
[0015] Preferably, obtaining the correlation parameter between any two keywords according to the repetition of the distribution of any two keywords in the medical record data specifically includes:
[0016] For any two keywords in the keyword set, the number of times the two keywords appear in the same medical record data is obtained and recorded as the first coefficient, the number of times each keyword appears in different medical record data is obtained and recorded as the feature quantity of each keyword, and the product of the feature quantities of the two keywords is squared to obtain the second coefficient;
[0017] A correlation parameter between the arbitrary two keywords is obtained according to the first coefficient and the second coefficient, wherein the first coefficient and the correlation parameter are positively correlated, and the second coefficient and the correlation parameter are negatively correlated.
[0018] Preferably, the method of using the importance degree to screen all keywords in the keyword set to obtain a medical tag sequence for each medical record data specifically includes:
[0019] For any keyword set, all keywords are arranged in descending order of importance, and a preset number of keywords are selected in the arrangement order to form a medical label sequence of the medical record data corresponding to the keyword set.
[0020] Preferably, the classifying of the medical record data according to the distribution differences between the medical label sequences of different medical record data specifically includes:
[0021] For any two medical record data, according to the position distribution of each keyword in the medical tag sequence of the medical record data in the medical tag sequence of the other medical record data of the any two medical record data, obtain the similarity index between the any two medical record data;
[0022] The medical record data is classified based on the similarity index to obtain a classification result of the medical record data.
[0023] Preferably, the method for obtaining the similarity index is specifically as follows:
[0024] Record any two medical record data as first medical record data and second medical record data respectively, and record any keyword in the medical tag sequence of the first medical record data as the first keyword;
[0025] If the medical tag sequence of the second medical record data does not include the first keyword, the value of the indication coefficient of the first keyword is 0; if the medical tag sequence of the second medical record data includes the first keyword, the value of the indication coefficient of the first keyword is 1;
[0026] Calculate the ratio between the position number of the first keyword in the medical label sequence of the first medical record data and the position number of the first keyword in the medical label sequence of the second medical record data, and perform negative correlation normalization processing on the absolute value of the difference between the ratio and the constant 1 to obtain the similarity feature coefficient of the first keyword;
[0027] The similarity index between the first medical record data and the second medical record data is obtained by calculating the average value of the product between the similarity feature coefficients and the indication coefficients of all keywords in the medical tag sequence of the first medical record data.
[0028] Preferably, the step of extracting lesion features according to the characteristic distribution of the lesion part in each medical image to obtain the pathological characteristic value of each medical image specifically includes:
[0029] For any medical image, a neural network is used to process the medical image to obtain the lesion area in the medical image;
[0030] The product of the edge length of each lesion region and the area of the lesion region is calculated to obtain the distribution characteristic factor of each lesion region; the mean value of the distribution characteristic factors of all lesion regions in the medical image is taken as the pathological characteristic value of the medical image.
[0031] Preferably, the step of constructing a disease label sequence for each medical image according to the pathological feature value, disease information and basic information of the patient of each medical image specifically includes:
[0032] The disease information includes the organ where each lesion area in the medical image is located and the name of the disease in the lesion area; the basic information of the patient includes the patient's gender and age;
[0033] For any medical image, the pathological feature value of the medical image, the disease information of the medical image, and the basic information of the patient corresponding to the medical image are combined to form a disease label sequence of the medical image.
[0034] Preferably, the classifying of medical images according to the similarities between the disease label sequences of different medical images specifically includes:
[0035] The word vector model is used to obtain the feature vector of the disease label sequence of each medical image, and the clustering algorithm is used to classify the medical images based on the Euclidean distance between any two feature vectors to obtain the classification results of the medical images.
[0036] In a second aspect, the present invention provides a statistical data processing system based on intelligent medicine, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the steps of a statistical data processing method based on intelligent medicine when executed by the processor.
[0037] The embodiments of the present invention have at least the following beneficial effects:
[0038] The present invention first collects medical images, medical record data, basic information of patients and symptom information of medical images, and provides a data basis for subsequent enrichment of data for multi-dimensional similar feature analysis. Then, the pathological feature value is obtained by extracting the feature information of the lesion part in the medical image, which characterizes the severity of the lesion part in the medical image, and the symptom label sequence of each medical image is constructed in combination with information from multiple aspects, which characterizes the lesion feature label information corresponding to each medical image. Further, the medical record data is subjected to feature analysis, and firstly, medical keyword extraction is performed, and the difference between the distribution of keywords in the keyword set in different medical record data is analyzed to obtain the importance of each keyword. The importance degree characterizes the importance of the core degree of the corresponding keyword in the keyword set and the corresponding medical record data. The keywords can be screened in combination with the core degree and the importance, and the final medical label sequence is determined, which characterizes the label information with a strong core degree in the medical record data. Finally, the label information adaptively obtained from the medical image and the medical record data is used for classification, and a classification result with a better effect can be obtained. When doctors use the classification result for information comparison and retrieval, the efficiency and accuracy of medical information retrieval can be effectively improved, and better support and service can be provided for medical workers. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention 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 invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0040] Figure 1 is a flowchart of a method for processing statistical data based on intelligent medicine provided by an embodiment of the present invention;
[0041] Figure 2 It is a flowchart of the steps of the method for obtaining a medical label sequence of medical record data provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0042] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the statistical data processing method and system based on intelligent medicine proposed by the present invention, its specific implementation method, structure, features and effects are described in detail as follows 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.
[0043] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0044] The specific scheme of the statistical data processing method and system based on intelligent medicine provided by the present invention is described in detail below with reference to the accompanying drawings.
[0045] An embodiment of a statistical data processing method based on intelligent medicine:
[0046] See also Figure 1 , which shows a flowchart of a statistical data processing method based on intelligent medicine provided by an embodiment of the present invention, the method comprising the following steps:
[0047] Step S100, collecting medical records and medical images of different patients as well as basic information of the patients and symptom information of the medical images.
[0048] Specifically, this embodiment mainly performs corresponding data processing on medical statistical data, so firstly, it is necessary to obtain relevant medical data, and different medical records and medical images of different patients can be obtained from the hospital database. At the same time, it is necessary to obtain the basic information of each patient in the record and the symptom information corresponding to the medical image. Among them, the medical image can be a CT grayscale image or an X-ray grayscale image of the patient.
[0049] In this embodiment, the basic information of the patient includes the patient's gender and age, and the disease information of the medical image can be the disease name of the lesion part contained in the medical image and the corresponding organ. For example, the disease name corresponding to the medical image is myocardial infarction, and the corresponding organ is heart; the disease name corresponding to the medical image is thyroid nodule, and the corresponding organ is thyroid gland.
[0050] Step S200, extracting lesion features according to the feature distribution of the lesion part in each medical image to obtain the pathological feature value of each medical image; constructing the symptom label sequence of each medical image according to the pathological feature value of each medical image, symptom information and basic information of the patient.
[0051] First, the medical images are feature analyzed, the lesion features of each medical image are extracted, each medical image is labeled, and the distribution of pathological features of each medical image is quantified. Specifically, for any medical image, the medical image is processed using a neural network to obtain the lesion area in the medical image; the product of the edge length of each lesion area and the area of the lesion area is calculated to obtain the distribution characteristic factor of each lesion area; the mean value of the distribution characteristic factors of all lesion areas in the medical image is taken as the pathological characteristic value of the medical image.
[0052] In this embodiment, in order to extract the lesion features of the medical image more accurately, the medical image may be preprocessed, such as denoising or enhancement processing, and then the lesion area may be extracted from the preprocessed medical image using a machine learning method.
[0053] The recognition of lesion areas in medical images based on neural networks is a common task in the field of medical imaging. It aims to accurately locate and identify lesion areas from medical images. Here we only give a brief introduction. Lesion areas can be annotated by professional doctors or medical imaging experts as training data for training neural networks. The network model can be a convolutional neural network. So far, the neural network is used to extract lesion features from each medical image, and the lesion area in each medical image can be identified. At the same time, combined with the disease information of the medical image, the name of the disease corresponding to the lesion area and the organ where the lesion area is located can be obtained.
[0054] Furthermore, after obtaining the lesion part in each medical image, considering that the more obvious the distribution of the lesion characteristics corresponding to the lesion area, the larger the distribution area of the lesion area, and the longer the edge, the more obvious the characteristic distribution of the lesion part is, the more serious the disease corresponding to the lesion part is. It should be noted that in order to perform feature analysis on the edge of the lesion area, the canny operator can be used to perform edge detection on the lesion area in each medical image to obtain the edge part of each lesion area.
[0055] Based on this, in this embodiment, taking any medical image as an example for explanation, the calculation formula of the pathological feature value of the medical image can be expressed as:
[0056]
[0057] in, represents the pathological feature value of the i-th medical image, represents the total number of lesion areas contained in the i-th medical image, represents the edge length of the nth lesion area in the i-th medical image, Represents the area of the nth lesion region in the i-th medical image.
[0058] It represents the distribution characteristic factor of the nth lesion area in the ith medical image. The larger the distribution area of the lesion area in the medical image, the longer the edge, and the larger the value of the distribution characteristic factor, the more obvious the characteristic distribution of the lesion area, and the more serious the corresponding disease condition. At the same time, considering the different number of lesion areas in the medical image, the mean method is used to quantify the lesion characteristics of the medical image. The pathological characteristic value represents the characteristic distribution value of the lesion part in the medical image. Different values of the pathological characteristic value represent different severity of the disease conditions of the lesion characteristics contained in different medical images.
[0059] Finally, considering that some diseases have certain gender and age characteristics, we can consider adding the basic information of the corresponding patients to the medical images while analyzing the lesion characteristics of the medical images. That is, we can construct the disease label sequence of each medical image based on the pathological feature value, disease information and basic information of the patients.
[0060] Specifically, for any medical image, the disease parameters corresponding to the medical image are used to construct a disease label sequence of the medical image in the order of pathological feature values, disease information of the medical image, and basic information of the patient. In this embodiment, the disease information of the medical image includes the organ where the lesion area in the medical image is located and the name of the disease of the lesion area. The basic information of the patient includes the patient's gender and age.
[0061] At this point, each medical image can obtain the corresponding disease label sequence. Each element in the disease label sequence can be regarded as a disease label, which represents the lesion feature part in the medical image and the patient feature information corresponding to the lesion feature, providing a data basis for the subsequent comparative classification of medical images.
[0062] Step S300, extracting medical keywords from each medical record data to form a keyword set for each medical record data; obtaining the importance of each keyword in the keyword set based on the difference between the distribution of each keyword in the keyword set in the medical record data and the distribution of other keywords in the medical record data; and using the importance to screen all keywords in the keyword set to obtain a medical label sequence for each medical record data.
[0063] In step S200, feature analysis is performed on different medical images of different patients to add labels of lesion features to the medical images for subsequent data classification. It is then necessary to extract features from the medical records of different patients in combination with the text information in the medical records, and to adaptively add label data of different features to the medical records of different patients, thereby avoiding confusion in the data classification process. Figure 2 The figure is a flowchart of the steps of the method for obtaining the medical label sequence of each different medical record data in this embodiment.
[0064] Step S301: Extract the medical keywords of each medical record data to form a keyword set for each medical record data.
[0065] First, it is necessary to preprocess each medical record data. Preprocessing can improve data quality, accuracy, and usability, providing a better foundation and support for medical data analysis, mining, and application. In this embodiment, text cleaning and word segmentation operations are performed on the medical record data. Among them, text cleaning of the medical record data can remove irrelevant characters, punctuation marks, and numbers, such as tab characters, line breaks, and spaces; filter out stop words, that is, remove common stop words, such as "of", "is", "in", "and", etc. The word segmentation operation on the medical record data is specifically to split the text into words or phrases.
[0066] Then, after the medical record data is preprocessed, the medical record data is transformed into the form of words or phrases. Furthermore, keyword extraction technology can be used to obtain keywords in the aspect of medical terms, forming a keyword set. In other embodiments, an existing medical term matching thesaurus can also be used to perform term matching on the word segmentation results corresponding to each medical record data, and medical keywords matching the medical terms can be obtained, forming a keyword set.
[0067] For example, the medical record data of a certain patient is "The patient went to the hospital for treatment due to chest pain and shortness of breath. Myocardial infarction was found through CT examination. Vasodilators and anticoagulant drugs were used during the treatment process". Extracting the medical keywords in the medical record data can obtain the keyword set of this medical record data as {chest pain, shortness of breath, CT examination, myocardial infarction, vasodilator, anticoagulant drug}.
[0068] It should be noted that each medical record data of each patient contains the patient's own attribute information. When storing the medical record data, the own attribute information is the basis for searching for medical record information, such as basic information required for medical clinical practice such as the name, gender, age, blood type, etc. of the patient corresponding to the medical record data.
[0069] Step S302: Obtain the importance degree of each keyword in the keyword set according to the difference between the distribution of each keyword in the keyword set in the medical record data and the distribution of other keywords in the medical record data.
[0070] Since the types of diseases are diverse, and the drugs and examination items for treating diseases are also different, by analyzing the distribution of the keywords extracted from each medical record data in the medical record data, the importance of the keywords in each keyword set can be quantified, which can help identify keywords with high information content and influence in the medical record data, so as to better understand the content and characteristics of the medical record data.
[0071] Specifically, for any keyword set, the correlation parameter between any two keywords is obtained according to the repetition of the distribution of any two keywords in the medical record data. That is, for any two keywords in the keyword set, the number of times the two keywords appear in the same medical record data is obtained and recorded as the first coefficient, the number of times each keyword appears in different medical record data is obtained and recorded as the feature quantity of each keyword, and the product of the feature quantities of the two keywords is squared to obtain the second coefficient.
[0072] The correlation parameter between the two arbitrary keywords is obtained according to the first coefficient and the second coefficient, wherein the first coefficient and the correlation parameter are positively correlated, and the second coefficient and the correlation parameter are negatively correlated. The positive correlation may be a multiplication relationship, an addition relationship, etc., and the negative correlation may be mapped by a negative exponential power function, a reciprocal form, etc.
[0073] In this embodiment, taking any two keywords in any keyword set as an example for explanation, the calculation formula of the correlation parameter between the uth keyword and the vth keyword in the keyword set can be expressed as:
[0074]
[0075] in, Represents the correlation parameter between the u-th keyword and the v-th keyword in the keyword set, It represents the number of times the u-th keyword and the v-th keyword appear in the same medical record data, which is also the first coefficient; represents the number of times the u-th keyword appears in different medical records, that is, the number of features of the u-th keyword; represents the number of times the vth keyword appears in different medical records, that is, the number of features of the vth keyword. is the second coefficient.
[0076] It should be noted that, in this embodiment, the u-th keyword and the v-th keyword are searched in all collected medical record data. When the u-th keyword and the v-th keyword appear in the same medical record data multiple times, the number of times they appear in the same medical record data at the same time is only recorded once.
[0077] The larger the value of , the greater the frequency of the two keywords appearing in the same medical record, and the stronger the correlation between the two keywords, the greater the value of the corresponding correlation parameter. The ratio of the first coefficient to the second coefficient represents the proportion of the frequency of the two keywords appearing at the same time relative to the frequency of their separate appearances. The larger the value, the greater the correlation strength between the two keywords.
[0078] Furthermore, any keyword in the keyword set is recorded as a target keyword, and all keywords in the keyword set except the target keyword are recorded as reference keywords; any reference keyword is recorded as a first reference keyword, and the next reference keyword adjacent to the first reference keyword is recorded as a second reference keyword.
[0079] Considering that adjacent keywords in the keyword set corresponding to the medical record data also represent the position distribution relationship in the medical record data, that is, adjacent keywords in the keyword set may belong to the same sentence or the same paragraph in the medical record data, that is, the position distribution is relatively close. Therefore, the importance of each keyword in the keyword set can be evaluated by analyzing the differences between the correlation degrees corresponding to adjacent keywords in the keyword set.
[0080] Specifically, the correlation parameters between the target keyword and each reference keyword are normalized to obtain the weight coefficient corresponding to each reference keyword; the absolute value of the difference between the correlation parameter between the target keyword and the first reference keyword and the correlation parameter between the target keyword and the second reference keyword is calculated, and the absolute value of the difference is negatively mapped to obtain the associated characteristic index of the first reference keyword; the product of the weight coefficient corresponding to each reference keyword and the associated characteristic index is used as the characteristic degree of each reference keyword, and the average value of the characteristic degrees of all reference keywords is calculated to obtain the importance degree of the target keyword.
[0081] In this embodiment, the uth keyword in the keyword set is recorded as the target keyword, and all keywords except the uth keyword in the keyword set are reference keywords corresponding to the uth keyword. The rth reference keyword is recorded as the first reference keyword, and the r+1th reference keyword is the second reference keyword. The calculation formula for the importance of the uth keyword in the keyword set, that is, the target keyword, can be expressed as:
[0082]
[0083] in, Indicates the importance of the u-th keyword in the keyword set, Indicates the total number of keywords contained in the keyword set. Indicates the total number of reference keywords. represents the correlation parameter between the u-th keyword and the r-th reference keyword in the keyword set, represents the correlation parameter between the uth keyword and the r+1th reference keyword in the keyword set, Norm represents the linear normalization function, and exp() represents the exponential function with the natural constant e as the base.
[0084] is the weight coefficient corresponding to the rth reference keyword, which reflects the proportion of the correlation parameters between the target keyword and the reference keyword. The larger the correlation parameter, the greater the correlation between the reference keyword and the target keyword, and the larger the corresponding weight coefficient value, the greater the data reference value of the reference keyword.
[0085] is the correlation feature index of the first reference keyword, It indicates the difference between the correlation parameters corresponding to two adjacent reference keywords. The larger the difference, the greater the difference in the correlation distribution between the reference keywords and the target keywords. In turn, the smaller the correlation characteristic degree of the target keywords in the first reference keyword dimension, the smaller the corresponding importance value.
[0086] is the characteristic degree of the rth reference keyword. The more reference keywords are associated with the target keyword, and the larger the correlation parameter corresponding to the associated reference keyword, the higher the correlation degree, indicating that the core degree of the target keyword is higher, the corresponding characteristic degree value is larger, and the importance value is larger. The importance degree of the target keyword reflects the core degree and importance of the target keyword from the proportion of correlation parameters between keywords and the distribution of differences.
[0087] Step S303: Screen all keywords in the keyword set using the importance level to obtain a medical tag sequence for each medical record data.
[0088] The importance of each keyword in each keyword set can be obtained in the same way as the importance of the target keyword. The larger the importance value, the more core the keyword is in the keyword set, that is, the greater the core of the keyword in the corresponding medical record data, and the greater the importance of the keyword in the keyword set and the corresponding medical record data.
[0089] Based on this, in order to avoid information redundancy and improve the efficiency of information retrieval, the most representative label data is added to each medical record data. This can be done by screening multiple keywords with greater importance and core degree in the keyword set corresponding to the medical record data as labels for the medical record data.
[0090] Specifically, for any keyword set, all keywords are arranged in descending order of importance, and a preset number of keywords are selected in the arrangement order to form a medical label sequence of the medical record data corresponding to the keyword set. In this embodiment, the preset number is 5. Each keyword in the medical label sequence is a keyword with greater importance and core degree in the corresponding medical record data.
[0091] It should be noted that when the number of keywords in the keyword set is less than the preset number, all the keywords in the keyword set are directly used to form a medical label sequence corresponding to the medical record data.
[0092] Step S400, classifying the medical images according to the similarities between the disease label sequences of different medical images, classifying the medical record data according to the distribution differences between the medical label sequences of different medical record data, and obtaining the classification results of the medical images and the medical record data.
[0093] The disease label sequence of each medical image represents the lesion feature label information corresponding to each medical image, and then the classification and sorting operations of the medical images can be achieved by analyzing the similarities between the lesion feature label information corresponding to different medical images.
[0094] Specifically, the word vector model is used to obtain the feature vector of the disease label sequence of each medical image, and the clustering algorithm is used to classify the medical image based on the Euclidean distance between any two feature vectors to obtain the classification result of the medical image. Among them, the disease label sequence of each medical image needs to be converted into a vector form. This embodiment uses the Word2Vec word vector model to obtain the feature vector of the disease label sequence. Then, clustering algorithms such as DBSCAN can be used to classify medical images by calculating the Euclidean distance between the feature vectors corresponding to any two medical images.
[0095] Furthermore, the medical label sequence of each medical record data represents the label information with a strong core degree in the medical record data, and then the classification and sorting operation of the medical record data can be achieved by analyzing the similarities and position distribution differences between the core keywords corresponding to different medical record data.
[0096] For any two medical record data, the similarity index between the any two medical record data is obtained according to the position distribution of each keyword in the medical tag sequence of the medical record data in the medical tag sequence of the other medical record data of the any two medical record data.
[0097] Specifically, any two medical record data are recorded as first medical record data and second medical record data, and any keyword in the medical tag sequence of the first medical record data is recorded as the first keyword;
[0098] If the first keyword is not included in the medical label sequence of the second medical record data, the value of the indication coefficient of the first keyword is 0; when the first keyword is not included in the medical label sequence of the second medical record data, it means that multiple keywords with a higher core degree in the second medical record data do not include the first keyword, that is, none of the keywords with a higher core degree in the first medical record data include this keyword, which means that under the dimension corresponding to the first keyword, the correlation and similarity between the two medical record data are smaller, so the indication coefficient is set to a smaller value at this time, which further means that under the dimension corresponding to the first keyword, the reference value of the data position distribution difference with other medical record data is smaller. It can also be understood that the position distribution of the first keyword in the medical label sequence corresponding to other medical record data is not considered.
[0099] If the first keyword is included in the medical tag sequence of the second medical record data, the value of the indicator coefficient of the first keyword is 1. When the first keyword is included in the medical tag sequence of the second medical record data, it means that the first medical record data and the second medical record data have repeated or similar core keywords in the dimension corresponding to the keyword. Therefore, setting a larger indicator coefficient than when there is no such indicator coefficient can further analyze the position distribution differences between the repeated core keywords, reflecting the data similarity of the two medical record data in each keyword dimension.
[0100] Furthermore, the ratio of the position number of the first keyword in the medical label sequence of the first medical record data to the position number of the first keyword in the medical label sequence of the second medical record data is calculated, and the absolute value of the difference between the ratio and the constant 1 is negatively correlated normalized to obtain the similarity feature coefficient of the first keyword; the average of the products of the similarity feature coefficients of all keywords in the medical label sequence of the first medical record data and the indicator coefficient is calculated to obtain the similarity index between the first medical record data and the second medical record data.
[0101] In this embodiment, the medical label sequences corresponding to any two medical record data are taken as an example for explanation, and the t-th keyword in the medical label sequence corresponding to the first medical record data A is used as the first keyword. The calculation formula of the similarity index between the first medical record data and the second medical record data can be expressed as:
[0102]
[0103] in, represents the similarity index between the first medical record data A and the second medical record data B, represents the total number of keywords contained in the medical tag sequence corresponding to the first medical record data A, represents the indicator coefficient of the t-th keyword in the medical tag sequence corresponding to the first medical record data A, that is, the indicator coefficient of the first keyword; represents the position number of the t-th keyword in the medical tag sequence corresponding to the first medical record data A, that is, the position number of the first keyword in the medical tag sequence of the first medical record data; It represents the position number of the first keyword in the medical tag sequence of the second medical record data, that is, the position number of the keyword that is the same as the first keyword element in the medical tag sequence of the second medical record data.
[0104] When the indicator coefficient is 0, it means that there are elements in the medical label sequence of the second medical record data that are different from the medical label sequence of the first medical record data. At this time, it is not necessary to analyze the position distribution of the first keyword in the second medical record data. It should be noted that in other embodiments, when there is no element in the second medical record data that is the same as the first keyword, The value is set to , indicating a numerical value greater than the maximum value of the position number of the first keyword in the second medical record data.
[0105] When the indicator coefficient takes the value of 1, it means that there are elements in the medical label sequence of the second medical record data that are the same as the elements in the medical label sequence of the first medical record data. At this time, further analysis is needed to determine the differences in the position distribution of the same elements in the medical label sequences of different medical record data.
[0106] is the similarity feature coefficient of the first keyword, It reflects the ratio of the position distribution of the first keyword in the medical label sequence of the first medical record data and the medical label sequence of the second medical record data. The closer the ratio is to 1, the more it means that the same keyword has the same position distribution in the medical label sequence of the two medical record data, which further indicates that the core degree of the same keyword in the two medical record data is the same. The larger the value of the corresponding similarity feature coefficient, the greater the similarity between the two medical record data.
[0107] Furthermore, the similarity index between any two medical record data reflects the similarity between the two medical record data. The medical record data is classified based on the similarity index to obtain the classification result of the medical record data. In this embodiment, the similarity index between any two medical record data is normalized, and the two medical record data corresponding to the normalized similarity index greater than or equal to the preset similarity threshold are classified into the same category. Among them, the value of the similarity threshold is 0.7, and the implementer can set it according to the specific implementation scenario. Since the value range of the normalized similarity index is (0,1), the value range of the similarity threshold can be (0,1). When the value of the similarity threshold is closer to 0, the standard for classification using the similarity threshold is looser. When the value of the similarity threshold is closer to 1, the standard for classification using the similarity threshold is stricter.
[0108] In other embodiments, the implementer may also use other appropriate clustering algorithms to cluster the medical record data. At the same time, it should be noted that in other embodiments, when classifying medical images, the same method as the medical record data can be selected for processing, that is, each medical image corresponds to a disease label sequence, and by analyzing the distribution of elements in the disease label sequence of any two medical images, the similarity index between any two medical images can be obtained, and then classification is performed to obtain the classification result.
[0109] By classifying and processing different medical images and different medical records before storing them, fast retrieval and query functions can be achieved to improve work efficiency. At the same time, it can also help doctors better understand the patient's medical history and treatment, thereby providing personalized medical services, so that relevant medical staff can formulate corresponding diagnosis and treatment plans and monitoring plans for different types of medical records, thereby improving the quality and effectiveness of medical services.
[0110] It should be noted that when a doctor needs to find medical images or medical records similar to a certain patient, as a specific implementation method, the doctor can obtain the feature vector of the medical label sequence of the medical record data of the patient to be found and the disease label sequence of the medical image, and compare the similarity index between the medical record data of the patient to be found and the stored medical record data to obtain the category corresponding to the data with the greatest similarity in the classification results of the stored medical record data of the patient to be found. Similarly, by calculating the Euclidean distance between the feature vector corresponding to the patient to be found and the feature vector corresponding to the stored medical image, the category corresponding to the data with the smallest Euclidean distance in the classification results of the stored medical image of the patient to be found can be obtained, thereby realizing the retrieval and query functions.
[0111] An embodiment of a statistical data processing system based on intelligent medicine:
[0112] This embodiment provides a statistical data processing system based on intelligent medicine, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the steps of a statistical data processing method based on intelligent medicine are implemented. Since an embodiment of a statistical data processing method based on intelligent medicine has been described in detail, it will not be described in detail here.
[0113] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.
Claims
1. A statistical data processing method based on intelligent medicine, characterized in that: The method comprises the following steps: Collect medical records and medical images of different patients as well as basic information of patients and symptom information of medical images; Extracting lesion features according to the feature distribution of the lesion part in each medical image to obtain the pathological feature value of each medical image; constructing the symptom label sequence of each medical image according to the pathological feature value of each medical image, symptom information and basic information of the patient; Extracting medical keywords from each medical record data to form a keyword set for each medical record data; obtaining the importance of each keyword in the keyword set according to the difference between the distribution of each keyword in the keyword set in the medical record data and the distribution of other keywords in the medical record data; and screening all keywords in the keyword set using the importance to obtain a medical tag sequence for each medical record data; Classify the medical images according to the similarities between the disease label sequences of different medical images, and classify the medical record data according to the distribution differences between the medical label sequences of different medical record data, and obtain the classification results of the medical images and medical record data; The obtaining of the importance of each keyword in the keyword set according to the difference between the distribution of each keyword in the keyword set in the medical record data and the distribution of other keywords in the medical record data specifically includes: For any keyword set, the correlation parameter between any two keywords is obtained according to the repetition of the distribution of any two keywords in the medical record data; Record any one of the keywords in the keyword set as a target keyword, and record all the keywords in the keyword set except the target keyword as reference keywords; record any one of the reference keywords as a first reference keyword, and record the next reference keyword adjacent to the first reference keyword as a second reference keyword; Normalizing the correlation parameters between the target keyword and each reference keyword to obtain a weight coefficient corresponding to each reference keyword; Calculating the absolute value of the difference between the correlation parameter between the target keyword and the first reference keyword and the correlation parameter between the target keyword and the second reference keyword, and performing negative correlation mapping on the absolute value of the difference to obtain the correlation characteristic index of the first reference keyword; The product of the weight coefficient corresponding to each reference keyword and the associated characteristic index is used as the characteristic degree of each reference keyword, and the average value of the characteristic degrees of all reference keywords is calculated to obtain the importance degree of the target keyword; The repetition of the distribution of any two keywords in the medical record data is used to obtain the correlation parameter between any two keywords, specifically including: For any two keywords in the keyword set, the number of times the two keywords appear in the same medical record data is obtained and recorded as the first coefficient, the number of times each keyword appears in different medical record data is obtained and recorded as the feature quantity of each keyword, and the product of the feature quantities of the two keywords is squared to obtain the second coefficient; A correlation parameter between the arbitrary two keywords is obtained according to the first coefficient and the second coefficient, wherein the first coefficient and the correlation parameter are positively correlated, and the second coefficient and the correlation parameter are negatively correlated.
2. The statistical data processing method based on intelligent medicine according to claim 1, characterized in that: The step of using the importance level to screen all keywords in the keyword set to obtain a medical tag sequence for each medical record data specifically includes: For any keyword set, all keywords are arranged in descending order of importance, and a preset number of keywords are selected in the arrangement order to form a medical label sequence of the medical record data corresponding to the keyword set.
3. The statistical data processing method based on intelligent medicine according to claim 1, characterized in that: The classifying of the medical record data according to the distribution differences between the medical label sequences of different medical record data specifically includes: For any two medical record data, according to the position distribution of each keyword in the medical tag sequence of the medical record data in the medical tag sequence of the other medical record data of the any two medical record data, obtain the similarity index between the any two medical record data; The medical record data is classified based on the similarity index to obtain a classification result of the medical record data.
4. The statistical data processing method based on intelligent medicine according to claim 3 is characterized in that: The method for obtaining the similarity index is specifically as follows: Record any two medical record data as first medical record data and second medical record data respectively, and record any keyword in the medical tag sequence of the first medical record data as the first keyword; If the medical tag sequence of the second medical record data does not include the first keyword, the value of the indication coefficient of the first keyword is 0; if the medical tag sequence of the second medical record data includes the first keyword, the value of the indication coefficient of the first keyword is 1; Calculate the ratio between the position number of the first keyword in the medical label sequence of the first medical record data and the position number of the first keyword in the medical label sequence of the second medical record data, and perform negative correlation normalization processing on the absolute value of the difference between the ratio and the constant 1 to obtain the similarity feature coefficient of the first keyword; The similarity index between the first medical record data and the second medical record data is obtained by calculating the average value of the product between the similarity feature coefficients and the indication coefficients of all keywords in the medical tag sequence of the first medical record data.
5. The statistical data processing method based on intelligent medicine according to claim 1, characterized in that: The step of extracting the lesion features according to the characteristic distribution of the lesion part in each medical image to obtain the pathological characteristic value of each medical image specifically includes: For any medical image, a neural network is used to process the medical image to obtain the lesion area in the medical image; The product of the edge length of each lesion region and the area of the lesion region is calculated to obtain the distribution characteristic factor of each lesion region; the mean value of the distribution characteristic factors of all lesion regions in the medical image is taken as the pathological characteristic value of the medical image.
6. The statistical data processing method based on intelligent medicine according to claim 5, characterized in that: The step of constructing a disease label sequence for each medical image based on the pathological feature value, disease information and basic information of the patient of each medical image specifically includes: The disease information includes the organ where each lesion area in the medical image is located and the name of the disease in the lesion area; the basic information of the patient includes the patient's gender and age; For any medical image, the pathological feature value of the medical image, the disease information of the medical image, and the basic information of the patient corresponding to the medical image are combined to form a disease label sequence of the medical image.
7. The statistical data processing method based on intelligent medicine according to claim 1, characterized in that: The classifying of medical images according to the similarities between the disease label sequences of different medical images specifically includes: The word vector model is used to obtain the feature vector of the disease label sequence of each medical image, and the clustering algorithm is used to classify the medical images based on the Euclidean distance between any two feature vectors to obtain the classification results of the medical images.
8. A statistical data processing system based on intelligent medicine, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the computer program is executed by a processor, the steps of a statistical data processing method based on intelligent medicine as described in any one of claims 1 to 7 are implemented.