Analysis method, system and equipment for nuclear medicine image radiation report and medium

Abstract

The invention relates to the technical field of medical images, in particular to an analysis method, system and device for a nuclear medical image radiation report and a medium. According to the method, the comprehensive correlation coefficient of the disease of the patient and the historical medical image of the patient is evaluated, the individual trajectory of the disease development of the patient can be fully considered, and the group deviation degree of the clinical data of the patient is calculated by associating the extracted clinical data feature value of the patient with the clinical data feature value of the same type of patient group; the model is constructed based on the comprehensive correlation coefficient of the patient's disease and the historical medical image and the group deviation degree of the patient's clinical data, the disease progress category of the patient is predicted, the patient's disease can be analyzed more comprehensively, and the disease progress of the patient can be managed better.

Description

Technical Field

[0001] The present invention relates to the field of nuclear medicine imaging technology, and in particular to an analysis method, system, device and medium for nuclear medicine imaging radiological reports. Background Art

[0002] As an important branch of the medical field, nuclear medicine, with the help of radionuclide tracer technology, can provide in-depth insights into the physiological and pathological processes of the human body at the molecular and cellular levels. Nuclear medicine imaging radiology reports are the key summary and interpretation of these examination results, providing an indispensable basis for clinical diagnosis, treatment plan formulation and disease monitoring.

[0003] With the rapid development of medical technology, nuclear medicine imaging is increasingly used in clinical practice, bringing revolutionary breakthroughs in the diagnosis and treatment of many diseases. However, the existing nuclear medicine imaging analysis technology still has many shortcomings that need to be addressed.

[0004] For example, the existing Chinese patent with application number 202410933655.8 discloses an analysis method, system, equipment and medium for nuclear medicine image radiology reports. The solution ensures that the data in nuclear medicine and radiology reports can be referenced to each other by establishing a corresponding relationship, performs two-way search and matching of data between nuclear medicine and radiology reports, and combines radiology structured report information with nuclear medicine reference navigation maps to enhance the accuracy and efficiency of diagnosis. After the nuclear medicine report is completed, it is compared with the radiology report, solving the problem of low efficiency and easy omissions in the current manual analysis of nuclear medicine image radiology reports.

[0005] However, the above patent has the following problems: First, the solution analyzes and evaluates the imaging data of a single examination, and does not involve a comprehensive analysis of the patient's own historical nuclear medicine imaging data corresponding to the data. It may be difficult to detect subtle progression of the disease at the individual level, and cannot fully consider the individual trajectory of the patient's disease development, and has limited ability to detect new changes in the patient's condition at an early stage.

[0006] Second, this solution does not involve placing the individual patient medical imaging data corresponding to the report in the context of a similar patient group, making it difficult to determine whether the symptoms reflected in the report generated based on the individual patient's medical imaging data are within the normal fluctuation range of the group or are abnormal. During the analysis process, this solution does not involve the differences in medical imaging data between individual patients, and uses a single standard to analyze the medical imaging data of all individual patients. As a result, important clues to the development of the disease may be missed, and the unique progression of the patient's disease cannot be accurately identified.

[0007] For example, the existing Chinese patent application number 202410966597.9 discloses a method and system for interpreting medical imaging reports based on a multi-layer deep learning model. The solution collects historical reports for preprocessing, determines the impact of text data, and then builds a model with a specific structure. The data is divided into training, verification, and test sets. The optimal model is obtained through training, evaluation, parameter optimization and fine-tuning. Finally, the preprocessed data of the report to be interpreted is input into the model to obtain the interpretation result, which solves the problem of poor readability of medical imaging reports in the prior art.

[0008] However, the above patent has the following problems: the solution focuses on the medical imaging report to be interpreted, and is not combined with the patient's real-time data or other relevant clinical information, making it difficult to fully reflect the patient's condition. The medical imaging report is only part of the basis for diagnosis. Only by combining the patient's real-time data and other relevant clinical information can the patient's overall condition be more accurately grasped, thereby making more appropriate medical decisions. Summary of the invention

[0009] In order to overcome the shortcomings of the background technology, the embodiments of the present invention provide a method, system, device and medium for analyzing nuclear medicine imaging radiology reports, which can effectively solve the problems involved in the above-mentioned background technology.

[0010] The purpose of the present invention can be achieved through the following technical solutions: The first aspect of the present invention provides an analysis method for nuclear medicine imaging radiology reports, the method comprising the following steps: S1. Acquire the patient's nuclear medicine imaging data and clinical data.

[0011] S2. Extract feature values ​​from the patient's nuclear medicine imaging data and clinical data respectively to obtain feature values ​​of the patient's nuclear medicine imaging data and clinical data.

[0012] S3. Correlate the extracted nuclear medicine imaging data feature values ​​of the patient with the feature values ​​of the patient's own historical nuclear medicine imaging data, and evaluate the comprehensive correlation coefficient between the patient's symptoms and his own historical nuclear medicine imaging by assigning correlation factors.

[0013] S4. Associate the extracted clinical data characteristic values ​​of the patient with the clinical data characteristic values ​​of the same type of patient group, and calculate the group deviation of the patient's clinical data.

[0014] S5. Build a model based on the comprehensive correlation coefficient between the patient's symptoms and his or her historical nuclear medicine images and the group deviation of the patient's clinical data to predict the patient's disease progression category.

[0015] The patient clinical data includes the patient's gene sequencing data, physical data, and medical history text data.

[0016] According to an achievable method of the first aspect of the present invention, the specific analysis method of step S2 is: S21. Obtain historical nuclear medicine imaging data of each part from the management database, input it into the deep learning model for training, use the trained model to analyze the patient's nuclear medicine images, identify the various parts in the patient's nuclear medicine images, perform model segmentation, and extract the feature representation corresponding to each part at the same time to form a feature vector of each part, which is regarded as the patient's nuclear medicine imaging data feature value.

[0017] S22. Obtain the patient's genetic sequencing data, compare it with the reference genome, and determine the position of each sequencing fragment in the reference genome. If the position of a sequencing fragment in the reference genome coincides with a known disease-related gene region, and there is a difference between the sequencing fragment and the reference genome, it is judged to be a gene mutation site. Screen out gene mutation sites associated with specific diseases, and calculate the frequency of occurrence of the patient's specific mutation site.

[0018] S23. Collect the patient's physical data and perform derivative feature calculations to obtain the patient's BMI and body fat percentage.

[0019] S24. Read the patient's medical history text data, structure the unstructured medical history text data, count the number of attacks of a specific disease from the structured medical history data, and calculate the frequency of disease attacks in combination with time information.

[0020] S25. Extract the degree of symptom improvement of the patient after each treatment, quantify it into specific indicators and classify it. The degree of symptom improvement is divided into significant improvement, partial improvement and no improvement.

[0021] S26. The frequency of occurrence of the patient’s specific mutation sites, BMI, body fat percentage, disease attack frequency, and degree of symptom improvement are regarded as the patient’s clinical data characteristic values.

[0022] According to an achievable method of the first aspect of the present invention, the specific operation method of assigning the correlation factor is: A1. Obtain the patient's own historical nuclear medicine images, mark the corresponding shooting time points for them, sort the patient's own historical nuclear medicine images according to the time points according to the clinical demand time period, assign the maximum time correlation factor to the historical nuclear medicine image with the closest time point, and use the linear attenuation function to calculate the time correlation factors of the patient's other historical nuclear medicine images.

[0023] A2. Determine the target site for the current disease diagnosis, set the basic site association factor, and adjust the site association factor based on the degree of influence of other sites that have an impact on the target site.

[0024] According to an achievable method of the first aspect of the present invention, the specific operation method of the comprehensive correlation coefficient between the patient's symptoms and his own historical nuclear medicine images is: using the feature vectors of each part, matching the target part in each historical nuclear medicine image of the patient, obtaining each historical nuclear medicine image of the patient corresponding to the target part, extracting the characteristic changes of the target part, recording them as the characteristic changes of each target part, and calculating the comprehensive correlation coefficient between the patient's symptoms and his own historical nuclear medicine images : F = ∑ p = 1 l { C p ∗ [ F max ∗ e − k ∗ ( T now − T p Δ T ) ] + ∑ i = 1 n ( F target ∗ I i ) } ,in Indicates The characteristic change of the target part, For the The number of the target site, , Indicates the time point at which the current patient's nuclear medicine image was taken. Indicates The time point at which the patient's historical nuclear medicine images were taken for each target part. is the maximum time correlation factor set, Indicates the preset time decay factor, Indicates the preset reference shooting interval duration. is a natural constant, represents the basic site association factor, Indicates the preset patient's own The influence factor of the corresponding part of the historical nuclear medicine image on the target part, Indicates that the patient The serial number of your own historical nuclear medicine images, , when When the corresponding part of the historical nuclear medicine image is the target part, .

[0025] According to an achievable method of the first aspect of the present invention, the specific analysis method of step S4 is: S41. Based on the patient's clinical data, determine the inclusion criteria for patients of the same type, obtain the clinical data of the same type of patient group that meets the inclusion criteria from the management database, and perform feature extraction on the clinical data of the same type of patient group.

[0026] S42. Quantify and encode the degree of symptom improvement in the patient's clinical data and the clinical data of the same type of patient group, and then calculate the means and standard deviations of the clinical data of the same type of patient group.

[0027] S43. Calculate the difference between each characteristic value of the patient's clinical data and the mean of the clinical data corresponding to the same type of patient group, and then divide it by the standard deviation of the mutation site occurrence frequency, BMI, body fat percentage, disease onset frequency, and symptom improvement degree of the same type of patient group to confirm the relative position of each characteristic value of the patient's clinical data among the characteristic values ​​of the clinical data of the same type of patient group.

[0028] S44. Calculate the group deviation of the patient's clinical data based on the relative position of each characteristic value of the patient's clinical data in the clinical data of the same type of patient group : ,in They respectively represent the relative positions of the occurrence frequency of a patient's specific mutation site, BMI, body fat percentage, disease attack frequency, and symptom improvement degree in the corresponding characteristic values ​​of the clinical data of the same type of patient group. They respectively represent the weight factors corresponding to the frequency of occurrence of specific mutation sites, BMI, body fat percentage, disease onset frequency, and degree of symptom improvement.

[0029] According to an achievable method of the first aspect of the present invention, the specific operation method of step S5 is: S51. Analyze each patient of the same type based on the comprehensive correlation coefficient between the patient's symptoms and his or her own historical nuclear medicine images and the group deviation of the patient's clinical data, and divide the patients into "low risk", "medium risk" and "high risk" disease progression categories according to the set classification criteria, thereby constructing a sample set of each disease progression category.

[0030] S52. Take the features and feature values ​​of the clinical data in each disease progression category sample set as splitting points, calculate the Gini index value of each feature in each disease progression category sample set, select the feature and feature value with the smallest Gini index, record the feature as the optimal splitting feature, and record the feature value as the optimal splitting point. Divide the disease progression category sample set according to the optimal splitting point, and recursively repeat the calculation of the Gini index, select the optimal splitting feature and splitting point for the obtained subsets until the number of samples in the subset is less than the set threshold, thereby constructing a random forest model composed of each decision tree.

[0031] S53. Input the nuclear medicine imaging data and clinical data of the selected patient into each decision tree in the random forest in turn, and output the disease progression category prediction of the selected patient.

[0032] The second aspect of the present invention provides an analysis system for nuclear medicine imaging radiology reports, the system specifically comprising the following modules: a medical data acquisition module, for acquiring nuclear medicine imaging data and clinical data of a patient.

[0033] The feature extraction module is used to extract feature values ​​based on the acquired patient nuclear medicine imaging data and clinical data.

[0034] The historical data association module is used to associate the extracted nuclear medicine imaging data characteristic values ​​of the patient with the patient's own historical nuclear medicine imaging data characteristic values, and to evaluate the comprehensive correlation coefficient between the patient's symptoms and his own historical nuclear medicine imaging by assigning correlation factors.

[0035] The patient data association module is used to associate the extracted clinical data characteristic values ​​of the patient with the clinical data characteristic values ​​of the same type of patient group, and calculate the group deviation of the patient's clinical data.

[0036] The model building module is used to build a model based on the comprehensive correlation coefficient between the patient's symptoms and his or her own historical nuclear medicine images and the group deviation of the patient's clinical data to predict the patient's disease progression category.

[0037] A third aspect of the present invention provides an analysis device for nuclear medicine imaging radiology reports, the device comprising a processor, a memory and a communication bus, wherein the memory stores a computer-readable program executable by the processor.

[0038] The communication bus realizes the connection and communication between the processor and the memory.

[0039] When the processor executes the computer-readable program, the processor implements the steps of a method for analyzing nuclear medicine imaging radiology reports according to the present invention.

[0040] A fourth aspect of the present invention provides an analysis medium for nuclear medicine imaging radiology reports, wherein the medium is burned with a computer program, and when the computer program is executed in the memory of a server, the analysis method for nuclear medicine imaging radiology reports described in the present invention is implemented.

[0041] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: First, the present invention evaluates the comprehensive correlation coefficient between the patient's symptoms and his or her own historical nuclear medicine images by correlating the extracted characteristic values ​​of the patient's nuclear medicine imaging data with the characteristic values ​​of the patient's own historical nuclear medicine imaging data. This can fully take into account the individual trajectory of the patient's disease development. Even if the changes in symptoms are relatively subtle, it is more likely to detect new changes in the patient's condition early compared to methods that only use a single image or do not consider individual historical conditions.

[0042] 2. The present invention calculates the group deviation of the patient's clinical data by associating the extracted clinical data characteristic values ​​of the patient with the clinical data characteristic values ​​of the same type of patient group, which can measure the degree of abnormality of the patient relative to the same type of patient group, and helps to accurately identify those individuals at high risk in the group.

[0043] 3. The present invention constructs a model based on the comprehensive correlation coefficient between the patient's symptoms and his or her own historical nuclear medicine images and the group deviation of the patient's clinical data to predict the patient's disease progression category. Compared with the single-dimensional prediction model, the multi-dimensional analysis method can more comprehensively consider the factors affecting disease progression, thereby improving the accuracy of disease progression category prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0045] Figure 1 The present invention is a flowchart of the steps for implementing the method.

[0046] Figure 2 It is a schematic diagram of system module connection of the present invention.

[0047] Figure 3 The figure is a schematic diagram of an embodiment of a device in an example of the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Example 1

[0050] See also Figure 1 As shown, the present invention provides an analysis method for nuclear medicine imaging radiology reports, comprising the following steps: S1. Acquire nuclear medicine imaging data and clinical data of a patient.

[0051] The patient clinical data includes the patient's gene sequencing data, physical data, and medical history text data.

[0052] S2. Extract feature values ​​from the patient's nuclear medicine imaging data and clinical data respectively to obtain feature values ​​of the patient's nuclear medicine imaging data and clinical data.

[0053] The specific analysis method of step S2 is: S21. Obtain historical nuclear medicine imaging data of each part from the management database, input it into the deep learning model for training, use the trained model to analyze the patient's nuclear medicine images, identify the various parts in the patient's nuclear medicine images, perform model segmentation, and extract the feature representation corresponding to each part at the same time to form a feature vector of each part, which is regarded as the patient's nuclear medicine imaging data feature value; it is helpful for early detection of the disease, and when there is a slight change in the characteristics of certain parts, it can be recognized by the model, thereby achieving early intervention.

[0054] It should be noted that the specific training steps of the deep learning model are: S211. Divide the historical nuclear medicine imaging data of each part into a training set, a validation set and a test set according to a set ratio, and perform a normalized grayscale value operation to map the grayscale value of the historical nuclear medicine imaging data of each part to a specific range.

[0055] S212. Select a loss function and an optimization algorithm, input the preprocessed historical nuclear medicine imaging data of each part into the selected deep learning model, calculate the prediction results of the model through forward propagation, use the loss function to predict the loss value between the result and the true label, calculate the gradient of the loss value to the model parameters through the back propagation algorithm, and use the optimization algorithm to update the parameters of the model according to the calculated gradient.

[0056] S213. After each training round, calculate the performance index of the model on the validation set (a part of the data reserved during data division, which does not participate in training), record the loss value and validation set performance index during the training process, draw the loss curve and performance index curve, and when the validation set performance index no longer improves, the model training is considered to have converged and the training process is terminated.

[0057] It should be noted that, in a specific embodiment, the patient's lung nuclear medicine image is input into the trained deep learning model. The model identifies the left upper lobe of the lung by analyzing the patient's lung nuclear medicine image. The segmentation function of the model is used to outline the boundary range of the left upper lobe of the lung, and obtain its feature representation, such as the average grayscale value and texture complexity of the boundary area of ​​the left upper lobe of the lung. The average grayscale value after normalization is 120, and the texture complexity is 0.8 according to the texture calculation method. The shape feature is expressed by a three-dimensional vector. [ 0 . 2 , 0 . 1 , 0 . 3 ] It represents the quantitative value of different shape features, and the feature vector of the left upper lobe is [ 120 , 0 . 8 , 0 . 2 , 0 . 1 , 0 . 3 ] .

[0058] S22. Obtain the patient's genetic sequencing data, compare it with the reference genome, and determine the position of each sequencing fragment in the reference genome. If the position of a sequencing fragment in the reference genome coincides with a known disease-related gene region, and there is a difference between the sequencing fragment and the reference genome, it is judged to be a gene mutation site. The gene mutation sites associated with specific diseases are screened out, and the frequency of occurrence of the patient's specific mutation sites is calculated, which can clarify the patient's genetic-level pathogenic factors or disease risk factors.

[0059] It should be noted that the specific operation method for calculating the frequency of occurrence of a patient's specific mutation site is as follows: for each determined gene mutation site associated with a specific disease, count the number of times it appears in the patient's gene sequencing data, divide the number of occurrences of each mutation site by the total number of sequenced fragments, and obtain the frequency of occurrence of the mutation site in the patient's gene.

[0060] S23. Collect the patient's physical data and perform derivative feature calculations to obtain the patient's BMI and body fat percentage; it can quantify the patient's body composition, reflect the patient's physical condition from a macro perspective, and assess the patient's overall health level and the risk of obesity-related diseases. For example, an excessively high BMI may be associated with an increased risk of various chronic diseases such as cardiovascular disease and diabetes.

[0061] It should be noted that physical fitness data include basic physiological indicators such as height, weight, waist circumference, etc., which are recorded as , , , .

[0062] S24. Read the patient's medical history text data, structure the unstructured medical history text data, count the number of attacks of a specific disease from the structured medical history data, and calculate the frequency of disease attacks in combination with time information; this helps to better judge the development trend of the disease and the effect of treatment.

[0063] It should be noted that the specific operation method for structuring unstructured medical history text data is: S241. Eliminate irrelevant symbols, special characters, and garbled codes in the patient's medical history text data, unify the format of dates, numbers, units, etc. in the text, and use natural language processing tools to segment continuous text into independent phrases.

[0064] S242. Through named entity recognition technology, key entities in the text are identified, such as basic information of patients, disease names, symptoms, examination items, treatment methods and drug names, the relationship between each entity is determined, and the structured information is stored in the corresponding database according to the designed data structure.

[0065] It should be noted that, in a specific embodiment, the relationship between "patient" and "disease" is a disease relationship, the relationship between "disease" and "symptoms" is a manifestation relationship, and the relationship between "treatment means" and "disease" is a treatment relationship, etc.

[0066] It should be noted that the specific operating method for calculating the frequency of disease attacks is: extracting time information related to disease attacks from structured medical history data, including the time of consultation, the time when symptoms first appeared, and the time of disease diagnosis, unifying the extracted time information into a standard time format, and determining the independent events of each disease attack based on the disease name and symptom description records in the structured data, counting the number of disease attacks within a set period, and dividing the number of disease attacks by the corresponding period length to obtain the frequency of disease attacks.

[0067] S25. Extract the degree of symptom improvement of patients after each treatment, quantify it into specific indicators and classify it. The degree of symptom improvement is divided into significant improvement, partial improvement and no improvement, which can intuitively evaluate the treatment effect.

[0068] It should be noted that the specific analysis method for the degree of symptom improvement is: counting the frequency of disease attacks within a set time period before and after each treatment of the patient, dividing the frequency of disease attacks within a set time period before each treatment of the patient by the frequency of disease attacks within a set time period after each treatment of the patient, to obtain the disease attack reduction rate of each treatment of the patient, setting the disease attack reduction rate range for significant improvement, partial improvement, and no improvement respectively, and obtaining the degree of symptom improvement of the patient after each treatment through matching.

[0069] S26. The frequency of occurrence of the patient's specific mutation sites, BMI, body fat percentage, disease attack frequency, and symptom improvement degree are regarded as the patient's clinical data characteristic values; integrating data from multiple aspects into clinical data characteristic values ​​can comprehensively describe the patient's clinical condition.

[0070] The specific operation method for assigning the correlation factor is as follows: A1. Obtain the patient's own historical nuclear medicine images, mark the corresponding shooting time points for them respectively, sort the patient's own historical nuclear medicine images according to the time points according to the clinical demand time period, assign the maximum time correlation factor to the historical nuclear medicine image with the latest time point, and use the linear attenuation function to calculate the time correlation factors of the patient's other historical nuclear medicine images; in the diagnosis process, recent images can often reflect the latest status of the patient's current condition and should be given more attention, while earlier images may also contain useful information, but the importance is relatively low. In this way, image information at different times can be systematically integrated to improve the accuracy of diagnosis.

[0071] It should be noted that the specific analysis method for calculating the time correlation factor of the patient's other historical nuclear medicine images is: setting the slope , through the formula Calculate the time correlation factor of the patient's own remaining historical nuclear medicine images, Indicates the time interval between the specified nuclear medicine image acquisition time and the most recent nuclear medicine image acquisition time. is the intercept; as the time interval The increase of time correlation factor It will gradually decrease according to a linear law, showing that the farther the image is from the current time, the lower the degree of correlation.

[0072] A2. Determine the target site for the current disease diagnosis, set the basic site association factor, and adjust the site association factor based on the degree of influence of other sites that have an impact on the target site; this can improve the comprehensiveness of the diagnosis. The occurrence and development of many diseases are not only limited to the target site itself, but also related to surrounding or associated sites. For example, heart disease may be related to lung function, vascular condition, etc. Considering these associated sites and adjusting the association factors can more comprehensively evaluate the condition of the target site.

[0073] It should be noted that the specific operation method of adjusting the part correlation factor is: setting the basic part correlation factor , based on the normal physiological functions of the human body, analyze the internal connection between the corresponding parts and the target parts, and thus set the influence factors of each part on the target part.

[0074] S3. Correlate the extracted nuclear medicine imaging data feature values ​​of the patient with the feature values ​​of the patient's own historical nuclear medicine imaging data, and evaluate the comprehensive correlation coefficient between the patient's symptoms and his own historical nuclear medicine imaging by assigning correlation factors.

[0075] The specific operation method of the comprehensive correlation coefficient between the patient's symptoms and his own historical nuclear medicine images is as follows: using the characteristic vectors of each part, the target part in the patient's own historical nuclear medicine images is matched, and the patient's own historical nuclear medicine images corresponding to the target part are obtained, and the characteristic changes of the target part are extracted, which are recorded as the characteristic changes of each target part; the situation of the target part in different historical images can be accurately located, which helps to more accurately judge the development process of the disease in the target part.

[0076] It should be noted that, in a specific embodiment, a total of 5 nuclear medicine images of the lungs of a designated patient taken every year in the past 5 years are obtained, and the 5 nuclear medicine images are respectively marked with corresponding timestamps. For example, the earliest one was taken on January 1, 2020, recorded as t1, and the one taken on January 5, 2021 was recorded as t2, and so on to obtain the nuclear medicine images of the lungs of the patient at each time point.

[0077] The method in step S21 is used to obtain the feature vectors of each part. For the patient's right middle lobe, the same right middle lobe in the patient's lung nuclear medicine images at each time point is matched, and then for the matched right middle lobe, the feature changes of the patient's right middle lobe over time are extracted. For details, please refer to Table 1, which lists some representative data.

[0078] Table 1. Some of the feature vectors of the right middle lobe collected

[0079]

[0080] Calculate the comprehensive correlation coefficient between the patient's symptoms and their historical nuclear medicine images : F = ∑ p = 1 l { C p ∗ [ F max ∗ e − k ∗ ( T now − T p Δ T ) ] + ∑ i = 1 n ( F target ∗ I i ) } ,in Indicates The characteristic change of the target part, For the The number of the target site, , Indicates the time point at which the current patient's nuclear medicine image was taken. Indicates The time point at which the patient's historical nuclear medicine images were taken for each target part. is the maximum time correlation factor set, Indicates the preset time decay factor, Indicates the preset reference shooting interval duration. is a natural constant, represents the basic site association factor, Indicates the preset patient's own The influence factor of the corresponding part of the historical nuclear medicine image on the target part, Indicates that the patient The serial number of your own historical nuclear medicine images, , when When the corresponding part of the historical nuclear medicine image is the target part, ; It can be used to analyze the commonalities and differences in the association between symptoms and historical images in different patient groups, which helps to discover the inherent laws and characteristics of the disease.

[0081] It should be noted that the time correlation and location correlation of the target part are two important influencing factors in the comprehensive correlation coefficient between the patient's symptoms and his or her own historical nuclear medicine images. In the first part, It reflects the degree of change in the characteristics of the target area from history to the present. This reflects the influence of time factors on the degree of association. Indicates the time point at which the current patient's nuclear medicine image was taken. Indicates The time points of the patient's own historical nuclear medicine imaging corresponding to the target part are The increase of The value of will decrease, which means that the correlation between historical images and current images will decrease over time. In the second part, It is right The patient's own historical nuclear medicine images are traversed. It is a value set according to the actual situation and is used to measure the influence of the historical image part on the target part.

[0082] Assumptions , , , the simulation results can be obtained. Please refer to Table 2 for details. Table 2 lists some representative data.

[0083] Table 2. Some of the collected target site related data and calculation results

[0084]

[0085] The above simulation data is used to calculate: , that is, the comprehensive correlation coefficient between the patient's symptoms and his or her historical nuclear medicine images is 3.06.

[0086] S4. Associate the extracted clinical data characteristic values ​​of the patient with the clinical data characteristic values ​​of the same type of patient group, and calculate the group deviation of the patient's clinical data.

[0087] The specific analysis method of step S4 is: S41. Based on the patient's clinical data, determine the inclusion criteria for patients of the same type, obtain the clinical data of the same type of patient groups that meet the inclusion criteria from the management database, and extract features from the clinical data of the same type of patient groups; this can ensure that the same type of patient groups studied have a high correlation and improve the accuracy of subsequent analysis.

[0088] S42. Quantitatively encode the degree of symptom improvement in the patient's clinical data and the clinical data of the same type of patient group, and then calculate the means and standard deviations of the clinical data of the same type of patient group; quantitative coding provides a basis for scientific analysis and can more objectively compare the differences in symptom improvement between different patients.

[0089] It should be noted that the specific operating method for the means and standard deviations of the clinical data of the same type of patient groups is to set quantitative intervals for "significant improvement", "partial improvement" and "no improvement" respectively, and for the target patients, encode the degree of symptom improvement after each treatment according to the above-mentioned quantitative correspondence, and encode the degree of symptom improvement for the clinical data of the same type of patient groups according to a unified quantitative correspondence. For the symptom improvement degree data of the same type of patient groups after quantitative encoding, calculate the mean, and on the basis of the calculated mean, further calculate the standard deviation. According to this method, the means and standard deviations of the clinical data of the same type of patient groups are calculated.

[0090] S43. Obtain the difference between each characteristic value of the patient's clinical data and the mean of the clinical data corresponding to the same type of patient group, and then divide it by the standard deviation of the mutation site occurrence frequency, BMI, body fat percentage, disease onset frequency, and symptom improvement degree of the same type of patient group to confirm the relative position of each characteristic value of the patient's clinical data among the characteristic values ​​of the clinical data of the same type of patient group; each characteristic value of the patient can be converted to a relative scale with the same type of patient group as a reference, which helps to identify the difference between the patient and the same type of patient group, and provide a reference for personalized diagnosis and treatment.

[0091] It should be noted that the specific calculation method for the difference between each characteristic value of the patient's clinical data and the mean of the clinical data corresponding to the same type of patient group is: obtain each characteristic value of the patient's clinical data and the mean of the clinical data corresponding to the same type of patient group, match the clinical data belonging to each characteristic value of the patient one by one to obtain the clinical data corresponding to the same type of patient group, and obtain the difference between each characteristic value of the patient's clinical data and the mean of the clinical data corresponding to the same type of patient group by subtracting them.

[0092] S44. Calculate the group deviation of the patient's clinical data based on the relative position of each characteristic value of the patient's clinical data in the clinical data of the same type of patient group : ,in They respectively represent the relative positions of the occurrence frequency of a patient's specific mutation site, BMI, body fat percentage, disease attack frequency, and symptom improvement degree in the corresponding characteristic values ​​of the clinical data of the same type of patient group. They respectively represent the frequency of occurrence of specific mutation sites, BMI, body fat percentage, disease attack frequency, and the weight factors corresponding to the degree of symptom improvement; they can more accurately reflect the patient's special circumstances, help to better locate the patient's disease characteristics in the group context, and provide an important basis for subsequent disease progression predictions.

[0093] It should be noted that It can be set to 0.3. It can be set to 0.2. It can be set to 0.2. It can be set to 0.15. It can be set to 0.15. Specific mutation sites are often closely related to the pathogenesis of the disease. For example, in some genetic diseases, specific gene mutations are the root cause of the disease, and changes in their frequency of occurrence may directly reflect the genetic tendency, risk of disease and potential pathological and physiological processes of the disease. BMI and body fat percentage are important indicators for measuring the overall health status and metabolic level of patients. They are closely related to a variety of chronic diseases, such as cardiovascular disease and diabetes. In patients with the same type of disease, abnormal BMI and body fat percentage may affect the progression of the disease and treatment response. Although the frequency of disease attacks and the degree of symptom improvement are also important aspects of evaluating the patient's condition, they are relatively more susceptible to The influence of multiple factors, including treatment plans, patient compliance, lifestyle changes, etc., are not as closely related to the intrinsic nature of the disease as specific mutation sites, BMI and body fat percentage. The degree of symptom improvement may be subjective, and different patients may have different perceptions and descriptions of symptoms. At the same time, the frequency of disease attacks and the degree of symptom improvement are often stage-by-stage manifestations in the development of the disease, with a certain lag, and cannot directly reflect the potential characteristics of the disease like specific mutation sites. The weight corresponding to the frequency of occurrence of the patient's specific mutation site is the largest, followed by the weights corresponding to BMI and body fat percentage, and the weights corresponding to the frequency of disease attacks and the degree of symptom improvement are the smallest.

[0094] It should be noted that there are five influencing factors in the group deviation of the patient's clinical data, namely, the frequency of occurrence of specific mutation sites, BMI, body fat percentage, disease attack frequency, and symptom improvement. The role of different clinical characteristics in assessing the degree of patient deviation from the group is comprehensively considered, and the deviation information of multiple dimensions is integrated together by weighted summation. Finally, the absolute value is taken because the focus is on the degree of deviation, rather than the direction of deviation (whether it is higher or lower than the group mean). The value obtained in this way can fully reflect the degree of deviation of the patient relative to the same type of patient group under the combined effect of multiple clinical characteristics. For details, please refer to Table 3, which lists some representative data.

[0095] Table 3. Corresponding characteristic values ​​of some collected clinical data

[0096]

[0097] Through the above simulation calculation, patient 4's group deviation of 1.5 is the highest among all patients, indicating that this patient has the greatest difference from the group in comprehensive clinical characteristics, while patient 1 has the lowest group deviation of 0.241, which is relatively small, indicating that patient 4's condition or physical condition is more special among this group of patients, and more personalized considerations may be needed when formulating treatment plans or judging the progression of the disease. Compared with patients 2 and 3, although both have higher group deviations, patient 2 has higher frequency of occurrence of specific mutation sites, BMI, and body fat percentage. The value of patient 3 contributes more to the deviation, while patient 3 mainly contributes to the frequency of occurrence of specific mutation sites and the frequency of disease onset. The value has a significant impact on the degree of deviation, which indicates that the main driving characteristics of deviations in different patients and groups are different.

[0098] S5. Build a model based on the comprehensive correlation coefficient between the patient's symptoms and his or her historical nuclear medicine images and the group deviation of the patient's clinical data to predict the patient's disease progression category.

[0099] The specific operation method of step S5 is: S51. Analyze each similar patient based on the comprehensive correlation coefficient between the patient's symptoms and his own historical nuclear medicine images and the group deviation analysis method of the patient's clinical data, and divide the patients into "low risk", "medium risk" and "high risk" disease progression categories according to the set classification criteria, thereby constructing a sample set for each disease progression category; this helps to have a preliminary and intuitive judgment on the severity and development trend of the patient's condition.

[0100] S52. Take the features and eigenvalues ​​of the clinical data in each disease progression category sample set as splitting points, calculate the Gini index value of each feature in each disease progression category sample set, select the feature and eigenvalue with the smallest Gini index, record the feature as the optimal splitting feature, and record the eigenvalue as the optimal splitting point. Divide the disease progression category sample set according to the optimal splitting point, and recursively repeat the calculation of the Gini index, the selection of the optimal splitting feature and the splitting point for the obtained subsets until the number of samples in the subset is less than the set threshold, thereby constructing a random forest model composed of each decision tree; the minimum Gini index means that when the sample set is divided according to the feature and eigenvalue, the impurity of the sample set can be minimized, so that the divided subsets are more "pure" in terms of category, which is conducive to in-depth research on diseases at different stages of progression, such as exploring the differences in pathological mechanisms, treatment responses, etc. of diseases of different risk categories.

[0101] It should be noted that the specific operation method for calculating the Gini index value of each feature in each disease progression category sample set is: for each disease progression category sample set, calculate its overall Gini index, and the calculation formula of the Gini index is: ,in Indicates categories, , It is The proportion of class samples in the sample set is calculated by dividing the sample set based on the splitting point, and the Gini index of each sub-sample set after division is calculated. The weighted summation is performed according to the size of the sub-sample set to obtain the characteristic Gini index based on the splitting point.

[0102] S53. Input the nuclear medicine imaging data and clinical data of the selected patient into each decision tree in the random forest in turn, and output the disease progression category prediction of the selected patient; different decision trees may have different preliminary judgments on the same patient, but by combining the results of each decision tree, the erroneous predictions made by individual decision trees due to data noise or local characteristics can be reduced.

[0103] Example 2

[0104] Reference Figure 2 As shown, the present invention proposes an analysis system for nuclear medicine imaging radiology reports, comprising the following modules: a medical data acquisition module, for acquiring nuclear medicine imaging data and clinical data of patients.

[0105] The feature extraction module is used to extract feature values ​​based on the acquired patient nuclear medicine imaging data and clinical data.

[0106] The historical data association module is used to associate the extracted nuclear medicine imaging data characteristic values ​​of the patient with the patient's own historical nuclear medicine imaging data characteristic values, and to evaluate the comprehensive correlation coefficient between the patient's symptoms and his own historical nuclear medicine imaging by assigning correlation factors.

[0107] The patient data association module is used to associate the extracted clinical data characteristic values ​​of the patient with the clinical data characteristic values ​​of the same type of patient group, and calculate the group deviation of the patient's clinical data.

[0108] The model building module is used to build a model based on the comprehensive correlation coefficient between the patient's symptoms and his or her own historical nuclear medicine images and the group deviation of the patient's clinical data to predict the patient's disease progression category.

[0109] Management database, used to store historical nuclear medicine imaging data of various parts.

[0110] The management database is connected to the feature extraction module, the historical data association module, the patient data association module, and the model construction module; the model construction module is connected to the historical data association module and the patient data association module; the feature extraction module is connected to the medical data acquisition module, the historical data association module, and the patient data association module.

[0111] Example 3

[0112] See also Figure 3 As shown, the present invention proposes an analysis device for nuclear medicine imaging radiology reports, the device comprising a processor, a memory and a communication bus, the memory storing a computer-readable program executable by the processor.

[0113] The communication bus realizes the connection and communication between the processor and the memory.

[0114] When the processor executes the computer-readable program, the processor implements the steps of a method for analyzing nuclear medicine imaging radiology reports according to the present invention.

[0115] Example 4

[0116] The present invention proposes an analysis medium for nuclear medicine imaging radiology reports, wherein the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in an analysis method for nuclear medicine imaging radiology reports as described in the present invention.

[0117] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations on the present invention. A person skilled in the art may make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention and they are still covered by the protection scope of the present invention.

Claims

1. A method for analyzing nuclear medicine imaging radiology reports, characterized in that: The steps include: S1. Obtain the patient's nuclear medicine imaging data and clinical data; S2. Extracting feature values ​​from the patient's nuclear medicine imaging data and clinical data respectively to obtain feature values ​​from the patient's nuclear medicine imaging data and clinical data; S3. Correlating the extracted nuclear medicine imaging data characteristic values ​​of the patient with the characteristic values ​​of the patient's own historical nuclear medicine imaging data, and evaluating the comprehensive correlation coefficient between the patient's symptoms and his own historical nuclear medicine imaging by assigning correlation factors; S4. Associating the extracted clinical data characteristic values ​​of the patient with the clinical data characteristic values ​​of the same type of patient group, and calculating the group deviation of the patient's clinical data; S5. Build a model based on the comprehensive correlation coefficient between the patient's symptoms and his or her historical nuclear medicine images and the group deviation of the patient's clinical data to predict the patient's disease progression category.

2. The method for analyzing nuclear medicine radiology reports according to claim 1, characterized in that: The patient clinical data includes the patient's gene sequencing data, physical data, and medical history text data.

3. The method for analyzing nuclear medicine radiology reports according to claim 2, characterized in that: The specific analysis method of step S2 is: S21. Obtain historical nuclear medicine image data of each part from the management database, input it into the deep learning model for training, use the trained model to analyze the patient's nuclear medicine images, identify each part in the patient's nuclear medicine images, perform model segmentation, and extract feature representations corresponding to each part to form feature vectors of each part, which are regarded as feature values ​​of the patient's nuclear medicine image data; S22. Obtain the patient's gene sequencing data, compare it with the reference genome, determine the position of each sequencing fragment in the reference genome, and if the position of a sequencing fragment in the reference genome coincides with a known disease-related gene region, and the sequencing fragment is different from the reference genome, it is determined to be a gene mutation site, and the gene mutation sites associated with a specific disease are screened out, and the frequency of occurrence of the patient's specific mutation site is calculated; S23. Collect the patient's physical data and perform derived feature calculation to obtain the patient's BMI and body fat percentage; S24. Read the patient's medical history text data, structure the unstructured medical history text data, count the number of attacks of a specific disease from the structured medical history data, and calculate the frequency of disease attacks in combination with time information; S25. Extract the degree of symptom improvement after each treatment, quantify it into specific indicators and classify it into significant improvement, partial improvement and no improvement; S26. The frequency of occurrence of the patient’s specific mutation sites, BMI, body fat percentage, disease attack frequency, and degree of symptom improvement are regarded as the patient’s clinical data characteristic values.

4. The method for analyzing nuclear medicine radiology reports according to claim 1, characterized in that: The specific operation method of allocating the correlation factor is: A1. Obtain the patient's own historical nuclear medicine images, mark the corresponding shooting time points for each of them, sort the patient's own historical nuclear medicine images according to the time points according to the clinical demand time period, assign the maximum time correlation factor to the historical nuclear medicine image with the closest time point, and use the linear attenuation function to calculate the time correlation factors of the patient's other historical nuclear medicine images; A2. Determine the target site for the current disease diagnosis, set the basic site association factor, and adjust the site association factor based on the degree of influence of other sites that have an impact on the target site.

5. The method for analyzing nuclear medicine radiology reports according to claim 3, characterized in that: The specific operation method of the comprehensive correlation coefficient between the patient's symptoms and his / her historical nuclear medicine images is as follows: Using the feature vectors of each part, the target parts in the patient's own historical nuclear medicine images are matched to obtain the patient's own historical nuclear medicine images corresponding to the target part, and the characteristic changes of the target part are extracted, recorded as the characteristic changes of each target part, and the comprehensive correlation coefficient between the patient's symptoms and his own historical nuclear medicine images is calculated. : ,in Indicates The characteristic change of the target part, For the The number of the target site, , Indicates the time point at which the current patient's nuclear medicine image was taken. Indicates The time point at which the patient's historical nuclear medicine images were taken for each target part. is the maximum time correlation factor set, Indicates the preset time decay factor, Indicates the preset reference shooting interval duration. is a natural constant, represents the basic site association factor, Indicates the preset patient's own The influence factor of the corresponding part of the historical nuclear medicine image on the target part, Indicates that the patient The serial number of your own historical nuclear medicine images, , when When the corresponding part of the historical nuclear medicine image is the target part, .

6. The method for analyzing nuclear medicine radiology reports according to claim 5, characterized in that: The specific analysis method of step S4 is: S41. Determine the inclusion criteria for patients of the same type based on the patient's clinical data, obtain clinical data of a group of patients of the same type that meet the inclusion criteria from the management database, and perform feature extraction on the clinical data of the group of patients of the same type; S42. Quantify and encode the degree of symptom improvement in the clinical data of the patient and the clinical data of the same type of patient group, and then calculate the means and standard deviations of the clinical data of the same type of patient group; S43. Calculate the difference between each characteristic value of the patient's clinical data and the mean of the clinical data corresponding to the same type of patient group, and then divide it by the standard deviation of the mutation site occurrence frequency, BMI, body fat percentage, disease attack frequency, and symptom improvement degree of the same type of patient group to confirm the relative position of each characteristic value of the patient's clinical data among the characteristic values ​​of the clinical data of the same type of patient group; S44. Calculate the group deviation of the patient's clinical data based on the relative position of each characteristic value of the patient's clinical data in the clinical data of the same type of patient group : ,in They respectively represent the relative positions of the occurrence frequency of a patient's specific mutation site, BMI, body fat percentage, disease attack frequency, and symptom improvement degree in the corresponding characteristic values ​​of the clinical data of the same type of patient group. They respectively represent the weight factors corresponding to the frequency of occurrence of specific mutation sites, BMI, body fat percentage, disease onset frequency, and degree of symptom improvement.

7. The method for analyzing nuclear medicine radiology reports according to claim 1, characterized in that: The specific operation method of step S5 is: S51. Analyze the patients of the same type based on the comprehensive correlation coefficient between the patient's symptoms and their own historical nuclear medicine images and the group deviation analysis method of the patient's clinical data, and divide the patients into "low risk", "medium risk" and "high risk" disease progression categories according to the set classification criteria, thereby constructing a sample set for each disease progression category; S52. Taking each feature and feature value of the clinical data in each disease progression category sample set as the splitting point, calculate the Gini index value of each feature in each disease progression category sample set, select the feature and feature value with the smallest Gini index, record the feature as the optimal splitting feature, and record the feature value as the optimal splitting point, divide each disease progression category sample set according to the optimal splitting point, and recursively repeat the calculation of the Gini index, the selection of the optimal splitting feature and the splitting point for the obtained subsets, until the number of samples in the subset is less than the set threshold, thereby constructing a random forest model composed of each decision tree; S53. Input the nuclear medicine imaging data and clinical data of the selected patient into each decision tree in the random forest in turn, and output the disease progression category prediction of the selected patient.

8. An analysis system for nuclear medicine imaging radiology reports, characterized in that: The system specifically includes the following modules: Medical data acquisition module, used to obtain nuclear medicine imaging data and clinical data of patients; A feature extraction module is used to extract feature values ​​from the patient's nuclear medicine imaging data and clinical data, respectively, to obtain the patient's nuclear medicine imaging data feature values ​​and clinical data feature values; A historical data association module is used to associate the extracted nuclear medicine imaging data characteristic values ​​of the patient with the characteristic values ​​of the patient's own historical nuclear medicine imaging data, and to evaluate the comprehensive correlation coefficient between the patient's symptoms and his own historical nuclear medicine imaging by assigning correlation factors; A patient data association module is used to associate the extracted clinical data characteristic values ​​of the patient with the clinical data characteristic values ​​of the same type of patient group, and calculate the group deviation of the patient's clinical data; A model building module is used to build a model based on the comprehensive correlation coefficient between the patient's symptoms and his or her own historical nuclear medicine images and the group deviation of the patient's clinical data to predict the patient's disease progression category; Management database, used to store historical nuclear medicine imaging data of various parts.

9. An analysis device for nuclear medicine imaging radiology reports, the device comprising a processor, a memory and a communication bus, the memory storing a computer readable program executable by the processor; The communication bus realizes the connection and communication between the processor and the memory; When the processor executes the computer-readable program, the processor implements the steps of the method for analyzing nuclear medicine imaging radiology reports as described in any one of claims 1 to 7.

10. An analytical medium for nuclear medicine imaging radiological reports, characterized in that: The medium is burned with a computer program, and when the computer program is run in the memory of the server, the method described in any one of claims 1 to 7 is implemented.

Images