A medical case image processing and analysis system

Abstract

The present invention discloses a medical case image processing and analysis system, including a model recognition end, which is used to use a number of existing case images with object names marked as training sets to train a deep learning model, and obtain an approved recognition model after the training, and use the approved recognition model to recognize the object to be tested photographed by the user in real time, and output the names of the objects to be confirmed of all organs recognized in the image to be tested, where the object to be tested is also a pathological image; synchronously, according to the previously confirmed symptom descriptions and associated objects, when the user photographs the object to be tested, the real-time symptom description recorded by the doctor when diagnosing the owner of the image to be tested will be automatically called, and the real-time symptom description and the symptom description will be matched according to the rules of the relevant matching model to obtain a number of potential associated objects, and re-marked as potential objects; the key position in the case image is locked according to the potential object, that is, the observation target is confirmed.

Description

Technical Field

[0001] The invention belongs to the field of image processing and analysis, and in particular is a medical case image processing and analysis system. Background Art

[0002] The patent with the publication number CN109716445A discloses a method for retrieving similar cases based on the distribution of lesions in diseases with lesions distributed in organs such as diffuse lung disease. The present invention is a similar case image retrieval program, which enables a computer to perform the following processing: extracting an organ region from a medical image and dividing the organ region into a plurality of regions; counting the number of pixels representing the lesions in each of the divided regions; and determining a similar case image corresponding to the similarity of the number of pixels representing the lesions by referring to a storage unit storing the number of pixels representing the lesions in each region.

[0003] However, for medical case images, how to build a system that can possibly identify the target location in the case image and circle it out, so that some interns or novice doctors can quickly lock the target area and put teaching into practice, and at the same time, it can also organize the user's related consultation situation according to the corresponding situation of the case image, so that the doctor can grasp the progress of the corresponding disease in time, whether it is getting better or worse, there is a lack of a method. Based on this, a solution is now provided. Summary of the invention

[0004] The purpose of the present invention is to provide a medical case image processing and analysis system.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A medical case image processing and analysis system, comprising:

[0007] The model recognition end is used to train the deep learning model with the help of several existing case images with object names labeled as training sets. The trained model is marked as a recognition model, and then the recognition model is tested with the help of a test set. After the test, the recognition model that can ensure that the recognition accuracy exceeds the set value X1 is marked as an approved recognition model.

[0008] Then, the object to be tested photographed by the user in real time is identified with the help of the verification recognition model, and the names of all organs identified in the image to be tested are output, and the object to be tested is also the pathological image;

[0009] Then, a matching model is established with the help of association analysis. The matching model is used based on previously confirmed disease descriptions and associated objects. Disease descriptions refer to descriptions of the patient's physical symptoms for any disease condition, while associated objects refer to the names of damaged organs that may correspond to the disease descriptions. Then, when the user takes a photo of the object to be tested, the real-time disease description recorded by the doctor when diagnosing the owner of the image to be tested is automatically called. The real-time disease description and the disease description are matched according to the rules of the relevant matching model to obtain several potential associated objects, and are re-marked as potential objects.

[0010] The potential object is compared with all the names of the objects to be confirmed output in the image to be verified, and the names of the objects to be confirmed that are consistent with the potential object are marked as confirmed observation targets, so as to obtain the objects to be verified that contain the confirmed observation targets.

[0011] Furthermore, the model recognition end includes:

[0012] A data import unit is used to import training data for model training. The training data is a film of a human body undergoing an imaging examination. The film is marked as a case image, and the object name of the corresponding examination item on each case image is marked;

[0013] The data import unit is used to transfer all the training data to the model building unit, and the model building unit is used to perform model training on the training data to obtain an approved recognition model.

[0014] Furthermore, the specific steps of model training are:

[0015] First, all case images are acquired, and classified according to the object names marked on the case images, and case images with the same object name are divided into an analysis data group, which contains several case images with the same object name;

[0016] Select any analysis data group and put it into the deep learning model as a training set for training, so as to train the model according to the case images in the analysis data group, and mark the model as a recognition model after the training is completed;

[0017] After the training is completed, several test sets will be automatically set to test the accuracy of the recognition model;

[0018] Several case images with the same object name but not labeled are used as test images, which are transmitted to the recognition model. Then, the input result of the recognition model is verified to be correct, and the accuracy of the recognition model output is automatically obtained. It is ensured that an approved recognition model with an accuracy exceeding the set value X1 can be obtained.

[0019] The approved recognition model is stored in the model building unit.

[0020] Furthermore, during the process of building the approval recognition model, when the accuracy is lower than the set value X1, the case images that have been labeled with the corresponding object name will continue to be acquired, and the number of re-acquired case images that is not less than half of the number of case images acquired for the first time will be marked as a supplementary training set;

[0021] The recognition model is trained again according to the supplementary training set. After the training is completed, the accuracy of the recognition model after the second training needs to be re-verified with the help of the test image. If it is still lower than X1, the second training will be repeated according to the above steps until the accuracy is not lower than X1, and then the training will be stopped.

[0022] The same training method is used to train the recognition model for all other case images with different object names, until all case images with different object names can be recognized, and the accuracy is not less than X1;

[0023] At this time, the qualified recognition model will be automatically obtained and marked as the approved recognition model.

[0024] Furthermore, the object name is the name of the organ being examined, and the organ corresponding to the object name will be circled separately on the case image and attached with the object name.

[0025] Furthermore, the deep learning model uses a convolutional neural network (CNN) model to identify several training sets. Of course, the deep learning model also includes vector machines (SVM) and random forests.

[0026] Furthermore, the data import unit is used to import the case image generated by the user at the latest moment, mark it as the object to be tested, and transmit the object to be tested to the comparison unit, and the comparison unit is used to identify the object to be tested with the help of the approved recognition model stored in the model building unit, output the object names of all identifiable organs in the image to be tested, and mark it as the name of the object to be confirmed;

[0027] Through association analysis, the required target case images are analyzed. The specific method of association analysis is as follows:

[0028] First, a recognition model library will be established. The recognition model library stores several targeted disease descriptions and associated objects. Disease descriptions refer to the descriptions of the patient's physical symptoms for any disease condition, while associated objects refer to the names of damaged organs that may correspond to the disease descriptions. In other words, the corresponding associated objects are damaged, which leads to the appearance of content related to the disease descriptions. Here, all disease descriptions and associated objects have been verified manually or by other means in the past, and the disease descriptions are also entered by doctors;

[0029] When the user's latest image to be examined is imported, the real-time description of the disease recorded by the doctor when diagnosing the owner of the image to be examined will be automatically called;

[0030] Then, the real-time symptom description is segmented to obtain a number of real-time segmented words, and then the duplicates in the real-time segmented words are removed, and the remaining number of real-time segmented words is marked as the number of sub-items;

[0031] Select any one of the symptom descriptions and perform word segmentation on the symptom description to obtain several symptom word segmentations;

[0032] The number of real-time segmented words that are consistent with the symptom segmented words is obtained, and the value obtained by dividing the number by the number of segmented words is marked as the symptom similarity number;

[0033] Then the real-time segmentation is compared with other symptom descriptions in the same way to obtain the symptom similarity number between the real-time segmentation and all symptom descriptions;

[0034] Mark the associated objects corresponding to the disease descriptions whose disease similarity number exceeds B1 as potential objects, where B1 is a preset value;

[0035] A number of potential objects are obtained, and then compared with all the names of the objects to be confirmed output in the image to be verified, and the names of the objects to be confirmed that are consistent with the potential objects are marked as confirmed observation targets, so as to obtain the objects to be verified that contain the confirmed observation targets.

[0036] Furthermore, a fusion end is also included, and the fusion end includes:

[0037] The data accumulation unit is used to collect the medical records of any user each time they see a doctor, and to form targeted treatment information. The treatment information includes consultation time, consultation description and treatment record. The treatment record is used to describe the specific situation of each treatment.

[0038] The data accumulation unit is connected with the data fusion unit;

[0039] The comparison unit is also used to transmit the object to be tested including the confirmed observation target, the real-time description of the symptoms and its real-time segmentation to the data fusion unit. The data fusion unit is used to describe the pathological process in combination with the data accumulation unit. The specific method of describing the pathological process is as follows:

[0040] Obtain the real-time description of the symptoms and its real-time segmentation, and compare the real-time segmentation with the medical inquiry description;

[0041] Perform word segmentation on the medical inquiry description to obtain a number of medical inquiry word segmentations, obtain the number of real-time word segmentations that are consistent with the medical inquiry word segmentations, divide the number by the number of segmentation items, and mark the value obtained as the medical inquiry similarity value. Mark the treatment information corresponding to the medical inquiry description whose medical inquiry similarity value exceeds B2 as reference information;

[0042] Arrange several reference information in order from morning to night according to the consultation time marked therein, and then form a diagnosis and treatment status list in chronological order for all treatment records in the reference information;

[0043] The data fusion unit is used to transmit the diagnosis and treatment status list to the display unit for real-time display.

[0044] Beneficial effects of the present invention:

[0045] The present invention uses a model recognition end to train a deep learning model with the help of a number of existing case images with object names already marked as training sets, and obtains an approved recognition model after training. The approved recognition model is used to recognize the object to be tested photographed by the user in real time, and the names of all organs to be confirmed in the image to be tested are output, and the object to be tested is a pathological image.

[0046] The system will also automatically call the real-time symptom description recorded by the doctor when diagnosing the owner of the image to be tested based on the previously confirmed symptom descriptions and associated objects. The real-time symptom description and the symptom description will be matched according to the rules of the relevant matching model to find several potential associated objects, and re-mark them as potential objects. The key positions in the case image will be locked based on the potential objects, that is, the observation target will be confirmed.

[0047] At the same time, the user's past related disease diagnosis records will be integrated with this one with the help of the fusion terminal, so that the doctor can better judge the patient's condition this time; the present invention is simple, effective, and easy to use. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0049] Figure 1 This is a connection structure diagram of the model recognition end and the fusion end of the present invention.

[0050] Figure 2 It is a specific structural diagram of the system block diagram of the present invention. DETAILED DESCRIPTION

[0051] like Figure 1-Figure 2 As shown, a medical case image processing and analysis system, as embodiment 1 of the present application, specifically includes:

[0052] Model recognition end;

[0053] The model recognition end specifically includes a data import unit, a model building unit and a comparison unit;

[0054] The data import unit is used to import training data for model training. The training data is a film of a human body undergoing ultrasound imaging or other imaging examinations. At this time, the film is marked as a case image. The object name of the corresponding examination item on each case image is marked to facilitate training. The object name is the name of the corresponding organ being examined. The organ corresponding to the object name will be circled separately on the case image and attached with the object name;

[0055] The data import unit is used to transfer all training data to the model building unit and perform the model training steps:

[0056] First, all case images are acquired, and classified according to the object names marked on the case images, and case images with the same object name are divided into an analysis data group, which contains several case images with the same object name;

[0057] Select any analysis data group and put it into the neural network model as a training set for training. Here, a convolutional neural network CNN model is used to identify several training sets. Of course, other models can also be used here, such as support vector machine SVM, random forest and other deep learning models, to train the model according to the case images in the analysis data group. After the training is completed, the model is marked as a recognition model;

[0058] After the training is completed, several test sets will be automatically set to test the accuracy of the recognition model;

[0059] Several case images with the same object name but not labeled are used as test images, which are transmitted to the recognition model. Then, the correctness of the recognition model input result is verified, and the accuracy of the recognition model output is automatically obtained.

[0060] When the accuracy is lower than the set value X1, the case images that have been labeled with the corresponding object name will continue to be acquired, and the number of re-acquired case images is not less than half of the number of case images acquired for the first time, and they will be marked as supplementary training sets;

[0061] The recognition model is trained again according to the supplementary training set. After the training is completed, the accuracy of the recognition model after the second training needs to be re-verified with the help of the test image. If it is still lower than X1, the second training will be repeated according to the above steps until the accuracy is not lower than X1, and then the training will be stopped.

[0062] The same training method is used to train the recognition model for all other case images with different object names, until all case images with different object names can be recognized, and the accuracy is not less than X1;

[0063] At this time, a qualified recognition model will be automatically obtained and marked as an approved recognition model;

[0064] storing the approved recognition model in the model building unit;

[0065] The data import unit is also used to import the case image generated by the user at the latest moment, and mark it as the object to be tested. The data import unit is also used to transmit the object to be tested to the comparison unit, and the comparison unit is used to identify the object to be tested with the help of the approved recognition model stored in the model building unit, output the object names of all identifiable organs in the image to be tested, and mark it as the name of the object to be confirmed;

[0066] Through association analysis, the required target case images are analyzed. The specific method of association analysis is as follows:

[0067] First, a recognition model library will be established. The recognition model library stores several targeted disease descriptions and associated objects. Disease descriptions refer to the descriptions of the patient's physical symptoms for any disease condition, while associated objects refer to the names of damaged organs that may correspond to the disease descriptions. In other words, the corresponding associated objects are damaged, which leads to the appearance of content related to the disease descriptions. Here, all disease descriptions and associated objects have been verified manually or by other means in the past, and the disease descriptions are also entered by doctors;

[0068] When the user's latest image to be examined is imported, the real-time description of the disease recorded by the doctor when diagnosing the owner of the image to be examined will be automatically called;

[0069] Then, the real-time symptom description is segmented to obtain a number of real-time segmented words, and then the duplicates in the real-time segmented words are removed, and the remaining number of real-time segmented words is marked as the number of sub-items;

[0070] Select any one of the symptom descriptions and perform word segmentation on the symptom description to obtain several symptom word segmentations;

[0071] The number of real-time segmented words that are consistent with the symptom segmented words is obtained, and the value obtained by dividing the number by the number of segmented words is marked as the symptom similarity number;

[0072] Then the real-time segmentation is compared with other symptom descriptions in the same way to obtain the symptom similarity number between the real-time segmentation and all symptom descriptions;

[0073] Mark the associated objects corresponding to the disease descriptions whose disease similarity number exceeds B1 as potential objects, where B1 is a preset value;

[0074] A number of potential objects are obtained, and then compared with all the names of the objects to be confirmed outputted in the image to be verified, and the names of the objects to be confirmed that are consistent with the potential objects are marked as confirmed observation targets, so as to obtain the objects to be verified that contain the confirmed observation targets;

[0075] As the second embodiment of the present application, the present application also includes

[0076] The fusion end includes:

[0077] Data integration unit, data accumulation unit and display unit;

[0078] The data accumulation unit is used to collect the medical records of any user each time they see a doctor, and to form targeted treatment information. The treatment information includes consultation time, consultation description and treatment records. The treatment records are used to describe the specific circumstances of each treatment, what medicines were used, the corresponding disease progression, etc.

[0079] The data accumulation unit is connected with the data fusion unit;

[0080] The comparison unit is also used to transmit the object to be tested including the confirmed observation target, the real-time description of the symptoms and its real-time segmentation to the data fusion unit. The data fusion unit is used to describe the pathological process in combination with the data accumulation unit. The specific method of describing the pathological process is as follows:

[0081] Obtain the real-time description of the symptoms and its real-time segmentation, and compare the real-time segmentation with the medical inquiry description;

[0082] Perform word segmentation on the medical inquiry description to obtain a number of medical inquiry word segmentations, obtain the number of real-time word segmentations that are consistent with the medical inquiry word segmentations, divide the number by the number of segmentation items, and mark the value obtained as the medical inquiry similarity value. Mark the treatment information corresponding to the medical inquiry description whose medical inquiry similarity value exceeds B2 as reference information;

[0083] Arrange several reference information in order from morning to night according to the consultation time marked therein, and then form a diagnosis and treatment status list in chronological order for all treatment records in the reference information;

[0084] The data fusion unit is used to transmit the diagnosis and treatment status list to the display unit for real-time display.

[0085] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A medical case image processing and analysis system, characterized in that: include: The model recognition end is used to train the deep learning model with the help of several existing case images with object names labeled as training sets. The trained model is marked as a recognition model, and then the recognition model is tested with the help of a test set. After the test, the recognition model that can ensure that the recognition accuracy exceeds the set value X1 is marked as an approved recognition model. The model recognition end includes: The data import unit is used to import training data for model training. The training data is a film of a human body undergoing an imaging examination. The film is marked as a case image. The object name of the corresponding examination item on each case image is marked. The object name is the name of the corresponding organ being examined. The organ corresponding to the object name will be circled separately on the case image and attached with the object name. The data import unit is also used to import the case image generated by the user at the latest moment, and mark it as the object to be tested. The data import unit is also used to transmit the object to be tested to the comparison unit, and the comparison unit is used to identify the object to be tested with the help of the approved recognition model stored in the model building unit, output the object names of all identifiable organs in the image to be tested, and mark it as the name of the object to be confirmed; The object to be tested photographed by the user in real time is identified by the verification recognition model, and the name of the object to be confirmed of all organs identified in the image to be tested is output. The object to be tested is also the pathological image; A matching model is established with the help of association analysis. The matching model is used to analyze several potential associated objects based on the previously confirmed disease descriptions and associated objects, and the real-time disease description recorded by the doctor when the user takes the photo of the object to be tested and the disease description when diagnosing the owner of the image to be tested according to the rules of the relevant matching model, and mark the associated objects here as potential objects; the disease description refers to the description of the patient's physical symptoms for any disease condition, and the associated object refers to the name of the damaged organ that may correspond to the disease description; Through association analysis, the required target case images are analyzed. The specific method of association analysis is as follows: First, a recognition model library will be established, which stores several targeted disease descriptions and associated objects. Disease descriptions refer to the descriptions of the patient's physical symptoms for any disease condition, while associated objects refer to the names of damaged organs that may correspond to the disease descriptions. Disease descriptions are entered by doctors. When the user's latest image to be examined is imported, the real-time description of the disease recorded by the doctor when diagnosing the owner of the image to be examined will be automatically called; Then, the real-time symptom description is segmented to obtain a number of real-time segmented words, and then the duplicates in the real-time segmented words are removed, and the remaining number of real-time segmented words is marked as the number of sub-items; Select any one of the symptom descriptions and perform word segmentation on the symptom description to obtain several symptom word segmentations; The number of real-time segmented words that are consistent with the symptom segmented words is obtained, and the value obtained by dividing the number by the number of segmented words is marked as the symptom similarity number; Then the real-time segmentation is compared with other symptom descriptions in the same way to obtain the symptom similarity number between the real-time segmentation and all symptom descriptions; Mark the associated objects corresponding to the disease descriptions whose disease similarity number exceeds B1 as potential objects, where B1 is a preset value; A number of potential objects are obtained, and then compared with all the names of the objects to be confirmed outputted in the image to be verified, and the names of the objects to be confirmed that are consistent with the potential objects are marked as confirmed observation targets, so as to obtain the objects to be verified that contain the confirmed observation targets; Compare the potential object with all the names of the objects to be confirmed output in the image to be verified, mark the names of the objects to be confirmed that are consistent with the potential object as confirmed observation targets, and obtain the objects to be verified that contain the confirmed observation targets; It also includes the fusion end, which includes: The data accumulation unit is used to collect the medical records of any user each time they see a doctor, and to form targeted treatment information. The treatment information includes consultation time, consultation description and treatment record. The treatment record is used to describe the specific situation of each treatment. The data accumulation unit is connected with the data fusion unit; The comparison unit is also used to transmit the object to be tested including the confirmed observation target, the real-time description of the symptoms and its real-time segmentation to the data fusion unit. The data fusion unit is used to describe the pathological process in combination with the data accumulation unit. The specific method of describing the pathological process is as follows: Obtain the real-time description of the symptoms and its real-time segmentation, and compare the real-time segmentation with the medical inquiry description; Perform word segmentation on the medical inquiry description to obtain a number of medical inquiry word segmentations, obtain the number of real-time word segmentations that are consistent with the medical inquiry word segmentations, divide the number by the number of segmentation items, and mark the value obtained as the medical inquiry similarity value. Mark the treatment information corresponding to the medical inquiry description whose medical inquiry similarity value exceeds B2 as reference information; Arrange several reference information in order from morning to night according to the consultation time marked therein, and then form a diagnosis and treatment status list in chronological order for all treatment records in the reference information; The data fusion unit is used to transmit the diagnosis and treatment status list to the display unit for real-time display.

2. A medical case image processing and analysis system according to claim 1, characterized in that: The data import unit is used to transfer all the training data to the model building unit, and the model building unit is used to perform model training on the training data to obtain an approved recognition model.

3. A medical case image processing and analysis system according to claim 1, characterized in that: The specific steps of model training are: First, all case images are acquired, and classified according to the object names marked on the case images, and case images with the same object name are divided into an analysis data group, which contains several case images with the same object name; Select any analysis data group and put it into the deep learning model as a training set for training, so as to train the model according to the case images in the analysis data group, and mark the model as a recognition model after the training is completed; After the training is completed, several test sets will be automatically set to test the accuracy of the recognition model; Several case images with the same object name but without labels are used as test images, which are transmitted to the recognition model, and then the input result of the recognition model is verified to be correct, and the accuracy of the recognition model output is automatically obtained; when the accuracy does not reach X1, the recognition model is retrained to ensure that an approved recognition model with an accuracy exceeding the set value X1 is obtained; the approved recognition model is stored in the model construction unit.

4. A medical case image processing and analysis system according to claim 3, characterized in that: During the process of building the verification recognition model, when the accuracy is lower than the set value X1, the case images that have been labeled with the corresponding object name will continue to be acquired, and the number of re-acquired case images that is not less than half of the number of case images acquired for the first time will be marked as a supplementary training set; The recognition model is trained again according to the supplementary training set. After the training is completed, the accuracy of the recognition model after the second training needs to be re-verified with the help of the test image. If it is still lower than X1, the second training will be repeated according to the above steps until the accuracy is not lower than X1, and then the training will be stopped. The same training method is used to train the recognition model for all other case images with different object names, until all case images with different object names can be recognized, and the accuracy is not less than X1; At this time, the qualified recognition model will be automatically obtained and marked as the approved recognition model.

5. A medical case image processing and analysis system according to claim 1, characterized in that: The deep learning model uses a convolutional neural network (CNN) model to identify several training sets.

6. A medical case image processing and analysis system according to claim 1, characterized in that: The deep learning model is any one of the convolutional neural network (CNN), vector machine (SVM), and random forest.

Images