Artificial intelligence ultrasonic auxiliary system for liver

Abstract

The invention discloses an artificial intelligence ultrasonic auxiliary system for a liver. The artificial intelligence ultrasonic auxiliary system comprises ultrasonic scanning equipment and a model analysis server, a database, a human skeleton analysis model, a hepatic lesion analysis model and a data processing module are arranged in the model analysis server; a human skeleton analysis model is used for forming a normal texture map of the liver and the surrounding environment of a normal person after ultrasonic scanning; applying the obtained liver ultrasonic image data to a liver lesion analysis model to obtain a lesion liver and a peripheral lesion texture map; the data processing module comprises a data storage unit, a data early warning unit and a data sending unit. According to the method, each patient establishes a self-contrast healthy liver problem map and compares the healthy liver problem map with a lesion texture map, the liver texture map error is greatly reduced, a model is established in combination with factors of other diseases influencing the liver texture map for analysis, whether the liver texture map is abnormal due to other diseases or not is checked, and the liver texture map error is judged. And the possibility of misjudgment is further reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of liver detection devices, and in particular, to an artificial intelligence ultrasound assistance system for the liver. Background Art

[0002] Diffuse liver diseases refer to diffuse pathological changes in the liver parenchyma, and typical ones are fatty liver and liver fibrosis. Liver histopathological examination is the gold standard for diagnosing fatty liver and liver fibrosis, but it has certain invasiveness and poor patient compliance. After the liver shows diffuse lesions, the acoustic impedance of the lesion site also changes accordingly, and this change will be reflected in the ultrasound image. Therefore, ultrasound doctors can observe the change of the echo condition in the ultrasound image and diagnose whether it is fatty liver or liver fibrosis based on their experience. This method is relatively convenient, but has great subjectivity and requires many years of clinical experience. Therefore, an artificial intelligence method for classifying diffuse liver diseases based on ultrasound images has important significance in clinical practice.

[0003] For the degree of liver lesions and whether there are lesions, it is inaccurate to judge only by observing the change of the echo condition in the ultrasound image, and it is easily affected by factors such as self-obesity (high fat content), high blood lipid, and having other diseases. Therefore, it is very difficult to determine liver lesions and the degree of lesions only through the texture map obtained by ultrasound observation. Only experts who have treated liver diseases for many years can accurately judge whether a person has liver diseases and the degree of lesions. If the personnel in the ultrasound room do not have a high medical level, it is very difficult to quickly give the scanning results. If the patient does not receive timely treatment, delaying treatment after scanning will aggravate the degree of liver lesions. Existing ultrasound scanning rooms cannot arrange an expert to check the liver texture maps of each scanned patient every day. Experts cannot know the data of patients with relatively serious liver diseases in time and cannot give expert treatment suggestions in time. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems in the related art to some extent. For this reason, an object of the present invention is to provide an artificial intelligence ultrasound assistance system for the liver, which solves the problem that existing ultrasound scans cannot accurately give a relatively intuitive and accurate liver lesion texture map and requires expert analysis for accurate judgment.

[0005] An artificial intelligence ultrasound assistance system for the liver according to the present invention includes an ultrasound scanning device and a model analysis server; The ultrasound scanning device scans the human body to obtain liver ultrasound image data of the examinee; The model analysis server is provided with a database, a human skeleton analysis model, a liver lesion analysis model, and a data processing module; The model analysis server is connected to the hospital system network, obtains the upper body skeleton diagram data of the patient from the CT room of the hospital, inputs the patient's height, weight, and fat content data into the model analysis server, and uses the human skeleton analysis model to comprehensively analyze the upper body skeleton diagram data and the patient's height, weight, and fat content data to obtain the internal organ distribution and size structure diagram of the patient. Then, combined with the data of the normal liver and surrounding organs in the database, a normal texture map of the liver and the surrounding environment after ultrasound scanning of normal people is formed; Apply the obtained liver ultrasound image data to the liver lesion analysis model to obtain the lesion texture map of the diseased liver and the surrounding environment. Divide the normal texture map and the lesion texture map in the same way, compare the same regions, analyze the texture map differences between the normal texture map and the lesion texture map in different regions, mark the significance of the differences, and the marked comparison results include the normal range, mild range, moderate range, and severe range. Mark the marked comparison results on the lesion texture map according to the regions; The data processing module includes a data storage unit, a data warning unit, and a data sending unit. Each time the lesion texture map formed by the patient's ultrasound scan of the diseased liver and its corresponding normal texture map will be automatically saved to a folder and transmitted to the data storage unit. When comparing the lesion texture map with the normal texture map, a warning message will be sent if it is in the moderate range or above. At the same time, the lesion texture map and the normal texture map for which the warning message is sent will be sent to the terminal device of the reserved liver expert in the model analysis server so that the liver expert can timely see the information of patients with more serious liver lesions.

[0006] In some embodiments of the present invention, when the liver lesion analysis model uses the lesion texture map for analysis, a convolutional neural network based on the GoogleNet architecture is used to implement texture map classification by applying a deep learning algorithm.

[0007] In some other embodiments of the present invention, the database stores texture map data of thousands of normal people's livers, and these texture map data of thousands of normal people's livers are obtained by orthogonal implementation using the height, weight, and fat content data of different people to form a texture map database.

[0008] In some other embodiments of the present invention, if the patient uses the liver ultrasound image data of the ultrasound scanning device and inputs it into the liver lesion analysis model for analysis, and the obtained texture map is normal, but the patient's height, weight, and fat content data are different from the data stored in the database, the system will automatically incorporate the normal texture map data of the patient and its corresponding height, weight, and fat content data into the database for comparative analysis.

[0009] In some other embodiments of the present invention, the texture features of the image are extracted, specifically including the histogram features of the image, the features based on the gray-level co-occurrence matrix, the features based on the gray-level gradient co-occurrence matrix, the features based on the wavelet multi-subgraph co-occurrence matrix, the features based on the gray-level run-length matrix, the tamura features, and the laws features.

[0010] In some other embodiments of the present invention, an other-disease database of patients is established. Through liver experts, the influence of other diseases on the formation of texture maps of the liver by ultrasonic scanning is established, and an influence model of other diseases on the liver ultrasonic texture map is established. After obtaining the normal texture map using the human skeleton analysis model, if the patient has other diseases that affect the formation of the normal texture map of the liver by ultrasonic scanning, the patient's disease data is input into the model analysis server and applied to the influence model of other diseases on the liver ultrasonic texture map to generate an abnormal texture map of the normal liver affected by other diseases. If the diseased texture map of the diseased liver and its surrounding environment obtained from the liver lesion analysis model is very similar to the abnormal texture map, it indicates that the patient's liver is okay and the diseased texture map is generated under the influence of other diseases.

[0011] In some other embodiments of the present invention, a patient personal data storage unit is established in the model analysis server. The patient's historical diseased texture maps will be saved in his own personal data storage unit. Each time a new diseased texture map formed by ultrasonic scanning of the patient's liver is compared with the historical diseased texture maps to analyze whether the patient's diseased liver has worsened or improved.

[0012] In some other embodiments of the present invention, a lesion factor analysis model is set in the model analysis server. Liver experts establish a lesion factor analysis model based on the factors affecting liver lesions and the degree of influence. Patients input influence factor data according to the lesion factor analysis model to obtain liver lesion analysis cause data, and in combination with the diseased texture map obtained by ultrasonic scanning, to check whether the liver lesion is caused by influencing factors or abnormal body lesions.

[0013] In some other embodiments of the present invention, the ultrasonic scanning device scans the human body to obtain liver ultrasonic image data. One liver ultrasonic image is extracted from each certain number of frames in the video and used to construct an experimental data set with static images. The constructed liver ultrasonic data set is cleaned, and the liver structures that are not clear and non-liver diagnostic sections in the extracted data are removed, and finally a liver ultrasonic classification data set is formed.

[0014] In the present invention, according to the differences in the upper body skeletons and fat contents of each patient, it will affect the impact of ultrasound on the liver position texture atlas, as well as the impact of other diseases on the liver texture atlas. These factors are built into the model for analysis and comparison. In this way, each patient will establish a healthy liver problem atlas for self-comparison and compare it with the lesion texture atlas, greatly reducing the error of the liver texture atlas. Then, combined with the factors affecting the liver texture atlas of other diseases, a model is constructed for analysis to check whether other diseases cause abnormalities in the liver texture atlas, further reducing the possibility of misjudgment. It ensures that when non-liver experts view the texture atlas, they can accurately judge the liver texture atlas after ultrasound scanning, and can timely treat patients with severe liver lesions, greatly solving the situation where the patient's condition worsens due to delayed treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the drawings: Figure 1 FIG. is a schematic diagram of the principle of an artificial intelligence ultrasound-assisted system for the liver proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments.

[0017] As Figure 1 shown, an artificial intelligence ultrasound-assisted system for the liver proposed by the present invention includes an ultrasound scanning device and a model analysis server; The ultrasound scanning device scans the human body to obtain the liver ultrasound image data of the examinee; the ultrasound scanning device is the same as the existing scanning room and collects the same data, but the model analysis server has the function of analyzing the ultrasound scanning data to form a texture atlas, but it has the functions set below.

[0018] The model analysis server is provided with a database, a human skeleton analysis model, a liver lesion analysis model, and a data processing module; The model analysis server is connected to the hospital system network, obtains the upper body skeleton map data of the patient from the hospital CT room, inputs the patient's height, weight, and fat content data into the model analysis server, and uses the human skeleton analysis model to comprehensively analyze the upper body skeleton map data and the patient's height, weight, and fat content data to obtain the internal organ distribution and size structure diagram of the patient. Then, combined with the data of the normal liver and surrounding organs in the database, the normal texture atlas of the liver and the surrounding environment after ultrasound scanning of normal people is formed; A normal person can construct a healthy internal organ diagram (with a normal liver size and positional relationships with other internal organs) based on factors such as the human skeleton and height. By combining with a large number of liver texture atlases of normal people in the comparison database, data that is very similar to the patient's body type, skeleton, and weight is found, and the normal texture atlas is extracted for later comparison with the liver texture atlas obtained by ultrasonic scanning to analyze the degree of lesion.

[0019] The obtained liver ultrasonic image data is applied to the liver lesion analysis model to obtain the lesion texture atlas of the diseased liver and its surrounding environment. The normal texture atlas and the lesion texture atlas are partitioned in the same way, the same regions are compared, the differences in the texture atlases of the normal texture atlas and the lesion texture atlas in different regions are analyzed, and the significance of the differences is marked. The marked comparison results include the normal range, the mild range, the moderate range, and the severe range. The marked comparison results are marked on the lesion texture atlas according to the regions. Partitioning is carried out in the same way to enable more accurate comparison and analysis of each region, more effectively analyze the causes and degrees of liver lesions, and mark the degree of lesions in each region, so that even non-experts can complete accurate judgments.

[0020] The data processing module includes a data storage unit, a data warning unit, and a data sending unit. Each time the lesion texture atlas formed by the ultrasonic scan of the patient's diseased liver and its corresponding normal texture atlas will be automatically saved to a folder and transmitted to the data storage unit. When comparing the lesion texture atlas with the normal texture atlas, a warning message will be sent if it is in the moderate range or above. At the same time, the lesion texture atlas and the normal texture atlas for which the warning message is sent will be sent to the terminal devices of the reserved liver experts in the model analysis server so that the liver experts can timely see the information of patients with relatively severe liver lesions.

[0021] The results of each patient's scan are saved. For patients with relatively severe conditions, the data will be sent to the mobile terminal or computer terminal of the liver expert, and the expert can see it in a timely manner. If the expert sees it, for patients with relatively severe conditions, the hospital-related personnel will be notified by phone for corresponding emergency treatment, and then wait for the expert to come to the hospital for relevant treatment, without delaying the treatment.

[0022] When the liver lesion analysis model uses the lesion texture atlas for analysis, it adopts a convolutional neural network based on the GoogleNet architecture to apply deep learning algorithms to achieve texture atlas classification. It belongs to the commonly used analysis and processing methods for ultrasonic scanning, and the specific details will not be described. The hospital is equipped with relatively mature analysis methods.

[0023] The database stores thousands of texture atlas data of normal human livers. These thousands of texture atlas data of normal human livers are obtained through orthogonal implementation using height, weight, and fat content data of different people, forming a texture atlas database. The skeleton size and shape, weight, fat, etc. are used as a variable to search for the best database data, and the closest existing normal liver texture atlas in the database is used as a reference to reduce the comparison error.

[0024] When the liver ultrasound image data of a patient is input into the liver lesion analysis model for analysis, if the obtained texture atlas is normal, but the height, weight, and fat content data of the patient are different from the data stored in the database, the system will automatically include the normal texture atlas data of this patient and its corresponding height, weight, and fat content data into the database for comparative analysis.

[0025] For database update, the data of each scanned patient is compared with the database. If there is no very similar data in the database, the data of this patient (a person without liver disease) will be added to the database to enrich or increase the database capacity.

[0026] The texture features of the image are extracted, specifically including the histogram features of the image, features based on the gray-level co-occurrence matrix, features based on the gray-level gradient co-occurrence matrix, features based on the wavelet multi-subimage co-occurrence matrix, features based on the gray-level run-length matrix, tamura features, and laws features. Multiple commonly used features are incorporated into the analysis model to more comprehensively analyze and generate the texture atlas.

[0027] An other-disease database for patients is established. Liver experts establish the influence of other diseases on the formation of texture atlases of ultrasound-scanned livers, and establish a model for the influence of other diseases on liver ultrasound texture atlases; After obtaining a normal texture atlas using the human skeleton analysis model, if a patient has other diseases that affect the formation of a normal texture atlas of the ultrasound-scanned liver, the disease data of the patient is input into the model analysis server and applied to the model for the influence of other diseases on liver ultrasound texture atlases to generate an abnormal texture atlas of the normal liver affected by other diseases; If the lesion texture atlas of the diseased liver and its surrounding environment obtained from the liver lesion analysis model is very similar to the abnormal texture atlas, it indicates that the patient's liver is okay, and the lesion texture atlas is generated under the influence of other diseases.

[0028] Other diseases can also affect the texture atlas of the liver in the re-scanning room. Different diseases have different degrees of influence. Liver experts incorporate the concentrated diseases with greater influence into the system, and also incorporate the degree of influence of the diseases, facilitating accurate analysis of the influence of other diseases on the liver texture atlas to eliminate the influencing factors of other diseases.

[0029] A patient personal data storage unit is established within the model analysis server. The historical lesion texture atlas of the patient will be stored in their own personal data storage unit. Each time a patient undergoes an ultrasound scan of the liver to form a new lesion texture atlas, it is compared with the historical lesion texture atlas to analyze whether the patient's diseased liver has worsened or improved.

[0030] The patient personal data storage unit facilitates the storage of each scan data for later comparison when the patient comes again, so as to form a disease trend chart.

[0031] A lesion factor analysis model is set within the model analysis server. Liver experts establish a lesion factor analysis model based on the factors affecting liver lesions and the degree of influence. Patients input influence factor data according to the lesion factor analysis model to obtain liver lesion analysis cause data, and in combination with the lesion texture atlas obtained from the ultrasound scan, to check whether the liver lesion is caused by influencing factors or by abnormal body lesions.

[0032] Some other factors that affect the lesion or the liver scan results are input into the lesion factor analysis model for analysis to check whether they will affect the liver lesion, so as to eliminate the influence of other factors on the liver lesion or the texture atlas.

[0033] The ultrasound scanning device scans the human body to obtain liver ultrasound image data. One liver ultrasound image is extracted from each certain number of frames in the video and used to construct an experimental dataset with static images. The constructed liver ultrasound dataset is cleaned, and unclear liver structures and non-liver diagnostic sections in the extracted data are excluded. Finally, a liver ultrasound classification dataset is formed. Unclear or poor-quality ultrasound image data is excluded to form a more accurate texture atlas.

[0034] As mentioned above, the above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and all should be covered by the protection scope of the present invention.

Claims

1. An artificial intelligence ultrasound-assisted system for the liver, characterized in that: Includes ultrasound scanning equipment and model analysis server; The ultrasonic scanning device scans the human body to obtain ultrasonic image data of the liver of the examined person; The model analysis server is provided with a database, a human skeleton analysis model, a liver lesion analysis model and a data processing module; The model analysis server is connected to the hospital system network, and obtains the patient's upper body skeleton data from the hospital CT room, inputs the patient's height, weight and fat content data into the model analysis server, and uses the human skeleton analysis model to perform a comprehensive analysis on the upper body skeleton data and the patient's height, weight and fat content data to obtain the patient's internal organs distribution and size structure diagram, and then combines the normal human liver and surrounding organ data in the database to form a normal texture map of the liver and surrounding environment of a normal person after ultrasound scanning; Apply the acquired liver ultrasound image data to the liver lesion analysis model to obtain the lesion texture atlas of the lesion liver and the surrounding environment, divide the normal texture atlas and the lesion texture atlas in the same way, compare the same area, analyze the difference in texture atlas between the normal texture atlas and the lesion texture atlas in different areas, mark the significance of the difference, and give the results of the mark comparison in the normal range, mild range, moderate range and severe range, and mark the mark comparison results on the lesion texture atlas according to the area; The data processing module includes a data storage unit, a data warning unit and a data sending unit. Each time a patient completes an ultrasound scan of the diseased liver, a lesion texture map formed and its corresponding normal texture map will be automatically saved in a folder and transmitted to the data storage unit. The lesion texture map is compared with the normal texture map, and a warning message will be issued if a moderate range or above is found. At the same time, the lesion texture map and the normal texture map that issue the warning message will be sent to the terminal device reserved for liver experts in the model analysis server, so that liver experts can see the information of patients with more serious liver lesions in time.

2. The artificial intelligence ultrasound-assisted system for liver according to claim 1, characterized in that: When the liver lesion analysis model uses the lesion texture atlas for analysis, a convolutional neural network based on the GoogleNet architecture is used to apply a deep learning algorithm to achieve texture atlas classification.

3. The artificial intelligence ultrasound-assisted system for liver according to claim 1, characterized in that: The database stores thousands of texture atlas data of normal human livers, which are texture atlas databases obtained by orthogonally realizing the height, weight and fat content data of different people.

4. The artificial intelligence ultrasound-assisted system for liver according to claim 3, characterized in that: If the patient's liver ultrasound image data is input into the liver lesion analysis model using the ultrasound scanning device, and the texture map obtained is normal, but the patient's height, weight and fat content data are different from the data stored in the database, the system will automatically include the patient's normal texture map data and its corresponding height, weight and fat content data into the database for comparative analysis.

5. The artificial intelligence ultrasound-assisted system for liver according to claim 1, characterized in that: The texture features of the image are extracted, including the histogram features of the image, the features based on the gray level co-occurrence matrix, the features based on the gray level gradient co-occurrence matrix, the features based on the wavelet multi-subgraph co-occurrence matrix, the features based on the gray level run length matrix, the Tamura features and the Laws features.

6. The artificial intelligence ultrasound-assisted system for liver according to claim 1, characterized in that: Establish a database of other diseases of patients, establish the influence of other diseases on the texture map of the liver formed by ultrasound scanning through liver experts, and establish a model of the influence of other diseases on the ultrasound texture map of the liver; After obtaining a normal texture map using the human skeleton analysis model, if the patient has other diseases that affect the formation of a normal texture map of the liver by ultrasonic scanning, the patient's disease data is input into the model analysis server and applied to the model of the impact of other diseases on the ultrasonic texture map of the liver to generate an abnormal texture map of the normal liver affected by other diseases; If the lesion texture maps of the diseased liver and the surrounding environment obtained by the liver lesion analysis model are very similar to the abnormal texture map, it means that there is nothing wrong with the patient's liver, and the lesion texture map is generated under the influence of other diseases.

7. The artificial intelligence ultrasound-assisted system for liver according to claim 1, characterized in that: A patient personal data storage unit is established in the model analysis server, and the patient's historical lesion texture map will be saved in his or her own personal data storage unit. Each time the patient's liver is scanned by ultrasound, a new lesion texture map is formed and compared with the historical lesion texture map to analyze whether the patient's liver lesions have worsened or alleviated.

8. The artificial intelligence ultrasound-assisted system for liver according to claim 1, characterized in that: A lesion factor analysis model is set in the model analysis server. Liver experts establish the lesion factor analysis model based on the factors affecting liver lesions and the degree of influence. Patients input the influencing factor data according to the lesion factor analysis model to obtain liver lesion analysis cause data. The lesion texture atlas obtained by ultrasound scanning is combined to check whether the liver lesions are caused by the influencing factors or by abnormal lesions in the body.

9. The artificial intelligence ultrasound-assisted system for liver according to claim 1, characterized in that: The ultrasonic scanning device scans the human body to obtain liver ultrasonic image data, which is to extract a liver ultrasonic image at every certain frame interval in the video, and construct an experimental data set with the static image. The constructed liver ultrasonic data set is cleaned, and unclear liver structures and non-liver diagnostic sections in the extracted data are eliminated, and finally a liver ultrasonic classification data set is formed.

Images