A method and system for generating ultrasonic image examination report based on artificial intelligence

Abstract

The present invention relates to the technical field of image processing, and discloses a method and system for generating ultrasonic image inspection reports based on artificial intelligence. The method collects historical ultrasonic images to construct a training set and establishes ultrasonic image report standards. At the same time, by processing the historical ultrasonic images in the training set, the collected historical ultrasonic images and the real-time collected ultrasonic images are subjected to feature extraction and comparison using a feature extraction algorithm, and the type of the real-time collected ultrasonic images is determined, thereby improving the efficiency and accuracy of ultrasonic image inspection.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an ultrasonic image examination report generation method and system based on artificial intelligence. Background Art

[0002] As the complexity and specialization of ultrasound image interpretation tasks continue to increase, users have higher and higher requirements for the detection accuracy and efficiency of ultrasound image reports. The demand for using artificial intelligence image recognition to generate ultrasound image reports as an auxiliary tool for physicians' diagnosis and treatment is also becoming increasingly urgent.

[0003] Existing technologies include Chinese patent application CN112529857A, which predicts the position information of the organ area by inputting the image into the target detection model, and extracts the feature code of the organ area; at the same time, the extracted feature code is input into the language generation model, and the features at different times are decoded and encoded by the language generation model to generate words, and finally the generated words are combined into a sentence sequence, which is the final output diagnosis report. However, since the quality problem of the image itself is ignored, errors will be caused in the report generation, which has great limitations. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the shortcomings of the prior art, the present invention provides an ultrasound image examination report generation method and system based on artificial intelligence, which has the advantages of high efficiency and accuracy, and solves the problem of insufficient detection accuracy and efficiency of ultrasound image reports.

[0006] (II) Technical solution

[0007] In order to solve the technical problems of insufficient detection accuracy and efficiency of the above-mentioned ultrasound image report, the present invention provides the following technical solutions:

[0008] This embodiment discloses a method for generating an ultrasound image examination report based on artificial intelligence, which specifically includes the following steps:

[0009] S1. Collect historical ultrasound images and historical ultrasound image report data to build a training set, and establish ultrasound image report standards based on the collected historical ultrasound images and historical ultrasound image report data;

[0010] S2, processing the historical ultrasound images in the training set to obtain processed historical ultrasound images;

[0011] S3, extracting features from the processed historical ultrasound images by using a feature extraction algorithm, obtaining features of the processed historical ultrasound images and saving them;

[0012] S4, collecting and processing ultrasonic images in real time, and extracting features from the processed ultrasonic images collected in real time by using a feature extraction algorithm;

[0013] S5, comparing the features extracted in real time with the features of the historical ultrasound images, processing the ultrasound images corresponding to the features extracted in real time based on the comparison results, and determining the type of the ultrasound images collected in real time;

[0014] S6. After determining the type of the ultrasound image collected in real time, extract features based on the established ultrasound image report standard and the processed ultrasound image collected in real time to generate an ultrasound report.

[0015] The present invention constructs a training set by collecting historical ultrasound images and establishes ultrasound image reporting standards. At the same time, by processing the historical ultrasound images in the training set, the type of the real-time collected ultrasound images is determined by using a feature extraction algorithm to extract features from the collected historical ultrasound images and the real-time collected ultrasound images and comparing them, thereby improving the efficiency and accuracy of ultrasound image inspection.

[0016] Preferably, the establishing of ultrasound image report standards based on the collected historical ultrasound images and historical ultrasound image report data comprises the following steps:

[0017] The ultrasound image report standard only includes name, gender, ultrasound image type and ultrasound image analysis, which cannot be missing.

[0018] The present invention establishes an ultrasound image report standard based on collected historical ultrasound images and historical ultrasound image report data, determines various parameters in the ultrasound image report standard, and ensures the standardization of the ultrasound image report.

[0019] Preferably, the processing of the historical ultrasound images in the training set to obtain the processed historical ultrasound images comprises the following steps:

[0020] S21, filtering the historical ultrasound images in the training set by using a mean filtering algorithm to obtain filtered historical ultrasound images;

[0021] S22. Fitting the pixel points in the filtered historical ultrasonic image by using an image fitting algorithm to obtain a processed historical ultrasonic image.

[0022] Preferably, filtering the historical ultrasound images in the training set by a mean filtering algorithm to obtain the filtered historical ultrasound images comprises the following steps:

[0023] S211, setting a window, wherein the window includes the historical ultrasonic image pixels to be processed and the neighborhood around the historical ultrasonic image pixels;

[0024] S212, for each pixel in the window, calculating the product of the spatial domain weight and the gray value weight as the weight of the pixel;

[0025] The pixel weight calculation formula is as follows:

[0026] ;

[0027] ;

[0028] ;

[0029] in, represents the spatial domain weight, represents the gray value weight, represents the pixel weight, represents the natural logarithm, represents the pixel value, represents pixel coordinates, represents the neighborhood pixels, represents the neighborhood coordinates, represents the Gaussian function parameters;

[0030] S213, calculating the weight of each pixel respectively;

[0031] S214, multiplying the weight of each pixel by its corresponding pixel value, and summing the weighted values ​​of all pixels;

[0032] S215 . Divide the weighted sum by the weighted sum to obtain a filtered historical ultrasound image.

[0033] Preferably, fitting the pixel points in the filtered historical ultrasound image by an image fitting algorithm to obtain the processed historical ultrasound image comprises the following steps:

[0034] S221, randomly selecting m pixel points in the filtered historical ultrasound image;

[0035] S222, fitting the selected m pixel points to construct a fitting contour;

[0036] S223, calculating the distance from other pixel points in the selected filtered historical ultrasound image to the fitted contour, setting a distance threshold, and when the distance from other pixel points in the selected filtered historical ultrasound image to the fitted contour is less than the set distance threshold, setting the pixel point as an inner point; when the distance from other pixel points in the selected filtered historical ultrasound image to the fitted contour is greater than or equal to the set distance threshold, setting it as a noise point, and counting the number of inner points;

[0037] S224, repeat steps S221-S223 to select the contour with the largest number of inner points;

[0038] S225 , setting the contour with the largest number of internal points as the filtered historical ultrasonic image, and setting the filtered historical ultrasonic image as the processed historical ultrasonic image.

[0039] The present invention filters historical ultrasonic images by using a mean filtering algorithm, processes the filtered historical ultrasonic images by an image fitting algorithm, fits pixel points in the historical ultrasonic images by the image fitting algorithm, completes the processing of the historical ultrasonic images, and ensures the accuracy of the historical ultrasonic image processing.

[0040] Preferably, extracting features from the processed historical ultrasound images by a feature extraction algorithm, obtaining features of the processed historical ultrasound images and saving the features comprises the following steps:

[0041] S31, dividing the processed historical ultrasound image into 16×16 image data blocks, and inputting the divided image data blocks into a convolutional neural network;

[0042] S32, extracting features from the divided image data blocks through a convolutional neural network;

[0043] The feature extraction of the divided image data blocks by convolutional neural network includes the following steps:

[0044] The convolutional neural network is set to include: input layer, convolution layer, pooling layer and fully connected layer;

[0045] S321, inputting the image data blocks divided from the processed historical ultrasound images into the input layer of the neural network;

[0046] S322, after receiving the image data blocks divided by the processed historical ultrasound images, the input layer transmits the received image data blocks to the convolution layer, and the convolution layer extracts local features in the image data blocks through convolution operations;

[0047] S323, after the convolution layer extracts the local features in the image data block, the local features in the extracted image data block are processed by the pooling layer, and the features are downsampled by the pooling layer to reduce the data dimension;

[0048] S324, extracting features from the divided image data blocks through continuous convolution and pooling until the extracted features converge, the convolution stops, and the extracted features are summarized and input into the fully connected layer;

[0049] S325, the fully connected layer integrates the extracted features and outputs the final feature extraction result;

[0050] The convolution operation formula is as follows:

[0051]

[0052] in, represents the input image data block, represents the weight of the corresponding convolution kernel, b Indicates the bias value, f represents the output features;

[0053] S33, connecting the feature extraction results in each image data block to form a feature vector, and saving the feature vectors of all processed historical ultrasound images;

[0054] The feature vector is set as the representation of the ultrasound image features.

[0055] The present invention uses a feature extraction algorithm to divide historical ultrasonic images, inputs the divided historical ultrasonic images into a convolutional neural network for feature extraction, and integrates the extracted features, thereby realizing the extraction of feature vectors of all processed historical ultrasonic images and improving the accuracy of ultrasonic image feature vector extraction.

[0056] Preferably, the step of processing the ultrasound image corresponding to the feature extracted in real time based on the comparison result to determine the type of the ultrasound image collected in real time comprises the following steps:

[0057] The features extracted in real time are compared with the features of historical ultrasound images through a similarity calculation formula;

[0058] The similarity calculation formula is as follows:

[0059] ;

[0060] in, Represents the feature vector and the eigenvector The similarity value between represents the real-time extraction of feature vectors of ultrasound images, A feature vector representing a historical ultrasound image;

[0061] A similarity threshold is set. When the calculated similarity is less than the set threshold, it indicates that the real-time collected ultrasound image is not similar to the historical ultrasound image; when the calculated similarity is greater than or equal to the set threshold, it indicates that the real-time collected ultrasound image is similar to the historical ultrasound image, and the two ultrasound images are of the same type.

[0062] The present invention compares the features extracted in real time by using a similarity calculation formula with the features of historical ultrasound images, and at the same time determines the type of ultrasound images collected in real time by setting a similarity threshold, thereby ensuring the accuracy of ultrasound image inspection report generation.

[0063] Preferably, after determining the type of the ultrasound image collected in real time, generating an ultrasound report by extracting features based on the established ultrasound image report standard and the processed ultrasound image collected in real time comprises the following steps:

[0064] Extract keywords of ultrasound image analysis from historical ultrasound image report data, and match the extracted keywords in Wikipedia entries. Calculate semantic relevance of keywords by using document connection information graph and classification structure graph of Wikipedia knowledge base, and construct a keyword set in current extracted ultrasound image analysis based on the semantic relevance calculation result.

[0065] The semantic relevance calculation result includes the semantic relevance between the extracted keywords and the Wikipedia entries;

[0066] The semantic relevance calculation is as follows:

[0067]

[0068] in, Indicates the semantic relevance value between the extracted keywords and the Wikipedia entries. represents the jth keyword in the i-th ultrasound image analysis, represents the first entry in Wikipedia, represents the breadth-first traversal path information of the concept node of the jth keyword in the i-th ultrasound image analysis on the classification structure graph, Represents the breadth-first traversal path information of the concept node of the first entry in Wikipedia on the classification structure graph, Indicates the nearest common root node between the keyword and the Wikipedia entry in the classification structure diagram. represents the depth information of the concept node of the jth keyword in the i-th ultrasound image analysis on the classification structure graph, Represents the depth information of the concept node of the first entry in Wikipedia on the classification structure graph;

[0069] The keywords extracted from the ultrasound image analysis are used as data in the ultrasound image report standard and combined with the ultrasound images collected in real time to generate an ultrasound report.

[0070] The present invention calculates the semantic relevance of keywords by using the document connection information graph and classification structure graph of the Wikipedia knowledge base, and constructs a keyword set in the current extracted ultrasound image analysis based on the semantic relevance calculation results, and uses the keywords extracted in the ultrasound image analysis as data in the ultrasound image report standard, and combines them with the ultrasound images collected in real time to generate an ultrasound report, thereby improving the efficiency of ultrasound report generation.

[0071] This embodiment also discloses an ultrasound image examination report generation system based on artificial intelligence, including: a data collection module, a database, a data processing module, a feature extraction module and a report generation module;

[0072] The data collection module is used to collect ultrasound images and transmit the collected ultrasound images to a database;

[0073] The database is used to store the collected ultrasound images;

[0074] The data processing module is used to process the ultrasound images stored in the database;

[0075] The feature extraction module is used to extract features from the classified ultrasound images;

[0076] The report generation module is used to analyze and compare ultrasound images and generate ultrasound image reports.

[0077] (III) Beneficial effects

[0078] Compared with the prior art, the present invention provides an artificial intelligence-based ultrasound image examination report generation method and system, which has the following beneficial effects:

[0079] 1. The present invention constructs a training set by collecting historical ultrasound images and establishes ultrasound image reporting standards. At the same time, by processing the historical ultrasound images in the training set and using a feature extraction algorithm to extract features from the collected historical ultrasound images and the real-time collected ultrasound images and compare them, the type of the real-time collected ultrasound images is determined, thereby improving the efficiency and accuracy of ultrasound image inspection.

[0080] 2. The present invention establishes ultrasound image report standards based on the collected historical ultrasound images and historical ultrasound image report data, determines various parameters in the ultrasound image report standards, and ensures the standardization of ultrasound image reports.

[0081] 3. The present invention filters the historical ultrasonic images by using a mean filtering algorithm, processes the filtered historical ultrasonic images by an image fitting algorithm, fits the pixel points in the historical ultrasonic images by the image fitting algorithm, completes the processing of the historical ultrasonic images, and ensures the accuracy of the historical ultrasonic image processing.

[0082] 4. The present invention realizes the extraction of feature vectors of all processed historical ultrasonic images by using a feature extraction algorithm, dividing the historical ultrasonic images, inputting the divided historical ultrasonic images into a convolutional neural network for feature extraction, and integrating the extracted features, thereby improving the accuracy of ultrasonic image feature vector extraction.

[0083] 5. The present invention compares the features extracted in real time by using a similarity calculation formula with the features of historical ultrasound images, and at the same time determines the type of ultrasound images collected in real time by setting a similarity threshold, thereby ensuring the accuracy of the generated ultrasound image examination report.

[0084] 6. The invention calculates the semantic relevance of keywords by using the document connection information graph and classification structure graph of the Wikipedia knowledge base, and constructs a keyword set in the current extracted ultrasound image analysis based on the semantic relevance calculation results, and uses the keywords extracted in the ultrasound image analysis as data in the ultrasound image report standard, and combines them with the ultrasound images collected in real time to generate an ultrasound report, thereby improving the efficiency of ultrasound report generation, and providing the resulting report to physicians for reference as an auxiliary tool for doctors, thereby improving the efficiency of physicians and the accuracy of diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 The present invention is a schematic diagram of the ultrasonic image examination report generation process structure. DETAILED DESCRIPTION

[0086] 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.

[0087] Example 1

[0088] This embodiment discloses a method for generating an ultrasound image examination report based on artificial intelligence, which specifically includes the following steps:

[0089] S1. Collect historical ultrasound images and historical ultrasound image report data to build a training set, and establish ultrasound image report standards based on the collected historical ultrasound images and historical ultrasound image report data;

[0090] The historical ultrasound image report data includes the name, gender, ultrasound image type and ultrasound image analysis corresponding to the historical ultrasound image;

[0091] Establishing ultrasound image standards based on the collected historical ultrasound images and historical ultrasound image parameters includes the following steps:

[0092] The ultrasound image report standard is set to include only name, gender, ultrasound image type and ultrasound image analysis, which cannot be missing;

[0093] S2, processing the historical ultrasound images in the training set to obtain processed historical ultrasound images;

[0094] Processing the historical ultrasound images in the training set to obtain the processed historical ultrasound images includes the following steps:

[0095] S21, filtering the historical ultrasound images in the training set by using a mean filtering algorithm to obtain filtered historical ultrasound images;

[0096] S211, setting a window, wherein the window includes the historical ultrasonic image pixels to be processed and the neighborhood around the historical ultrasonic image pixels;

[0097] S212, for each pixel in the window, calculating the product of the spatial domain weight and the gray value weight as the weight of the pixel;

[0098] The pixel weight calculation formula is as follows:

[0099] ;

[0100] ;

[0101] ;

[0102] in, represents the spatial domain weight, represents the gray value weight, represents the pixel weight, represents the natural logarithm, represents the pixel value, represents pixel coordinates, represents the neighborhood pixels, represents the neighborhood coordinates, represents the Gaussian function parameters;

[0103] S213, calculating the weight of each pixel respectively;

[0104] S214, multiplying the weight of each pixel by its corresponding pixel value, and summing the weighted values ​​of all pixels;

[0105] S215, dividing the weighted sum by the weighted sum to obtain a filtered historical ultrasound image;

[0106] S22, fitting the pixel points in the filtered historical ultrasound image by using an image fitting algorithm to obtain a processed historical ultrasound image;

[0107] S221, randomly selecting m pixel points in the filtered historical ultrasound image;

[0108] S222, fitting the selected m pixel points to construct a fitting contour;

[0109] S223, calculating the distance from other pixel points in the selected filtered historical ultrasound image to the fitted contour, setting a distance threshold, and when the distance from other pixel points in the selected filtered historical ultrasound image to the fitted contour is less than the set distance threshold, setting the pixel point as an inner point; when the distance from other pixel points in the selected filtered historical ultrasound image to the fitted contour is greater than or equal to the set distance threshold, setting it as a noise point, and counting the number of inner points;

[0110] S224, repeat steps S221-S223 to select the contour with the largest number of inner points;

[0111] S225. Set the contour with the largest number of internal points as the filtered historical ultrasonic image, and record the contour parameters of the filtered historical ultrasonic image.

[0112] S3, extracting features from the processed historical ultrasound images by using a feature extraction algorithm, obtaining features of the processed historical ultrasound images and saving them;

[0113] S31, dividing the processed historical ultrasound image into 16×16 image data blocks, and inputting the divided image data blocks into a convolutional neural network;

[0114] S32, extracting features from the divided image data blocks through a convolutional neural network;

[0115] The feature extraction of the divided image data blocks by convolutional neural network includes the following steps:

[0116] The convolutional neural network is set to include: input layer, convolution layer, pooling layer and fully connected layer;

[0117] S321, inputting the image data blocks divided from the processed historical ultrasound images into the input layer of the neural network;

[0118] S322, after receiving the image data blocks divided by the processed historical ultrasound images, the input layer transmits the received image data blocks to the convolution layer, and the convolution layer extracts local features in the image data blocks through convolution operations;

[0119] S323, after the convolution layer extracts the local features in the image data block, the local features in the extracted image data block are processed by the activation function, and the local features in the image data block processed by the activation function are transmitted to the pooling layer, and the features are downsampled by the pooling layer to reduce the data dimension;

[0120] S324, extracting features from the divided image data blocks through continuous convolution and pooling until the extracted features converge, the convolution stops, and the extracted features are summarized and input into the fully connected layer;

[0121] S325, the fully connected layer integrates the extracted features and outputs the final feature extraction result;

[0122] Furthermore, the convolution operation formula is as follows:

[0123]

[0124] in, represents the input image data block, represents the weight of the corresponding convolution kernel, b Indicates the bias value, f represents the output features;

[0125] S33, connecting the feature extraction results in each image data block to form a feature vector, and saving the feature vectors of all processed historical ultrasound images;

[0126] The feature vector is set as the representation of the ultrasound image features.

[0127] S4, collecting and processing ultrasonic images in real time, and extracting features from the processed ultrasonic images collected in real time by using a feature extraction algorithm;

[0128] S5, comparing the features extracted in real time with the features of the historical ultrasound images, processing the ultrasound images corresponding to the features extracted in real time based on the comparison results, and determining the type of the ultrasound images collected in real time;

[0129] Comparing the features extracted in real time with the features in each decision tree, and processing the ultrasound image corresponding to the features extracted in real time based on the comparison results includes the following steps:

[0130] The features extracted in real time are compared with the features of historical ultrasound images through a similarity calculation formula;

[0131] The similarity calculation formula is as follows:

[0132] ;

[0133] in, Represents the feature vector and the eigenvector The similarity value between represents the real-time extraction of feature vectors of ultrasound images, A feature vector representing a historical ultrasound image;

[0134] A similarity threshold is set. When the calculated similarity is less than the set threshold, it indicates that the real-time collected ultrasound image is not similar to the historical ultrasound image; when the calculated similarity is greater than or equal to the set threshold, it indicates that the real-time collected ultrasound image is similar to the historical ultrasound image and the two ultrasound images are of the same type.

[0135] When the ultrasound image collected in real time is not similar to the historical ultrasound image, determining that the ultrasound image collected in real time is abnormal;

[0136] When the ultrasound image collected in real time is similar to the historical ultrasound image, determining that the ultrasound image collected in real time is normal, and determining the type of the ultrasound image collected in real time based on the ultrasound image type in the corresponding historical ultrasound image report data;

[0137] S6, after determining the type of the ultrasound image collected in real time, extracting features based on the established ultrasound image report standard and the processed ultrasound image collected in real time to generate an ultrasound report;

[0138] Extract keywords of ultrasound image analysis from historical ultrasound image report data, and match the extracted keywords in Wikipedia entries. Calculate semantic relevance of keywords by using document connection information graph and classification structure graph of Wikipedia knowledge base, and construct a keyword set in current extracted ultrasound image analysis based on the semantic relevance calculation result.

[0139] The semantic relevance calculation result includes the semantic relevance between the extracted keywords and the Wikipedia entries;

[0140] The semantic relevance calculation is as follows:

[0141]

[0142] in, Indicates that the semantic relevance value of the extracted keywords and Wikipedia entries is in the interval [-1,1], represents the jth keyword in the i-th ultrasound image analysis, represents the first entry in Wikipedia, represents the breadth-first traversal path information of the concept node of the jth keyword in the i-th ultrasound image analysis on the classification structure graph, Represents the breadth-first traversal path information of the concept node of the first entry in Wikipedia on the classification structure graph, Indicates the nearest common root node between the keyword and the Wikipedia entry in the classification structure diagram. represents the depth information of the concept node of the jth keyword in the i-th ultrasound image analysis on the classification structure graph, Represents the depth information of the concept node of the first entry in Wikipedia on the classification structure graph;

[0143] When the semantic relevance value between the extracted keywords and the Wikipedia entry is in the interval [-1,0], it means that the extracted keywords do not match the Wikipedia entry, and the keywords in the ultrasound image analysis are re-extracted;

[0144] When the semantic relevance value between the extracted keyword and the Wikipedia entry is in the interval of (0,1], it means that the extracted keyword matches the Wikipedia entry, and the extracted keyword is added to the keyword set in the current extraction ultrasound image analysis;

[0145] The keywords extracted from the ultrasound image analysis are used as data in the ultrasound image report standard and combined with the ultrasound images collected in real time to generate an ultrasound report.

[0146] The above is an optimal and feasible solution. In the prior art, the text content of the ultrasound report can also be completely generated by generating a model through a large model language, which will not be described in detail here.

[0147] Example 2

[0148] This embodiment also discloses an ultrasound image examination report generation system based on artificial intelligence, including: a data collection module, a database, a data processing module, a feature extraction module and a report generation module;

[0149] The data collection module is used to collect ultrasound images and transmit the collected ultrasound images to a database;

[0150] The database is used to store the collected ultrasound images;

[0151] The data processing module is used to process the ultrasound images stored in the database;

[0152] The feature extraction module is used to extract features from the classified ultrasound images;

[0153] The report generation module is used to analyze and compare ultrasound images and generate ultrasound image reports.

[0154] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for generating an ultrasound image examination report based on artificial intelligence, characterized in that: The following steps are involved: S1. Collect historical ultrasound images and historical ultrasound image report data to build a training set, and establish ultrasound image report standards based on the collected historical ultrasound images and historical ultrasound image report data; S2, processing the historical ultrasound images in the training set to obtain processed historical ultrasound images; S21, filtering the historical ultrasound images in the training set by using a mean filtering algorithm to obtain filtered historical ultrasound images; S211, setting a window, wherein the window includes the historical ultrasonic image pixels to be processed and the neighborhood around the historical ultrasonic image pixels; S212, for each pixel in the window, calculating the product of the spatial domain weight and the gray value weight as the weight of the pixel; The pixel weight calculation formula is as follows: w=w1×w2; Among them, w1 represents the spatial domain weight, w2 represents the gray value weight, w represents the pixel weight, e represents the natural logarithm, f(α,η) represents the pixel value, α,η represents the pixel coordinates, f(β,γ) represents the neighborhood pixel, β,γ represents the neighborhood coordinates, and σ represents the Gaussian function parameter; S213, calculating the weight of each pixel respectively; S214, multiplying the weight of each pixel by its corresponding pixel value, and summing the weighted values ​​of all pixels; S215, dividing the weighted sum by the weighted sum to obtain a filtered historical ultrasound image; S22, fitting the pixel points in the filtered historical ultrasound image by using an image fitting algorithm to obtain a processed historical ultrasound image; S3, extracting features from the processed historical ultrasound images by using a feature extraction algorithm, obtaining features of the processed historical ultrasound images and saving them; S4, collecting and processing ultrasonic images in real time, and extracting features from the processed ultrasonic images collected in real time by using a feature extraction algorithm; S5, comparing the features extracted in real time with the features of the historical ultrasound images, processing the ultrasound images corresponding to the features extracted in real time based on the comparison results, and determining the type of the ultrasound images collected in real time; S6. After determining the type of the ultrasound image collected in real time, extract features based on the established ultrasound image report standard and the processed ultrasound image collected in real time to generate an ultrasound report.

2. The method for generating an ultrasonic image examination report based on artificial intelligence according to claim 1, characterized in that: The establishment of ultrasound image report standards based on the collected historical ultrasound images and historical ultrasound image report data comprises the following steps: The ultrasound image report standard only includes name, gender, ultrasound image type and ultrasound image analysis, which cannot be missing.

3. The method for generating an ultrasonic image examination report based on artificial intelligence according to claim 1, characterized in that: The step of fitting the pixel points in the filtered historical ultrasound image by using an image fitting algorithm comprises the following steps: S221, randomly selecting m pixel points in the filtered historical ultrasound image; S222, fitting the selected m pixel points to construct a fitting contour; S223, calculating the distance from other pixel points in the selected filtered historical ultrasound image to the fitted contour, setting a distance threshold, and when the distance from other pixel points in the selected filtered historical ultrasound image to the fitted contour is less than the set distance threshold, setting the pixel point as an inner point; when the distance from other pixel points in the selected filtered historical ultrasound image to the fitted contour is greater than or equal to the set distance threshold, setting it as a noise point, and counting the number of inner points; S224, repeat steps S221-S223 to select the contour with the largest number of inner points; S225 , setting the contour with the largest number of internal points as the filtered historical ultrasonic image, and setting the filtered historical ultrasonic image as the processed historical ultrasonic image.

4. The method for generating an ultrasonic image examination report based on artificial intelligence according to claim 1, characterized in that: The feature extraction of the processed historical ultrasound image by the feature extraction algorithm to obtain the features of the processed historical ultrasound image and save the features comprises the following steps: S31, dividing the processed historical ultrasound image into 16×16 image data blocks, and inputting the divided image data blocks into a convolutional neural network; S32, extracting features from the divided image data blocks through a convolutional neural network; The convolutional neural network is set to include: input layer, convolution layer, pooling layer and fully connected layer; S321, inputting the image data blocks divided from the processed historical ultrasound images into the input layer of the neural network; S322, after receiving the image data blocks divided by the processed historical ultrasound images, the input layer transmits the received image data blocks to the convolution layer, and the convolution layer extracts local features in the image data blocks through convolution operations; S323, after the convolution layer extracts the local features in the image data block, the local features in the extracted image data block are processed by the pooling layer, and the features are downsampled by the pooling layer to reduce the data dimension; S324, extracting features from the divided image data blocks through continuous convolution and pooling until the extracted features converge, the convolution stops, and the extracted features are summarized and input into the fully connected layer; S325, the fully connected layer integrates the extracted features and outputs the final feature extraction result; The convolution operation formula is as follows: f=I×ω+b Among them, I represents the input image data block, ω represents the weight of the corresponding convolution kernel, b represents the bias value, and f represents the output feature; S33, connecting the feature extraction results in each image data block to form a feature vector, and saving the feature vectors of all processed historical ultrasound images; The feature vector is set as the representation of the ultrasound image features.

5. The method for generating an ultrasonic image examination report based on artificial intelligence according to claim 1, characterized in that: The step of processing the ultrasonic image corresponding to the feature extracted in real time based on the comparison result to determine the type of the ultrasonic image collected in real time comprises the following steps: The features extracted in real time are compared with the features of historical ultrasound images through a similarity calculation formula; The similarity calculation formula is as follows: in, Represents the feature vector and the eigenvector The similarity value between represents the real-time extraction of feature vectors of ultrasound images, A feature vector representing a historical ultrasound image; A similarity threshold is set. When the calculated similarity is less than the set threshold, it indicates that the real-time collected ultrasound image is not similar to the historical ultrasound image; when the calculated similarity is greater than or equal to the set threshold, it indicates that the real-time collected ultrasound image is similar to the historical ultrasound image, and the two ultrasound images are of the same type.

6. The method for generating an ultrasonic image examination report based on artificial intelligence according to claim 1, characterized in that: After determining the type of the ultrasound image collected in real time, generating an ultrasound report by extracting features based on the established ultrasound image report standard and the processed ultrasound image collected in real time includes the following steps: Extract keywords of ultrasound image analysis from historical ultrasound image report data, and match the extracted keywords in Wikipedia entries. Calculate semantic relevance of keywords by using document connection information graph and classification structure graph of Wikipedia knowledge base, and construct a keyword set in current extracted ultrasound image analysis based on the semantic relevance calculation result. The semantic relevance calculation result includes the semantic relevance between the extracted keywords and the Wikipedia entries; The semantic relevance calculation is as follows: Among them, sim(x ij ,y1) represents the semantic relevance value between the extracted keywords and the Wikipedia entries, x ij represents the jth keyword in the i-th ultrasound image analysis, y1 represents the first entry in Wikipedia, Lenx ij represents the breadth-first traversal path information of the concept node of the jth keyword in the i-th ultrasound image analysis on the classification structure diagram, Leny1 represents the breadth-first traversal path information of the concept node of the first entry in Wikipedia on the classification structure diagram, Dep pub Depx represents the nearest common root node between the concept node of the keyword and the Wikipedia entry in the classification structure diagram. ij Depy1 represents the depth information of the concept node of the jth keyword in the i-th ultrasound image analysis on the classification structure diagram, and Depy2 represents the depth information of the concept node of the first entry in Wikipedia on the classification structure diagram; The keywords extracted from the ultrasound image analysis are used as data in the ultrasound image report standard and combined with the ultrasound images collected in real time to generate an ultrasound report.

7. A system for implementing the method for generating an ultrasonic image examination report based on artificial intelligence as claimed in any one of claims 1 to 6, characterized in that: include: Data collection module, database, data processing module, feature extraction module and report generation module; The data collection module is used to collect ultrasound images and transmit the collected ultrasound images to a database; The database is used to store the collected ultrasound images; The data processing module is used to process the ultrasound images stored in the database; The feature extraction module is used to extract features from the classified ultrasound images; The report generation module is used to analyze and compare ultrasound images and generate ultrasound image reports.

Images