Image data analysis method, system and equipment based on artificial intelligence

Abstract

The invention relates to the technical field of image recognition, and particularly provides an artificial intelligence-based image data analysis method, system and equipment, and the method comprises the steps: obtaining a medical image, and recognizing organ types, feature points and abnormal regions in the medical image through a target recognition algorithm; generating a coordinate system based on the feature points, and generating coordinates of the abnormal region in the coordinate system; extracting local image features of the abnormal region, and screening out a local image feature vector with the highest confidence coefficient by using a classifier; inputting the organ type, the local image feature vector and the coordinate of the abnormal region into a full connection layer to obtain a comprehensive feature of the abnormal region; and inputting the comprehensive features of the plurality of abnormal regions into a pre-training random forest model to obtain an analysis result. According to the invention, more comprehensive, accurate and efficient analysis of medical images can be realized.

Description

Technical Field

[0001] The present invention belongs to the field of image recognition technology, and specifically relates to an image data analysis method, system and equipment based on artificial intelligence. Background Art

[0002] With the rapid advancement of medical technology, medical imaging, as an indispensable part of modern medicine, has become increasingly prominent in the accurate diagnosis and effective treatment of diseases. It not only provides patients with a detailed view of the internal body but also provides valuable evidence for doctors to formulate treatment plans. However, traditional medical image analysis relies heavily on the physician's personal experience and professional judgment. While this model embodies profound human wisdom, it also inevitably introduces subjectivity, and the analysis process is often time-consuming, leaving room for improvement in efficiency.

[0003] To address this challenge, image recognition algorithms have emerged and are gradually penetrating the field of medical image analysis, aiming to reduce the burden on doctors through automation and improve diagnostic speed and accuracy. Although existing algorithms have made significant progress in extracting and classifying image features, enabling basic identification of lesions, most of them are limited to surface-level feature analysis, failing to fully explore the complex relationship between lesions and diagnostic results, and ignoring the profound impact of lesion morphology, size, location, and even dynamic changes on diagnostic conclusions. Summary of the Invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides an image data analysis method, system and device based on artificial intelligence to solve the above-mentioned technical problems.

[0005] In a first aspect, the present invention provides an image data analysis method based on artificial intelligence, comprising: Acquire medical images and use target recognition algorithms to identify organ types, feature points, and abnormal areas in the medical images; Generate a coordinate system based on the feature points, and generate coordinates of the abnormal area in the coordinate system; Extract local image features of abnormal areas and use the classifier to filter out the local image feature vector with the highest confidence; Inputting the organ type, the local image feature vector, and the coordinates of the abnormal region into a fully connected layer to obtain comprehensive features of the abnormal region; The comprehensive features of multiple abnormal areas are input into the pre-trained random forest model to obtain the analysis results.

[0006] In an optional embodiment, obtaining a medical image and using an object recognition algorithm to identify organ types, feature points, and abnormal areas in the medical image include: Using a pre-trained YOLO model to identify organ types, organ feature points, and abnormal areas in the medical image; The training methods of the YOLO model include: Collect historical medical images and annotate organ types, characteristic points of each organ type, and normal organs or abnormal areas in the historical medical images; Import annotated historical medical images into the dataset; The YOLO model is trained using the dataset.

[0007] In an optional embodiment, generating a coordinate system based on the feature points and generating the coordinates of the abnormal area in the coordinate system includes: Pre-set the coordinate values ​​of the feature points; Locating the origin position and coordinate axis position of the coordinate system based on the position and coordinate values ​​of the feature points, and generating a plane coordinate system on the medical image according to the origin position and coordinate axis position; Acquire coordinate data of the abnormal area in the plane coordinate system, wherein the coordinate data includes a horizontal coordinate range and a vertical coordinate range; The medical image is upsampled based on the abscissa range and the ordinate range to segment an abnormal area.

[0008] In an optional embodiment, extracting local image features of the abnormal area and using a classifier to filter out the local image feature vector with the highest confidence level includes: Extracting overall image features of the abnormal area using multiple convolutional networks, wherein the overall image features include edge contour features, texture features, and shape features; The overall image features are continuously reduced using multiple convolutional layers, and the output features of each convolutional layer are respectively input into a 1×1 convolutional layer to obtain multiple feature maps of different sizes; Use maximum suppression to remove duplicate feature maps; Use the classifier to identify the remaining feature maps and obtain the recognition confidence; The feature maps whose confidence reaches the set confidence threshold are filtered out, and the filtered feature maps are output as local image feature vectors.

[0009] In an optional embodiment, the organ type, the local image feature vector, and the coordinates of the abnormal region are input into a fully connected layer to obtain comprehensive features of the abnormal region, including: Map organ types to organ type vectors; Determine the area of ​​the abnormal region and the functional region in the organ according to the coordinates of the abnormal region, and map the area and the functional region into an area vector and a functional region vector; The organ type vector, area vector, functional region vector and local image feature vector are input into a pre-designed fully connected layer, which performs linear transformation and nonlinear activation on the input vector to extract the comprehensive features of the abnormal area.

[0010] In a second aspect, the present invention provides an image data analysis system based on artificial intelligence, comprising: The basic recognition module is used to acquire medical images and use target recognition algorithms to identify organ types, feature points, and abnormal areas in medical images; A coordinate generation module, configured to generate a coordinate system based on the feature points, and generate the coordinates of the abnormal area in the coordinate system; Feature extraction module, used to extract local image features of abnormal areas and use the classifier to filter out the local image feature vector with the highest confidence; a feature fusion module, configured to input the organ type, the local image feature vector, and the coordinates of the abnormal region into a fully connected layer to obtain comprehensive features of the abnormal region; The feature classification module is used to input the comprehensive features of multiple abnormal areas into the pre-trained random forest model to obtain analysis results.

[0011] In an optional embodiment, the basic identification module includes: an object recognition unit, configured to recognize organ types, organ feature points, and abnormal regions in the medical image using a pre-trained YOLO model; The training methods of the YOLO model include: Collect historical medical images and annotate organ types, characteristic points of each organ type, and normal organs or abnormal areas in the historical medical images; Import annotated historical medical images into the dataset; The YOLO model is trained using the dataset.

[0012] In an optional embodiment, the coordinate generation module includes: A coordinate setting unit, used to pre-set the coordinate values ​​of feature points; A coordinate positioning unit, configured to locate the origin position and coordinate axis position of the coordinate system based on the position and coordinate value of the feature point, and generate a plane coordinate system on the medical image according to the origin position and coordinate axis position; A coordinate acquisition unit, configured to acquire coordinate data of the abnormal area in the plane coordinate system, wherein the coordinate data includes a horizontal coordinate range and a vertical coordinate range; The region segmentation unit is configured to upsample the medical image based on the horizontal coordinate range and the vertical coordinate range to segment out abnormal regions.

[0013] In an optional embodiment, the feature extraction module includes: A first processing unit is configured to extract overall image features of the abnormal area using multiple convolutional networks, where the overall image features include edge contour features, texture features, and shape features; A second processing unit is configured to continuously reduce the overall image features using multiple convolutional layers, and input the output features of each convolutional layer into a 1×1 convolutional layer to obtain multiple feature maps of different sizes; Feature deduplication unit, used to remove duplicate feature maps using maximum value suppression; The classification and recognition unit is used to use the classifier to identify the remaining feature maps and obtain the recognition confidence; The feature screening unit is used to screen out feature maps whose confidence reaches a set confidence threshold, and output the screened feature maps as local image feature vectors.

[0014] According to a third aspect, a device is provided, comprising: A memory for storing an artificial intelligence-based image data analysis program; A processor is used to implement the steps of the artificial intelligence-based image data analysis method provided in the first aspect when executing the artificial intelligence-based image data analysis program.

[0015] In a fourth aspect, a computer-readable storage medium is provided, on which an artificial intelligence-based image data analysis program is stored. When the artificial intelligence-based image data analysis program is executed by a processor, the steps of the artificial intelligence-based image data analysis method provided in the first aspect are implemented.

[0016] The beneficial effect of the present invention is that the artificial intelligence-based image data analysis method, system and equipment provided by the present invention can globally identify abnormal areas, extract the location of abnormal areas, the organ type to which they belong and the local feature vectors, and fuse and identify all the extracted features, thereby achieving a more comprehensive, accurate and efficient analysis of medical images.

[0017] In addition, the present invention has a reliable design principle, a simple structure and a very broad application prospect. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0019] Figure 1is a schematic flow chart of a method according to an embodiment of the present invention.

[0020] Figure 2 FIG. 4 is a schematic block diagram of a system according to an embodiment of the present invention.

[0021] Figure 3 A schematic structural diagram of a device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of 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 making creative efforts should fall within the scope of protection of the present invention.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0024] The key terms appearing in the present invention are explained below.

[0025] The YOLO algorithm (You Only Look Once: Unified, Real-Time Object Detection) is a real-time object detection algorithm with widespread applications in computer vision. The YOLO algorithm transforms the object detection problem into a regression problem, using a single convolutional neural network (CNN) to directly predict bounding boxes and class probabilities from the input image. This approach avoids the complex multi-stage processing required by traditional object detection algorithms, such as region proposal and feature extraction, resulting in higher detection speed and accuracy.

[0026] Random Forest (RF) is an ensemble learning method. It is a type of supervised learning algorithm and a classifier or regressor composed of multiple decision trees.

[0027] Working principle: Bootstrap Sampling: Random Forest uses bootstrap sampling to randomly sample the dataset, generating multiple training sets. This means each decision tree is trained on a different subset of the original dataset. This sampling method means that some samples will appear repeatedly within a decision tree, while others may not be sampled.

[0028] Random feature selection: When splitting a node in each decision tree, instead of considering all possible features, a subset of features is randomly selected as candidate split features. This increases the diversity of the model and reduces the correlation between trees.

[0029] Ensemble of decision trees: In classification problems, random forests use majority voting to determine the final prediction result; for regression problems, the final prediction is obtained by calculating the average of the prediction results of all trees.

[0030] Convolutional layers are a fundamental operation and a core component of convolutional neural networks (CNNs). They extract local features from input data by performing convolution operations on it. Convolution is a mathematical operation that slides one function (the convolution kernel or filter) over another function (the input data) and performs a dot product operation at each location to produce an output feature map.

[0031] The artificial intelligence-based image data analysis method provided in the embodiment of the present invention is executed by a computer device. Accordingly, the artificial intelligence-based image data analysis system runs in the computer device.

[0032] Figure 1 is a schematic flow chart of a method according to an embodiment of the present invention. Figure 1 The execution subject can be an image data analysis system based on artificial intelligence. According to different needs, the order of the steps in the flowchart can be changed, and some can be omitted.

[0033] like Figure 1 As shown, the method includes: S1. Obtain medical images and use target recognition algorithms to identify organ types, feature points, and abnormal areas in the medical images.

[0034] High-quality medical images, such as X-rays, CT scans, or MRIs, are acquired from medical devices. Advanced object recognition algorithms (such as YOLO and Faster R-CNN) are then used to perform in-depth analysis of these images, accurately identifying organ types, key features, and potential abnormal areas. This step requires not only high algorithm accuracy but also the ability to quickly process large amounts of data to meet clinical needs.

[0035] S2. Generate a coordinate system based on the feature points, and generate the coordinates of the abnormal area in the coordinate system.

[0036] Based on the identified feature points, a precise coordinate system is constructed. This coordinate system allows the precise location of the abnormal area in the image and generates corresponding coordinate information. This step provides critical spatial positioning information for subsequent detailed analysis and treatment.

[0037] S3. Extract local image features of the abnormal area and use the classifier to filter out the local image feature vector with the highest confidence.

[0038] For each abnormal region, local image features are extracted. These features may include texture, color, shape, and other information. These features are then filtered using a classifier (such as a support vector machine or random forest) to select the local image feature vector with the highest confidence. This step helps reduce noise interference and improve the accuracy of subsequent analysis.

[0039] S4. Input the organ type, the local image feature vector and the coordinates of the abnormal area into a fully connected layer to obtain comprehensive features of the abnormal area.

[0040] The identified organ type, the selected local image feature vectors, and the coordinates of the abnormal area are input into a fully connected layer (a type of neural network layer). Through complex nonlinear transformations, a comprehensive feature of the abnormal area is generated. This step integrates multiple pieces of information and provides comprehensive data support for subsequent model analysis.

[0041] S5. Input the comprehensive features of multiple abnormal areas into the pre-trained random forest model to obtain analysis results.

[0042] The comprehensive features of multiple abnormal regions are fed into a pre-trained random forest model. The random forest model is a powerful machine learning algorithm capable of processing high-dimensional data with high accuracy and robustness. This model enables in-depth analysis and diagnosis of abnormal regions, yielding final results. These results can provide doctors with valuable information, helping them make more accurate diagnostic and treatment decisions.

[0043] In an embodiment of the present invention, based on step S1, a possible embodiment will be given below to illustrate its specific implementation scheme in a non-limiting manner.

[0044] 1. Detailed explanation of the training process of the YOLO model Data collection and annotation: Collect historical medical images: Collect rich historical medical imaging data from major medical institutions, medical imaging libraries and other channels, including but not limited to X-rays, CT scans, MRI images, etc.

[0045] Labeling organ types: Professional doctors or radiologists carefully label the collected medical images and clearly mark the organ types appearing in each image.

[0046] Marking feature points: For each organ type, further mark its key feature points, such as the apex of the heart and the pulmonary valve, which play a key role in subsequent analysis.

[0047] Marking abnormal areas: Experts also mark abnormal areas in the image, including tumors, cysts, inflammation and other pathological areas.

[0048] Dataset construction: Import annotated medical images: Import annotated historical medical images and their related information (organ type, feature points, abnormal areas) into a dedicated dataset.

[0049] Data preprocessing: Preprocess the dataset, including image enhancement (such as rotation, scaling, flipping, etc.), normalization, etc., to improve the generalization ability of the model.

[0050] YOLO model training: Select the model architecture: Choose the version of the YOLO model (such as YOLOv3, YOLOv4, etc.) according to specific needs, and determine its network structure, hyperparameters, etc.

[0051] Model training: The YOLO model is trained using the preprocessed dataset, and the model parameters are continuously optimized through the back-propagation algorithm, enabling it to accurately identify organ types, feature points, and abnormal areas in medical images.

[0052] Model evaluation: During the training process, the model is regularly evaluated using the validation set to monitor changes in its performance and adjust the training strategy based on the evaluation results.

[0053] 2. Using the Pre-trained YOLO Model for Medical Image Recognition Loading pre-trained model: After training is completed, save the trained YOLO model as a weight file and load it when needed.

[0054] Input medical images: Input the medical images to be identified into the loaded YOLO model.

[0055] Model prediction: The YOLO model performs forward propagation on the input medical image and outputs a prediction result containing information about organ type, feature points, and abnormal areas.

[0056] Result analysis: Analyze the model's prediction results to extract information such as the type of each organ, the location of feature points, and the location and size of abnormal areas.

[0057] In an embodiment of the present invention, based on step S2, a possible embodiment will be given below to illustrate its specific implementation scheme in a non-limiting manner.

[0058] S201. Accurate setting of feature point coordinate values Select feature points: First, based on the specific type of medical image (such as CT, MRI, etc.) and diagnostic requirements, select feature points with anatomical or landmark significance. These feature points can be vascular bifurcations, organ boundaries, bone intersections, etc.

[0059] Determine coordinate values: By manually measuring or using tools in image processing software, accurately record the location of each feature point in the image and convert it into coordinate values. This step requires a high degree of accuracy and expertise to ensure the accuracy of subsequent steps.

[0060] S202. Establishment of plane coordinate system Determine the origin: Based on the coordinate values ​​of the feature points, the origin of the plane coordinate system is calculated and determined through an algorithm. Usually, the origin can be selected at the center of the image or a specific feature point.

[0061] Determine the direction of the coordinate axis: The specific direction of the coordinate axis is determined based on the anatomical structure and the positional relationship of the feature points in the medical image. For example, in a CT image, the direction perpendicular to the bed surface can be selected as the Z axis, and the directions perpendicular to the Z axis and passing through the origin can be selected as the X axis and the Y axis.

[0062] Plane coordinate system generation: After determining the origin and axis orientation, image processing software or programming tools are used to dynamically generate a clear plane coordinate system on the medical image. This coordinate system should accurately reflect the spatial layout of the medical image and facilitate subsequent data processing and analysis.

[0063] S203. Acquisition of coordinate data of abnormal area Preprocessing: Before obtaining the coordinate data of the abnormal area, medical images may need to be preprocessed, such as denoising and contrast enhancement, to improve the accuracy of subsequent processing.

[0064] Automatic detection: Utilize advanced image processing techniques and algorithms, such as threshold segmentation, morphological operations, and edge detection, to automatically detect and identify abnormal areas in images.

[0065] Coordinate data extraction: After identifying the abnormal area, the coordinate values ​​in the coordinate system are used to extract the precise coordinate data of the abnormal area in the plane coordinate system. This includes the center point coordinates, boundary coordinates, horizontal coordinate range, and vertical coordinate range of the abnormal area.

[0066] S204. Sampling and processing of medical images Determine the sampling range: Based on the coordinate data of the abnormal area, determine the specific sampling range. This range should be able to completely include the abnormal area and leave a certain margin for subsequent analysis and diagnosis.

[0067] Sampling method selection: Choose an appropriate sampling method based on the characteristics of the abnormal area and the diagnostic requirements. For example, for a small abnormal area, point-by-point sampling can be used; for a large abnormal area, block sampling or regional sampling can be used.

[0068] Sampling and processing: Using image processing software or programming tools, the medical images are sampled and processed. This step should be able to accurately segment abnormal areas and retain sufficient image information for subsequent analysis.

[0069] In an embodiment of the present invention, based on step S3, a possible embodiment will be given below to illustrate its specific implementation scheme in a non-limiting manner.

[0070] S301. Extracting Global Image Features Using Multiple Convolutional Networks Edge profile features: In convolutional networks, edge detection layers typically use specific convolution kernels (such as the Sobel operator and the Prewitt operator) to detect edge information in images. These convolution kernels can capture areas in the image where pixel values ​​change significantly, thereby extracting the edge contours of abnormal areas.

[0071] Activation functions (such as ReLU, Sigmoid, etc.) are used to enhance the representation ability of edge features, making the edge contours clearer and more prominent.

[0072] Texture features: The texture analysis layer captures the texture information in the image by designing different convolution kernels. These convolution kernels can have different sizes, directions, and weights to extract various texture features.

[0073] Through convolution operations, the texture analysis layer can extract texture changes and detailed information inside the abnormal area, such as texture thickness, direction, periodicity, etc.

[0074] Shape characteristics: The shape recognition layer mainly uses convolution kernels and pooling operations to extract the shape features of abnormal areas. The convolution kernel can capture the edges and contours of the shape, while the pooling operation is used to reduce the resolution of the feature map while retaining important shape information.

[0075] Through the shape recognition layer, the overall shape and contour information of the abnormal area can be extracted, such as circle, ellipse, irregular shape, etc.

[0076] S302. Multi-layer convolutional layer feature reduction and feature map generation After extracting the overall image features, multiple convolutional layers further process and extract the features. Each convolutional layer reduces the size of the feature map through convolution operations and extracts deeper feature information.

[0077] The number of convolution kernels in the convolution layer will gradually increase to capture more feature information. At the same time, the size of the convolution kernel may gradually decrease to extract finer features.

[0078] The output features of each convolutional layer are then fed into a 1×1 convolutional layer. The 1×1 convolutional layer is primarily used to change the number of channels in the feature map while maintaining the spatial size of the feature map. This step yields multiple feature maps of varying sizes and channel counts, each containing rich image feature information.

[0079] S303. Maximum suppression to remove duplicate feature maps After obtaining multiple feature maps of different sizes, there may be some duplicate or redundant feature maps. These feature maps may contain similar feature information, resulting in increased computational complexity and reduced recognition efficiency.

[0080] To solve this problem, the maximum suppression algorithm is used to remove duplicate feature maps. The maximum suppression algorithm compares the similarities and differences between adjacent feature maps, retaining feature maps with lower similarity and greater difference, thereby removing duplicate or redundant feature maps.

[0081] When performing maximum value suppression, a similarity threshold can be set. When the similarity between two feature maps exceeds this threshold, they are considered to be duplicates and one of them is retained as the representative.

[0082] S304. Classifier Identification and Confidence Screening The remaining feature maps are input into the classifier for recognition. The classifier can use algorithms such as support vector machines and neural networks to classify and recognize feature maps.

[0083] The classifier outputs the recognition confidence of each feature map, which indicates the probability that the feature map belongs to a certain category. The higher the confidence, the greater the possibility that the feature map belongs to that category.

[0084] By setting a confidence threshold, feature maps with higher confidence can be selected as local image feature vectors. These feature vectors can accurately describe the image characteristics of abnormal areas and provide strong support for subsequent diagnosis and analysis.

[0085] S305. Local Image Feature Vector Output and Subsequent Processing The filtered feature maps are further processed and encoded to obtain local image feature vectors. These feature vectors are usually high-dimensional and sparse, and require dimensionality reduction and compression to improve computational and storage efficiency.

[0086] Dimensionality reduction methods can use algorithms such as principal component analysis (PCA) and linear discriminant analysis (LDA) to reduce high-dimensional feature vectors to a low-dimensional space. Compression methods can use algorithms such as sparse coding and hash coding to compress feature vectors into a more compact representation.

[0087] The obtained local image feature vectors can be used in subsequent diagnosis and analysis tasks, such as disease classification, lesion detection, condition assessment, etc. By combining with other medical data and information, it can provide doctors with a more comprehensive and accurate diagnosis basis.

[0088] In an embodiment of the present invention, based on step S4, a possible embodiment will be given below to illustrate its specific implementation scheme in a non-limiting manner.

[0089] S401. Organ type mapping to organ type vector In medical image processing, different organs have different anatomical structures and functional characteristics. In order to accurately describe the relationship between abnormal regions and organs, it is necessary to map organ types into organ type vectors.

[0090] Organ type vectors can be implemented through encoding methods such as one-hot encoding (IHE) or embedding vectors. IHE represents each organ type as a separate binary vector with only one element set to 1 and all other elements set to 0. Embedding vectors, through trained low-dimensional representations, can capture the underlying relationships between organ types.

[0091] The introduction of organ type vectors helps subsequent models take into account the influence of organ type when processing abnormal areas, thereby improving the accuracy of diagnosis.

[0092] S402. Mapping of abnormal area and functional area Based on the coordinate information of the abnormal area, the area of ​​the abnormal area can be calculated. This area information is of great significance for assessing the severity and possible impact of the abnormal area.

[0093] It is also necessary to determine the functional domain of the abnormal area within the organ. A functional domain refers to an area within an organ that has a specific physiological function, such as the left ventricle or right atrium of the heart. By determining the functional domain of the abnormal area, we can more accurately understand its impact on organ function.

[0094] Map the area of ​​the abnormal region and the functional region into an area vector and a functional region vector. The area vector can be simply represented as a numerical representation of the area of ​​the abnormal region, while the functional region vector can be implemented through encoding methods such as one-hot encoding or embedding vectors.

[0095] S403. Comprehensive Feature Extraction of Fully Connected Layers The organ type vector, area vector, functional region vector, and local image feature vector are input together into a pre-designed fully connected layer. A fully connected layer is a neural network layer in which each neuron is connected to all neurons in the previous layer.

[0096] The fully connected layer performs linear transformation and nonlinear activation on the input vector to extract comprehensive features of the abnormal region. This linear transformation is implemented through a weight matrix and bias term, mapping the input vector into a new feature space. Nonlinear activation functions (such as ReLU and Sigmoid) are used to introduce nonlinear factors and enhance the model's expressiveness.

[0097] Through processing in the fully connected layer, a comprehensive feature vector of the abnormal region can be obtained. This vector contains information such as organ type, abnormal area, functional area, and local image features, which can provide strong support for subsequent diagnosis and analysis.

[0098] In an embodiment of the present invention, based on step S5, a possible embodiment will be given below to illustrate its specific implementation scheme in a non-limiting manner.

[0099] 1. Data Preparation Extraction of comprehensive features of abnormal areas: First, for each abnormal region, according to the steps described above, its organ type is mapped to an organ type vector, the area of ​​the abnormal region is calculated and mapped to an area vector, and the functional area of ​​the abnormal region is determined and mapped to a functional area vector.

[0100] Then, deep learning models such as convolutional neural networks (CNN) are used to extract local image feature vectors of abnormal areas.

[0101] Finally, these four vectors are concatenated to form the comprehensive feature vector of the abnormal area.

[0102] Dataset construction: Collect a large amount of medical imaging data, including normal and abnormal images.

[0103] Each image is annotated, including information such as the coordinates of the abnormal area, organ type, area, functional area, and diagnosis results.

[0104] Based on the annotation information, the comprehensive feature vector of each abnormal area is extracted and the dataset is constructed.

[0105] 2. Model Training Construction of random forest model: Random forest is selected as the classification or regression model because it has good anti-overfitting ability, is insensitive to outliers, and can handle high-dimensional data.

[0106] Depending on the size and complexity of the dataset, set the parameters of the random forest, such as the number of trees, maximum depth, minimum number of samples, etc.

[0107] Model training: Divide the dataset into training and testing sets.

[0108] The random forest model is trained using the training set, the comprehensive feature vector of the abnormal area is input, and the diagnostic results or related analysis indicators are output.

[0109] During the training process, the performance of the model is evaluated through methods such as cross-validation, and parameters are adjusted to optimize the model.

[0110] 3. Model Application Input of comprehensive features of abnormal areas: For new medical imaging data, the comprehensive feature vector of each abnormal region is extracted according to the steps described previously.

[0111] Obtaining analysis results: The extracted comprehensive feature vector is input into the pre-trained random forest model.

[0112] The model will output analysis results based on the input feature vector, such as the type, severity, and possible cause of the abnormal area.

[0113] Result interpretation and auxiliary diagnosis: Doctors can conduct further diagnosis and analysis based on the analysis results output by the model, combined with the patient's medical history, clinical manifestations and other information.

[0114] The analysis results of the model can provide doctors with a basis for auxiliary diagnosis and improve the accuracy and efficiency of diagnosis.

[0115] In some embodiments, the AI-based image data analysis system may include multiple functional modules composed of computer program segments. The computer program of each program segment in the AI-based image data analysis system may be stored in a memory of a computer device and executed by at least one processor to perform (see Figure 1 Description) Functions of image data analysis based on artificial intelligence.

[0116] In this embodiment, the image data analysis system based on artificial intelligence can be divided into multiple functional modules according to the functions it performs, such as Figure 2 As shown. The functional modules of system 200 may include: a basic recognition module 210, a coordinate generation module 220, a feature extraction module 230, a feature fusion module 240, and a feature classification module 250. As used herein, a module refers to a series of computer program segments that can be executed by at least one processor and perform fixed functions, and is stored in a memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0117] The basic recognition module is used to acquire medical images and use target recognition algorithms to identify organ types, feature points, and abnormal areas in medical images; A coordinate generation module, configured to generate a coordinate system based on the feature points, and generate the coordinates of the abnormal area in the coordinate system; Feature extraction module, used to extract local image features of abnormal areas and use the classifier to filter out the local image feature vector with the highest confidence; a feature fusion module, configured to input the organ type, the local image feature vector, and the coordinates of the abnormal region into a fully connected layer to obtain comprehensive features of the abnormal region; The feature classification module is used to input the comprehensive features of multiple abnormal areas into the pre-trained random forest model to obtain analysis results.

[0118] Optionally, as an embodiment of the present invention, the basic identification module includes: an object recognition unit, configured to recognize organ types, organ feature points, and abnormal regions in the medical image using a pre-trained YOLO model; The training methods of the YOLO model include: Collect historical medical images and annotate organ types, characteristic points of each organ type, and normal organs or abnormal areas in the historical medical images; Import annotated historical medical images into the dataset; The YOLO model is trained using the dataset.

[0119] Optionally, as an embodiment of the present invention, the coordinate generation module includes: A coordinate setting unit, used to pre-set the coordinate values ​​of feature points; A coordinate positioning unit, configured to locate the origin position and coordinate axis position of the coordinate system based on the position and coordinate value of the feature point, and generate a plane coordinate system on the medical image according to the origin position and coordinate axis position; A coordinate acquisition unit, configured to acquire coordinate data of the abnormal area in the plane coordinate system, wherein the coordinate data includes a horizontal coordinate range and a vertical coordinate range; The region segmentation unit is used to upsample the medical image based on the horizontal coordinate range and the vertical coordinate range to segment out abnormal regions.

[0120] Optionally, as an embodiment of the present invention, the feature extraction module includes: A first processing unit is configured to extract overall image features of the abnormal area using multiple convolutional networks, where the overall image features include edge contour features, texture features, and shape features; A second processing unit is configured to continuously reduce the overall image features using multiple convolutional layers, and input the output features of each convolutional layer into a 1×1 convolutional layer to obtain multiple feature maps of different sizes; Feature deduplication unit, used to remove duplicate feature maps using maximum value suppression; The classification and recognition unit is used to use the classifier to identify the remaining feature maps and obtain the recognition confidence; The feature screening unit is used to screen out feature maps whose confidence reaches a set confidence threshold, and output the screened feature maps as local image feature vectors.

[0121] Figure 3 The artificial intelligence-based image data analysis method provided for the embodiment of the present application can be applied to a device. Those skilled in the art will understand that the device structure involved in the embodiment of the present invention does not constitute a limitation on the device, and the device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently. In the embodiment of the present invention, the device includes but is not limited to a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of the present application described and / or required herein.

[0122] The device 300 may include a processor 310, a memory 320, and a communication unit 330. These components communicate via one or more buses. Those skilled in the art will appreciate that the server structure shown in the figure does not limit the present invention. The server structure may be a bus structure or a star structure, and may include more or fewer components than shown, or combine certain components, or arrange the components differently.

[0123] The memory 320 can be used to store execution instructions of the processor 310. The memory 320 can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk. When the execution instructions in the memory 320 are executed by the processor 310, the device 300 can perform some or all of the steps in the above-described method embodiments.

[0124] The processor 310 is the control center of the storage device, which uses various interfaces and lines to connect various parts of the entire electronic device. It executes various functions of the electronic device and / or processes data by running or executing software programs and / or modules stored in the memory 320, and calling data stored in the memory. The processor can be composed of an integrated circuit (IC), for example, it can be composed of a single packaged IC, or it can be composed of multiple packaged ICs with the same or different functions. For example, the processor 310 can only include a central processing unit (CPU). In an embodiment of the present invention, the CPU can be a single computing core or multiple computing cores.

[0125] The communication unit 330 is configured to establish a communication channel so that the storage device can communicate with other devices, receive user data sent by other devices, or send user data to other devices.

[0126] The present invention also provides a computer storage medium, wherein the computer storage medium may store a program that, when executed, may include some or all of the steps of each embodiment provided herein. The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0127] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software and a necessary general-purpose hardware platform. Based on this understanding, the technical solutions in the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, among other media capable of storing program code, and includes instructions for causing a computer device (which can be a personal computer, a server, or a second device, a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention.

[0128] In this specification, the same or similar parts between the various embodiments can be referred to each other. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment.

[0129] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the modules is merely a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of systems or modules, and can be electrical, mechanical or other forms.

[0130] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of the present embodiment according to actual needs.

[0131] In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0132] Although the present invention has been described in detail with reference to the accompanying drawings and in conjunction with preferred embodiments, the present invention is not limited thereto. Without departing from the spirit and essence of the present invention, persons of ordinary skill in the art may make various equivalent modifications or substitutions to the embodiments of the present invention, and such modifications or substitutions shall be within the scope of the present invention. Any changes or substitutions that can be easily conceived by persons skilled in the art within the technical scope disclosed in the present invention shall be within the scope of protection of the present invention.

Claims

1. An image data analysis method based on artificial intelligence, characterized in that: include: Acquire medical images and use target recognition algorithms to identify organ types, feature points, and abnormal areas in the medical images; Generate a coordinate system based on the feature points, and generate coordinates of the abnormal area in the coordinate system; Extract local image features of abnormal areas and use the classifier to filter out the local image feature vector with the highest confidence; Inputting the organ type, the local image feature vector, and the coordinates of the abnormal region into a fully connected layer to obtain comprehensive features of the abnormal region; The comprehensive features of multiple abnormal areas are input into the pre-trained random forest model to obtain the analysis results.

2. The method according to claim 1, characterized in that Acquire medical images and use object recognition algorithms to identify organ types, feature points, and abnormal areas in the medical images, including: Using a pre-trained YOLO model to identify organ types, organ feature points, and abnormal areas in the medical image; The training methods of the YOLO model include: Collect historical medical images and annotate organ types, characteristic points of each organ type, and normal organs or abnormal areas in the historical medical images; Import annotated historical medical images into the dataset; The YOLO model is trained using the dataset.

3. The method according to claim 1, characterized in that Generating a coordinate system based on the feature points and generating coordinates of the abnormal area in the coordinate system includes: Pre-set the coordinate values ​​of the feature points; Locating the origin position and coordinate axis position of the coordinate system based on the position and coordinate values ​​of the feature points, and generating a plane coordinate system on the medical image according to the origin position and coordinate axis position; Acquire coordinate data of the abnormal area in the plane coordinate system, wherein the coordinate data includes a horizontal coordinate range and a vertical coordinate range; The medical image is upsampled based on the abscissa range and the ordinate range to segment an abnormal area.

4. The method according to claim 1, wherein Extract local image features of abnormal areas and use the classifier to filter out the local image feature vectors with the highest confidence, including: Extracting overall image features of the abnormal area using multiple convolutional networks, wherein the overall image features include edge contour features, texture features, and shape features; The overall image features are continuously reduced using multiple convolutional layers, and the output features of each convolutional layer are respectively input into a 1×1 convolutional layer to obtain multiple feature maps of different sizes; Use maximum suppression to remove duplicate feature maps; Use the classifier to identify the remaining feature maps and obtain the recognition confidence; The feature maps whose confidence reaches the set confidence threshold are filtered out, and the filtered feature maps are output as local image feature vectors.

5. The method according to claim 1, wherein The organ type, the local image feature vector, and the coordinates of the abnormal region are input into a fully connected layer to obtain comprehensive features of the abnormal region, including: Map organ types to organ type vectors; Determine the area of ​​the abnormal region and the functional region in the organ according to the coordinates of the abnormal region, and map the area and the functional region into an area vector and a functional region vector; The organ type vector, area vector, functional region vector and local image feature vector are input into a pre-designed fully connected layer, which performs linear transformation and nonlinear activation on the input vector to extract the comprehensive features of the abnormal area.

6. An image data analysis system based on artificial intelligence, characterized in that: include: The basic recognition module is used to acquire medical images and use target recognition algorithms to identify organ types, feature points, and abnormal areas in medical images; A coordinate generation module, configured to generate a coordinate system based on the feature points, and generate the coordinates of the abnormal area in the coordinate system; Feature extraction module, used to extract local image features of abnormal areas and use the classifier to filter out the local image feature vector with the highest confidence; a feature fusion module, configured to input the organ type, the local image feature vector, and the coordinates of the abnormal region into a fully connected layer to obtain comprehensive features of the abnormal region; The feature classification module is used to input the comprehensive features of multiple abnormal areas into the pre-trained random forest model to obtain analysis results.

7. The system according to claim 6, characterized in that The basic identification module includes: an object recognition unit, configured to recognize organ types, organ feature points, and abnormal regions in the medical image using a pre-trained YOLO model; The training methods of the YOLO model include: Collect historical medical images and annotate organ types, characteristic points of each organ type, and normal organs or abnormal areas in the historical medical images; Import annotated historical medical images into the dataset; The YOLO model is trained using the dataset.

8. The system according to claim 6, wherein: The coordinate generation module includes: A coordinate setting unit, used to pre-set the coordinate values ​​of feature points; A coordinate positioning unit, configured to locate the origin position and coordinate axis position of the coordinate system based on the position and coordinate value of the feature point, and generate a plane coordinate system on the medical image according to the origin position and coordinate axis position; A coordinate acquisition unit, configured to acquire coordinate data of the abnormal area in the plane coordinate system, wherein the coordinate data includes a horizontal coordinate range and a vertical coordinate range; The region segmentation unit is configured to upsample the medical image based on the horizontal coordinate range and the vertical coordinate range to segment out abnormal regions.

9. The system according to claim 6, wherein: The feature extraction module includes: A first processing unit is configured to extract overall image features of the abnormal area using multiple convolutional networks, where the overall image features include edge contour features, texture features, and shape features; A second processing unit is configured to continuously reduce the overall image features using multiple convolutional layers, and input the output features of each convolutional layer into a 1×1 convolutional layer to obtain multiple feature maps of different sizes; Feature deduplication unit, used to remove duplicate feature maps using maximum value suppression; The classification and recognition unit is used to use the classifier to identify the remaining feature maps and obtain the recognition confidence; The feature screening unit is used to screen out feature maps whose confidence reaches a set confidence threshold, and output the screened feature maps as local image feature vectors.

10. A device, characterized in that include: A memory for storing an artificial intelligence-based image data analysis program; A processor, configured to implement the steps of the artificial intelligence-based image data analysis method as described in any one of claims 1 to 5 when executing the artificial intelligence-based image data analysis program.

Images