An intelligent medical analysis method and system integrating local and global features of images

Abstract

The present invention relates to a smart medical analysis method and system for integrating local and global features of an image, the method comprising: dividing an image into regions to obtain one or more image regions, arranging the image regions in order to obtain a corresponding image region sequence; performing numerical data transformation on the pixel values ​​of each image region in the image region sequence to obtain a corresponding numerical region sequence and a numerical relationship region sequence, and then obtaining the weight of an analysis model; obtaining an analysis result based on the analysis model; and obtaining a final analysis result by synthesizing the corresponding analysis results based on the weight value. The present invention localizes the global image region, so that the analysis of local features can guide the analysis of global features, and the analysis results can be comprehensively used to greatly improve the efficiency of the subsequent analysis model.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to an intelligent medical analysis method and system that integrates local and global features of an image. Background Art

[0002] In recent years, in the era of medical big data and artificial intelligence, precision medicine has become a hot topic in the field of medical research. The traditional imaging diagnosis model that relies on visual form has gradually failed to meet the requirements of accurate diagnosis and treatment of diseases. With the rapid development of artificial intelligence and big data technology, people have realized that medical data and intelligent analysis models can help medical data analysis; medical clinical data is like a treasure buried in the mountains, which has a huge impact on pharmaceutical companies, medical service providers, medical payers and patients.

[0003] On the other hand, big models have been widely used in many fields; in some tests, medical big models are comparable to or even surpass doctors, showing their application advantages and value in ensuring the accuracy and fairness of medical services and improving the efficiency of medical systems. Based on the advantages of field expertise, high customization, high accuracy, and strong generalization, vertical multi-task big models based on medical scenarios integrate multimodal data to provide accurate and fast auxiliary solutions for medical decision-making, helping doctors and patients achieve high-quality medical experience.

[0004] Big models are naturally compatible with the medical industry. There is a large amount of data with rich modalities in the medical field, and it is multidisciplinary and cross-domain. The strength of big models is to integrate, summarize, analyze, judge and automatically summarize multiple types of data. Driven by cutting-edge research and corporate layout, big models have taken root in the medical field and have shown a clear trend.

[0005] At present, the vertical application of big models is in a state of independent development and flourishing, and the application of big models themselves is still questionable; the analysis results of each analysis model are relatively independent, and because the input of the big model is multimodal and diverse, they are often different and incompatible, and the analysis results can only be manually borrowed from each other, and ultimately rely on manual analysis. Therefore, the comprehensive application and effective use of the analysis results of intelligent analysis models are technical issues to be solved.

[0006] Based on the above problems, the present invention localizes the global image area, so that the analysis of local features can guide the analysis of global features, and the analysis results can be comprehensively used to greatly improve the efficiency of subsequent analysis models. Summary of the invention

[0007] In order to solve the above problems in the prior art, the present invention proposes an intelligent medical analysis method and system that integrates local and global features of an image. The method comprises:

[0008] Step S1: dividing the image into regions to obtain one or more image regions, and arranging the image regions in order to obtain a corresponding image region sequence;

[0009] Step S2: Performing a numerical data transformation on the pixel values ​​of each image region in the image region sequence to obtain a corresponding numerical region sequence; the numerical region sequence contains one or more numerical regions, and each element in the numerical region corresponds to a numerical data transformation result; performing a numerical relationship transformation on the relationship between the pixel values ​​of each image region in the image region sequence to obtain a numerical relationship region sequence; the numerical region relationship sequence contains one or more numerical relationship regions, and each element in the numerical relationship region corresponds to a numerical relationship transformation result; performing K-bit picking on the numerical data transformation result and the numerical relationship transformation result respectively to obtain a K-bit value; at this time, each numerical region in the numerical region sequence and the data type of the elements contained in the numerical relationship region in the numerical relationship region sequence are all K-bit values; correspondingly, a numerical string sequence {d1, ..., dp1, ..., d p1 , ..., d P1} and the numerical relation string sequence {dr1, ..., dr p2 , ..., dr P2}; They are the numbers of the numerical strings in the numerical string sequence or the numerical relation strings in the numerical relation string sequence;

[0010] Step S3: Determine the weights of the analysis models for different global feature types according to the numerical string sequence and the numerical relationship string sequence; specifically:

[0011] Step S31: obtaining a numerical substring sample or a numerical relationship substring sample to be compared; and obtaining the local accuracy of the analysis model when the numerical substring sample or the numerical relationship substring sample is included;

[0012] Constructing a sample library including numerical substring samples and numerical relationship substring samples in advance; the sample library pre-stores the corresponding relationship between the numerical substring samples or numerical relationship substring samples and the local accuracy of the analysis model when including the numerical substring samples or numerical relationship substring samples;

[0013] Step S32: compare the numerical substring samples to be compared with each numerical string in the numerical string sequence respectively. If the numerical substring samples hit, the numerical sensitivity is incremented, and the numerical accuracy of the analysis model is set equal to the current numerical accuracy plus the local accuracy of the analysis model when the numerical substring samples are included; compare the numerical relationship substring samples to be compared with each numerical relationship string in the numerical relationship string sequence respectively. If the numerical relationship substring samples hit, the numerical relationship sensitivity is incremented, and the relationship accuracy of the analysis model is set equal to the current relationship accuracy plus the local accuracy of the analysis model when the numerical relationship substring samples are included;

[0014] Step S33: determine whether there are still numerical substring samples or numerical relationship substring samples to be compared, if yes, return to step S31; otherwise, proceed to the next step;

[0015] Step S34: Determine the numerical accuracy of the analysis model and the current numerical accuracy Divide by the number of sample hits for the numeric substring; determine the relational accuracy of the analytical model Equal to the current relation accuracy divided by the number of sample hits of the numeric relation substring;

[0016] Step S35: Calculate the weight of the analysis model based on the numerical accuracy and the relationship accuracy, respectively, so that the higher the numerical accuracy and / or the relationship accuracy, the larger the corresponding weight value; specifically: Calculate the numerical weight value of the analysis model m based on the numerical accuracy ; Calculate the relationship weight value of the analysis model m based on the relationship accuracy in: is the number of the analysis model; M is the number of analysis models;

[0017] Step S4: Input each type of global feature into the analysis model related to the global feature, and obtain the corresponding analysis results ; Based on the weight value, the corresponding analysis results are synthesized Get the final analysis results.

[0018] Furthermore, the step S1 specifically includes: partitioning the region involved in the object of interest in the image based on the size of the area to obtain one or more image regions.

[0019] Furthermore, the image comes from medical data, and the image contains an object of interest.

[0020] Furthermore, the object of interest is an analysis object of an analysis model.

[0021] Furthermore, the initial values ​​of the numerical sensitivity and numerical relationship sensitivity, the numerical accuracy and the relationship accuracy are set to 0.

[0022] A smart medical analysis system for fusing local and global features of an image, comprising: an intelligent server and a mobile terminal; the intelligent server is used to implement the above-mentioned smart medical analysis method for fusing local and global features of an image; the mobile terminal is used to send a smart medical analysis request for fusing local and global features of an image to the intelligent server, and obtain and present the analysis results returned by the intelligent server.

[0023] Furthermore, the intelligent server is an artificial intelligence server.

[0024] A smart medical analysis platform for fusing local and global features of an image comprises a processor, wherein the processor is coupled to a memory, wherein the memory stores program instructions, and when the program instructions stored in the memory are executed by the processor, the smart medical analysis method for fusing local and global features of an image is implemented.

[0025] A computer-readable storage medium includes a program, which, when running on a computer, enables the computer to execute the intelligent medical analysis method for fusing local and global features of an image.

[0026] An intelligent server is configured to execute the intelligent medical analysis method for fusing local and global features of an image.

[0027] The beneficial effects of the present invention include:

[0028] (1) For image data with large data volume, the data volume can be reduced by partitioning, and the global image area can be localized, so that the analysis of local features can guide the analysis of global features, which greatly improves the efficiency of subsequent analysis models; through the fusion of local features and global features, various existing vertical application large models can be effectively and comprehensively utilized;

[0029] (2) Local features in the image area are discovered through numerical transformation and numerical relationship changes. Combined with K-bit data collection, local features are dynamically converted into numerical string sequences and numerical relationship string sequences that are easy to compare numerically. This provides a basis for the rapid processing of image data. Local features are used as guiding information to quickly screen the analysis results from the two aspects of sensitivity and accuracy of the analysis model, thereby effectively combining the local features of the image area with the subsequent global feature analysis, greatly improving the processing efficiency and the availability of the analysis results. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present application, but do not constitute an improper limitation of the present invention. In the drawings:

[0031] Figure 1Schematic diagram of the intelligent medical analysis method for fusing local and global features of images provided by the present invention. DETAILED DESCRIPTION

[0032] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments, wherein the illustrative embodiments and descriptions are only used to explain the present invention but are not intended to limit the present invention.

[0033] The present invention proposes an intelligent medical analysis method and system that integrates local and global features of images. Figure 1 As shown, the method comprises the following steps:

[0034] Step S1: performing region division on the image to obtain one or more image regions, and arranging the image regions in order to obtain a corresponding image region sequence; specifically: performing area-based region division on the region involved in the object of interest in the image to obtain one or more image regions; for example: a simple region division method is to divide the image from top to bottom into a plurality of long strip regions, and the heights of the long strip regions are the same;

[0035] Preferably: the image is a human medical-related image, a fundus image, etc.;

[0036] Alternatively, the step S1 specifically includes: performing space-based partitioning on the three-dimensional region involved in the object of interest in the image to obtain one or more image slices involved in the object of interest; taking each image slice as an image region, thereby obtaining one or more image regions;

[0037] Preferably: based on the positional relationship of the image regions, the image regions are arranged in sequence in a certain order to obtain a corresponding image region sequence;

[0038] Preferably: there is an inherent spatial relationship between each image region in the region sequence; however, this spatial relationship may be based on different times or the same time;

[0039] Through the above processing, the image data with large data volume can be reduced by partitioning, and the global image area can be localized, so that the analysis of local features can guide the analysis of global features, so that the efficiency of subsequent analysis models can be greatly improved;

[0040] Step S2: transform the pixel values ​​of each image region in the image region sequence into numerical data to obtain a corresponding numerical region sequence; the numerical region sequence contains one or more numerical regions; transform the relationship between the pixel values ​​of each image region in the image region sequence into numerical relationship to obtain a numerical relationship region sequence; the numerical region relationship sequence contains one or more numerical relationship regions; wherein: each numerical region in the numerical region sequence, and the data type of the elements contained in the numerical relationship region in the numerical relationship region sequence are all K-bit numerical values; arranging these K-bit numerical values ​​according to their positions in the numerical region or the numerical relationship region constitutes a numerical string or numerical relationship string corresponding to each image region; thus, a numerical string sequence {d1, ..., d2, ..., d3, ..., d4, ..., d5, ..., d6, ..., d7, ..., d8, ..., d9, ..., d10, ..., d11, ..., d12, ..., d13, ..., d14, ..., d15, ..., d16, ..., d17, ..., d18, ..., d19, ..., d21, ..., d22, ..., d19, ..., d23, ..., d14, ..., d15, ..., d16, ..., d17, ..., d18, ..., d19, ..., d19, ..., d24, ..., d19, ..., d11, ..., d11, ..., d11 p1 , ..., d P1} and the numerical relation string sequence {dr1, ..., dr p2 , ..., dr P2};

[0041] Preferably, K is a preset value, such as 32, 64, 128, etc. The adjustment of the K value can directly affect the size of the numerical string sequence and the numerical relationship string sequence, resulting in a larger or smaller amount of information loss, thereby increasing or reducing the subsequent processing amount; therefore, the K value can be dynamically adjusted according to the type of the object of interest and the need for local guidance;

[0042] The local features in the image area are discovered through numerical transformation and numerical relationship changes. Combined with K-bit data collection, the local features are dynamically converted into numerical string sequences and numerical relationship string sequences that are convenient for numerical comparison. The local features of the image area are combined with the subsequent global feature analysis, thereby greatly increasing the processing efficiency.

[0043] The step of converting the pixel values ​​of each image in the image region sequence into numerical data to obtain a corresponding numerical region sequence specifically comprises the following steps:

[0044] Step SA21: sequentially acquiring an unprocessed image region in the image region sequence as the current image region;

[0045] Step SA22: Processing the case of numerical data transformation and the case of numerical relationship transformation respectively;

[0046] For the case of numerical data transformation: binarize each pixel value in the current image area, if the pixel value is greater than the first numerical threshold, set the corresponding numerical value to 1; if the pixel value is less than the second numerical threshold, set the corresponding numerical value to 0; otherwise, set the corresponding pixel value to the default value; after all pixel values ​​are processed, proceed to the next step; wherein: the first numerical threshold and the second numerical threshold are both preset values ​​and the first numerical threshold is greater than the second numerical threshold;

[0047] For the case of numerical relational transformation: binarize each pixel value in the current image area, if the pixel value difference between the pixel value and any of its adjacent pixel values ​​is greater than the first relational threshold, set the corresponding value to 1; if the pixel value is less than the second relational threshold, set the corresponding value to 0; otherwise, set the corresponding pixel value to the default value; after all pixel values ​​are processed, proceed to the next step; wherein: the first relational threshold and the second relational threshold are both preset values ​​and the first relational threshold is greater than the second relational threshold; the greater the distance between the first relational threshold and the second relational threshold, or the greater the distance between the first numerical threshold and the second numerical threshold, the higher the degree of sparseness; otherwise, the lower;

[0048] Preferably: the pixel value is one of a grayscale value, a brightness value, a specific color value, etc.;

[0049] Step SA23: respectively process the values ​​obtained in the case of numerical data transformation and numerical relationship transformation; if there is a row or a column in the current image area where all the values ​​corresponding to the pixel values ​​are the default values, delete the row or the column; that is, the 0 and 1 values ​​in the area are retained; the value area obtained at this time is the 0 and 1 area, which can be used for subsequent Kbit-based value picking;

[0050] Step SA24: respectively process the values ​​obtained in the case of numerical data transformation and numerical relationship transformation; pick up the values ​​corresponding to each pixel in the image area in a specific order, every K bits, to form a value string and a value relationship string corresponding to the image area; the image area sequence corresponds to the data string sequence and the value relationship string sequence; thus, the value string sequence {d1, ..., d p1 , ..., d P1} and the numerical relation string sequence {dr1, ..., dr p2 , ..., dr P2};in: are the numbers of the numerical strings or numerical relation strings in the sequence; P1 and P2 are the lengths of the numerical string sequence and the numerical relation string sequence respectively;

[0051] Preferably: the specific order is from top to bottom and from left to right;

[0052] Preferably: P1 and P2 are dynamic values ​​obtained in the above calculation process, and P1 and P2 are equal or unequal;

[0053] Step S3: Determine the weights of the analysis models for different global feature types according to the numerical string sequence and the numerical relationship string sequence; specifically: compare the numerical substring samples and the numerical relationship substring samples with the numerical string sequence and the numerical relationship string sequence respectively to determine the weights of the analysis models for different global feature types; pre-store the corresponding relationship between the numerical substring samples or the numerical relationship substring samples and the local accuracy of the analysis model when the numerical substring samples or the numerical relationship substring samples are included; the analysis model is a model used to analyze the global features to obtain the analysis results;

[0054] Wherein: the global features include image features, ultrasound features, observable features and other aspects; wherein: each aspect of the global features includes one or more sub-features;

[0055] Preferably: the image features include multiple features for overall image evaluation, such as edge, texture, shape, grayscale, etc.; the ultrasound features include multiple ultrasound evaluation features, such as echo, boundary, smoothness, etc.; the observable features may be subjective evaluation features; set according to the doctor's experience or big data analysis;

[0056] Preferably: the analysis model is an artificial intelligence model;

[0057] The step S3 specifically includes the following steps:

[0058] Step S31: obtaining a numerical substring sample or a numerical relationship substring sample to be compared; and obtaining the local accuracy of the analysis model when the numerical substring sample or the numerical relationship substring sample is included;

[0059] Before performing step S1, a sample library including numerical substring samples and numerical relationship substring samples is constructed in advance; the sample library pre-stores the correspondence between the numerical substring samples or numerical relationship substring samples and the local accuracy of the analysis model when the numerical substring samples or numerical relationship substring samples are included; wherein: the numerical substring samples and the numerical relationship substring samples are extracted from big data; the image samples are processed similarly to the above step S2 to obtain the numerical string and the numerical relationship string corresponding to the image sample, the global features corresponding to the image sample are input into the analysis model to obtain the corresponding analysis result, and then the analysis result is obtained The correctness of the image sample; the numerical string and the numerical relationship string are divided to obtain one or more numerical substrings and numerical relationship substrings; obviously, the substrings here can be true substrings or false substrings; the numerical substrings and numerical relationship substrings are clustered to obtain one or more numerical substring samples and numerical relationship substring samples; for each numerical substring sample or numerical relationship substring sample, the local accuracy of the analysis model when the numerical substring sample or numerical relationship substring sample is included is calculated and saved, thereby constructing a sample library; the above-mentioned clustering method can be replaced by manual screening, screening according to specific rules, mean screening, random screening and other methods;

[0060] Step S32: compare the numerical substring samples to be compared with each numerical string in the numerical string sequence respectively. If the numerical substring samples hit, the numerical sensitivity is incremented, and the numerical accuracy of the analysis model is set equal to the current numerical accuracy plus the local accuracy of the analysis model when the numerical substring samples are included; compare the numerical relationship substring samples to be compared with each numerical relationship string in the numerical relationship string sequence respectively. If the numerical relationship substring samples hit, the numerical relationship sensitivity is incremented, and the relationship accuracy of the analysis model is set equal to the current relationship accuracy plus the local accuracy of the analysis model when the numerical relationship substring samples are included;

[0061] Preferably: the initial values ​​of the numerical sensitivity and numerical relationship sensitivity, the numerical accuracy and the relationship accuracy are set to 0;

[0062] Step S33: determine whether there are still numerical substring samples or numerical relationship substring samples to be compared, if yes, return to step S31; otherwise, proceed to the next step;

[0063] Step S34: Determine the numerical accuracy of the analysis model and the current numerical accuracy Divide by the number of sample hits for the numeric substring; determine the relational accuracy of the analytical model Equal to the current relation accuracy divided by the number of sample hits of the numeric relation substring;

[0064] Step S35: Calculate the weight of the analysis model based on the numerical accuracy and the relationship accuracy, respectively, so that the higher the numerical accuracy and / or the relationship accuracy, the larger the corresponding weight value; specifically: Calculate the numerical weight value of the analysis model m based on the numerical accuracy ; Calculate the relationship weight value of the analysis model m based on the relationship accuracy in: is the number of the analysis model; M is the number of analysis models;

[0065] Step S4: Obtain global features; input each type of global feature into the analysis model related to the global feature, and obtain the corresponding analysis results ; Determine the final analysis result based on the numerical weight value or the relationship weight value; Select the final analysis result based on the numerical accuracy and the relationship accuracy;

[0066] The step S4 specifically includes the following steps:

[0067] Step S41: when the numerical sensitivity is greater than the numerical relationship sensitivity, proceed to step S42, otherwise, proceed to step S43;

[0068] Step S42: Correspondence analysis results based on various types of global features And the numerical weight value Determine the final analysis result Rf1; proceed to step S44; specifically: use the following formula (1) to calculate the final analysis result Rf1;

[0069] (1);

[0070] Step S43: Correspondence analysis results based on various types of global features And the relationship weight value Determine the final analysis result Rf2; specifically: use the following formula (2) to calculate the final analysis result Rf2;

[0071] (2);

[0072] Step S44: outputting the final analysis result; the final analysis result is a comprehensive result obtained according to various types of global feature corresponding analysis models;

[0073] Based on the same inventive concept, the present invention also provides an intelligent medical analysis system for fusing local and global features of an image, the system comprising: an intelligent server and an institutional terminal; the intelligent server is used to implement the intelligent medical analysis method for fusing local and global features of an image; the institutional terminal is used to send an intelligent medical analysis request for fusing local and global features of an image to the intelligent server;

[0074] Preferably: the intelligent server is an artificial intelligence server, in which analysis models for different types of global features are stored; the analysis model is a vertical application large model for global features;

[0075] Preferably: the institutional terminal is a mobile phone, a computer, or a mobile terminal;

[0076] A computer program (also referred to as a program, software, software application, script, or code) can be written in any form of a programming language, including an assembled or interpreted language, a declarative or procedural language, and it can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, object, or other unit suitable for use in a computing environment. A computer program may, but need not, correspond to a file in a file system. A program can be stored in a portion of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to the program, or in multiple collaborative files (e.g., files storing one or more modules, subroutines, or code portions). A computer program can be deployed to execute on one computer or on multiple computers located at one site or distributed across multiple sites and interconnected by a communication network.

[0077] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0078] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0079] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0080] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. An intelligent medical analysis method integrating local and global features of an image, characterized in that: The method comprises: Step S1: dividing the image into regions to obtain one or more image regions, and arranging the image regions in order to obtain a corresponding image region sequence; Step S2: Performing a numerical data transformation on the pixel values ​​of each image region in the image region sequence to obtain a corresponding numerical region sequence; the numerical region sequence contains one or more numerical regions, and each element in the numerical region corresponds to a numerical data transformation result; performing a numerical relationship transformation on the relationship between the pixel values ​​of each image region in the image region sequence to obtain a numerical relationship region sequence; the numerical region relationship sequence contains one or more numerical relationship regions, and each element in the numerical relationship region corresponds to a numerical relationship transformation result; performing K-bit picking on the numerical data transformation result and the numerical relationship transformation result respectively to obtain a K-bit value; at this time, each numerical region in the numerical region sequence and the data type of the elements contained in the numerical relationship region in the numerical relationship region sequence are all K-bit values; correspondingly, a numerical string sequence {d1, ..., dp1, ..., d p1 , ..., d P1 } and the numerical relation string sequence {dr1, ..., dr p2 , ..., dr P2 }; p1∈1~P1, p2∈1~P2 are the numbers of the numerical strings in the numerical string sequence or the numerical relation strings in the numerical relation string sequence respectively; Step S3: Determine the weights of the analysis models for different global feature types according to the numerical string sequence and the numerical relationship string sequence; specifically: Step S31: obtaining a numerical substring sample or a numerical relationship substring sample to be compared; and obtaining the local accuracy of the analysis model when the numerical substring sample or the numerical relationship substring sample is included; Constructing a sample library including numerical substring samples and numerical relationship substring samples in advance; the sample library pre-stores the corresponding relationship between the numerical substring samples or numerical relationship substring samples and the local accuracy of the analysis model when including the numerical substring samples or numerical relationship substring samples; Step S32: compare the numerical substring samples to be compared with each numerical string in the numerical string sequence respectively. If the numerical substring samples hit, the numerical sensitivity is incremented, and the numerical accuracy of the analysis model is set equal to the current numerical accuracy plus the local accuracy of the analysis model when the numerical substring samples are included; compare the numerical relationship substring samples to be compared with each numerical relationship string in the numerical relationship string sequence respectively. If the numerical relationship substring samples hit, the numerical relationship sensitivity is incremented, and the relationship accuracy of the analysis model is set equal to the current relationship accuracy plus the local accuracy of the analysis model when the numerical relationship substring samples are included; Step S33: determine whether there are still numerical substring samples or numerical relationship substring samples to be compared, if yes, return to step S31; Otherwise, proceed to the next step; Step S34: Determine whether the numerical accuracy of the analysis model is equal to the current numerical accuracy Acd m Divide by the number of hits of the numerical substring sample; determine the relational accuracy Acr of the analysis model m It is equal to the current relation accuracy divided by the number of sample hits of the numeric relation substring; Step S35: Calculate the weight of the analysis model based on the numerical accuracy and the relationship accuracy, respectively, so that the higher the numerical accuracy and / or the relationship accuracy, the larger the corresponding weight value; specifically: Calculate the numerical weight value Wd of the analysis model m based on the numerical accuracy m =Acd m / ∑ m A m ; Calculate the relationship weight value Wr of the analysis model m based on the relationship accuracy m =Acr m / ∑ m Acr m ; Where: m∈1~M is the number of the analysis model; M is the number of analysis models; Step S4: Input each type of global feature into the analysis model related to the global feature, and obtain the corresponding analysis result R m ; Based on the weight value, the corresponding analysis results R m Get the final analysis results.

2. The intelligent medical analysis method for fusing local and global features of an image according to claim 1, characterized in that: The step S1 specifically includes: partitioning the region involved in the object of interest in the image based on the size of the area to obtain one or more image regions.

3. The intelligent medical analysis method for fusing local and global features of an image according to claim 2, characterized in that: The image comes from medical data, and the image contains an object of interest.

4. The intelligent medical analysis method for fusing local and global features of an image according to claim 3, characterized in that: The object of interest is an analysis object of the analysis model.

5. The intelligent medical analysis method for fusing local and global features of an image according to claim 4, characterized in that: Set the initial values ​​of numeric sensitivity and numeric relationship sensitivity, numeric accuracy and relationship accuracy to 0.

6. An intelligent medical analysis system integrating local and global features of an image, characterized in that: include: An intelligent server and a mobile terminal; the intelligent server is used to implement the intelligent medical analysis method for fusing local and global features of an image as described in any one of claims 1 to 5; the mobile terminal is used to send an intelligent medical analysis request for fusing local and global features of an image to the intelligent server, and obtain and present the analysis results returned by the intelligent server.

7. The intelligent medical analysis system for integrating local and global features of an image according to claim 6, characterized in that: The intelligent server is an artificial intelligence server.

8. An intelligent medical analysis platform integrating local and global features of images, characterized in that: It includes a processor, the processor is coupled to a memory, the memory stores program instructions, and when the program instructions stored in the memory are executed by the processor, the intelligent medical analysis method for fusing local and global features of an image as described in any one of claims 1-5 is implemented.

9. A computer-readable storage medium, characterized in that: The invention comprises a program which, when running on a computer, enables the computer to execute the intelligent medical analysis method for fusing local and global features of an image as described in any one of claims 1 to 5.

10. An intelligent server, characterized in that: The intelligent server is configured to execute the intelligent medical analysis method for fusing local and global features of an image as described in any one of claims 1-5.

Images