Intelligent processing system for medical images
Through an intelligent processing system for medical images, encrypted transmission, significant feature extraction and abnormal detection technology, the problems of low accuracy and reliability of existing medical image processing technologies are solved, and more efficient and accurate diagnostic results are achieved.
Patent Information
- Application Number
- CN202411639869.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-18
- Publication Date
- 2025-03-11
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing medical imaging processing technology has low accuracy and reliability, resulting in insufficient accuracy and reliability of diagnostic results.
It provides an intelligent processing system for medical images, transmits medical images through encryption, performs significant feature extraction and distribution discrete analysis, generates medical significant feature maps, and outputs comprehensive detection results through abnormal detection through dual channels, and finally returns to the user for auxiliary diagnosis.
It improves the accuracy and reliability of image processing, thereby improving diagnostic efficiency and accuracy.
Smart Images
Figure CN119601177A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of medical image processing, and in particular to an intelligent processing system for medical images. Background Art
[0002] With the continuous advancement of medical technology and the rapid development of digital medicine, medical imaging plays an increasingly important role in the diagnosis, treatment and prevention of diseases. Traditional medical image processing and analysis methods mainly rely on the doctor's experience and intuitive judgment, which is not only time-consuming and labor-intensive, but also easily affected by subjective factors, making it difficult to ensure the accuracy and consistency of the diagnosis results. Most existing computer-aided diagnosis systems can only perform simple processing and analysis on a single medical image, lacking in-depth extraction and comprehensive analysis of medical image features, resulting in the need to improve the accuracy and reliability of the diagnosis results.
[0003] In the current related technologies, medical image processing has the technical problem of low accuracy and reliability, which leads to insufficient accuracy and reliability of the final diagnosis results. Summary of the invention
[0004] The present application provides an intelligent processing system for medical images, encrypts and transmits medical images, extracts significant features of medical images, and performs distribution discrete analysis on multiple significant feature areas. If the distribution discreteness is less than a predetermined discrete scalar, a medical significant feature map is generated based on the multiple significant feature areas, and the medical significant feature map is transferred to the abnormality detection dual channel, and a first abnormality detection result and a second abnormality detection result are output. After performing an intersection operation, a comprehensive detection result is obtained, and the comprehensive detection result is encrypted and returned to the user end, and auxiliary diagnosis and other technical means are used to achieve the technical effect of improving the accuracy and reliability of image processing, thereby improving the efficiency and accuracy of diagnosis.
[0005] This application provides an intelligent processing system for medical images, which includes a user end and a cloud processing platform, including:
[0006] An image transmission module is used to receive a medical image to be identified at a user end, and encrypt and transmit the medical image to a cloud processing platform; a significant feature extraction module is used to extract significant features of the medical image at the cloud processing platform, and perform distribution discrete analysis on multiple significant feature areas. If the distribution discreteness is less than a predetermined discrete scalar, a medical significant feature map is generated according to the multiple significant feature areas; an abnormality detection module is used to transfer the medical significant feature map to an abnormality detection dual channel, output a first abnormality detection result and a second abnormality detection result, and obtain a comprehensive detection result after performing an intersection operation; an auxiliary diagnosis module is used to encrypt the comprehensive detection result and return it to the user end for auxiliary diagnosis.
[0007] In a possible implementation, significant features are extracted from the medical image, and the following processing is performed:
[0008] The medical image is input into a Gaussian pyramid for multi-scale Gaussian downsampling to obtain a multi-scale image set, wherein the Gaussian pyramid includes a brightness extraction unit, a color extraction unit and a direction extraction unit; the center-surrounding difference calculation is performed on the multi-scale image set to generate a brightness feature map, a color feature map and a direction feature map; the brightness feature map, the color feature map and the direction feature map are superimposed to generate an initial significant image; the initial significant image is extracted according to an initial significant feature threshold to obtain a plurality of significant feature regions that meet the initial significant feature threshold, wherein the initial significant feature threshold is determined based on the medical image type and the abnormal part.
[0009] In a possible implementation, the following processing is performed:
[0010] If the distribution dispersion is greater than or equal to the predetermined discrete scalar, the initial significant feature threshold is incremented according to a predetermined step size to determine a first significant feature threshold; the initial significant image is extracted according to the first significant feature threshold until a plurality of significant feature regions smaller than the predetermined discrete scalar are obtained.
[0011] In a possible implementation, a medically significant feature map is generated and the following processing is performed:
[0012] The plurality of part coordinates of the plurality of significant feature regions are read; the plurality of significant feature regions are marked according to the plurality of part coordinates, and are fused to generate the medical significant feature map.
[0013] In a possible implementation, the medically significant feature map is transferred to the abnormality detection dual channel, the first abnormality detection result and the second abnormality detection result are output, and the comprehensive detection result is obtained after the intersection operation is performed, and the following processing is performed:
[0014] A dual channel for abnormality detection is constructed, wherein a first abnormality detection branch is constructed based on a convolutional neural network, and a second abnormality detection branch is constructed based on a similarity comparison library; the medical significant feature map is respectively input into the first abnormality detection branch and the second abnormality detection branch, a first abnormality detection category and a second abnormality detection category are output, and an intersection operation is performed on the first abnormality detection category and the second abnormality detection category, and the comprehensive detection result is output.
[0015] In a possible implementation, a first anomaly detection branch is constructed based on a convolutional neural network, and the following processing is performed:
[0016] The image type and abnormal part of the medical image are read, and the image type and abnormal part are used as retrieval constraints to query the medical image processing log to obtain a first sample significant feature atlas and a first sample abnormality detection category set; the first sample significant feature map is used as input, and the first sample abnormality detection category is used as output, and the first sample significant feature atlas and the first sample abnormality detection category set are used to supervise the convolutional neural network training until a first abnormality detection branch that meets convergence is obtained.
[0017] In a possible implementation, a similarity comparison library is constructed and the following processing is performed:
[0018] Taking the image type and abnormal part as constraints, positive sample retrieval is performed based on big data to obtain a second sample abnormality detection category set, and the second sample abnormality detection category set is deduplicated to obtain a standard detection category set; a first standard detection category is randomly selected from the standard detection category set, and a first sample medical image set is retrieved according to the image type, abnormal part and the first standard detection category; high-frequency features are extracted from the first sample medical image set to obtain a first standard image set; a first similarity discrimination threshold is configured, and a first similarity comparison unit is constructed according to the first standard detection category, the first standard image set and the first similarity discrimination threshold, and the unit is added to the similarity comparison library, wherein if the comparison similarity satisfies the first similarity discrimination threshold, the first standard detection category is output.
[0019] In a possible implementation, high-frequency features are extracted from the first sample medical image set to obtain a first standard image set, and the following processing is performed:
[0020] Randomly select any image in the first sample medical image set as the first medical image, wherein the first medical image includes a first structural feature, a first texture feature and a first shape feature; perform a similarity comparison between the first structural feature and other image structural features in the first sample medical image set, and set the amount of data whose statistical similarity is less than a predetermined similarity scalar as a first difference; if the first difference is greater than a difference threshold, determine a second difference based on the first texture feature comparison; if the second difference is greater than the difference threshold, determine a third difference based on the first shape feature comparison; if the third difference is greater than the difference threshold, add the first medical image to the first standard image set.
[0021] The intelligent processing system for medical images proposed in this application is intended to receive the medical image to be identified at the user end through the image transmission module, and encrypt the medical image and transmit it to the cloud processing platform. The significant feature extraction module is used to extract the significant features of the medical image on the cloud processing platform, and the distribution discrete analysis is performed on multiple significant feature areas. If the distribution discreteness is less than the predetermined discrete scalar, a medical significant feature map is generated according to the multiple significant feature areas. The medical significant feature map is transferred to the abnormality detection dual channel through the abnormality detection module, and the first abnormality detection result and the second abnormality detection result are output. The comprehensive detection result is obtained after the intersection operation, and the comprehensive detection result is encrypted and returned to the user end through the auxiliary diagnosis module for auxiliary diagnosis, thereby achieving the technical effect of improving the accuracy and reliability of image processing, and thus improving the efficiency and accuracy of diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the accompanying drawings of the embodiment of the present invention will be briefly introduced below. A flow chart is used in the present application to illustrate the operations performed by the system according to the embodiment of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more operations can be removed from these processes.
[0023] Figure 1 A schematic diagram of the structure of an intelligent processing system for medical images provided in an embodiment of the present application.
[0024] Figure 2 A schematic diagram of the process of extracting significant features from medical images in an intelligent processing system for medical images provided in an embodiment of the present application.
[0025] Explanation of the reference numerals: image transmission module 10 , significant feature extraction module 20 , anomaly detection module 30 , auxiliary diagnosis module 40 . DETAILED DESCRIPTION
[0026] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.
[0027] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.
[0028] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.
[0029] The embodiment of the present application provides an intelligent processing system for medical images, including a user end and a cloud processing platform, for efficiently and accurately processing and analyzing medical images through intelligent means, and assisting doctors in making diagnoses, such as Figure 1 As shown, the system comprises:
[0030] The image transmission module 10 is used to receive the medical image to be identified at the user end, and encrypt and transmit the medical image to the cloud processing platform. Specifically, the user uploads the medical image (such as CT, MRI, etc.) to the user end (such as a computer, mobile phone or tablet, etc.) interface. After the system receives the image, it automatically encrypts it (such as using encryption algorithms such as AES and RSA), and then transmits it to the cloud processing platform through a secure network channel (such as HTTPS). The cloud processing platform is a server cluster located on the Internet, which is used to receive, store and process the medical image data uploaded by the user.
[0031] The significant feature extraction module 20 is used to extract significant features from the medical image on the cloud processing platform, and to perform distribution discrete analysis on multiple significant feature areas. If the distribution discreteness is less than a predetermined discrete scalar, a medical significant feature map is generated based on the multiple significant feature areas. Specifically, the cloud processing platform receives and decrypts the medical image uploaded by the user. The image processing algorithm is used to extract significant features from the medical image, and these features include brightness, color, texture, shape, etc. in the image. The extracted significant features are subjected to distribution discrete analysis to evaluate the distribution of these features in the image and determine whether the distribution of the features is concentrated or dispersed. If the distribution discreteness (the degree of difference in the spatial distribution of the significant feature area in the image, used to measure the degree of dispersion of the feature area) is less than a predetermined discrete scalar (a preset threshold value, used to determine whether the distribution of the significant feature area is reasonable), that is, the feature distribution is relatively concentrated, then a medical significant feature map is generated based on these significant features, which highlights the key areas in the medical image and helps in abnormality detection or diagnosis.
[0032] like Figure 2 As shown, in a possible implementation, performing significant feature extraction on the medical image includes: inputting the medical image into a Gaussian pyramid for multi-scale Gaussian downsampling to obtain a multi-scale image set, wherein the Gaussian pyramid includes a brightness extraction unit, a color extraction unit, and a direction extraction unit; performing center-surround difference calculation on the multi-scale image set to generate a brightness feature map, a color feature map, and a direction feature map; superimposing the brightness feature map, the color feature map, and the direction feature map to generate an initial significant image; extracting the initial significant image according to an initial significant feature threshold to obtain multiple significant feature regions that meet the initial significant feature threshold, wherein the initial significant feature threshold is determined based on the medical image type and the abnormal site.
[0033] Specifically, the medical image to be processed is input into the Gaussian pyramid model. The Gaussian pyramid is an image processing technology used to represent the image at multiple scales. By applying Gaussian filters of different scales, a series of images from the original image to gradually blurred images of different levels can be obtained, namely, a multi-scale image set, which contains the information of the image at different scales. In the Gaussian pyramid, a brightness extraction unit, a color extraction unit and a direction extraction unit are respectively set, and these units are used to extract the brightness, color and direction information of the image from the multi-scale image set. For each image in the multi-scale image set, the center-surrounding difference calculation is performed to highlight the significant features in the image. By comparing the difference between each pixel in the image and its surrounding pixels, a brightness feature map, a color feature map and a direction feature map are generated, and these feature maps reflect the significant features of the image in brightness, color and direction respectively. The brightness feature map, the color feature map and the direction feature map are superimposed to generate an initial significant image. The superimposition processing can be different methods such as weighted summation and maximum value selection to comprehensively reflect the significant features in the image. Through the superimposition processing, an initial significant image containing all the significant features of the image is obtained. The initial significant feature threshold is determined according to the type of medical image and the abnormal part. This threshold is used to screen out the significant feature areas that meet the conditions. The significant threshold can be fine-tuned according to different image features to ensure that it can adapt to different types of medical images (such as CT, MRI, etc.) and images of different parts. The initial significant image is extracted according to the initial significant feature threshold to obtain multiple significant feature areas that meet the threshold conditions. These areas are the key information in the image and are used for abnormality detection and auxiliary diagnosis. This implementation method effectively extracts significant features in medical images through steps such as Gaussian pyramid multi-scale Gaussian downsampling, center-surrounding difference calculation, feature map overlay processing and significant feature area extraction, providing important information support for abnormality detection and auxiliary diagnosis.
[0034] In a possible implementation, if the distribution dispersion is greater than or equal to the predetermined discrete scalar, the initial significant feature threshold is incremented according to a predetermined step size to determine a first significant feature threshold; and the initial significant image is extracted according to the first significant feature threshold until a plurality of significant feature regions smaller than the predetermined discrete scalar are obtained.
[0035] Specifically, if the distribution dispersion is greater than or equal to a predetermined discrete scalar, it indicates that the distribution of the significant feature area is too dispersed and there is a possibility of recognition error. At this time, the initial significant feature threshold is incremented according to a predetermined step size (a fixed value used to control the amplitude of the threshold adjustment) to determine a higher first significant feature threshold. The initial significant image is re-extracted using the new first significant feature threshold. The extraction process is the same as the previous step, but a higher threshold is used, so fewer but more significant feature areas will be screened out. Repeat the above steps, that is, calculate the distribution dispersion, determine whether the dispersion meets the conditions, adjust the significant feature threshold, and re-extract the significant feature area, until the distribution dispersion of the obtained significant feature area is less than the predetermined discrete scalar. This implementation method ensures that the extracted feature area is more accurate and concentrated by adjusting the significant feature threshold. When the distribution dispersion is high, increasing the threshold can reduce the feature area that is misidentified and improve the accuracy of feature extraction.
[0036] In a possible implementation, generating a medical significant feature map includes: reading multiple part coordinates of the multiple significant feature regions; marking the multiple significant feature regions according to the multiple part coordinates, and fusing and generating the medical significant feature map.
[0037] Specifically, the coordinates of multiple significant feature areas are stored in a CSV file, a database record or a JSON object, and these coordinate information are read. After reading the coordinates, preprocessing is performed, such as unifying the coordinate format and verifying the coordinate values. The corresponding significant feature areas are marked on the medical image using the read coordinates. The marking method can be highlighting, border marking or other forms of visual identification. Through image overlay, transparency adjustment, etc., all marked significant feature areas are fused in one image to generate a comprehensive medical significant feature map. The medical significant feature map is a medical image that comprehensively displays multiple significant feature areas, which is used to help identify and analyze key information in the image. This implementation method improves the efficiency and accuracy of medical image analysis by automatically marking and fusing significant feature areas.
[0038] The abnormality detection module 30 is used to transfer the medically significant feature map to the abnormality detection dual channel, output the first abnormality detection result and the second abnormality detection result, and obtain the comprehensive detection result after performing an intersection operation. Specifically, the medically significant feature map is input into the abnormality detection module 30, and the abnormality detection module 30 includes two detection channels (such as a detection algorithm based on machine learning, a detection algorithm based on rules, etc.) for analyzing the medically significant feature map. The medically significant feature map is subjected to abnormality detection through these two detection channels to identify abnormal areas or lesions that may exist in the image, and these results are subjected to an intersection operation to obtain the final comprehensive detection result, including the location, size, shape and other information of the abnormal area, which is used to assist doctors in making a diagnosis.
[0039] In a possible implementation, the medically significant feature map is transferred to a dual channel for abnormality detection, a first abnormality detection result and a second abnormality detection result are output, and an intersection operation is performed to obtain a comprehensive detection result, including: constructing a dual channel for abnormality detection, wherein a first abnormality detection branch is constructed based on a convolutional neural network, and a second abnormality detection branch is constructed based on a similarity comparison library; the medically significant feature map is respectively input into the first abnormality detection branch and the second abnormality detection branch, a first abnormality detection category and a second abnormality detection category are output, and an intersection operation is performed on the first abnormality detection category and the second abnormality detection category to output the comprehensive detection result.
[0040] Specifically, a dual channel for anomaly detection is constructed, in which the first anomaly detection branch is constructed based on a convolutional neural network (CNN). CNN automatically extracts features from images through multi-layer convolution and pooling operations, and learns the mapping relationship between these features and anomalies; the second anomaly detection branch is constructed based on a similarity comparison library. This branch uses a pre-established database containing normal and abnormal medical image features to determine whether there is an anomaly by calculating the similarity between the image to be detected and the image in the library.
[0041] The medical significant feature map processed by the significant feature extraction module 20 is input into the first anomaly detection branch and the second anomaly detection branch respectively. The first anomaly detection branch outputs one or more possible anomaly detection categories through multi-layer processing of CNN, which are based on the patterns learned in the training data; the second anomaly detection branch outputs one or more categories that are most similar to the abnormal images in the library as anomaly detection categories through similarity comparison. The anomaly detection categories output by the two branches are subjected to intersection operation, that is, the categories that are considered abnormal by both branches are found as the comprehensive detection results. This implementation adopts a dual-channel design for anomaly detection. By performing an intersection operation on the results of the two branches, the advantages of both branches can be combined to reduce false positives and false negatives, thereby improving the accuracy and reliability of medical image anomaly detection.
[0042] In a possible implementation, a first anomaly detection branch is constructed based on a convolutional neural network, including: reading the image type and abnormal part of the medical image, using the image type and abnormal part as retrieval constraints, querying the medical image processing log, and obtaining a first sample significant feature atlas and a first sample anomaly detection category set; using the first sample significant feature map as input and the first sample anomaly detection category as output, and using the first sample significant feature atlas and the first sample anomaly detection category set to supervise the convolutional neural network until a first anomaly detection branch that meets convergence is obtained.
[0043] Specifically, the image type (such as X-ray, CT scan, MRI, etc.) and abnormal part (such as lung, heart, brain, etc.) information are read from the medical image received from the user end. The read image type and abnormal part are used as retrieval constraints to query the medical image processing log database, which records the medical images processed in the past and their related information, including significant feature maps, abnormal detection categories, etc. According to the retrieval constraints, the first sample significant feature map set and the first sample abnormal detection category set that match the current medical image type and abnormal part are screened out from the log. The screened first sample significant feature map set is used as the input data of the convolutional neural network, and the corresponding first sample abnormal detection category set is used as the expected output data (i.e., label) of the convolutional neural network. The convolutional neural network is supervised and trained using the prepared training data. During the training process, the network will continuously adjust its weights and bias parameters to minimize the error between the input features and the output categories, so that the convolutional neural network can learn the mapping relationship between the significant feature map and the abnormal detection category. During the training process, the performance of the network on the validation set is regularly evaluated to determine whether the network has converged, that is, the performance of the network on the validation set is no longer significantly improved, and the error is no longer significantly reduced. When the network meets the convergence conditions, the training is stopped and the current network model is saved as the first anomaly detection branch. This implementation method obtains historical data that matches the current medical image type and abnormal part as training samples by querying the medical image processing log. These samples have higher relevance and representativeness, which helps the convolutional neural network learn more accurate mapping relationships, thereby improving the accuracy of detection.
[0044] In a possible implementation, a similarity comparison library is constructed, including: taking the image type and abnormal part as constraints, performing positive sample retrieval based on big data, obtaining a second sample abnormality detection category set, and performing deduplication processing on the second sample abnormality detection category set to obtain a standard detection category set; randomly selecting a first standard detection category from the standard detection category set, and retrieving and obtaining a first sample medical image set according to the image type, abnormal part and the first standard detection category; performing high-frequency feature extraction on the first sample medical image set to obtain a first standard image set; configuring a first similarity discrimination threshold, constructing a first similarity comparison unit according to the first standard detection category, the first standard image set and the first similarity discrimination threshold, and adding it to the similarity comparison library, wherein if the comparison similarity satisfies the first similarity discrimination threshold, the first standard detection category is output.
[0045] Specifically, a big data platform or database is used to retrieve medical image samples that are marked as abnormal and match these conditions according to the image type and abnormal part, forming a second sample abnormality detection category set, that is, the "positive sample" here refers to an image that has been confirmed to have a certain abnormality. Since there are duplications in the retrieved second sample abnormality detection category set (that is, multiple images belong to the same abnormal category), it is necessary to remove these categories to ensure that each category only appears once to form a standard detection category set. A category is randomly selected from the standard detection category set as the basis for the current similarity comparison unit to be constructed. According to the image type, abnormal part and the selected first standard detection category, a search is performed again to obtain a medical image set belonging to this category, that is, the first sample medical image set. Feature extraction is performed on each image in the first sample medical image set to extract high-frequency features (such as specific shapes, textures, color distribution, etc.) that can represent the abnormal category to form a first standard image set. In order to judge the similarity between the new image and the standard image set, a similarity threshold is set. This threshold determines the degree of similarity between the new image and the standard image set when they can be considered to belong to the same category. The first standard detection category, the first standard image set, and the first similarity discrimination threshold are combined into a unit, which is an element in the similarity comparison library. When a new image needs to be detected for anomalies, it is compared with each unit in the similarity comparison library to calculate the similarity. If the similarity of a unit meets the set similarity discrimination threshold, the first standard detection category corresponding to the unit is output as the detection result. This implementation method uses big data and high-frequency feature extraction technology to ensure that the standard image set in the similarity comparison library can accurately represent various anomaly categories, thereby improving the accuracy of detection.
[0046] In a possible implementation, high-frequency features are extracted from the first sample medical image set to obtain a first standard image set, including: randomly selecting any image in the first sample medical image set as a first medical image, wherein the first medical image includes a first structural feature, a first texture feature, and a first shape feature; performing a similarity comparison between the first structural feature and other image structural features in the first sample medical image set, and setting the amount of data whose statistical similarity is less than a predetermined similarity scalar as a first difference; if the first difference is greater than a difference threshold, determining a second difference based on the first texture feature comparison, and if the second difference is greater than the difference threshold, determining a third difference based on the first shape feature comparison; if the third difference is greater than the difference threshold, adding the first medical image to the first standard image set.
[0047] Specifically, an image is randomly selected from the first sample medical image set and is set as the first medical image. This image represents the candidate image currently under consideration. Feature extraction is performed on the first medical image to obtain its first structural feature (features such as the layout and spatial relationship of objects in the image), first texture feature (pattern or law of grayscale or color change in a local area of the image) and first shape feature (shape or contour feature of the object in the image). The first structural feature of the first medical image is similarly compared with the structural features of all other images in the first sample medical image set, and the amount of data with a similarity less than a predetermined similarity scalar (a threshold used to determine whether the similarity of two images in a certain feature is low enough) is counted, that is, the degree of difference in structural features between the first medical image and other images in the set is calculated and set as the first difference. Check whether the first difference is greater than a difference threshold. The difference threshold is a preset threshold used to determine whether the difference is large enough to determine whether to continue with subsequent feature comparison. If the first difference is not greater than the difference threshold, it means that the first medical image is relatively similar to other images in the set in terms of structural features, so it is directly discarded and no subsequent comparison is performed. If the first difference is greater than the difference threshold, continue to compare the texture features. Perform a similarity comparison on the first texture feature of the first medical image and other images in the first sample medical image set, and calculate the second difference. Check whether the second difference is greater than the difference threshold. If the second difference is not greater than the difference threshold, it also means that the first medical image is similar to other images in the set in terms of texture features, so it is directly discarded. If the second difference is greater than the difference threshold, continue to compare the shape features. Perform a similarity comparison on the first shape feature of the first medical image and other images in the first sample medical image set, calculate the third difference, and check whether the third difference is greater than the difference threshold. If the third difference is greater than the difference threshold, it means that the first medical image is significantly different from other images in the set in terms of structure, texture and shape features, so it is added to the first standard image set. This implementation method ensures that the selected images are significantly different from other images in the set in multiple dimensions by gradually comparing structural features, texture features, and shape features, thereby increasing the diversity and representativeness of the first standard image set. In addition, it reduces the number of images while ensuring comparison accuracy, improves comparison efficiency, and provides a reliable standard image set for similarity comparison and anomaly detection.
[0048] The auxiliary diagnosis module 40 is used to encrypt the comprehensive detection result and return it to the user end for auxiliary diagnosis. Specifically, the comprehensive detection result output by the abnormality detection module 30 is encrypted to ensure the security of the data. The encrypted comprehensive detection result is returned to the user end through a secure network channel. The doctor views the encrypted comprehensive detection result at the user end, and performs a comprehensive analysis in combination with other diagnostic information (such as the patient's medical history, symptoms, etc.) to assist in diagnosis, improve diagnostic efficiency, and ensure the accuracy and security of the final judgment. The embodiment of the present application encrypts and transmits the medical image, extracts significant features of the medical image, and performs distribution discrete analysis on multiple significant feature areas. If the distribution discreteness is less than a predetermined discrete scalar, a medical significant feature map is generated according to multiple significant feature areas, and the medical significant feature map is transferred to the abnormality detection dual channel, and the first abnormality detection result and the second abnormality detection result are output, and the comprehensive detection result is obtained after the intersection operation, and the comprehensive detection result is encrypted and returned to the user end, and the technical means such as auxiliary diagnosis are performed to achieve the technical effect of improving the accuracy and reliability of image processing, thereby improving the efficiency and accuracy of diagnosis.
[0049] The above specific implementation manner does not constitute a limitation to the protection scope of the present application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application. In some cases, the actions or steps recorded in the present application can be performed in an order different from that in the embodiment and can still achieve the desired results. In addition, the process depicted in the accompanying drawings does not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
Claims
1. Intelligent processing system for medical images, characterized by: The system includes a user-side and cloud processing platform, including: An image transmission module is used to receive the medical image to be identified at the user end, and encrypt and transmit the medical image to the cloud processing platform; A significant feature extraction module is used to extract significant features from the medical image on a cloud processing platform, and to perform distribution discrete analysis on a plurality of significant feature regions, and to generate a medical significant feature map based on the plurality of significant feature regions if the distribution discreteness is less than a predetermined discrete scalar; An anomaly detection module, used for transferring the medically significant feature map to the anomaly detection dual channel, outputting a first anomaly detection result and a second anomaly detection result, and obtaining a comprehensive detection result after performing an intersection operation; The auxiliary diagnosis module is used to encrypt the comprehensive detection result and return it to the user end for auxiliary diagnosis.
2. The intelligent processing system for medical images according to claim 1, characterized in that: Extracting significant features from the medical image includes: Inputting the medical image into a Gaussian pyramid for multi-scale Gaussian downsampling to obtain a multi-scale image set, wherein the Gaussian pyramid includes a brightness extraction unit, a color extraction unit, and a direction extraction unit; Performing center-surround difference calculations on the multi-scale image set to generate a brightness feature map, a color feature map, and a direction feature map; Performing superposition processing on the brightness feature map, the color feature map and the direction feature map to generate an initial salient image; The initial salient image is extracted according to an initial salient feature threshold to obtain a plurality of salient feature regions that meet the initial salient feature threshold, wherein the initial salient feature threshold is determined based on the medical image type and the abnormal part.
3. The intelligent processing system for medical images according to claim 2, characterized in that: If the distribution dispersion is greater than or equal to the predetermined discrete scalar, the initial significant feature threshold is incremented according to a predetermined step length to determine a first significant feature threshold; The initial salient image is extracted according to the first salient feature threshold until a plurality of salient feature regions smaller than the predetermined discrete scalar are obtained.
4. The intelligent processing system for medical images according to claim 1, characterized in that: Generate medically significant feature maps, including: Reading multiple coordinates of the multiple significant feature areas; The plurality of significant feature regions are marked according to the plurality of part coordinates, and are fused to generate the medical significant feature map.
5. The intelligent processing system for medical images according to claim 4, characterized in that: The medically significant feature map is transferred to the abnormality detection dual channel, the first abnormality detection result and the second abnormality detection result are output, and the comprehensive detection result is obtained after the intersection operation, including: Constructing a dual channel for anomaly detection, wherein a first anomaly detection branch is constructed based on a convolutional neural network, and a second anomaly detection branch is constructed based on a similarity comparison library; The medically significant feature map is input into the first abnormality detection branch and the second abnormality detection branch respectively, a first abnormality detection category and a second abnormality detection category are output, and an intersection operation is performed on the first abnormality detection category and the second abnormality detection category to output the comprehensive detection result.
6. The intelligent processing system for medical images according to claim 5, characterized in that: The first anomaly detection branch is constructed based on the convolutional neural network, including: Reading the image type and abnormal part of the medical image, using the image type and abnormal part as retrieval constraints, querying the medical image processing log, and obtaining a first sample significant feature atlas and a first sample abnormality detection category set; Taking the first sample salient feature map as input and the first sample anomaly detection category as output, the convolutional neural network is supervisedly trained using the first sample salient feature map set and the first sample anomaly detection category set until a first anomaly detection branch that meets convergence is obtained.
7. The intelligent processing system for medical images according to claim 6, characterized in that: Build a similarity comparison library, including: Taking the image type and abnormal part as constraints, performing positive sample retrieval based on big data to obtain a second sample abnormality detection category set, and performing deduplication processing on the second sample abnormality detection category set to obtain a standard detection category set; Randomly selecting a first standard detection category from the standard detection category set, and retrieving and acquiring a first sample medical image set according to the image type, abnormal part, and the first standard detection category; Extracting high-frequency features from the first sample medical image set to obtain a first standard image set; A first similarity discrimination threshold is configured, and a first similarity comparison unit is constructed according to the first standard detection category, the first standard image set and the first similarity discrimination threshold, and added to the similarity comparison library, wherein if the comparison similarity satisfies the first similarity discrimination threshold, the first standard detection category is output.
8. The intelligent processing system for medical images according to claim 7, characterized in that: Extracting high-frequency features from the first sample medical image set to obtain a first standard image set includes: Randomly selecting any image from the first sample medical image set as a first medical image, wherein the first medical image includes a first structural feature, a first texture feature, and a first shape feature; Performing a similarity comparison between the first structural feature and other image structural features in the first sample medical image set, and setting the amount of data whose statistical similarity is less than a predetermined similarity scalar as a first difference degree; If the first difference is greater than a difference threshold, determining a second difference according to the first texture feature comparison; if the second difference is greater than the difference threshold, determining a third difference according to the first shape feature comparison; If the third difference is greater than the difference threshold, the first medical image is added to the first standard image set.