Medical image labeling system and method
By performing multi-feature extraction and Euclidean distance method feature matching on marked medical images, automatic labeling of medical images is achieved, solving the problems of inefficiency and lack of flexibility on existing platforms, and improving the labeling efficiency and user experience.
Patent Information
- Application Number
- CN202411466388.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-18
- Publication Date
- 2025-02-11
AI Technical Summary
At this stage, medical image data labeling platforms are difficult to meet the labeling needs of medical images of various forms, lack a complete feature matching process for labeling and similarity measurement, relying on a large number of professional doctors to assist in judgment, which is inefficient and lacks flexibility.
By extracting labeled medical images multi-features, using the Euclidean distance method to match the feature similarity measurement of image features, automatic labeling is realized, and the keywords of the image with the smallest Euclidean distance are assigned to the image to be marked, displaying and labeling The most similar images.
There is no need to rely on a large number of professional doctors to make auxiliary judgments, which improves the efficiency and flexibility of medical image labeling and brings convenience to users' use and retrieval.
Smart Images
Figure CN119418878A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of medical image processing, and specifically relates to a medical image annotation system and method. Background Art
[0002] Medical imaging refers to the technology and process of obtaining images of internal tissues of the human body or a part of the human body in a non-invasive manner for the purpose of medical treatment or medical research.
[0003] As artificial intelligence technology becomes more mature and its application scope becomes wider, the medical AI field is relatively slow compared to the rapid development of other fields. The main reasons include the small data scale, poor labeling quality, and the need for cross-domain and cross-disciplinary professionals. Collecting, organizing, segmenting and labeling high-quality medical imaging data and converting them into structured data that can be recognized and further processed by computers is the basis of artificial intelligence algorithms.
[0004] At present, the medical image data annotation platform is unable to meet the annotation needs of various forms of medical images. The annotation platform is only used as a tool and has not simulated a complete feature matching process of annotation and similarity measurement. It still needs to rely on a large number of professional doctors for auxiliary judgment, which makes it inefficient and lacks flexibility in the annotation of medical images. Summary of the invention
[0005] In order to overcome the above-mentioned defects, the present invention provides a medical image annotation system and method, which solves the problem that the current medical image data annotation platform is difficult to meet the annotation needs of various forms of medical images. The annotation platform is only used as a tool and does not simulate a complete feature matching process of annotation and similarity measurement. It still needs to rely on a large number of professional doctors for auxiliary judgment, which makes it inefficient and lacks flexibility in the annotation of medical images.
[0006] The purpose of the present invention is to extract multiple features from annotated medical images, use the Euclidean distance method to perform feature matching of image feature similarity measurement and perform automatic annotation, assign the keyword of the image with the smallest Euclidean distance to the image to be annotated, and display multiple images that are most similar to the image to be annotated, so that the visual features of the image can be converted into the annotation word information of the image without relying on a large number of professional doctors for auxiliary judgment.
[0007] To achieve the above-mentioned object, the present invention provides the following technical solutions: a medical image annotation system, comprising an image preprocessing module, a task transmission module, an annotation module, an image feature extraction unit, an automatic annotation and similar image output module, a data storage management module and an image feature library, wherein the output end of the image preprocessing module is connected to the input end of the task transmission module, and the output end of the task transmission module is connected to the input end of the annotation module;
[0008] The output end of the labeling module is connected to the input end of the automatic labeling and similar image output module through the image feature extraction unit, the output end of the automatic labeling and similar image output module is connected to the input end of the data storage management module, the input end of the data storage management module is connected to the output end of the image feature library, and the image feature library is bidirectionally connected to the image feature extraction unit.
[0009] As a further solution of the present invention: the image preprocessing module is used to parse and preprocess the uploaded medical images, and the image preprocessing module includes image classification, image enhancement, image registration and image denoising processing.
[0010] As a further solution of the present invention: the task transmission module is used to distribute the annotation work of the pre-processed medical images to designated personnel according to a preset task allocation strategy.
[0011] As a further solution of the present invention: the labeling module is used to provide a labeling process to designated personnel, and complete the labeling work of medical images according to the labeling process.
[0012] As a further solution of the present invention: the image feature extraction unit includes a color feature extraction module and a texture feature extraction module;
[0013] Color feature extraction module: convert the color space of the labeled medical images to the HSI color space, and extract color features of the labeled medical images;
[0014] The application steps of the color feature extraction module are as follows:
[0015] Through equations (1), (2) and (3), the RGB color space of the labeled medical image is converted to the HSI color space;
[0016]
[0017] S=MAX(R,G,B)-,MIN(R,G,B)(2)
[0018]
[0019] The red, green, and blue component values R, G, and B of each pixel in the image are converted into hue, saturation, and intensity component values H, S, and I (h[0,360], s[0,1], and υ[0,1]), and then non-uniformly spaced quantization is performed according to the color characteristics in the HSI color space, as follows:
[0020]
[0021] After quantizing the HSI space, the value range of H is [0, 1, ..., 7], the value range of S is [0, 1, 2], and the value range of I is [0, 1, 2]. The HSI is synthesized into a one-dimensional feature vector by equation (7), and the value range of B is [0, 7, 1], which is a one-dimensional histogram with 72 bins. Then, a histogram is constructed for the feature vector B as the color feature of the image:
[0022] B=9H+3S+I(7);
[0023] Texture feature extraction module: uses wavelet decomposition to extract texture features of annotated medical images;
[0024] The application steps of the texture feature extraction module are as follows:
[0025] The annotated medical images are decomposed by Daubechies4 wavelet, and then the mean and variance of the four-band wavelet coefficients of the first layer decomposition are extracted as the texture features of the image. The four-band distribution of a single image is shown in I LL ,I LH ,I HL ,I HH express;
[0026] After one layer of transformation, the wavelet coefficient of point (i, j) is C ij , then the calculation formulas for the mean and variance of any frequency band are:
[0027]
[0028] The texture feature vector is expressed as: T = [μ LL ,D LL ,μ HL ,D HL ,μ LH ,D LH ,μ HH ,D HH ].
[0029] As a further solution of the present invention: the image feature library is used to store standard image features of labeled images to form a training set.
[0030] As a further solution of the present invention: the automatic annotation and similar image output module uses the Euclidean distance method to perform feature matching of image feature similarity measurement and perform automatic annotation;
[0031] The specific application steps of the automatic annotation and similar image output module are as follows:
[0032] Since the smaller the Euclidean distance, the more the two groups of images match, and the larger the Euclidean distance, the less the two groups of images match, the annotation keyword of the image with the smallest Euclidean distance in the annotated image set is selected as the semantic keyword of the image to be annotated. The two n-dimensional vectors a(x 11 ,x 12 ,...,x 1n ) and b(x 21 ,x 22 ,...,x 2n ) is:
[0033]
[0034] Calculate the Euclidean distance between the features of the image to be labeled and all the image features in the training set, arrange all the distances in ascending order, assign the keyword of the image with the smallest Euclidean distance to the image to be labeled, and display multiple images that are closest to the image to be labeled.
[0035] As a further solution of the present invention: the data storage management module is used to store the obtained data after completing the color feature and texture feature extraction, automatic annotation of keywords and feature vector calculation for image feature matching.
[0036] A medical image annotation method, the method specifically comprising:
[0037] S1. Use the image preprocessing module to parse and preprocess the uploaded medical images, and the preprocessing methods include image classification, image enhancement, image registration and image denoising, according to the preset task allocation strategy;
[0038] S2. Distribute the labeling work of the pre-processed medical images to designated personnel, provide the designated personnel with the labeling process, and complete the labeling work of the medical images according to the labeling process;
[0039] S3, converting the color space of the labeled medical image to the HSI color space, and extracting color features of the labeled medical image, extracting texture features of the labeled medical image by wavelet decomposition, and storing standard image features of the labeled image in an image feature library to form a training set;
[0040] S4, using the automatic labeling and similar image output module to perform feature matching of image feature similarity measurement using the Euclidean distance method and perform automatic labeling;
[0041] S5. After the color feature and texture feature extraction, automatic annotation of keywords and feature vector calculation are completed for the image, the obtained data is stored for image feature matching.
[0042] Compared with the prior art, the beneficial effects of the present invention are as follows: by parsing and preprocessing the uploaded medical images, distributing the annotation work of the preprocessed medical images to designated personnel according to a preset task allocation strategy, providing the designated personnel with an annotation process, and completing the annotation work of the medical image tags according to the annotation process, and then extracting multiple features from the annotated medical images, using the Euclidean distance method to perform feature matching of image feature similarity measurement and automatically annotating, assigning the keyword of the image with the smallest Euclidean distance to the image to be annotated, and displaying multiple images that are closest to the image to be annotated, thereby being able to convert the visual features of the image into the annotation word information of the image, without relying on a large number of professional doctors for auxiliary judgment, bringing great convenience to user use and retrieval, and improving annotation efficiency and flexibility of use. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 It is a principle block diagram of the system of the present invention;
[0044] Figure 2 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION
[0045] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0046] Embodiment 1:
[0047] A medical image annotation system comprises an image preprocessing module, a task transmission module, an annotation module, an image feature extraction unit, an automatic annotation and similar image output module, a data storage management module and an image feature library, wherein the output end of the image preprocessing module is connected to the input end of the task transmission module, and the output end of the task transmission module is connected to the input end of the annotation module;
[0048] The output end of the annotation module is connected to the input end of the automatic annotation and similar image output module through the image feature extraction unit, the output end of the automatic annotation and similar image output module is connected to the input end of the data storage management module, the input end of the data storage management module is connected to the output end of the image feature library, and the image feature library is bidirectionally connected to the image feature extraction unit.
[0049] Embodiment 2:
[0050] The image preprocessing module is used to parse and preprocess the uploaded medical images. The image preprocessing module includes image classification, image enhancement, image registration and image denoising.
[0051] The task transmission module is used to distribute the annotation tasks of pre-processed medical images to designated personnel according to the preset task allocation strategy.
[0052] The labeling module is used to provide the labeling process to designated personnel and complete the labeling work of medical images according to the labeling process.
[0053] The image feature extraction unit includes a color feature extraction module and a texture feature extraction module;
[0054] Color feature extraction module: convert the color space of the labeled medical images to the HSI color space, and extract color features of the labeled medical images;
[0055] The application steps of the color feature extraction module are as follows:
[0056] Through equations (1), (2) and (3), the RGB color space of the labeled medical image is converted to the HSI color space;
[0057]
[0058] S=MAX(R,G,B)-,MIN(R,G,B)(2)
[0059]
[0060] The red, green, and blue component values R, G, and B of each pixel in the image are converted into hue, saturation, and intensity component values H, S, and I (h[0,360], s[0,1], and υ[0,1]), and then non-uniformly spaced quantization is performed according to the color characteristics in the HSI color space, as follows:
[0061]
[0062] After quantizing the HSI space, the value range of H is [0, 1, ..., 7], the value range of S is [0, 1, 2], and the value range of I is [0, 1, 2]. The HSI is synthesized into a one-dimensional feature vector by equation (7), and the value range of B is [0, 7, 1], which is a one-dimensional histogram with 72 bins. Then, a histogram is constructed for the feature vector B as the color feature of the image:
[0063] B=9H+3S+I(7);
[0064] Texture feature extraction module: uses wavelet decomposition to extract texture features of annotated medical images;
[0065] The application steps of the texture feature extraction module are as follows:
[0066] The annotated medical images are decomposed by Daubechies4 wavelet, and then the mean and variance of the four-band wavelet coefficients of the first layer decomposition are extracted as the texture features of the image. The four-band distribution of a single image is shown in I LL ,I LH ,I HL ,I HH express;
[0067] After one layer of transformation, the wavelet coefficient of point (i, j) is C ij , then the calculation formulas for the mean and variance of any frequency band are:
[0068]
[0069] The texture feature vector is expressed as: T = [μ LL ,D LL ,μ HL ,D HL ,μ LH ,D LH ,μ HH ,D HH ].
[0070] The image feature library is used to store the standard image features of labeled images to form a training set.
[0071] The automatic annotation and similar image output module uses the Euclidean distance method to measure the similarity of image features and perform feature matching and automatic annotation;
[0072] The specific application steps of the automatic annotation and similar image output module are as follows:
[0073] Since the smaller the Euclidean distance, the more the two groups of images match, and the larger the Euclidean distance, the less the two groups of images match, the annotation keyword of the image with the smallest Euclidean distance in the annotated image set is selected as the semantic keyword of the image to be annotated. The two n-dimensional vectors a(x 11 ,x 12 ,...,x 1n ) and b(x 21 ,x 22 ,...,x 2n ) is:
[0074]
[0075] Calculate the Euclidean distance between the features of the image to be labeled and all the image features in the training set, arrange all the distances in ascending order, assign the keyword of the image with the smallest Euclidean distance to the image to be labeled, and display multiple images that are closest to the image to be labeled.
[0076] The data storage management module is used to complete the color feature and texture feature extraction, automatic annotation of keywords and feature vector calculation of the image, and then store the obtained data for image feature matching.
[0077] Embodiment 3:
[0078] A medical image annotation method, the method specifically comprising:
[0079] S1. Use the image preprocessing module to parse and preprocess the uploaded medical images, and the preprocessing methods include image classification, image enhancement, image registration and image denoising, according to the preset task allocation strategy;
[0080] S2. Distribute the labeling work of the pre-processed medical images to designated personnel, provide the designated personnel with the labeling process, and complete the labeling work of the medical images according to the labeling process;
[0081] S3, converting the color space of the labeled medical image to the HSI color space, and extracting color features of the labeled medical image, extracting texture features of the labeled medical image by wavelet decomposition, and storing standard image features of the labeled image in an image feature library to form a training set;
[0082] S4, using the automatic labeling and similar image output module to perform feature matching of image feature similarity measurement using the Euclidean distance method and perform automatic labeling;
[0083] S5. After the color feature and texture feature extraction, automatic annotation of keywords and feature vector calculation are completed for the image, the obtained data is stored for image feature matching.
[0084] From the above we know:
[0085] By parsing and preprocessing the uploaded medical images, the labeling work of the preprocessed medical images is distributed to designated personnel according to the preset task allocation strategy, and the labeling process is provided to the designated personnel. The medical image labeling work is completed according to the labeling process. Secondly, multi-feature extraction is performed on the labeled medical images, and the Euclidean distance method is used to perform feature matching of image feature similarity measurement and automatic labeling. The keyword of the image with the smallest Euclidean distance is assigned to the image to be labeled, and multiple images closest to the image to be labeled are displayed. Therefore, the visual features of the image can be converted into the image annotation information without relying on a large number of professional doctors for auxiliary judgment, which brings great convenience to user use and retrieval, and improves labeling efficiency and flexibility of use.
[0086] The preferred implementation modes of the present application are described in detail above, but the present application is not limited to the above implementation modes, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present application.
Claims
1. A medical image annotation system, comprising an image preprocessing module, a task transmission module, an annotation module, an image feature extraction unit, an automatic annotation and similar image output module, a data storage management module and an image feature library, characterized in that: The output end of the image preprocessing module is connected to the input end of the task transmission module, and the output end of the task transmission module is connected to the input end of the labeling module; The output end of the labeling module is connected to the input end of the automatic labeling and similar image output module through the image feature extraction unit, the output end of the automatic labeling and similar image output module is connected to the input end of the data storage management module, the input end of the data storage management module is connected to the output end of the image feature library, and the image feature library is bidirectionally connected to the image feature extraction unit.
2. A medical image annotation system according to claim 1, characterized in that: The image preprocessing module is used to parse and preprocess the uploaded medical images, and the image preprocessing module includes image classification, image enhancement, image registration and image denoising processing.
3. A medical image annotation system according to claim 1, characterized in that: The task transmission module is used to distribute the annotation work of the pre-processed medical images to designated personnel according to a preset task allocation strategy.
4. A medical image annotation system according to claim 1, characterized in that: The labeling module is used to provide a labeling process to designated personnel and complete the labeling work of medical images according to the labeling process.
5. A medical image annotation system according to claim 4, characterized in that: The image feature extraction unit includes a color feature extraction module and a texture feature extraction module; Color feature extraction module: convert the color space of the labeled medical images to the HSI color space, and extract color features of the labeled medical images; The application steps of the color feature extraction module are as follows: Through equations (1), (2) and (3), the RGB color space of the labeled medical image is converted to the HSI color space; S=MAX(R,G,B)-,MIN(R,G,B)(2) The red, green, and blue component values R, G, and B of each pixel in the image are converted into hue, saturation, and intensity component values H, S, and I (h[0,360], s[0,1], and υ[0,1]), and then non-uniformly spaced quantization is performed according to the color characteristics in the HSI color space, as follows: After quantizing the HSI space, the value range of H is [0, 1, ..., 7], the value range of S is [0, 1, 2], and the value range of I is [0, 1, 2]. The HSI is synthesized into a one-dimensional feature vector by equation (7), and the value range of B is [0, 7, 1], which is a one-dimensional histogram with 72 bins. Then, a histogram is constructed for the feature vector B as the color feature of the image: B=9H+3S+I(7); Texture feature extraction module: uses wavelet decomposition to extract texture features of annotated medical images; The application steps of the texture feature extraction module are as follows: The annotated medical images are decomposed by Daubechies4 wavelet, and then the mean and variance of the four-band wavelet coefficients of the first layer decomposition are extracted as the texture features of the image. The four-band distribution of a single image is shown in I LL ,I LH ,I HL ,I HH express; After one layer of transformation, the wavelet coefficient of point (i, j) is C ij , then the calculation formulas for the mean and variance of any frequency band are: The texture feature vector is expressed as: T = [μ LL ,D LL ,μ HL ,D HL ,μ LH ,D LH ,μ HH ,D HH ].
6. A medical image annotation system according to claim 5, characterized in that: The image feature library is used to store standard image features of labeled images to form a training set.
7. A medical image annotation system according to claim 6, characterized in that: The automatic annotation and similar image output module uses the Euclidean distance method to perform feature matching of image feature similarity measurement and perform automatic annotation; The specific application steps of the automatic annotation and similar image output module are as follows: Since the smaller the Euclidean distance, the more the two groups of images match, and the larger the Euclidean distance, the less the two groups of images match, the annotation keyword of the image with the smallest Euclidean distance in the annotated image set is selected as the semantic keyword of the image to be annotated. The two n-dimensional vectors a(x 11 ,x 12 ,...,x 1n ) and b(x 21 ,x 22 ,...,x 2n ) is: Calculate the Euclidean distance between the features of the image to be labeled and all the image features in the training set, arrange all the distances in ascending order, assign the keyword of the image with the smallest Euclidean distance to the image to be labeled, and display multiple images that are closest to the image to be labeled.
8. A medical image annotation system according to claim 7, characterized in that: The data storage management module is used to store the obtained data after completing the extraction of color features and texture features, automatic annotation of keywords and calculation of feature vectors for image feature matching.
9. A medical image annotation method, characterized in that: The method is used in the medical image annotation system according to any one of claims 1 to 8, and the method specifically comprises: S1. Use the image preprocessing module to parse and preprocess the uploaded medical images, and the preprocessing methods include image classification, image enhancement, image registration and image denoising, according to the preset task allocation strategy; S2. Distribute the labeling work of the pre-processed medical images to designated personnel, provide the designated personnel with the labeling process, and complete the labeling work of the medical images according to the labeling process; S3, converting the color space of the labeled medical image to the HSI color space, and extracting color features of the labeled medical image, extracting texture features of the labeled medical image by wavelet decomposition, and storing standard image features of the labeled image in an image feature library to form a training set; S4, using the automatic labeling and similar image output module to perform feature matching of image feature similarity measurement using the Euclidean distance method and perform automatic labeling; S5. After the color feature and texture feature extraction, automatic annotation of keywords and feature vector calculation are completed for the image, the obtained data is stored for image feature matching.