Medical manifold vector set classification method for thyroid nodules in ultrasonic image

Through manifold mapping network and hierarchical clustering, efficient unsupervised classification of thyroid nodules is achieved, solving the problems of high labor costs and low accuracy in the prior art, and providing fast and explainable diagnostic results.

CN118887461AActive Publication Date: 2024-11-01QUZHOU XUANYAN BIOTECHNOLOGY CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202410919855.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-10
Publication Date
2024-11-01
Estimated Expiration
2044-07-10

AI Technical Summary

Technical Problem

The existing automatic analysis method for ultrasonic image characteristics of thyroid nodules relies on supervised learning, which consumes a lot of labor costs and time, and the cystic component analysis method is difficult to adapt to nodules in complex structures, and its accuracy is limited.

Method used

Use manifold mapping network to obtain high-dimensional vector coordinates, reduce dimensionality through Fourier transform, combine pixels or superpixels from bottom to top, generate manifold vector category distribution map, and measure vector sets within the global range Distance to achieve unsupervised thyroid nodules classification.

Benefits of technology

Reduce the amount of data labeling, improve the accuracy and interpretability of classification, avoid the black box limitations of supervised neural networks, and provide higher diagnostic reliability and efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118887461A_ABST
    Figure CN118887461A_ABST
Patent Text Reader

Abstract

The invention provides a medical manifold vector set classification method for thyroid nodules in ultrasonic images. The method comprises the following steps: making a thyroid nodule ultrasonic image data set; designing and training a manifold mapping network, and obtaining high-dimensional vector coordinates of each pixel point of the thyroid nodule image; carrying out dimension reduction on the high-dimensional vector space based on Fourier transform; in a single nodule, based on the spatial position information and the high-dimensional vector similarity, carrying out pixel or superpixel combination from bottom to top, and generating a manifold vector category distribution diagram corresponding to the nodule; based on the category distribution diagrams of all nodules and manifold spaces thereof, measuring distances between manifolds of different vector spaces in a global range, and combining vector sets with close distances to finally obtain a classification result of each vector set; and manifold vector set category distribution statistics is carried out for benign and malignant nodules. The method has the advantages of being small in data labeling amount, rapid, high in interpretability, visual in result, convenient for medical analysis and diagnosis and the like.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a medical manifold vector set classification method for thyroid nodules in ultrasonic images. Background Art

[0002] Thyroid nodules are very common in clinical practice. In the past two decades, the incidence of thyroid nodules has continued to rise. Part of the reason is that the development of medical detection technology has improved the detection rate. The use of a variety of medical imaging methods can detect some asymptomatic thyroid nodules. Among them, ultrasound examination has become the most commonly used method for thyroid nodule examination due to its safety, radiation-free, convenient operation, and low cost. Ultrasound examination diagnoses thyroid nodules by measuring important signs such as the size, boundary, morphology, echo and blood flow information of thyroid nodules. In two-dimensional ultrasound, the more comprehensive the research and analysis of ultrasound signs, the higher the accuracy, but it is also more time-consuming and labor-intensive. In addition, ultrasound examination has been limited in diagnostic accuracy, standardization and uniformity due to its subjective dependence on the operating physician.

[0003] Most of the existing automatic analysis methods for thyroid nodule ultrasound image features use supervised learning to train multi-task classification networks. For example, the Chinese patents with publication numbers CN111341443A and CN112270667A require annotation of nodule echoes, components, boundaries, point-like strong echoes and other features, and the production of data sets inevitably consumes a lot of manpower and time costs. Other methods, such as the cystic component analysis method proposed in the Chinese patent with publication number CN112102311A, only use pixel value information, and the accuracy of cystic and solid area segmentation is limited, making it difficult to adapt to nodules with complex structures. Summary of the invention

[0004] In view of the above-mentioned shortcomings in the prior art, the present invention provides a medical manifold vector set classification method for thyroid nodules.

[0005] According to one aspect of the present invention, a medical manifold vector set classification method for thyroid nodules in ultrasound images is provided, comprising:

[0006] Acquire a dataset of thyroid nodule ultrasound images;

[0007] Using a manifold mapping network, high-dimensional vector coordinates of each pixel point of the thyroid nodule ultrasound image are obtained to obtain a high-dimensional vector space;

[0008] Performing dimensionality reduction on the high-dimensional vector space based on Fourier transform;

[0009] In a single nodule, based on vector similarity, the high-dimensional vector space after dimensionality reduction is merged from bottom to top using a hierarchical clustering method to form a vector set and generate a manifold vector category distribution map corresponding to the nodule;

[0010] Based on the manifold vector category distribution map of all nodules, the distances between vector sets in different vector spaces are measured globally, and vector sets with similar distances are merged to finally obtain the classification results of each vector set.

[0011] Based on the classification results of each vector set, the distribution statistics of benign and malignant nodules are performed.

[0012] Preferably, the structure of the manifold mapping network includes an encoder part and a decoder part, and can accept thyroid nodule ultrasound images of any size as input;

[0013] The encoder part includes multiple convolution modules, each of which performs one or more of convolution, batch normalization, activation function, and pooling operations; the strides of each convolution module are 1; and multiple convolution modules are hollow convolutions of different scales, which can obtain different receptive fields;

[0014] The decoder part is symmetrical with the encoder part, and includes a plurality of deconvolution modules, each of which performs one or more of convolution, batch normalization, activation function, and pooling operations, and the strides of each of the deconvolution modules are 1;

[0015] The output of the encoder part is a high-dimensional vector space, which contains multi-dimensional nodule feature maps under receptive fields of different sizes, and the size of the feature map is the same as the input size of the manifold mapping network.

[0016] Preferably, the dimensionality reduction of the high-dimensional vector space based on Fourier transform comprises:

[0017] Count the proportion of non-zero values ​​in each dimension of the high-dimensional vector space, and remove coordinate dimensions where the proportion of non-zero values ​​is lower than the set threshold;

[0018] Obtain the two-dimensional Fourier transform spectrum of the feature map corresponding to each vector dimension, and calculate the score of each dimension;

[0019] Calculate the Pearson correlation coefficient between the two feature maps based on the spectrum map;

[0020] For the nodule image, the optimal n-dimensional vector space of the nodule is obtained by combining all dimension scores and the Pearson correlation coefficient.

[0021] Preferably, the obtaining of a two-dimensional Fourier transform spectrum diagram of a feature diagram corresponding to a vector dimension and calculating scores of each dimension includes:

[0022] The specific scoring method for each dimension is:

[0023] Score=Score lf +αScore hf

[0024]

[0025]

[0026] Among them, Score lf is the low-frequency area score, Score hf is the high frequency region fraction, α is a real constant, FL is the low frequency region matrix intercepted from the spectrum graph, FH is the high frequency region matrix intercepted from the spectrum graph, n 1 and n 2 is the size of the FL matrix or the FH matrix.

[0027] Preferably, for the nodule image, combining all dimension scores and the Pearson correlation coefficient to obtain the optimal n-dimensional vector space of the nodule includes:

[0028] For each nodule image, select the dimension with the smallest dimension score as the starting dimension;

[0029] The absolute value of the Pearson correlation coefficient is used as the correlation coefficient between the vector dimensions, and the correlation coefficient Score between the remaining dimensions in the vector space and the starting dimension is calculated. r , select Score+βScore r The smallest dimension, where β is a constant;

[0030] Calculate the average of the correlation coefficients of the remaining dimensions in the vector space with the selected dimension choose The minimum dimension is selected, and this step is repeated until the number of dimensions selected in the vector space reaches the preset number.

[0031] Preferably, within a single nodule, based on vector similarity, a hierarchical clustering method is used to merge pixels or superpixels from bottom to top in the high-dimensional vector space after dimensionality reduction to generate a manifold vector category distribution map corresponding to the nodule, including:

[0032] Using Euclidean distance as the vector similarity;

[0033] Number all the pixels in a single nodule as initial categories to obtain point sets of multiple categories;

[0034] For each category of point set, obtain the neighboring category numbers;

[0035] For each category of point set, calculate the distance between the category and all its adjacent categories, find and record the category number with the smallest distance;

[0036] If two categories are the categories with the smallest distance to each other, the two categories are considered to be matched successfully, and the matched category pairs are merged and renumbered;

[0037] Repeat the above operation until the termination condition is met: the number of adjacent categories is less than a specified threshold, or the total number of categories in the nodule is less than a specified threshold, or there is no successfully matched category;

[0038] When the above termination conditions are met, the AGNES method is used for hierarchical clustering to reduce the number of categories to the target number.

[0039] Preferably, the manifold vector category distribution map based on all nodules globally measures the distances between vector sets in different vector spaces, and merges vector sets with similar distances, and finally obtains the classification results of each vector set, including:

[0040] Randomly select a set number of samples from all vector sets for clustering;

[0041] Design measurement methods for vector set distance and category distance in different vector spaces;

[0042] A bottom-up hierarchical clustering method is used to merge categories until the number of global categories is reduced to a specified threshold;

[0043] The vector sets outside the sample are classified according to their distance.

[0044] Preferably, the method for measuring the distance between vector sets in different vector spaces includes:

[0045] For vector set ps i , calculate the distance between it and the rest of the vector sets in the sample, and record the minimum distance and the maximum distance:

[0046]

[0047] Among them, ps i and ps j For two different vector sets, fl i and fl j Respectively vector set ps i and ps j Vector space dimension list, Coordinate(ps j ,fl i ) represents the vector set ps j In fl i The coordinate matrix in dimension, Represents vector set psj In fl i The center point coordinates in dimension, Represents vector set ps i and ps j In vector set ps i The distance in the corresponding vector space dimension; mind i and maxd i Respectively represent the dimensions of the vector space fl i Next, vector set ps i The minimum and maximum distances to the rest of the vector set.

[0048] Preferably, the measurement method of category distance in different vector spaces includes:

[0049] For categories It includes one or more vector sets belonging to the same or different nodules, and calculates their correlation with the category Distance:

[0050]

[0051] Preferably, the classifying the vector sets outside the sample according to the vector set distance comprises:

[0052] Under the corresponding vector space dimension of the vector set outside the sample, the distance between it and all vector sets in the sample is calculated and the global category to which the vector set with the smallest distance belongs is selected as its classification result.

[0053] Compared with the prior art, the embodiments of the present invention have at least one of the following beneficial effects:

[0054] The medical manifold vector set classification method for thyroid nodules in ultrasound images introduced in the embodiment of the present invention performs mask annotation on thyroid nodules, and has many advantages in data annotation, such as small amount, fast speed, and intuitive results. The present invention adopts a clustering method, which belongs to unsupervised learning, and its distance calculation method between vectors is artificially designed to ensure the accuracy and reliability of classification. Compared with the existing supervised neural network classification method, this method avoids the limitations of the black box method and provides higher interpretability. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0056] Figure 1 It is a flow chart of a medical manifold vector set classification method for thyroid nodules in a preferred embodiment of the present invention;

[0057] Figure 2is a schematic diagram of a manifold mapping network structure in a preferred embodiment of the present invention;

[0058] Figure 3 is an example diagram of the category distribution of manifold vectors inside a single nodule in a preferred embodiment of the present invention;

[0059] Figure 4 This is an example diagram of global vector clustering results in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0060] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several variations and improvements may be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0061] Figure 1 It is a flowchart of a medical manifold vector set classification method for thyroid nodules in a preferred embodiment of the present invention, which specifically includes the following steps:

[0062] S1, a dataset of ultrasound images of thyroid nodules was obtained;

[0063] S2, using a manifold mapping network to obtain high-dimensional vector coordinates of each pixel point of the thyroid nodule ultrasound image to obtain a high-dimensional vector space;

[0064] S3, reducing the dimension of the high-dimensional vector space based on Fourier transform;

[0065] S4, within a single nodule, based on vector similarity, the high-dimensional vector space after dimensionality reduction is merged from bottom to top using a hierarchical clustering method to generate a vector set. The manifold vector category distribution map corresponding to the nodule is obtained, and each manifold vector category includes at least one vector set.

[0066] S5, based on the manifold vector category distribution map of all nodules and their manifold space, the distances between manifolds in different vector spaces are measured globally, and vector sets with similar distances are merged to finally obtain the classification results of each vector set;

[0067] S6, manifold vector set category distribution statistics for benign and malignant nodules respectively.

[0068] In a preferred embodiment of the present invention, the data set S1 includes the original thyroid ultrasound image, the thyroid nodule mask, and the true value annotation of the benign and malignant thyroid.

[0069] In a preferred embodiment of the present invention, the manifold mapping network structure of S2 is as follows: Figure 2 As shown in the figure, it includes the encoder part and the decoder part, and uses hole convolutions of different scales to obtain features under different receptive fields. The specific structure of the network is:

[0070] In the first layer, the input image passes through a convolution module with a window size of 5×5, 16 output channels, a stride of 1, and a dilation rate of 1;

[0071] In the second layer, the outputs of the first layer are connected to convolutional modules with a window size of 5×5, 32 output channels, a stride of 1, and dilation rates of 1 and 2 respectively;

[0072] In the third layer, the output of the convolution module with a dilation rate of 1 in the second layer is connected to a convolution module with a window size of 5×5, 64 output channels, a stride of 1, and a dilation rate of 1. The output of the convolution module with a dilation rate of 2 in the second layer is connected to convolution modules with a window size of 5×5, 64 output channels, a stride of 1, and dilation rates of 2 and 3 respectively.

[0073] The three sets of features obtained by encoding are the outputs of branches with three different receptive fields, namely group1, group2, and group3. The size of each set of features is the same as the original image size, and the number of features is 64.

[0074] Batch normalization layer and relu activation function are used in all the above convolution modules, and 2×2 and 4×4 pooling layers are connected after the dilated convolution modules with stride 2 and 3, respectively.

[0075] In the fourth layer, feature group 1 passes through a deconvolution module with a window size of 5×5, 32 output channels, a stride of 1, and a dilation rate of 1. Feature group 2 passes through a deconvolution module with a window size of 5×5, 32 output channels, a stride of 1, and a dilation rate of 2. Feature group 3 passes through a deconvolution module with a window size of 5×5, 32 output channels, a stride of 1, and a dilation rate of 3.

[0076] In the fifth layer, the output of the convolution module with a dilation rate of 1 in the fourth layer is connected to a deconvolution module with a window size of 5×5, 16 output channels, a stride of 1, and a dilation rate of 1. The output of the convolution module with a dilation rate of 2 in the fourth layer and the output of the convolution module with a dilation rate of 3 are concatenated and connected to a deconvolution module with a window size of 5×5, 16 output channels, a stride of 1, and a dilation rate of 2.

[0077] In the sixth layer, the two output parts of the fifth layer are concatenated and connected to a deconvolution module with a window size of 5×5, an output channel number of 1, a stride of 1, and a dilation rate of 1, and finally the reconstructed image is output;

[0078] The batch normalization layer and relu activation function are used in the above intermediate layer deconvolution modules, and the linear activation function is used in the output layer.

[0079] The core function of the manifold mapping network is to achieve accurate mapping from low-dimensional space to high-dimensional space. For the single-channel grayscale image of thyroid nodule ultrasound image, that is, one-dimensional space data, this embodiment uses a manifold mapping network to effectively convert it into a high-dimensional space to ensure that each pixel corresponds to an accurate high-dimensional vector. In view of the limitations of traditional feature extraction methods in feature mining and noise reduction, this embodiment specifically introduces a manifold mapping network to extract richer and more accurate features, thereby optimizing subsequent data analysis and processing processes.

[0080] Ultrasound images themselves have the characteristics of speckle noise. The above-mentioned manifold mapping network maps the one-dimensional space to a high-dimensional space. In addition to the effective information, this space also contains a lot of noise information. Some dimensions are greatly affected by noise and are not conducive to clustering. It is necessary to select a suitable dimension. Therefore, in a preferred embodiment of the present invention, S3 is implemented to reduce the dimension of the high-dimensional vector space based on Fourier transform, including the following steps:

[0081] Count the proportion of non-zero values ​​in each dimension, and remove coordinate dimensions where the proportion of non-zero values ​​is less than 80%;

[0082] For each nodule image, the two-dimensional Fourier transform spectrum of the feature map corresponding to all its vector dimensions is calculated, and the score of each dimension is calculated using the following formula:

[0083] Score=Score lf +α·Score hf

[0084]

[0085] Among them, Score lf is the low-frequency area score, Score hf is the high frequency region fraction, α is a real constant, FL is the low frequency region matrix intercepted from the spectrum graph, FH is the high frequency region matrix intercepted from the spectrum graph, n 1 and n 2 is the size of the FL matrix or the FH matrix;

[0086] For each nodule image, select the dimension with the smallest score in each set of vector spaces as the starting dimension;

[0087] The absolute value of the Pearson correlation coefficient of the feature spectrum graph corresponding to the vector dimension is used as the correlation coefficient between the vector dimensions, and the correlation coefficient Score of the remaining dimensions in the vector space and the starting dimension is calculated. r , select Score+βScore r The smallest dimension, where β is 5;

[0088] Similarly, calculate the average value of the correlation coefficients between the remaining dimensions in the vector space and the selected dimension choose The minimum dimension, where β is 5, is repeated until the number of dimensions selected in the vector space reaches 5.

[0089] The prior art usually extracts features from one-dimensional images, which makes it difficult to eliminate the influence of noise. The above embodiment maps a one-dimensional space to a high-dimensional space, and then obtains a vector space suitable for clustering by evaluating each dimension, which can greatly reduce the influence of noise on clustering.

[0090] In a preferred embodiment of the present invention, S4 is implemented to merge pixels or superpixels from bottom to top within a single nodule based on spatial position information and high-dimensional vector similarity to generate a manifold vector category distribution map corresponding to the nodule, including the following steps:

[0091] All pixels within a single nodule are numbered as initial categories;

[0092] For each category of point set, obtain the category numbers of its four neighborhoods;

[0093] Calculate two point sets according to the following formula and Distance between:

[0094]

[0095] Among them, psd 12 represents the distance between two point sets in the same node, represents the Euclidean distance between two points in the same node;

[0096] For each category’s vector set, calculate the distance between the category and all its adjacent categories, find and record the category number with the smallest distance;

[0097] In order to speed up clustering, the distance between the center points of the vector set is used as the distance measure between the two vector sets in the later stage of clustering;

[0098] If two categories are the categories with the smallest distance to each other, the two categories are considered to be matched successfully, and the matched category pairs are merged and renumbered;

[0099] Repeat the above steps until the termination condition is reached: the number of adjacent categories is less than 3 or there is no successful matching category;

[0100] When the above termination conditions are met, the Euclidean distance is used as the distance metric and the AGNES method is used for hierarchical clustering until the number of categories is reduced to 30, generating a manifold vector category result map of the nodule.

[0101] like Figure 3 The figure shows an example of the distribution of manifold vector categories inside a single nodule. The same color represents the same category, and the number of manifold vector categories in each nodule is 30.

[0102] In a preferred embodiment of the present invention, S5 is implemented to measure the distances between different vector space manifolds in a global scope based on the category distribution map of nodules and their manifold space, and merge vector sets with similar distances to finally obtain the classification results of each vector set. The implementation steps are as follows:

[0103] Randomly select 1000 vector sets from all vector sets as samples;

[0104] All vector sets in the sample are numbered, and the distance between each vector set in the sample and the remaining vector sets is calculated according to the following formula And record mind i and maxd i :

[0105]

[0106] Among them, ps i and ps j For two different vector sets, fl i and fl j Respectively vector set ps i and ps j Vector space dimension list, Coordinate(ps j ,fl i ) represents the vector set ps j In fl i The coordinate matrix in dimension, Represents vector set ps j In fl i The center point coordinates in dimension, Represents vector set ps i and ps j In vector set ps i The distance in the corresponding vector space dimension. mind i and maxd i Respectively represent the dimensions of the vector space fl i Next, vector set ps i The minimum and maximum distances to the rest of the vector set;

[0107] If two vector sets are the vector sets with the minimum distance to each other, the two vector sets are considered to be matched successfully, and the successfully matched vector sets are merged and renumbered;

[0108] The categories are calculated according to the following formula With category Distance:

[0109]

[0110] After calculating the distance of each category, the AGNES method is used for hierarchical clustering until the number of global categories is reduced to a specified threshold;

[0111] For a vector set outside the sample, the distance between it and all vector sets in the sample is calculated under the spatial dimension of the vector set, and the global category to which the vector set with the smallest distance belongs is selected as its classification result.

[0112] like Figure 4 This is an example of the global vector set clustering result. Two categories are given in the figure, and each category lists the corresponding vector set masks in three nodules.

[0113] In the experimental test set, the classification accuracy of this method for cystic / solid is 96.8%, the classification accuracy for hypoechoic / isoechoic / hyperechoic is 86.6%, and the classification accuracy for calcification / non-calcification is 95.8%.

[0114] Compared with the prior art, the medical manifold vector set classification method for thyroid nodules in ultrasound images provided by the above embodiment of the present invention obtains the feature dimensions under different receptive fields through the manifold mapping network to form a high-dimensional coordinate space; then uses an unsupervised method to reduce the dimension; and then realizes the clustering of vector sets in different nodules through the methods of internal hierarchical clustering of a single nodule and global hierarchical clustering; finally, the vector set category characteristics inside benign and malignant nodules can be analyzed based on the clustering results. It has the advantages of less data annotation, fast speed, high interpretability, intuitive results, and convenience for medical analysis and diagnosis.

[0115] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art may make various modifications or variations within the scope of the claims, which do not affect the essence of the present invention.

Claims

1. A medical manifold vector set classification method for thyroid nodules in ultrasound images, characterized in that: include: Acquire a dataset of thyroid nodule ultrasound images; Using a manifold mapping network, high-dimensional vector coordinates of each pixel point of the thyroid nodule ultrasound image are obtained to obtain a high-dimensional vector space; Performing dimensionality reduction on the high-dimensional vector space based on Fourier transform; In a single nodule, based on vector similarity, the high-dimensional vector space after dimensionality reduction is merged from bottom to top using a hierarchical clustering method to form a vector set and generate a manifold vector category distribution map corresponding to the nodule; Based on the manifold vector category distribution map of all nodules, the distances between vector sets in different vector spaces are measured globally, and vector sets with similar distances are merged to finally obtain the classification results of each vector set. Based on the classification results of each vector set, the distribution statistics of benign and malignant nodules are performed.

2. The method for classifying medical manifold vector sets of thyroid nodules in ultrasound images according to claim 1, characterized in that: The structure of the manifold mapping network includes an encoder part and a decoder part, and can accept thyroid nodule ultrasound images of any size as input; The encoder part includes multiple convolution modules, each of which performs one or more of convolution, batch normalization, activation function, and pooling operations; the strides of each convolution module are 1; and multiple convolution modules are hollow convolutions of different scales, which can obtain different receptive fields; The decoder part is symmetrical with the encoder part, and includes a plurality of deconvolution modules, each of which performs one or more of convolution, batch normalization, activation function, and pooling operations, and the strides of each of the deconvolution modules are 1; The output of the encoder part is a high-dimensional vector space, which contains multi-dimensional nodule feature maps under receptive fields of different sizes, and the size of the feature map is the same as the input size of the manifold mapping network.

3. The medical manifold vector set classification method for thyroid nodules in ultrasound images according to claim 1, characterized in that: The dimensionality reduction of the high-dimensional vector space based on Fourier transform includes: Count the proportion of non-zero values ​​in each dimension of the high-dimensional vector space, and remove coordinate dimensions where the proportion of non-zero values ​​is lower than the set threshold; Obtain the two-dimensional Fourier transform spectrum of the feature map corresponding to each vector dimension, and calculate the score of each dimension; Calculate the Pearson correlation coefficient between the two feature maps based on the spectrum map; For the nodule image, the optimal n-dimensional vector space of the nodule is obtained by combining all dimension scores and the Pearson correlation coefficient.

4. The method for classifying medical manifold vector sets of thyroid nodules in ultrasound images according to claim 3, characterized in that: The obtaining of the two-dimensional Fourier transform spectrum of the feature graph corresponding to the vector dimension and calculating the scores of each dimension include: The specific scoring method for each dimension is: Score=Score lf +α·Score hf Among them, Score lf is the low-frequency area score, Score hf is the high frequency region fraction, α is a real constant, FL is the low frequency region matrix intercepted from the spectrum graph, FH is the high frequency region matrix intercepted from the spectrum graph, and n1 and n2 are the sizes of the FL matrix or the FH matrix.

5. The method for classifying medical manifold vector sets of thyroid nodules in ultrasound images according to claim 3, characterized in that: The optimal n-dimensional vector space of the nodule is obtained by combining all dimension scores and the Pearson correlation coefficient for the nodule image, including: For each nodule image, select the dimension with the smallest dimension score as the starting dimension; The absolute value of the Pearson correlation coefficient is used as the correlation coefficient between the vector dimensions, and the correlation coefficient Score between the remaining dimensions in the vector space and the starting dimension is calculated. r , select Score+βScore r The smallest dimension, where β is a constant; Calculate the average of the correlation coefficients of the remaining dimensions in the vector space with the selected dimension choose The minimum dimension is selected, and this step is repeated until the number of dimensions selected in the vector space reaches the preset number.

6. The method for classifying medical manifold vector sets of thyroid nodules in ultrasound images according to claim 1, characterized in that: In a single nodule, based on vector similarity, the high-dimensional vector space after dimensionality reduction is merged from bottom to top using a hierarchical clustering method to generate a manifold vector category distribution map corresponding to the nodule, including: Using Euclidean distance as the vector similarity; Number all the pixels in a single nodule as initial categories to obtain point sets of multiple categories; For each category of point set, obtain the neighboring category numbers; For each category of point set, calculate the distance between the category and all its adjacent categories, find and record the category number with the smallest distance; If two categories are the categories with the smallest distance to each other, the two categories are considered to be matched successfully, and the matched category pairs are merged and renumbered; Repeat the above operation until the termination condition is met: the number of adjacent categories is less than a specified threshold, or the total number of categories in the nodule is less than a specified threshold, or there is no successfully matched category; When the above termination conditions are met, the AGNES method is used for hierarchical clustering to reduce the number of categories to the target number.

7. The method for classifying medical manifold vector sets of thyroid nodules in ultrasound images according to claim 1, characterized in that: The manifold vector category distribution map based on all nodules measures the distances between vector sets in different vector spaces globally, and merges vector sets with similar distances, and finally obtains the classification results of each vector set, including: Randomly select a set number of samples from all vector sets for clustering; Design measurement methods for vector set distance and category distance in different vector spaces; A bottom-up hierarchical clustering method is used to merge categories until the number of global categories is reduced to a specified threshold; The vector sets outside the sample are classified according to their distance.

8. The method for classifying medical manifold vector sets of thyroid nodules in ultrasound images according to claim 7, characterized in that: The measurement methods of vector set distance in different vector spaces include: For vector set ps i , calculate the distance between it and the rest of the vector sets in the sample, and record the minimum distance and the maximum distance: Among them, ps i and ps j For two different vector sets, fl i and fl j Respectively vector set ps i and ps j Vector space dimension list, Coordinate(ps j , fl i ) represents the vector set ps j In fl i The coordinate matrix in dimension, Represents vector set ps j In fl i The center point coordinates in dimension, Represents vector set ps i and ps j In vector set ps i The distance in the corresponding vector space dimension; mind i and maxd i Respectively represent the dimensions of the vector space fl i Next, vector set ps i The minimum and maximum distances to the rest of the vector set.

9. The method for classifying medical manifold vector sets of thyroid nodules in ultrasound images according to claim 7, characterized in that: The measurement methods of category distance in different vector spaces include: For categories It includes one or more vector sets belonging to the same or different nodules, and calculates their correlation with the category Distance:

10. The medical manifold vector set classification method for thyroid nodules in ultrasound images according to claim 7, characterized in that: The classifying the vector sets outside the sample according to the vector set distance includes: Under the corresponding vector space dimension of the vector set outside the sample, the distance between it and all vector sets in the sample is calculated and the global category to which the vector set with the smallest distance belongs is selected as its classification result.

Citation Information

Patent Citations

  • Thyroid nodule diagnosis method based on deep learning network

    CN112529894A

  • Thyroid myeloid cancer ultrasonic image recognition method based on clinical prior knowledge guidance

    CN114119458A

  • Diabetic retinopathy consultation method and system based on multi-image fusion

    CN116703915A

  • Thyroid nodule image processing method and apparatus, and computer device

    WO2021169452A1