A Method and System for Auxiliary Diagnosis of Lung Nodules Based on Multi-View Fuzzy Clustering

Abstract

A method and system for assisted diagnosis of pulmonary nodules based on multi-view fuzzy clustering, relating to the field of intelligent medical image analysis, addresses the problems of existing pulmonary nodule analysis methods, such as reliance on supervised classification, difficulty in discovering potential heterogeneity within the disease, and inability to express fuzzy transitional relationships between subtypes. The method includes acquiring DICOM data and clinical structured data of lung CT images of target patients, and segmenting and extracting regions of interest (ROIs) for pulmonary nodules; constructing multi-view features for the same pulmonary nodule to form a multi-view description of the nodule; constructing a multi-view fuzzy clustering model, introducing local manifold preservation constraints and view weight adaptive mechanisms; solving the constructed multi-view fuzzy clustering model, updating the view weight cluster centers v, view weights w, and fuzzy membership degrees U in an alternating iterative manner until the multi-view fuzzy clustering model converges; and outputting the multi-view fuzzy clustering results.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical image analysis, specifically to a method and system for auxiliary diagnosis of pulmonary nodules based on multi-view fuzzy clustering. Background Technology

[0002] In recent years, pulmonary nodules have become one of the most common and challenging imaging findings in clinical lung CT examinations. Due to the complex and diverse pathological nature of pulmonary nodules, although most are benign lesions, some may be potential manifestations of early-stage lung cancer. Traditional clinical diagnosis mainly relies on physician experience or risk classification methods based on fixed labels to assess the risk of pulmonary nodules. However, even pulmonary nodules at the same risk level may exhibit significant differences in their internal imaging characteristics, tissue structure, and pathological evolution. Therefore, automatically identifying potential disease subtypes within pulmonary nodules and providing auxiliary analysis of different disease evolution patterns has become a crucial problem urgently needing to be solved in the field of intelligent medical image analysis. Mining potential disease subtypes of pulmonary nodules helps to reveal the complex heterogeneity within pulmonary nodules, providing support for precise diagnosis and treatment and individualized decision support.

[0003] Chinese patent CN115375632A discloses an intelligent detection and analysis system and method for lung nodules. However, this method is a single-view approach. A single view is not only highly susceptible to interference from local noise or artifacts in lung CT images, but also lacks cross-validation of multi-dimensional information, leading to limitations in risk assessment results. Furthermore, the absolute hard classification mechanism severely violates the biological fuzzy nature of lesions; lung nodules are a progressive and continuous evolutionary process. Some boundary-state nodules often possess multiple risk characteristics simultaneously, and their classification is somewhat ambiguous. Traditional hard classification results easily lead to misclassification of nodules at the boundary, and the interpretability of hard classification results is poor, failing to explain the reasons for the classification and thus providing insufficient information for treatment reference. In addition, most current multi-view-based lung nodule risk stratification methods are based on supervised learning. For example, Chinese patent CN112990344A discloses a multi-view classification method for lung nodules. However, the occurrence and development of lung nodules is essentially a progressive evolutionary process encompassing multiple pathological stages. Supervised learning-based methods can only learn existing risk levels and cannot effectively discover and identify potential disease subtypes at different evolutionary stages. This lack of ability to discover potential subtypes means that existing methods cannot finely characterize the true multi-level risk status of nodules, making it difficult to provide more granular risk reference information for clinical practice. Finally, Chinese patent CN120015323A proposes a potential subtype discovery method based on multi-omics data fusion. Although this method utilizes multi-view data, it is merely a simple stitching together of different views without considering the differences in information contribution between different views. Therefore, when a view contains low-quality information, the model lacks the ability to dynamically suppress the contribution of that view, thus affecting the accuracy of the classification results and the expressive power of the model. Summary of the Invention

[0004] This invention aims to overcome the problems of existing methods for analyzing pulmonary nodules, such as reliance on supervised classification, difficulty in detecting potential heterogeneity within the disease, and inability to express fuzzy transitional relationships between subtypes. To address these issues, this invention proposes a method and system for auxiliary diagnosis of pulmonary nodules based on multi-view fuzzy clustering. The invention achieves these technical problems through the following technical solutions: Option 1: This invention proposes an auxiliary diagnostic method for lung nodules based on multi-view fuzzy clustering, the method comprising the following steps: Step 1: Obtain DICOM data and clinical structured data of the lung CT images of the target patient, and segment and extract the regions of interest for the lung nodules; Step 2: Construct multi-view features for the same lung nodule to form a multi-view description of the lung nodule; Step 3: Construct a multi-view fuzzy clustering model based on the multi-view features built in Step 2, and introduce a local manifold preservation constraint term and a view weight adaptive mechanism term; Step 4: Solve the multi-view fuzzy clustering model constructed in Step 3. Update the view weight cluster center v, view weight w, and fuzzy membership degree U in an alternating iterative manner until the multi-view fuzzy clustering model converges. Step 5: Output the multi-view fuzzy clustering results, which include soft membership degree U, view weight W, cluster center V, and corresponding auxiliary diagnostic interpretation information.

[0005] Furthermore, a preferred embodiment is provided, wherein the method for segmenting and extracting the region of interest of the lung nodule in step 1 is as follows: The trilinear interpolation algorithm was used to resample the cases of each target patient into isovoxels with a size of 1mm×1mm×1mm for feature extraction.

[0006] Furthermore, a preferred embodiment is provided in which the construction of multi-view features in step 2 includes constructing a radiomics view and a clinical structured feature view; The method for constructing the radiomics view is as follows: Based on the PyRadiomics toolkit, radiomics features of the lung nodules within the region of interest are extracted to obtain the feature vector of the sample radiomics view. ; The method for constructing the clinical structured feature view is as follows: One-hot encoding is used, and missing values ​​are filled in using the mean / mode to obtain the feature vector of the clinical structured feature view. .

[0007] Furthermore, in a preferred embodiment, step 2 further includes z-score standardization of each view sample to transform it into standard data with a mean of 0 and a standard deviation of 1, i.e.

[0008] Let j be the j-th feature of the i-th sample in the k-th view, where Let the mean and standard deviation of the j-th dimension of the k-th view be denoted as . This is a smoothing term.

[0009] Furthermore, a preferred embodiment is provided, wherein the method for constructing the multi-view fuzzy clustering model in step 3 is as follows: Step 3.1: Define the main objective function for multi-view fuzzy clustering. By defining the fuzzy coefficient m, adjust the smoothness of the fuzzy membership degree, quantify the membership degree loss between lung nodule samples and each cluster center, and complete the basic fitting measure of soft clustering of lung nodules under multi-view features. Step 3.2: Based on the multi-view fuzzy clustering main objective function defined in Step 3.1, add a local manifold regularization constraint term to achieve structural modification of the basic clustering model; Step 3.3: Based on the multi-view fuzzy clustering main objective function constructed in Step 3.1, introduce a weight entropy regularization term to achieve adaptive optimization of multi-view feature weights. Construct a weight entropy regularization term through hyperparameter γ to constrain the weight distribution state of each view. Step 3.4: Based on steps 3.1 to 3.3, construct the overall objective function to complete the construction of the multi-view fuzzy clustering model.

[0010] Furthermore, a preferred embodiment is provided, wherein step 4 uses an alternating iterative method to update the fuzzy membership degree U, view weight cluster center v, and view weight w until the multi-view fuzzy clustering model converges. Step 4.1: Update the fuzzy membership degree U: Iteratively update the fuzzy membership degree matrix U using the gradient descent method; obtain the membership degree gradient by taking the partial derivative of the total objective function; complete the iterative update of the membership degree parameters based on the negative gradient direction and the preset update step size η; and perform constraint correction on the updated membership degree through simplex projection operation. Step 4.2: Update the cluster centers v of the view weights: Iteratively update the cluster centers corresponding to each view using a weighted average method to obtain accurate cluster center parameters for each dimension; Step 4.3, Update view weights w: Calculate the weighted fitting error for each view, and use the Lagrange multiplier method to solve the optimization subproblem with weighted entropy regularization constraints to obtain the closed-form solution of the view weights; Step 4.4: Based on steps 4.1 to 4.3, set the maximum number of iterations T and the convergence threshold ε; when the number of iterations reaches the maximum number of iterations T, or the difference between the total objective function of two adjacent iterations is less than the preset minimum positive number ε, the model is determined to have converged and the iteration update process is terminated.

[0011] Furthermore, a preferred embodiment is provided, wherein the method described in step 4.4, which determines that the model iteration convergence and terminates the iteration update process when the number of iterations reaches the maximum number of iterations T, or the difference between the total objective function of two adjacent iterations is less than a preset minimum positive number ε, is as follows:

[0012] In the formula, The result is a positive number; finally, the fuzzy membership matrix is ​​obtained. Reliability weights for each view Cluster centers of each view .

[0013] Option 2: A lung nodule auxiliary diagnostic system based on multi-view fuzzy clustering, the method comprising the following steps: The data segmentation and extraction module is used to segment and extract the regions of interest of lung nodules from the DICOM data and clinical structured data of the lung CT images of the target patient. The multi-view description module is used to construct multi-view features for the same lung nodule, forming a multi-view description of the lung nodule. The multi-view fuzzy clustering model construction module is used to construct a multi-view fuzzy clustering model based on the multi-view features constructed by the multi-view description module, and introduces a local manifold preservation constraint term and a view weight adaptive mechanism term; The training module is used to solve the multi-view fuzzy clustering model constructed by the multi-view fuzzy clustering model construction module. It updates the view weight cluster center v, view weight w, and fuzzy membership degree U in an alternating iterative manner until the multi-view fuzzy clustering model converges. The output module is used to output the multi-view fuzzy clustering results, which include soft membership degree U, view weight W, cluster center V and corresponding auxiliary diagnostic explanation information.

[0014] Option 3: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in Option 1.

[0015] Option 4: A computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the method described in Option 1.

[0016] The advantages of this invention are: This invention discloses a lung nodule auxiliary diagnosis method and system based on multi-view fuzzy clustering. It achieves adaptive weighted fusion of multi-view information and complementary advantages of multi-view information. This invention not only utilizes multi-view information but also introduces view reliability weights into the objective function and performs alternating iterative updates. This dynamic adjustment allows views with more information to receive higher weights, thereby reducing interference from low-quality views and overcoming the shortcomings of single data sources—such as incomplete information and susceptibility to noise.

[0017] This invention discloses a lung nodule auxiliary diagnostic method and system based on multi-view fuzzy clustering. Addressing the common issue of blurred nodule boundaries in medical imaging, this invention employs a fuzzy clustering algorithm, which effectively expresses the fuzzy transitional relationships between different subtypes of lung nodules, better aligning with the continuous evolution of the disease and resulting in more accurate clustering results.

[0018] The lung nodule auxiliary diagnosis method and system based on multi-view fuzzy clustering described in this invention has a clear structure, well-defined parameters, is easy to implement, and is applicable to various other disease subtype analysis scenarios. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the lung nodule auxiliary diagnosis method based on multi-view fuzzy clustering as described in Implementation Method 1. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0021] Implementation Method 1, see [link] Figure 1 This embodiment describes a method for auxiliary diagnosis of lung nodules based on multi-view fuzzy clustering. The method specifically includes the following steps: Step 1: Lung Data Acquisition and Preprocessing The system receives multiple desensitized lung CT images and corresponding pathological labels in DICOM format, for example, with a resolution of 512×512, a slice thickness of 1mm, and a pixel pitch of [0.7mm, 0.7mm]. Simultaneously, it acquires corresponding clinical structured feature data (such as age, smoking history, etc.). A trilinear interpolation algorithm is used to resample each case into isovoxels, with a voxel size of 1mm×1mm×1mm, facilitating subsequent feature extraction. Based on the coordinates of the nodule center point as defined by the doctor, a three-dimensional ROI block is extracted, for example, 64×64×64 voxels, with any excess padded with zeros. Step 2: Multi-view feature extraction The multi-view architecture of this invention realizes spatial positive interaction between radiomics view and clinical prior view, fundamentally overcoming the shortcomings of single view information being one-sided and easily dominated by noise.

[0022] 1. View Figure 1 Radiomics View Radiomics features of pulmonary nodules within the region of interest were extracted using the PyRadiomics toolkit; Morphological characteristics: such as the three-dimensional volume, surface area, sphericity, and maximum 3D diameter of the nodules.

[0023] First-order statistical characteristics: based on the distribution of voxel intensity, such as mean, variance, skewness, kurtosis, etc.

[0024] Texture features: such as gray-level co-occurrence matrix, gray-level run-length matrix, etc.

[0025] Obtain sample view Figure 1 eigenvectors .

[0026] 2. View Two: Clinical Structured Feature View Structured variables including age, gender, smoking index, family history, and previous cancer history; Categorical variables were encoded using one-hot encoding; missing values ​​were filled in using the mean / mode to obtain the feature vector for View 2. .

[0027] For each view sample, z-score standardization is performed to transform it into standard data with a mean of 0 and a standard deviation of 1:

[0028] Let j be the j-th feature of the i-th sample in the k-th view, where Let the mean and standard deviation of the j-th dimension of the k-th view be denoted as . This is a smoothing term.

[0029] Step 3: Construction of Multi-View Fuzzy Clustering Model Define the main objective function for multi-view fuzzy clustering.

[0030] Where m is the fuzzy coefficient, which defines the smoothness of the membership degree. The value of m is usually set to 2. This is the standardized feature vector of the i-th lung nodule sample under the k-th view after preprocessing. Let be the cluster center vector of the k-th view and the c-th class.

[0031] Local manifold regularization constraints

[0032] in, It refers to the similarity of features between two lung nodule samples. For regularization parameters, Let be the fuzzy membership degree of the i-th lung nodule sample belonging to the c-th cluster. Let be the fuzzy membership degree of the j-th lung nodule sample to the c-th cluster. If nodules A and B are extremely similar in features, then their membership degrees to different clusters will also be extremely close.

[0033] for K-nearest neighbors are used to find samples. The similarity between a sample and its p nearest neighbors in each view is calculated using a Gaussian kernel function. The kernel function formula is as follows:

[0034] in, Let i be the standardized feature vector of the i-th lung nodule sample under the k-th view. Let j be the standardized feature vector of the j-th lung nodule sample in the k-th view. The bandwidth of the Gaussian kernel corresponding to the k-th view.

[0035] Finally, summing and averaging the results yields the similarity matrix.

[0036]

[0037] Weight Entropy Regularization

[0038] hyperparameters Used to control the smoothness of view weight distribution, assigning different weights to different views.

[0039] objective function

[0040] in, To ensure that clustering partitioning closely matches sample characteristics; To constrain the clustering results to conform to the local manifold distribution of the lung nodule samples; Adaptive balancing of clustering contribution weights across multiple views; Step 4: Iterative solution of the model This invention uses an alternating iterative method to update U, w, The steps are as follows: For the already obtained global fusion similarity matrix Calculate the similarity matrix Symmetric normalized Laplace matrix

[0041] Where I is the N*N identity matrix, and D is the degree matrix of S, calculated as follows:

[0042] For the obtained matrix L, perform eigenvalue decomposition to obtain N eigenvalues, arrange them in ascending order, calculate the difference between two adjacent eigenvalues, and find the index i corresponding to the largest difference as the subtype cluster number C.

[0043]

[0044] Preferably, the fuzziness coefficient m=2, T=100, and the view weight are set. .

[0045] Initialize cluster centers using k-means. .

[0046] Calculate the sum of distances from samples to each cluster under different views, and use the reciprocal of the distance to initialize the fuzzy membership degree, ensuring that samples closer to the center have a higher membership degree. The initialization formula is as follows:

[0047] in Let represent the Euclidean distance between the i-th sample in the k-th view and the c-th class, where This is a minimal constant to prevent the denominator from being 0.

[0048] Update membership degree U The membership matrix U is iteratively updated using the gradient descent method.

[0049]

[0050] Using the gradient described above, perform a one-step update along the negative gradient direction to obtain the membership degree.

[0051] in, >0 represents the update step size.

[0052] Finally, a simplex projection operation was performed to correct it.

[0053]

[0054] Update cluster center

[0055] With U and w fixed, the prototype is a weighted mean:

[0056] in It is a small constant to avoid a denominator of 0. Update view weights w: Calculate the weighted fitting error of the view:

[0057] Solving the subproblem with weighted entropy regularization terms using the Lagrange multiplier method yields a closed-form solution:

[0058] This update gives higher weight to views with smaller errors, thereby reducing the interference of low-quality views on the results.

[0059] Iteration stops when the number of iterations reaches T or the objective function satisfies the following condition:

[0060] in Given a positive number, N samples, K views, and the k-th view having the following feature count: The i-th sample in the k-th view is The number of clusters is C, and the fuzzy membership matrix is... , The membership degree of the i-th sample to the c-th cluster, and the view reliability weight. , This represents the weight of the k-th view. This represents the c-th cluster center of the k-th view; Finally, the fuzzy membership matrix is ​​obtained. Reliability weights for each view Cluster centers of each view .

[0061] Step 5: Clustering Results Output and Clinical Semantic Interpretation The output includes a fuzzy membership matrix U for each potential disease subtype of the lung nodule sample, view weights w, and cluster centers v. The fuzzy membership matrix represents the probability distribution of each lung nodule sample belonging to different potential disease subtypes. View weights reflect the contribution of different views to the clustering results, and cluster centers describe the typical feature patterns corresponding to each potential disease subtype. Furthermore, to enhance the medical interpretability of the clustering results, posterior statistical analysis is performed on samples within each cluster to establish the correspondence between cluster subtypes and true clinical semantics. The specific evaluation method is as follows: The proportion of different pathological labels within different clusters was statistically analyzed. Since different pathological types correspond to different degrees of malignancy and stages of tissue evolution, the differences in pathological composition between clusters were statistically analyzed. For example, when a cluster has a high proportion of benign nodules, it indicates that the cluster corresponds to a lower-risk pathological pattern.

[0062] The potential heterogeneity of different disease subtypes was analyzed, and the differences in important radiomic features and clinical features of different clusters were statistically analyzed. For example, although many clusters were dominated by high-risk nodules, some clusters may mainly show increased texture complexity and enhanced edge spiculation, while others mainly show increased nodule volume and abnormally increased density.

[0063] The fuzzy membership distribution of samples within different clusters was analyzed, and statistical analysis was performed on the average membership degree and the dispersion of membership degree for different clusters. For example, two clusters were dominated by low-risk nodules, but some samples within one cluster had a higher average membership degree to the low-risk subtype, and the membership degree distribution was more concentrated, indicating that the subtype has a more stable and typical low-risk characteristic. Meanwhile, samples within another cluster also had a high membership degree to other risk subtypes, indicating a clear transitional tendency between this subtype and other risk states. This can further reveal the continuous changing relationship and risk transition characteristics between different potential disease subtypes of pulmonary nodules, thereby enhancing the medical interpretability of the clustering results.

[0064] In summary, the lung nodule auxiliary diagnosis method and system based on multi-view fuzzy clustering described in this embodiment constructs multi-view data for the same object, overcoming the shortcomings of insufficient information in a single view and susceptibility to local noise. A multi-view weight adaptive mechanism is designed to give high weights to high-quality views, improving the clustering accuracy of the model. It achieves unsupervised discovery of potential disease subtypes, breaking through the limitation of traditional supervised learning, which can only learn known risk levels. The method described in this embodiment automatically mines and identifies the potential disease subtype structure at different evolutionary stages within lung nodules by constructing a multi-view fuzzy clustering model, using fuzzy clustering instead of traditional hard classification to effectively express the fuzziness of the lung nodule boundary data.

[0065] Those skilled in the art will understand that the above description is merely a preferred embodiment of the present invention, and the features described in the various embodiments and / or technical solutions of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. This is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0066] Although preferred embodiments of the present invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended technical solutions are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the present invention. Clearly, those skilled in the art can make various modifications and variations to the present invention without departing from its spirit and scope. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalents, the present invention also intends to include these modifications and variations.

Claims

1. A lung nodule auxiliary diagnostic method based on multi-view fuzzy clustering, characterized in that, The method includes the following steps: Step 1: Obtain DICOM data and clinical structured data of the lung CT images of the target patient, and segment and extract the regions of interest for the lung nodules; Step 2: Construct multi-view features for the same lung nodule to form a multi-view description of the lung nodule; Step 3: Construct a multi-view fuzzy clustering model based on the multi-view features built in Step 2, and introduce a local manifold preservation constraint term and a view weight adaptive mechanism term; Step 4: Solve the multi-view fuzzy clustering model constructed in Step 3. Update the view weight cluster center v, view weight w, and fuzzy membership degree U in an alternating iterative manner until the multi-view fuzzy clustering model converges. Step 5: Output the multi-view fuzzy clustering results, which include soft membership degree U, view weight W, cluster center V, and corresponding auxiliary diagnostic interpretation information.

2. The lung nodule auxiliary diagnosis method based on multi-view fuzzy clustering according to claim 1, characterized in that, The method for segmenting and extracting the region of interest for lung nodules in step 1 is as follows: the trilinear interpolation algorithm is used to resample the cases of each target patient into isovoxels with a voxel size of 1mm×1mm×1mm for feature extraction.

3. The lung nodule auxiliary diagnosis method based on multi-view fuzzy clustering according to claim 1, characterized in that, Step 2 involves constructing multi-view features, including constructing a radiomics view and a clinical structured feature view. The method for constructing the radiomics view is as follows: Based on the PyRadiomics toolkit, radiomics features of the lung nodules within the region of interest are extracted to obtain the feature vector of the sample radiomics view. ,in, for Vieuxiliary real space; The method for constructing the clinical structured feature view is as follows: One-hot encoding is used, and missing values ​​are filled in using the mean / mode to obtain the feature vector of the clinical structured feature view. .

4. The lung nodule auxiliary diagnosis method based on multi-view fuzzy clustering according to claim 3, characterized in that, Step 2 also includes z-score standardization of each view sample to transform it into standard data with a mean of 0 and a standard deviation of 1. in, Let j be the j-th feature of the i-th sample in the k-th view, where Let the mean and standard deviation of the j-th dimension of the k-th view be denoted as . This is a smoothing term.

5. The lung nodule auxiliary diagnosis method based on multi-view fuzzy clustering according to claim 1, characterized in that, The method for constructing the multi-view fuzzy clustering model in step 3 is as follows: Step 3.1: Define the main objective function for multi-view fuzzy clustering. By defining the fuzzy coefficient m, adjust the smoothness of the fuzzy membership degree, quantify the membership degree loss between lung nodule samples and each cluster center, and complete the basic fitting measure of soft clustering of lung nodules under multi-view features. Step 3.2: Based on the multi-view fuzzy clustering main objective function defined in Step 3.1, add a local manifold regularization constraint term to achieve structural modification of the basic clustering model; Step 3.3: Based on the multi-view fuzzy clustering main objective function constructed in Step 3.1, introduce a weight entropy regularization term to achieve adaptive optimization of multi-view feature weights. Construct a weight entropy regularization term through hyperparameter γ to constrain the weight distribution state of each view. Step 3.4: Based on steps 3.1 to 3.3, construct the overall objective function to complete the construction of the multi-view fuzzy clustering model.

6. The lung nodule auxiliary diagnosis method based on multi-view fuzzy clustering according to claim 1, characterized in that, Step 4 uses an alternating iterative approach to update the fuzzy membership degree U, view weight cluster center v, and view weight w until the multi-view fuzzy clustering model converges. Step 4.1: Update the fuzzy membership degree U: Iteratively update the fuzzy membership degree matrix U using the gradient descent method; obtain the membership degree gradient by taking the partial derivative of the total objective function; complete the iterative update of the membership degree parameters based on the negative gradient direction and the preset update step size η; and perform constraint correction on the updated membership degree through simplex projection operation. Step 4.2: Update the view weight cluster center v: Iteratively update the cluster center corresponding to each view using a weighted average method to obtain accurate cluster center parameters for each dimension; Step 4.3, Update view weights w: Calculate the weighted fitting error for each view, and use the Lagrange multiplier method to solve the optimization subproblem with weighted entropy regularization constraints to obtain the closed-form solution of the view weights; Step 4.4: Based on steps 4.1 to 4.3, set the maximum number of iterations T and the convergence threshold ε; when the number of iterations reaches the maximum number of iterations T, or the difference between the total objective function of two adjacent iterations is less than the preset minimum positive number ε, the model is determined to have converged and the iteration update process is terminated.

7. The lung nodule auxiliary diagnosis method based on multi-view fuzzy clustering according to claim 1, characterized in that, In step 4.4, when the number of iterations reaches the maximum number of iterations T, or the difference between the total objective function of two adjacent iterations is less than the preset minimum positive number ε, the method for determining that the model iteration has converged and terminating the iteration update process is as follows: In the formula, The result is a positive number; finally, the fuzzy membership matrix is ​​obtained. Reliability weights for each view Cluster centers of each view , Let be the overall objective function after the t-th iteration.

8. A lung nodule auxiliary diagnostic system based on multi-view fuzzy clustering, characterized in that, The system includes: The data segmentation and extraction module is used to segment and extract the regions of interest of lung nodules from the DICOM data and clinical structured data of the lung CT images of the target patient. The multi-view description module is used to construct multi-view features for the same lung nodule, forming a multi-view description of the lung nodule. The multi-view fuzzy clustering model construction module is used to construct a multi-view fuzzy clustering model based on the multi-view features constructed by the multi-view description module, and introduces a local manifold preservation constraint term and a view weight adaptive mechanism term; The training module is used to solve the multi-view fuzzy clustering model constructed by the multi-view fuzzy clustering model construction module. It updates the view weight cluster center v, view weight w, and fuzzy membership degree U in an alternating iterative manner until the multi-view fuzzy clustering model converges. The output module is used to output the multi-view fuzzy clustering results, which include soft membership degree U, view weight W, cluster center V and corresponding auxiliary diagnostic interpretation information.

9. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-7.

10. A computer device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the method of any one of claims 1-7.

Images