Multi-source heterogeneous medical data multi-view clustering method, device, medium and equipment

Abstract

The application discloses a multi-source heterogeneous medical data multi-view clustering method and device, a medium and equipment, the method comprises the following steps: obtaining original multi-source heterogeneous medical data; performing feature extraction on the data of each view to obtain hidden representations corresponding to each view; performing round-robin adversarial learning on the hidden representations to align the distribution of the hidden representations and to determine a plurality of new hidden representations; performing consistent hidden representation learning according to the new hidden representations to obtain consistent hidden representations; inputting the consistent hidden representations into a clustering layer to obtain clustering results output by the clustering layer; and the clustering layer is trained by using a clustering self-supervised loss function. The embodiment of the application solves the problem of difficult fusion caused by the misalignment of feature spaces of various views, can realize more effective feature fusion, obtain a more compact intra-class and sparse inter-class cluster structure, and finally improve the clustering effect of multi-source heterogeneous medical data.

Description

Technical Field

[0001] This invention relates to the field of multi-view clustering technology, and in particular to a method, apparatus, medium and equipment for multi-view clustering of multi-source heterogeneous medical data. Background Technology

[0002] The medical field contains a wide variety of data, often originating from different acquisition devices and appearing in different forms and structures. For example, CT images can show lesions, and blood analysis data can reflect the condition of the blood; this type of data is called multi-view data. These data describe the same sample from different perspectives, containing different information while always sharing the essential characteristics of the sample. Analyzing this medical data through multi-view clustering yields the cluster distribution of the samples, thereby enabling assistance in disease diagnosis, management of similar patients, and prognostic analysis.

[0003] In existing technologies, traditional multi-view clustering methods generally employ techniques such as subspace clustering, matrix factorization, and canonical correlation analysis followed by direct fusion operations to cluster multi-view data.

[0004] In existing technologies, since each view comes from a different domain, their feature spaces are often misaligned. Direct fusion will cause feature scale disorder. Furthermore, since each view is often not equally important, directly and evenly fusing all views will cause important information to be ignored and unimportant information to interfere with learning, thus affecting the clustering effect. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a multi-view clustering method, apparatus, medium, and device for multi-source heterogeneous medical data. This method can align the distribution of various views of multi-source heterogeneous data and employs a self-supervised clustering loss to guide feature learning, thereby enabling effective fusion of various views and achieving better clustering results.

[0006] To achieve the above objectives, embodiments of the present invention provide a multi-view clustering method for multi-source heterogeneous medical data, including:

[0007] Acquire raw, multi-source heterogeneous medical data; wherein, the raw, multi-source heterogeneous medical data includes several views;

[0008] Feature extraction is performed on the data of each view to obtain the feature of each view. Figure 1 One-to-one implicit representation;

[0009] The hidden representations are subjected to rotational adversarial learning to align their distribution and redetermine several new hidden representations.

[0010] Based on the new hidden representation, a consistent hidden representation is learned to obtain a consistent hidden representation.

[0011] The consistent implicit representation is input into the clustering layer to obtain the clustering result output by the clustering layer; wherein the clustering layer is trained using a clustering self-supervised loss function.

[0012] Furthermore, feature extraction is performed on the data of each view to obtain features related to each view. Figure 1 A corresponding implicit representation specifically includes: inputting each of the views to a corresponding view... Figure 1 One corresponding autoencoder network is used to obtain the hidden representation of the output of each autoencoder network.

[0013] Furthermore, the original multi-source heterogeneous medical data includes M views; where M ≥ 2;

[0014] The implicit representations are subjected to rotational adversarial learning to align their distribution and redetermine several new implicit representations. Specifically, this includes: selecting the v-th view as a reference view; where v = 1, 2, ..., M; inputting each implicit representation into a discriminator corresponding to the reference view to obtain an output result from the discriminator corresponding to each implicit representation; performing a calculation based on the output result and a preset rotational adversarial loss function to obtain a rotational adversarial loss result, and correcting the implicit representation based on the calculated rotational adversarial loss result; determining whether the rotational adversarial loss result satisfies a preset rotational adversarial learning condition; wherein the rotational adversarial learning condition includes that the rotational adversarial loss results obtained from the previous N calculations are all within a preset first stable interval, where N is a preset positive integer; if the rotational adversarial learning condition is satisfied, the corrected implicit representation is redetermined as a new implicit representation.

[0015] If the rotation adversarial learning condition is not met, then: if (v+1)≤M, then (v+1) is used as the new v to select a new reference view; if (v+1)>M, then (v+1-M) is used as the new v to select a new reference view; the new rotation adversarial loss result is recalculated based on the new reference view and the hidden representation is corrected until the rotation adversarial learning condition is met, so that all view distributions reach an aligned state before consistent representation learning is performed.

[0016] Furthermore, the rotational adversarial loss function is given by equation (1):

[0017]

[0018] Where G is the autoencoder network, D is the discriminator, M is the number of views, r is the reference view, i is the view other than the reference view, and z is the number of views.(v) Let E be the implicit representation corresponding to the v-th view, and let E be the mathematical expectation.

[0019] Furthermore, consistent hidden representation learning is performed based on the new hidden representation to obtain a consistent hidden representation, specifically including: along each of the views... Figure 1 For each corresponding feature dimension, the views are concatenated and input into a preset weight learning network to obtain the weight vector W output by the weight learning network; where W = (W1, W2, ..., W...). M ), and W t Let W be the weight of the t-th view; t Multiplying the implicit representation of the t-th view with the implicit representation of the t-th view yields zt; the consistent implicit representation is then calculated.

[0020] Furthermore, the method further includes: iteratively training the consistent implicit representation learning using a consistent implicit representation learning loss function until a preset consistent implicit representation learning condition is met, thereby obtaining a trained consistent implicit representation learning; wherein, the consistent implicit representation learning condition includes the number of iterative training iterations reaching a preset second threshold, or the value obtained by iterative calculation of the consistent implicit representation learning loss function being within a preset second stable interval;

[0021] The consistent latent representation learning loss function is given by equation (2):

[0022]

[0023] Where K() is an n×n matrix, n is the number of samples in each training iteration, and Z u For consistent implicit representation, W v Z is the weight of the t-th view. (v) For the implicit representation corresponding to the v-th view, The F-norm of a matrix;

[0024] Where, the element k in the i-th row and j-th column of the n-dimensional matrix K() i，j The calculation is as follows (3):

[0025] k i，j =exp(-||K()) i -K() j || 2 ) / 2σ 2 (3)

[0026] Where K()i is the i-th row of matrix K(), K() j Let σ be the j-th row of matrix K(), and let σ be a preset constant.

[0027] Furthermore, the clustering self-supervised loss function is given by equation (4):

[0028] min D CS +L ortho (4)

[0029] Among them, D CS For cluster constraints based on CS divergence, L ortho For clustering orthogonality constraints;

[0030]

[0031] L ortho =trisum(AA T );

[0032] Where C is the preset number of clusters, A is the clustering result, and α i Let A be the i-th row, trisum() be the sum of the elements of the upper triangular matrix, K() be an n×n matrix, and Z be the element of the upper triangular matrix. u This is a consistent implicit representation.

[0033] This invention also provides a multi-view clustering device for multi-source heterogeneous medical data, comprising:

[0034] The raw data acquisition module is used to acquire raw multi-source heterogeneous medical data; wherein, the raw multi-source heterogeneous medical data includes several views;

[0035] The feature extraction module is used to extract features from the data of each view to obtain features corresponding to each view. Figure 1 One-to-one implicit representation;

[0036] The rotating adversarial learning module is used to perform rotating adversarial learning on the hidden representations, align the distribution of the hidden representations, and redetermine several new hidden representations.

[0037] The consistent implicit representation learning module is used to learn the consistent implicit representation based on the new implicit representation to obtain the consistent implicit representation.

[0038] The clustering module is used to input the consistent implicit representation into the clustering layer to obtain the clustering result output by the clustering layer; wherein the clustering layer is trained using a clustering self-supervised loss function.

[0039] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the multi-view clustering method for multi-source heterogeneous medical data described in any of the preceding embodiments.

[0040] This invention also provides a computer device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the steps of the multi-source heterogeneous medical data multi-view clustering method described in any of the preceding claims.

[0041] In summary, the present invention has the following beneficial effects:

[0042] By employing embodiments of the present invention, rotational adversarial learning can be used to align the distributions of various views of multi-source heterogeneous medical data. A weighted learning network is then used to assign weights to the aligned features, solving the problem of difficulty in fusion caused by misalignment of feature spaces of various views. This enables more effective feature fusion and yields a consistent implicit representation. Furthermore, a self-supervised loss based on CS divergence and cluster orthogonality constraints is introduced to guide feature learning, resulting in a more compact intra-class and sparse inter-class cluster structure. Ultimately, this improves the clustering effect of multi-source heterogeneous medical data. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating an embodiment of a multi-view clustering method for multi-source heterogeneous medical data provided by the present invention;

[0044] Figure 2 This is a schematic diagram of an embodiment of a multi-view clustering device for multi-source heterogeneous medical data provided by the present invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] See Figure 1 This is a flowchart illustrating an embodiment of the multi-source heterogeneous medical data multi-view clustering method provided by the present invention. The method includes steps S1 to S5, as follows:

[0047] S1, acquire raw multi-source heterogeneous medical data; wherein, the raw multi-source heterogeneous medical data includes several views;

[0048] S2, perform feature extraction on the data of each view to obtain the feature of each view. Figure 1 One-to-one implicit representation;

[0049] Preferably, the step of extracting features from the data of each view to obtain a feature that matches the data of each view. Figure 1A corresponding implicit representation specifically includes: inputting each of the views to a corresponding view... Figure 1 One corresponding autoencoder network is used to obtain the hidden representation of the output of each autoencoder network.

[0050] For example, the original data X is processed by an autoencoder network where each view is independent to extract features, resulting in the latent representation Z = {Z...} for each view. (1) , ..., Z (v)}, v = 1, 2, ..., M.

[0051] S3, perform rotational adversarial learning on the hidden representations to align the distribution of the hidden representations and redetermine several new hidden representations;

[0052] Preferably, the original multi-source heterogeneous medical data includes M views; wherein, M≥2;

[0053] The implicit representations are subjected to rotational adversarial learning to align their distribution and redetermine several new implicit representations. Specifically, this includes: selecting the v-th view as a reference view; where v = 1, 2, ..., M; inputting each implicit representation into a discriminator corresponding to the reference view to obtain an output result from the discriminator corresponding to each implicit representation; performing a calculation based on the output result and a preset rotational adversarial loss function to obtain a rotational adversarial loss result, and correcting the implicit representation based on the calculated rotational adversarial loss result; determining whether the rotational adversarial loss result satisfies a preset rotational adversarial learning condition; wherein the rotational adversarial learning condition includes that the rotational adversarial loss results obtained from the previous N calculations are all within a preset first stable interval, where N is a preset positive integer; if the rotational adversarial learning condition is satisfied, the corrected implicit representation is redetermined as a new implicit representation.

[0054] If the rotation adversarial learning condition is not met, then: if (v+1)≤M, then (v+1) is used as the new v to select a new reference view; if (v+1)>M, then (v+1-M) is used as the new v to select a new reference view; the new rotation adversarial loss result is recalculated based on the new reference view and the hidden representation is corrected until the rotation adversarial learning condition is met, so that all view distributions reach an aligned state before consistent representation learning is performed.

[0055] As an improvement to the above scheme, the rotational adversarial loss function is given by equation (1):

[0056]

[0057] Where G is the autoencoder network, D is the discriminator, M is the number of views, r is the reference view, i is the view other than the reference view, and z is the number of views. (v) Let E be the implicit representation corresponding to the v-th view, and let E be the mathematical expectation.

[0058] For example, alternate adversarial learning is performed on the latent representations Z of each view to feed the learning results back into the previous latent representation generation process. The specific learning process is as follows: ① Select view v as the reference view, and the other views as fake views (initially v is the first). ② Use the latent representations of all views as input to the v-th discriminator, and obtain a numerical output in the interval [0, 1]. A discriminator output close to 0 indicates that it considers the input to be a fake view, while a output close to 1 indicates it is a reference view. ③ Switch the reference view to the next view v+1. If v+1 is greater than the total number of views M, take the remainder of v+1 divided by M. Repeat steps ①-③ until the alternate adversarial learning process is complete.

[0059] It should be noted that this application can run on current mainstream deep learning frameworks, which can automatically calculate the gradients based on the corresponding modules (such as generators and discriminators) according to the loss function, and then use gradient descent to update the parameters of the modules. When the discriminator outputs a number, the corresponding alternating adversarial loss function will calculate and output the corresponding value, which will then update the parameters of the corresponding modules and ultimately affect the generation of the hidden representation.

[0060] S4. Based on the new hidden representation, perform consistent hidden representation learning to obtain a consistent hidden representation;

[0061] Preferably, consistent hidden representation learning is performed based on the new hidden representation to obtain a consistent hidden representation, specifically including: along each of the views Figure 1 For each corresponding feature dimension, the views are concatenated and input into a preset weight learning network to obtain the weight vector W output by the weight learning network; where W = (W1, W2, ..., W...). M ), and W t Let W be the weight of the t-th view; t Multiplying the implicit representation of the t-th view with the implicit representation of the t-th view yields zt; the consistent implicit representation is then calculated.

[0062] As an improvement to the above scheme, the method further includes: iteratively training the consistent implicit representation learning using a consistent implicit representation learning loss function until a preset consistent implicit representation learning condition is met, thereby obtaining a trained consistent implicit representation learning; wherein, the consistent implicit representation learning condition includes the number of iterations reaching a preset second threshold, or the value obtained by iterative calculation of the consistent implicit representation learning loss function being within a preset second stable interval; wherein, the consistent implicit representation learning loss function is Equation (2):

[0063]

[0064] Where K() is an n×n matrix, n is the number of samples in each training iteration, and Z u For consistent implicit representation, W v Z is the weight of the t-th view. (v) For the implicit representation corresponding to the v-th view, The F-norm of a matrix;

[0065] Where, the element k in the i-th row and j-th column of the n-dimensional matrix K() i，j The calculation is as follows (3):

[0066] k i，j =exp(-||K()) i -K() j || 2 ) / 2σ 2 (3)

[0067] Among them, K() i Let K() be the i-th row of matrix K(). j Let σ be the j-th row of matrix K(), and let σ be a preset constant.

[0068] For example, consistent latent representation learning is performed on the latent representation Z of each view. The specific learning process is as follows: ① The views are concatenated according to their feature dimensions and used as the input to the weight learning network (Attention layer), i.e., Z. con ② The weighted learning network outputs an M-dimensional vector W, where a certain dimension W is a weighted vector. v The value is in the range [0, 1] and represents the importance of the v-th view after being measured by the network. ③ Each view W v With Z (v) The corresponding products are multiplied and summed to obtain the final consistent implicit representation Z. u .

[0069] S5, the consistent implicit representation is input into the clustering layer to obtain the clustering result output by the clustering layer; wherein, the clustering layer is trained using a clustering self-supervised loss function.

[0070] Preferably, the clustering self-supervised loss function is Equation (4):

[0071] min D cs +L ortho (4)

[0072] Among them, D CS For cluster constraints based on CS divergence, L ortho For clustering orthogonality constraints;

[0073]

[0074] L ortho =trisum(AA T );

[0075] Where C is the preset number of clusters, A is the clustering result, and α i Let A be the i-th row, trisum() be the sum of the elements of the upper triangular matrix, K() be an n×n matrix, and Z be the element of the upper triangular matrix. u This is a consistent implicit representation.

[0076] Accordingly, embodiments of the present invention also provide a multi-source heterogeneous medical data multi-view clustering device, which can implement all the processes of the multi-source heterogeneous medical data multi-view clustering method provided in the above embodiments.

[0077] See Figure 2 This is a schematic diagram of an embodiment of the multi-source heterogeneous medical data multi-view clustering device provided by the present invention.

[0078] The raw data acquisition module 101 is used to acquire raw multi-source heterogeneous medical data; wherein, the raw multi-source heterogeneous medical data includes several views;

[0079] Feature extraction module 102 is used to extract features from the data of each view to obtain features related to each view. Figure 1 One-to-one implicit representation;

[0080] The rotating adversarial learning module 103 is used to perform rotating adversarial learning on the hidden representations, align the distribution of the hidden representations, and redetermine several new hidden representations.

[0081] The consistent implicit representation learning module 104 is used to perform consistent implicit representation learning based on the new implicit representation to obtain a consistent implicit representation.

[0082] Clustering module 105 is used to input the consistent implicit representation into the clustering layer to obtain the clustering result output by the clustering layer; wherein the clustering layer is trained using a clustering self-supervised loss function.

[0083] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the multi-view clustering method for multi-source heterogeneous medical data described in any of the preceding embodiments.

[0084] This invention also provides a computer device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the steps of the multi-source heterogeneous medical data multi-view clustering method described in any of the preceding claims.

[0085] The computer device in this embodiment includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a multi-source heterogeneous medical data multi-view clustering program. When the processor executes the computer program, it implements the steps in the various embodiments of the multi-source heterogeneous medical data multi-view clustering method described above, for example... Figure 1 The steps S1 to S5 are shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as steps 101 to 105.

[0086] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the computer device.

[0087] The computer device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device may include, but is not limited to, a processor and memory.

[0088] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the computer device, connecting various parts of the computer device via various interfaces and lines.

[0089] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0090] Wherein, if the modules / units integrated into the computer device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0091] In summary, the present invention has the following beneficial effects:

[0092] By employing embodiments of the present invention, rotational adversarial learning can be used to align the distributions of various views of multi-source heterogeneous medical data. A weighted learning network is then used to assign weights to the aligned features, solving the problem of difficulty in fusion caused by misalignment of feature spaces of various views. This enables more effective feature fusion and yields a consistent implicit representation. Furthermore, a self-supervised loss based on CS divergence and cluster orthogonality constraints is introduced to guide feature learning, resulting in a more compact intra-class and sparse inter-class cluster structure. Ultimately, this improves the clustering effect of multi-source heterogeneous medical data.

[0093] Through the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary hardware platforms, and of course, it can also be implemented entirely by hardware. Based on this understanding, all or part of the technical solution of the present invention that contributes to the background art can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

[0094] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A multi-view clustering method for multi-source heterogeneous medical data, characterized in that, include: Acquire raw, multi-source heterogeneous medical data; wherein, the raw, multi-source heterogeneous medical data includes several views; Feature extraction is performed on the data of each view to obtain a hidden representation that corresponds one-to-one with each view; The hidden representations are subjected to rotational adversarial learning to align their distribution and redetermine several new hidden representations. Based on the new hidden representation, a consistent hidden representation is learned to obtain a consistent hidden representation. The consistent implicit representation is input into the clustering layer to obtain the clustering result output by the clustering layer; wherein, the clustering layer is trained using a clustering self-supervised loss function; The original multi-source heterogeneous medical data includes One view; among which ; The hidden representations are subjected to rotational adversarial learning to align their distribution and determine several new hidden representations, specifically including: Select the first One view serves as a reference view; among them... ; Each of the hidden representations is input to a discriminator corresponding to the reference view, and the discriminator outputs a result that corresponds one-to-one with each of the hidden representations. The implicit representation is then modified based on the output result and the preset rotation adversarial loss function. Determine whether the rotation adversarial loss result meets the preset rotation adversarial learning conditions; wherein, the rotation adversarial learning conditions include previous... The calculated rotational combat loss results are all within the preset first stability interval. The preset positive integer; If the aforementioned rotating adversarial learning conditions are met, the modified hidden representation will be redefined as the new hidden representation. If the aforementioned conditions for rotational adversarial learning are not met, then: like Then As a new To select a new reference view; like Then As a new To select a new reference view; The new rotation adversarial loss result is recalculated based on the new reference view, and the hidden representation is corrected until the rotation adversarial learning condition is met, so that the distribution of all views reaches an aligned state before consistent representation learning is performed.

2. The multi-view clustering method for multi-source heterogeneous medical data as described in claim 1, characterized in that, The step of extracting features from the data of each view to obtain a hidden representation corresponding to each view specifically includes: Each of the views is input into an autoencoder network corresponding to each view, to obtain the hidden representation output by each autoencoder network.

3. The multi-view clustering method for multi-source heterogeneous medical data as described in claim 1, characterized in that, The rotational adversarial loss function is given by equation (1): ； （1） in, For autoencoder networks, For discriminator, The number of views. For reference view, For views other than the reference view, In order to be with the first The implicit representation corresponding to each view. Let be the mathematical expectation.

4. The multi-view clustering method for multi-source heterogeneous medical data as described in claim 1, characterized in that, Based on the new hidden representation, a consistent hidden representation is learned to obtain a consistent hidden representation, specifically including: The views are concatenated along their respective feature dimensions and then input into a preset weight learning network to obtain the weight vector output by the weight learning network. ;in, ,and For the first The weight of each view; Will and the Multiplying the implicit representations of each view yields... ; The consistent implicit representation is calculated. .

5. The multi-view clustering method for multi-source heterogeneous medical data as described in claim 4, characterized in that, The method further includes: The consistent implicit representation learning loss function is used to iteratively train the consistent implicit representation learning until the preset consistent implicit representation learning conditions are met, and the trained consistent implicit representation learning is obtained; wherein, the consistent implicit representation learning conditions include the number of iterations reaching a preset second threshold, or the value obtained by the consistent implicit representation learning loss function through iterative calculation is within a preset second stable interval; Wherein, the consistent latent representation learning loss function is Equation (2): ；（2） in, It is a dimension of The matrix, The number of samples during each training iteration. This is a consistent implicit representation. For the first The weight of each view, In order to be with the first The implicit representation corresponding to each view. Represents the matrix Norm; in, 3D matrix The Line number Column elements The calculation is as follows (3): ；（3） in, For matrix The OK, For matrix The OK, This is a preset constant.

6. The multi-view clustering method for multi-source heterogeneous medical data as described in claim 1, characterized in that, The clustering self-supervised loss function is given by equation (4): ；（4） in, For cluster constraints based on CS divergence, For clustering orthogonality constraints; ； ； in, The preset number of clusters, For the clustering results, for The OK, Let be the sum of the elements of the upper triangular matrix. It is a dimension of The matrix, This is a consistent implicit representation.

7. A multi-view clustering device for multi-source heterogeneous medical data, characterized in that, include: The raw data acquisition module is used to acquire raw multi-source heterogeneous medical data; wherein, the raw multi-source heterogeneous medical data includes several views; The feature extraction module is used to extract features from the data of each view to obtain a hidden representation that corresponds one-to-one with each view. The rotating adversarial learning module is used to perform rotating adversarial learning on the hidden representations, align the distribution of the hidden representations, and redetermine several new hidden representations. The consistent implicit representation learning module is used to learn the consistent implicit representation based on the new implicit representation to obtain the consistent implicit representation. A clustering module is used to input the consistent implicit representation into a clustering layer to obtain the clustering result output by the clustering layer; wherein, the clustering layer is trained using a clustering self-supervised loss function; The original multi-source heterogeneous medical data includes One view; among which ; The hidden representations are subjected to rotational adversarial learning to align their distribution and determine several new hidden representations, specifically including: Select the first One view serves as a reference view; among them... ; Each of the hidden representations is input to a discriminator corresponding to the reference view, and the discriminator outputs a result that corresponds one-to-one with each of the hidden representations. The implicit representation is then modified based on the output result and the preset rotation adversarial loss function. Determine whether the rotation adversarial loss result meets the preset rotation adversarial learning conditions; wherein, the rotation adversarial learning conditions include previous... The calculated rotational combat loss results are all within the preset first stability interval. The preset positive integer; If the aforementioned rotating adversarial learning conditions are met, the modified hidden representation will be redefined as the new hidden representation. If the aforementioned conditions for rotational adversarial learning are not met, then: like Then As a new To select a new reference view; like Then As a new To select a new reference view; The new rotation adversarial loss result is recalculated based on the new reference view, and the hidden representation is corrected until the rotation adversarial learning condition is met, so that the distribution of all views reaches an aligned state before consistent representation learning is performed.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the multi-view clustering method for multi-source heterogeneous medical data as described in any one of claims 1 to 6.

9. A computer device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the multi-view clustering method for multi-source heterogeneous medical data as described in any one of claims 1 to 6.

Images