An alzheimer disease auxiliary diagnosis method based on cross-modal feature reuse

Abstract

The application provides an Alzheimer's disease auxiliary diagnosis method based on cross-modal feature reuse, and belongs to the technical field of medical data processing. The application comprises the following steps: acquiring MRI image data of a diagnosis object; extracting high-dimensional feature representation from the MRI image, separating the high-dimensional feature representation into shared features and MRI private features, and mapping the MRI private features into PET private features through feature conversion; performing spatial self-adaptive fusion on the shared features and the PET private features, retaining local structure information and global semantic information, and obtaining a reconstructed PET image; performing classification based on the shared features, the PET private features and the MRI private features, and obtaining an auxiliary diagnosis result; and outputting the reconstructed PET image and the auxiliary diagnosis result. Under the condition of missing PET modal, high-quality PET image reconstruction and accurate disease diagnosis are realized, task interference is effectively reduced, and model training stability and diagnosis accuracy are improved.

Description

Technical Field

[0001] This invention belongs to the field of medical data processing technology, specifically relating to an auxiliary diagnostic method for Alzheimer's disease based on cross-modal feature reuse. Background Technology

[0002] Alzheimer's disease (AD) is a common neurodegenerative disease, and early diagnosis is crucial for slowing disease progression and developing intervention strategies. Medical imaging technology is one of the important tools for assisting in the diagnosis of Alzheimer's disease. Among them, magnetic resonance imaging (MRI) provides information on brain structure, while positron emission tomography (PET) reflects changes in brain metabolism and function. Studies have shown that joint analysis of MRI and PET information helps improve the ability to differentiate between different types of Alzheimer's disease.

[0003] However, in practical clinical applications, PET imaging equipment is expensive, examination costs are high, and there are certain radiation risks, making PET image data difficult to obtain in most medical institutions. This limits the widespread application of PET images in the auxiliary diagnosis of Alzheimer's disease. To meet the needs of image visualization and disease discrimination in the absence of PET, existing technologies attempt to construct a multi-task learning framework, hoping to simultaneously achieve PET image reconstruction and Alzheimer's disease classification diagnosis.

[0004] However, when existing technologies attempt to simultaneously complete multi-task models of generation and diagnosis, they typically treat the two tasks as independent branches, extracting independent features for each. This process lacks effective feature sharing and information interaction, severing the inherent connection between the generation and classification tasks at the feature level, which in turn leads to a disconnect between the quality of modality conversion and the results of auxiliary diagnosis. Summary of the Invention

[0005] To address the shortcomings of existing technologies in the auxiliary diagnosis of Alzheimer's disease, this invention provides an auxiliary diagnostic method for Alzheimer's disease based on cross-modal feature reuse.

[0006] To achieve the above objectives, the present invention provides the following technical solution: An Alzheimer's disease auxiliary diagnostic method based on cross-modal feature reuse includes the following steps: Acquire MRI image data of the diagnostic subject; High-dimensional feature representations containing pathological semantics are extracted from the MRI images. A constraint based on minimizing mutual information is introduced to separate the extracted high-dimensional features by channel, outputting shared features characterizing common pathological conditions across modalities and MRI-specific texture structures. Feature transformation maps the MRI-specific features to PET-specific features. Spatial affine parameters are learned from the PET-specific features, and the shared features are calibrated using these parameters. Spatial adaptive fusion of the PET-specific features and the calibrated shared features is then performed to obtain the reconstructed PET image. Disease-aided classification prediction probabilities are obtained based on the shared features, PET-specific features, and MRI-specific features. Based on the reconstructed PET images and the disease-aided classification prediction probabilities, auxiliary diagnostic results are obtained.

[0007] Preferably, four consecutive residual convolutional units are used to extract high-dimensional feature representations containing pathological semantics from the MRI image; each residual convolutional unit contains a 3×3 convolutional layer, an instance normalization layer and a ReLU activation function, and downsampling is achieved between the residual convolutional units through max pooling.

[0008] Preferably, spatial affine parameters are learned from the PET private features, and the shared features are calibrated using the spatial affine parameters, specifically: The scaling and translation factors of the spatial affine transformation are learned from the aligned PET private features using a multilayer perceptron (MLP). The shared features are then normalized by instance, and the normalized shared features are subjected to an affine transformation based on the scaling and translation factors to obtain the calibrated shared features.

[0009] Preferably, the PET private features and the calibrated shared features are spatially adaptively fused to obtain the reconstructed PET image. Specifically, the PET private features and the calibrated shared features are input into two residual convolutional units to perform conventional upsampling. Between the residual units, the size of the feature map is gradually restored to the original spatial dimension through transpose convolution to obtain the reconstructed PET image.

[0010] Preferably, a constraint based on minimizing mutual information is introduced to separate the extracted high-dimensional features by channel, outputting shared features characterizing common pathologies across modalities and MRI-specific texture structures, specifically: The extracted high-dimensional features are split into two parts along the channel dimension. The two split features are projected onto a unified embedding space using a learnable mapping layer. The mutual information is minimized by minimizing the cosine similarity between the two features, resulting in shared features and MRI private features.

[0011] Preferably, the disease-aided classification prediction probability is obtained based on the shared features, PET private features, and MRI private features, specifically: The shared features, MRI private features, and PET private features are respectively input into a multilayer perceptron mechanism and projected onto their respective codebook spaces during the mapping process to achieve feature enhancement. The shared features are used as queries, and the features obtained by concatenating the PET private features and MRI private features along the channel dimension are used as keys and values, respectively. The discriminative information in the two private features is aggregated by a cross-attention mechanism to obtain fused features. The fused features are then passed through a global average pooling operation and a fully connected layer to obtain the disease auxiliary classification prediction probability.

[0012] Preferably, the MRI private features are processed sequentially using three consecutive convolutional layers, instance normalization, and ReLU activation function to perform feature transformation and map the MRI private features to PET private features.

[0013] This invention also provides an Alzheimer's disease auxiliary diagnostic system based on cross-modal feature reuse, specifically including: The data module is used to acquire MRI image data of the diagnostic subject.

[0014] The auxiliary diagnostic module is used to extract high-dimensional feature representations containing pathological semantics from the MRI image. It introduces a constraint based on minimizing mutual information to separate the extracted high-dimensional features by channel, outputting shared features characterizing common pathological conditions across modalities and MRI-specific features characterizing modal texture structures. The MRI-specific features are mapped to PET-specific features through feature transformation. Spatial affine parameters are learned from the PET-specific features, and the shared features are calibrated using these parameters. The PET-specific features and the calibrated shared features are then spatially adaptively fused to obtain a reconstructed PET image. Finally, the disease auxiliary classification prediction probability is obtained based on the shared features, PET-specific features, and MRI-specific features. The auxiliary diagnostic results are obtained based on the reconstructed PET images and the disease auxiliary classification prediction probabilities.

[0015] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps described in the Alzheimer's disease auxiliary diagnosis method based on cross-modal feature reuse.

[0016] The present invention also provides a computer-readable storage medium storing a computer program that, when loaded by a processor, is capable of executing the steps described in the Alzheimer's disease auxiliary diagnosis method based on cross-modal feature reuse.

[0017] The Alzheimer's disease auxiliary diagnosis method based on cross-modal feature reuse provided by this invention has the following beneficial effects: This invention introduces a constraint based on minimizing mutual information to separate extracted high-dimensional features by channel, decoupling MRI features into shared features carrying global anatomical structure and disease commonalities, and private features carrying modality-specific texture details. Subsequent image generation and auxiliary diagnostic classification share these decoupled features. One approach calibrates the shared features by learning channel-level affine transformation parameters based on the PET private features, injecting PET-specific information into the shared features that preserve global anatomical structure, generating high-quality reconstructed PET images with both structural fidelity and texture realism. The other approach, based on the mined modality feature information, losslessly transfers the dependent global pathological features to the diagnostic task, obtaining disease auxiliary classification prediction probabilities. In clinical decision-making, the output auxiliary classification prediction probabilities are cross-validated with the visualized reconstructed PET images, enhancing the interpretability and credibility of the auxiliary diagnosis. Attached Figure Description

[0018] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is an overall architecture diagram of the auxiliary diagnostic model in an embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram of the spatial adaptive fusion method in the PET generation module of this invention.

[0021] Figure 3 This is a structural diagram of the feature reuse module in an embodiment of the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0023] Example This invention provides an auxiliary diagnostic method for Alzheimer's disease based on cross-modal feature reuse, such as... Figure 1 As shown, the specific steps include: Step 1: Data Acquisition. Neuroimaging data, including MRI and PET data, from the Alzheimer's Disease Neuroimaging Initiative (ADNI), the Australian Imaging, Biomarkers and Lifestyle Flagship Study of Aging (AIBL), and the National Alzheimer's Coordinating Center (NACC) were used for subsequent cross-modal imaging analysis and Alzheimer's disease diagnostic validation. Data selection criteria included: ① only one scan record per subject; ② the time interval between MRI and PET scans for the same subject did not exceed six months. Following these criteria, 498 paired T1 and FDG-PET scans were obtained from the ADNI database, 147 paired T1 and ¹¹C-PIB-PET scans from the AIBL database, and 216 paired T1 and AV1451-PET scans from the NACC database.

[0024] Step 2: Data preprocessing.

[0025] (1) All MRI data were normalized for orientation using FSL software. Subsequently, the skull was dissected using the hd-bet tool. Finally, the MRI data were biased for field correction using ANTs software and registered to the MNI152 standard brain template space.

[0026] (2) For all PET data, orientation normalization was first performed using FSL. Then, ANTs software was used to register the PET data to the MNI152 standard brain template space.

[0027] (3) To balance computational efficiency, both MRI and PET images were resampled to a size of 160×160×160 voxels.

[0028] Step 3: Construct an auxiliary diagnostic model that includes a feature extraction module, a feature decoupling and transformation module, a PET generation module, a feature reuse module, and an auxiliary diagnostic module.

[0029] The feature extraction module extracts high-dimensional feature representations from MRI images; the feature decoupling and transformation module decomposes the MRI feature representation into MRI-specific and shared features, and maps the MRI-specific features to PET-specific features through feature transformation; the PET generation module reconstructs PET images using shared and PET-specific features as input; the feature reuse module embeds the MRI and PET-specific and shared features to provide discriminative features for the auxiliary diagnosis module; the auxiliary diagnosis module classifies the three types of features output by the feature reuse module, uses a cross-attention mechanism to achieve feature fusion, and obtains auxiliary diagnostic results. This enables preliminary diagnosis of normal subjects (NC), mild cognitive impairment (MCI), and Alzheimer's disease (AD) under missing modalities.

[0030] Step 4: Process the subject's MRI image data using an auxiliary diagnostic model to obtain reconstructed PET images and auxiliary diagnostic results. This includes the following steps: In the feature extraction module, a CNN consisting of four consecutive residual convolutional units, each containing a 3×3 convolutional layer, an instance normalization layer, and a ReLU activation function, is used to extract high-dimensional feature representations from MRI images. The number of convolutional kernels is 32, 64, 128, and 256, respectively; downsampling is achieved between the residual convolutional units through max pooling. The high-dimensional feature representations serve as input to the feature decoupling and transformation module.

[0031] In the feature decoupling and transformation module, high-dimensional features are separated into shared features and MRI-private features by channel. Shared features represent information such as brain anatomical structures, spatial layout, and global disease-related patterns with cross-modal consistency, while MRI-private features contain information such as tissue contrast, structural texture, and boundary details from MRI. To ensure their independence, a constraint based on minimizing mutual information is introduced into the feature space, forcing the private and shared features to be orthogonal. Specifically, a learnable mapping layer is used to map the decomposed features to a unified embedding space. By minimizing the cosine similarity between shared and private features (Equation 1), the two types of features remain orthogonal in the representation space, reducing statistical correlation and achieving semantic-level feature separation. Subsequently, a feature transformation module consisting of three consecutive convolutional layers, instance normalization, and the RULU activation function maps the MRI-private features to PET-private features. Shared features, MRI-private features, and PET-private features serve as inputs to the subsequent PET generation module, feature reuse module, and auxiliary diagnostic module.

[0032] (1); in, and These are the projected representations of shared features and private features, respectively. This constraint ensures that the two types of features remain orthogonal in the representation space, reducing their statistical correlation and thus promoting effective separation of features at the semantic level.

[0033] The PET generation module takes shared features and PET private features as input. During the first two layers of upsampling, the PET private features undergo adaptive spatial-aware style calibration on the shared features. First, the PET private features... With shared features Adaptive size matching and alignment are performed to maintain spatial dimension consistency between the two features. Subsequently, a multilayer perceptron (MLP) is used to learn the scaling factor γ and translation factor β of the spatial affine transformation from the aligned PET private features. Simultaneously, instance normalization is applied to the shared features. Finally, channel-level style calibration is performed on the shared features based on the scaling factor γ, translation factor β, and instance-normalized shared features, such as... Figure 2 As shown, the PET generation module simultaneously preserves local structural information and global semantic information, thereby generating PET images that retain the complete 3D spatial context. The module consists of four consecutive residual convolutional units and one mapping unit. Each residual convolutional unit includes a 3×3 convolutional layer, an instance normalization layer, and a ReLU activation function, with the number of convolutional kernels being 256, 128, 64, and 32 respectively. Upsampling is achieved between residual convolutional units through transposed convolution. The mapping unit is a 1×1 convolutional layer used to generate the final PET image. During training, the PET generation module is optimized using an independent loss function and its gradient is separated from that of the auxiliary diagnostic module to ensure training stability and reduce inter-task interference.

[0034] The feature reuse module is used to achieve functional coupling between the PET generation module and the auxiliary diagnostic module, while maintaining independent optimization of each task. This module receives shared features, PET-specific features, and MRI-specific features as input, and processes these three types of features independently, such as... Figure 3 As shown in the diagram. Specifically, the three types of features are input into their respective multilayer perceptrons (MLPs) and projected into their respective codebook spaces during the mapping process, achieving high-dimensional representation and semantic enhancement of the features. By mapping private and shared features, discriminability is maintained in the embedding space, while reducing redundant information and improving the reusability and stability of the features.

[0035] The auxiliary diagnosis module receives shared enhancement features, PET private enhancement features, and MRI private enhancement features output by the feature reuse module as input, and performs cross-attention fusion on the three types of features. Specifically, the shared enhancement features serve as the Query, and the fused representations of the PET private enhancement features and MRI private enhancement features serve as the Key and Value, respectively. Under the action of the cross-attention mechanism formula (2), information interaction and weighted fusion between features are completed, enabling the shared enhancement features to aggregate discriminative information from the private features while retaining their own global semantic information.

[0036] (2); in, This represents the fused features, and softmax(·) denotes the softmax function. It is a shared feature. and These are features resulting from splicing PET-specific enhancement features and MRI-specific enhancement features along the channel. It is the dimension of the key vector, and T represents the matrix transpose.

[0037] The fused features are compressed into vector representations using global average pooling to facilitate subsequent classification. These vectors are then input into a fully connected layer and pass through a softmax function to obtain the probability of each sample belonging to a different disease. Finally, the label corresponding to the highest probability is taken as the final classification result, thus enabling the prediction of classification probabilities for normal subjects (NC), mild cognitive impairment (MCI), and Alzheimer's disease (AD), thereby assisted in diagnosis.

[0038] During model training, the auxiliary diagnosis module is optimized using an independent loss function, and the gradient and PET generation modules are separated to ensure that the two tasks do not interfere with each other. Information transfer and collaboration are achieved through a feature reuse module. Based on cross-attention, the importance of different modalities can be dynamically adjusted according to the feature fusion results, thereby maintaining high diagnostic accuracy even under conditions of missing modalities.

[0039] This invention also provides an Alzheimer's disease auxiliary diagnostic system based on cross-modal feature reuse, comprising: The data module is used to acquire MRI image data of the diagnostic subject.

[0040] The auxiliary diagnostic module extracts high-dimensional feature representations containing pathological semantics from MRI images. It introduces a constraint based on minimizing mutual information to separate the extracted high-dimensional features by channel, outputting shared features characterizing common pathological conditions across modalities and MRI-specific texture structures. Feature transformation maps the MRI-specific features to PET-specific features. Spatial affine parameters are learned from the PET-specific features, and the shared features are calibrated using these parameters. Spatial adaptive fusion of the PET-specific features and the calibrated shared features yields a reconstructed PET image. Based on the shared features, PET-specific features, and MRI-specific features, the probability of disease-aided classification prediction is obtained. Auxiliary diagnostic results are obtained based on reconstructed PET images and disease-aided classification prediction probabilities.

[0041] The modules in the aforementioned Alzheimer's disease assisted diagnostic system based on cross-modal feature reuse can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0042] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps in an embodiment of an Alzheimer's disease auxiliary diagnosis method based on cross-modal feature reuse. Specific implementation methods can be found in the method embodiments, and will not be repeated here.

[0043] Furthermore, the present invention also provides a non-transitory computer-readable storage medium containing instructions on which a computer program is stored. For example, a memory containing instructions that can be executed by a processor of a computer device to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc. When the computer program is executed by the processor, it can implement the steps in an embodiment of an Alzheimer's disease auxiliary diagnosis method based on cross-modal feature multiplexing. Specific implementation methods can be found in the method embodiments, which will not be repeated here.

[0044] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0045] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0046] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0047] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0048] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the present invention patent. No reference numerals in the claims should be construed as limiting the scope of the claims. Any simple variations or equivalent substitutions of technical solutions that can be readily obtained by those skilled in the art within the scope of the technology disclosed in the present invention are within the protection scope of the present invention.

Claims

1. An auxiliary diagnostic method for Alzheimer's disease based on cross-modal feature reuse, characterized in that, Includes the following steps: Acquire MRI image data of the diagnostic subject; High-dimensional feature representations containing pathological semantics are extracted from the MRI images. Constraints based on minimizing mutual information are introduced to separate the extracted high-dimensional features by channel. Shared features representing common pathologies across modalities and MRI private features representing modal texture structures are output. MRI private features are mapped to PET private features through feature transformation. Spatial affine parameters are learned from the PET private features, and the shared features are calibrated using the spatial affine parameters. Spatial adaptive fusion is performed on the PET private features and the calibrated shared features to obtain a reconstructed PET image. Disease auxiliary classification prediction probability is obtained based on the shared features, PET private features, and MRI private features. The auxiliary diagnostic results are obtained based on the reconstructed PET images and the disease auxiliary classification prediction probabilities.

2. The Alzheimer's disease auxiliary diagnostic method based on cross-modal feature reuse according to claim 1, characterized in that, High-dimensional feature representations containing pathological semantics are extracted from the MRI image using four consecutive residual convolutional units; each residual convolutional unit contains a 3×3 convolutional layer, an instance normalization layer, and a ReLU activation function, and downsampling is achieved between the residual convolutional units through max pooling.

3. The Alzheimer's disease auxiliary diagnostic method based on cross-modal feature reuse according to claim 1, characterized in that, The shared features are calibrated using spatial affine parameters learned from the PET private features, specifically: The scaling and translation factors of the spatial affine transformation are learned from the aligned PET private features using a multilayer perceptron (MLP). The shared features are then normalized by instance, and the normalized shared features are subjected to an affine transformation based on the scaling and translation factors to obtain the calibrated shared features.

4. The Alzheimer's disease auxiliary diagnostic method based on cross-modal feature reuse according to claim 1, characterized in that, Spatial adaptive fusion is performed on the PET private features and the calibrated shared features to obtain the reconstructed PET image. Specifically, the PET private features and the calibrated shared features are input into two residual convolutional units to perform conventional upsampling. Between the residual units, the size of the feature map is gradually restored to the original spatial dimension through transpose convolution to obtain the reconstructed PET image.

5. The method for auxiliary diagnosis of Alzheimer's disease based on cross-modal feature reuse according to claim 1, characterized in that, By introducing a constraint based on minimizing mutual information, the extracted high-dimensional features are separated by channel, outputting shared features representing common pathological conditions across modalities and MRI-specific texture structures. Specifically: The extracted high-dimensional features are split into two parts along the channel dimension. The two split features are projected onto a unified embedding space using a learnable mapping layer. The mutual information is minimized by minimizing the cosine similarity between the two features, resulting in shared features and MRI private features.

6. The Alzheimer's disease auxiliary diagnosis method based on cross-modal feature reuse according to claim 1, characterized in that, Based on the shared features, PET private features, and MRI private features, the disease-aided classification prediction probability is obtained, specifically: The shared features, MRI private features, and PET private features are respectively input into a multilayer perceptron mechanism and projected into their respective codebook spaces during the mapping process to achieve feature enhancement. The shared features are used as queries, and the features obtained by concatenating the PET private features and MRI private features along the channel dimension are used as keys and values, respectively. The discriminative information in the two private features is aggregated by the shared features through a cross-attention mechanism to obtain fused features. The fused features are then passed through a global average pooling operation and a fully connected layer to obtain the disease-aided classification prediction probability.

7. The Alzheimer's disease auxiliary diagnostic method based on cross-modal feature reuse according to claim 5, characterized in that, The MRI private features are processed sequentially using three consecutive convolutional layers, instance normalization, and ReLU activation function to perform feature transformation and map the MRI private features to PET private features.

8. An Alzheimer's disease auxiliary diagnostic system based on cross-modal feature reuse, characterized in that, include: The data module is used to acquire MRI image data of the diagnostic subject; The auxiliary diagnostic module is used to extract high-dimensional feature representations containing pathological semantics from the MRI images. It introduces a constraint based on minimizing mutual information to separate the extracted high-dimensional features by channel, outputs shared features representing common pathologies across modalities and MRI private features representing modal texture structures, and maps MRI private features to PET private features through feature transformation. Spatial affine parameters are learned from the PET private features, and the shared features are calibrated using the spatial affine parameters. Spatial adaptive fusion is performed on the PET private features and the calibrated shared features to obtain a reconstructed PET image. Disease auxiliary classification prediction probability is obtained based on the shared features, PET private features, and MRI private features. The auxiliary diagnostic results are obtained based on the reconstructed PET images and the disease auxiliary classification prediction probabilities.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is loaded by the processor, it is able to perform the steps of the method according to any one of claims 1 to 7.

Images