A Medical Multimodal Classification Method Based on Quantum Measurement Reliability Constraint Modulation
By using a quantum measurement reliability constraint modulation method, quantum features are generated and reliability weighted, which solves the problem of the impact of quantum measurement instability on medical multimodal classification and improves the stability and accuracy of the model.
Patent Information
- Application Number
- CN202611271518.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-20
- Publication Date
- 2026-11-10
AI Technical Summary
Existing medical multimodal classification methods are insufficient in terms of the complex nonlinear representation of structured clinical metadata and the stability of quantum measurements, which affects the stability and accuracy of classification models.
By using a quantum measurement reliability-constrained modulation method, quantum features are generated and reliability-weighted constraints are applied. The original image features are then subjected to residual affine modulation using quantum modulation parameters, and the influence of quantum measurement fluctuations is reduced through gated residual fusion.
It improves the stability and robustness of multimodal fusion, and enhances the accuracy and interpretability of classification models.
Smart Images

Figure CN123069072A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of medical artificial intelligence, multimodal data fusion, and quantum machine learning, and particularly to a medical multimodal classification method based on quantum measurement reliability constraint modulation. Background Technology
[0002] Medical image classification is a crucial task in computer-aided medical image analysis, widely applied in scenarios such as lesion identification, disease screening, pathological image analysis, and image-assisted diagnosis. With the development of deep learning technology, medical image classification methods based on models such as convolutional neural networks and visual Transformer networks have been widely adopted, capable of automatically extracting medical image features and completing classification. In real clinical scenarios, disease states or lesion categories are not only related to the appearance of medical images but also to structured clinical metadata such as patient basic information, medical history, and examination indicators. Multimodal fusion of medical images and structured clinical metadata helps to utilize complementary information between different modalities, improving the classification accuracy and robustness of the classification model.
[0003] Existing multimodal fusion methods for medical images and structured clinical metadata include feature stitching, feature weighting, gated residual fusion, attention fusion, and classification result fusion. While these methods can utilize complementary information between different modalities to some extent, structured clinical metadata suffers from characteristics such as heterogeneous field types, numerous missing values and noisy records, and a mixture of categorical and continuous variables. Furthermore, it differs from medical image features in data form, feature scale, and semantic hierarchy. Therefore, traditional stitching, weighting, or ordinary gated fusion methods have limited ability to express complex nonlinear clinical factors and their combinations, easily leading to insufficient information exchange between modalities and potentially allowing noisy information to adversely affect classification results.
[0004] In quantum machine learning, parameterized quantum circuits can form nonlinear feature maps through quantum state encoding, circuit evolution, and measurement statistics, providing a new technical approach for the feature representation of structured clinical metadata. Existing quantum-classical hybrid learning methods typically generate quantum features based on parameterized quantum circuits and combine them with image features extracted by classical neural networks to complete classification tasks.
[0005] However, in multimodal classification scenarios involving medical images and clinical metadata, the introduction of quantum features still faces challenges related to quantum measurement stability and fusion reliability. Quantum measurement statistics can fluctuate due to factors such as a limited number of samplings, quantum device noise, and line parameter perturbations. If quantum features directly participate in feature stitching, modal fusion, or classification decisions, the unstable components may be amplified in subsequent networks, making the classification results highly sensitive to quantum measurement fluctuations, thus affecting the model's stability and robustness in medical applications.
[0006] Furthermore, existing quantum-classical hybrid multimodal classification methods typically focus on the generation of quantum features and the fusion of quantum and classical features, while rarely explicitly evaluating the stability of quantum measurement statistics or constraining the strength of the role of quantum features in feature modulation, modality fusion, or classification decision-making based on their reliability. When quantum measurement statistics are unstable or quantum feature reliability is low, existing methods struggle to promptly suppress the adverse effects of quantum features on classification results. If quantum features are directly used to generate fusion gating coefficients or control the multimodal fusion ratio, the fusion process may become overly reliant on unstable quantum outputs, thereby reducing the reliability and interpretability of medical classification models.
[0007] Therefore, there is an urgent need for a medical multimodal classification method that can enhance the nonlinear representation of structured clinical metadata using quantum features, assess the stability of quantum measurement statistics, and constrain the strength of the role of quantum features in subsequent feature interactions based on the reliability of quantum features. This would reduce the adverse effects of quantum measurement fluctuations on multimodal fusion and classification decisions, and improve the stability and accuracy of classification models. Summary of the Invention
[0008] To address the shortcomings of existing medical multimodal classification methods in terms of the complex nonlinear representation of structured clinical metadata, quantum measurement reliability constraints, and robustness of multimodal fusion, this invention provides a medical multimodal classification method based on quantum measurement reliability constraint modulation. This method is applicable to multimodal classification scenarios involving medical images and structured clinical metadata, reducing the adverse effects of quantum measurement fluctuations on multimodal fusion and classification results, and improving the stability and accuracy of medical multimodal classification.
[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A medical multimodal classification method based on quantum measurement reliability-constrained modulation includes the following steps: S100: Obtain the medical image and structured clinical metadata of the object to be processed, and perform preprocessing on each; obtain the preprocessed medical image and structured clinical metadata. S200. Extract original image features based on preprocessed medical images; extract clinical metadata features based on preprocessed structured clinical metadata; group the clinical metadata features to obtain multiple clinical feature groups. S300: Based on clinical characteristics, grouping is used to generate quantum measurement statistics and quantum characteristics using parameterized quantum circuits; S400. Generate quantum reliability weights based on quantum measurement statistics, and use the quantum reliability weights to apply reliability weighted constraints to the quantum features; S500. Based on the quantum features constrained by reliability, quantum modulation parameters are generated, and the original image features are subjected to residual affine modulation using the quantum modulation parameters to obtain modulated image features. S600: Gated residual fusion of modulated image features and original image features is performed through a gating network to obtain fused features; S700: Input the fused features into the classification network to obtain medical classification results.
[0010] Specifically, in step S200, the clinical metadata features are grouped to obtain multiple clinical feature groups. The grouping process includes classifying the clinical metadata features based on clinical field categories, data types, clinical function attributes, or combinations thereof.
[0011] Specifically, in step S300, clinical feature groups are input into the corresponding pre-grouping mapping network to generate grouping quantum coding parameters corresponding to each clinical feature group; the grouping quantum coding parameters are allocated to the corresponding qubit subsets in the parameterized quantum circuit, and quantum state encoding, variational evolution, and quantum measurement are performed through the parameterized quantum circuit to obtain quantum measurement statistics; the quantum measurement statistics are input into the post-mapping network to generate quantum features.
[0012] Preferably, in step S300, the parameterized quantum circuit includes a multi-layer parameterized quantum circuit structure, each layer including a group re-upload quantum coding sublayer, a parameterized rotation sublayer, an intra-group entanglement sublayer, and an inter-group entanglement sublayer. The group re-upload quantum coding sublayer is used to load group quantum coding parameters corresponding to clinical feature groups; the parameterized rotation sublayer is used to perform variational evolution on the quantum state; the intra-group entanglement sublayer is used to establish correlations within the same clinical feature group; and the inter-group entanglement sublayer is used to establish correlations between different clinical feature groups.
[0013] Preferably, in step S400, a stability index is generated based on the quantum measurement statistics, and a quantum reliability weight is generated based on the stability index; wherein, the stability index is calculated based on the differences between the quantum measurement statistics corresponding to multiple measurement subsets.
[0014] Preferably, in step S400, the quantum reliability weight is generated based on a stability index and is used to characterize the reliability of the quantum features; when the stability of the quantum measurement statistics decreases, the quantum reliability weight decreases to reduce the modulation effect of the quantum features on the original image features.
[0015] Preferably, in step S500, the quantum modulation parameters include quantum scaling modulation parameters and quantum offset modulation parameters, and are used to scale and offset the original image features.
[0016] Preferably, in step S600, the fusion gating coefficients are generated by a gating network. The gating network input includes the original image features and the modulated image features, but does not include quantum features, quantum features after reliability constraints, and quantum modulation parameters.
[0017] Compared with the prior art, the present invention has the following beneficial effects: First, the present invention generates quantum features from structured clinical metadata, which can improve the nonlinear characterization ability of structured clinical metadata for complex disease factors and their combination relationships.
[0018] Second, the present invention calculates stability index based on quantum measurement statistics and uses this to constrain the reliability of quantum features, which can reduce the influence of unstable quantum features in the subsequent modulation process and suppress the adverse effects of quantum measurement fluctuations on the classification results.
[0019] Third, this invention utilizes quantum features constrained by reliability to perform residual affine modulation on the original image features, and performs gated residual fusion by fusing gating coefficients with the original image features. This can introduce quantum modulation information while preserving the original image information, thereby improving the stability and robustness of multimodal fusion. Attached Figure Description
[0020] Figure 1 This is a flowchart illustrating a medical multimodal classification method based on quantum measurement reliability constraint modulation, as shown in an embodiment of the present invention. Figure 2 This is a schematic diagram of the quantum feature and quantum measurement statistical information generation module structure shown in an embodiment of the present invention. Figure 3 This is a schematic diagram of a parameterized quantum circuit structure shown in an embodiment of the present invention; Figure 4 This is a schematic diagram of a reliability-constrained modulation and gated residual fusion structure based on quantum measurement statistical information, as shown in an embodiment of the present invention. Detailed Implementation
[0021] 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. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the protection scope of the present invention.
[0022] like Figure 1As shown, this embodiment of the invention provides a medical multimodal classification method based on quantum measurement reliability constraint modulation, the method comprising the following steps: S100. Acquire and preprocess medical images and structured clinical metadata: Acquire the medical image and structured clinical metadata of the object to be processed, and preprocess them respectively to obtain the preprocessed medical image. I and preprocessed structured clinical metadata M The medical images include, but are not limited to, one or more of the following: CT images, MRI images, ultrasound images, X-ray images, PET images, fundus images, endoscopic images, and pathological images; the preprocessing of the medical images includes, or is not limited to, one or more of the following: image normalization, noise suppression, image enhancement, image cropping, and size unification; the preprocessing of the structured clinical metadata includes, or is not limited to, missing value imputation, outlier handling, categorical variable encoding, continuous variable normalization, and missing indicator variable construction.
[0023] S200. Extract original image features and clinical metadata features, and group the clinical metadata features: Preprocess the medical images... Input image feature extraction network To obtain the original image features ;Preprocessed structured clinical metadata Input clinical metadata feature extraction network Clinical metadata features were obtained. Based on the medical semantics, field type, action path, or variable correlation corresponding to the clinical metadata features, the clinical metadata features are analyzed. Grouping, resulting in G The clinical features are grouped as follows: in, Indicates the first g Grouped by clinical characteristics g = 1, 2, ..., G The image feature extraction network It can be implemented using convolutional neural networks, ResNet, DenseNet, EfficientNet, visual Transformer networks, or combinations thereof; the clinical metadata feature extraction network It can be implemented using fully connected networks, multilayer perceptrons, feature embedding networks, attention networks, or combinations thereof.
[0024] S300, Based on clinical characteristic grouping, generate quantum measurement statistics and quantum features using parameterized quantum circuits: Combine the multiple clinical metadata features obtained in step S200. Input quantum feature generation module Statistical information of quantum measurement was obtained. and quantum characteristics , represented as: in, The quantum feature generation module includes: a pre-mapping network, a parameterized quantum circuit, a quantum measurement statistics layer, and a post-mapping network. Its structure can be found in [reference needed]. Figure 2 .
[0025] S301. Generate grouped quantum coding parameters through the grouped pre-mapping network: For the parameters obtained in step S200... g Clinical feature grouping Input it into the corresponding pre-mapping network Generate grouped quantum coding parameters that match the input dimension of the parameterized quantum circuit, expressed as: in, Indicates the first g Grouping quantum coding parameters corresponding to each clinical feature group.
[0026] In this embodiment, the packet pre-mapping network A multilayer perceptron structure is adopted, and the grouped quantum coding parameters are restricted to the range of [-π, π] by using the hyperbolic tangent function to adapt to quantum rotation gate operations, specifically expressed as follows: in, and For the first g Trainable weight parameters and bias parameters of a grouped pre-mapping network.
[0027] S302. Parameterized quantum coding and circuit evolution based on grouped quantum coding parameters: Assume the parameterized quantum circuit includes... n A set of qubits, where the qubits are represented as... To correspond with the clinical feature grouping in step S200, the set of qubits is... Q Divided into G A subset of qubits is represented as: in, Indicates the first g A subset of qubits corresponding to each clinical feature group. g = 1, 2, ..., G .
[0028] The parameterized quantum circuit includes L The structure of the layer can be found in [reference]. Figure 3 . No. l The layer parameterized quantum circuit is represented as: in, Indicates the first l Layer grouping and re-uploading of quantum-coded sublayers, Indicates the first l Layer parameterization of rotatable sublayers, Indicates the first l Entangled sublayers within a layer group Indicates the first l Inter-layer entanglement of sub-layers.
[0029] It should be noted that in the order of action of parameterized quantum circuit operators, the right-hand operator acts on the quantum state first; therefore, the above expression represents the first... l Execute first in the layer Then execute in sequence , and .
[0030] The group re-upload quantum coding sublayer is used to load the group quantum coding parameters obtained in step S301 into the corresponding qubits to achieve repeated coding of clinical feature groups. For the first l The first in the layer g Each clinical feature group, and its corresponding re-upload quantum coding sublayer representation, is as follows: in, , They represent the first g The clinical feature group was loaded into the first... j The quantum encoding parameters of each qubit , and The first l The first in the layer g Learnable scaling factor for each clinical feature group.
[0031] No. l The overall grouping and re-uploading of the quantum-coded sublayer is represented as follows: The parameterized rotation sublayer is used to apply trainable rotation gates to each qubit to achieve parameterized transformation of the quantum state. l The parameterized rotatable sublayer is represented as follows: in, , and The first l Layer j Trainable quantum parameters corresponding to each qubit.
[0032] The intra-group entangled sublayer is used to establish the correlation between qubits within the same clinical feature group. For the first... g A subset of qubits , No. l Entangled sublayers within a layer group are represented as follows: in, Indicates the first g The set of entangled edges within a subset of qubits Indicates that the control bit is the 1st i The number of qubits, with the target bit being the 1st qubit. j A controlled NOT gate for 1 qubits. l The entangled sublayers within the overall layer group are represented as follows: The intergroup entanglement sublayer is used to establish the correlation between different clinical feature groups. Each clinical feature group selects one or more representative qubits, denoted as... . No. l Inter-layer entanglement sublayers are represented as follows: in, This represents the set of entangled edges between different clinical feature groups. The topology corresponding to the set of entangled edges can be one of the following: ring structure, chain structure, fully connected structure, or sparse connection structure determined based on clinical prior relations.
[0033] Let the initial quantum state be ,go through L After the layer parameterized quantum circuit evolves, the output quantum state is represented as: In this embodiment, group re-upload quantum coding, parameterized rotation gates, intra-group entanglement, and inter-group entanglement are preferred implementations. In other implementations, the parameterized quantum circuits may also employ other coding orders, rotation gate combinations, entangled topologies, or quantum variational circuit structures.
[0034] S303. Perform measurement and statistics on the output quantum state through a quantum measurement and statistics layer: The output state obtained in step S302 is then... Measurements were performed to obtain statistical information on quantum measurement. In this embodiment, the quantum measurement statistics... It consists of the Pauli-Z operator expectation values of each qubit; for the ...j The measurement statistics of qubits are expressed as follows: in, Indicates the action on the first j Pauli-Z operator for 1 qubit, Indicates the first j Measurement statistics corresponding to each qubit j = 1, 2, ⋯ , n .
[0035] By combining the measurement statistics corresponding to each qubit in sequence, we obtain the quantum measurement statistics: It should be noted that when the quantum feature generation module is running on a real quantum device or a simulator with a finite number of sampling times, the Pauli-Z expectation value of each qubit is estimated by the statistical mean of the results of a finite number of measurements.
[0036] S304. Map the quantum measurement statistics to quantum features using a post-mapping network: Map the quantum measurement statistics obtained in step S303 to quantum features. Input post-mapping network Quantum characteristics were obtained. In this embodiment, the post-mapping network includes a linear transformation layer and a nonlinear activation function, expressed as follows: in, and These are the trainable parameters of the post-mapping network. This represents a non-linear activation function.
[0037] During model training, the trainable parameters in the pre-mapping network, parameterized quantum circuit, post-mapping network, and other neural network modules are jointly optimized through end-to-end training; among them, the gradients of the trainable quantum parameters in the parameterized quantum circuit are calculated using the parameter offset rule.
[0038] S400. Apply reliability-weighted constraints to quantum characteristics based on quantum measurement statistics: based on the quantum measurement statistics obtained in step S303. Calculate the quantum reliability weights and apply them to the quantum features obtained in step S304. Weighted constraints are applied to reduce the impact of quantum measurement sampling fluctuations, quantum device noise, or circuit disturbances on subsequent processing. Figure 4 A schematic diagram of a reliability-constrained modulation and gated residual fusion structure based on quantum measurement statistics is shown.
[0039] S401. Calculate quantum reliability weights: For the same clinical metadata input, divide the measurement results obtained from a finite number of samplings into... R A statistical subset is used to obtain quantum measurement statistics. The mean of the quantum measurement statistics corresponding to each statistical subset is taken as... , represented as: Based on the measurement fluctuations among different statistical subsets, a stability index is calculated: in, Indicates the first r Quantum measurement statistics corresponding to a statistical subset Used to characterize the degree of fluctuation in statistical information of quantum measurements.
[0040] Based on the stability index The quantum reliability weights are calculated using an exponential decay function: in, This is the reliability attenuation coefficient. When... When it increases, Reduce the size to mitigate the impact of unstable quantum characteristics.
[0041] S402. Apply reliability weighting constraints to quantum characteristics: using the aforementioned quantum reliability weights The quantum features obtained in step S304 By applying reliability-weighted constraints, the reliability-constrained quantum characteristics are obtained, expressed as: The quantum reliability weight It can be a sample-level scalar or a feature-dimensional vector, when When the vector is a feature dimension vector, an element-wise weighting method is used.
[0042] Therefore, quantum features can be adaptively constrained based on the fluctuation of quantum measurement statistical information, thereby reducing the impact of unstable quantum features on subsequent image feature modulation or multimodal fusion processes.
[0043] S500, Generate quantum modulation parameters and perform residual affine modulation on the original image features: The reliability-constrained quantum features obtained in step S402 are then used to generate quantum modulation parameters. Input modulation parameter generation network The quantum modulation parameters are generated through the tanh amplitude constraint function, which mainly includes quantum scaling modulation parameters. and quantum offset modulation parameters , represented as: in, and These represent quantum modulation parameter generation networks. The network branches used to generate quantum scaling modulation parameters and quantum offset modulation parameters; and These are the amplitude constraint coefficients for the quantum scaling modulation parameter and the quantum offset modulation parameter, respectively.
[0044] Residual affine modulation is used to extract features from the original image. Modulation is performed to obtain the quantum-modulated image features: Here, ⊙ represents element-wise multiplication.
[0045] S600, Gated residual fusion of modulated image features and original image features is performed through a gated network: the original image features are... and clinical metadata features After splicing, input into the gating network The fusion gating coefficients are obtained as follows: .
[0046] in, This represents the Sigmoid activation function. , Representing the features of the original image The feature dimensions of the gated network. The input does not include quantum features Quantum characteristics after reliability constraints Quantum scaling modulation parameters and quantum shift modulation parameters This is to avoid quantum measurement fluctuations directly affecting fusion gating decisions.
[0047] According to the fusion gating coefficient Features of the original image and modulated image features Gated residual fusion is performed to obtain fused features: Where ⊙ denotes element-wise multiplication. g When it approaches 0, Approaching ;when g When it approaches 1, Approaching This enables controlled fusion between original image features and modulated image features.
[0048] S700. Input the fused features into the classification network to obtain the medical classification result: input the fused features obtained in step S600... Input classification network C The medical classification results are as follows: in, The classification network represents the classification prediction result output by the model. C It can be implemented using fully connected networks, multilayer perceptrons, sigmoid classifiers, or combinations thereof.
[0049] During the training phase, based on real labels y End-to-end joint training was performed on the image feature extraction network, clinical metadata feature extraction network, quantum feature generation module, quantum modulation parameter generation network, gating network, and classification network. Classification loss... Cross-entropy loss can be used, expressed as: in, K Indicates the number of categories. Indicates the true label in the first place c Values for each category To represent the model prediction results in the th... c The probabilities of each category. To limit excessive shift of quantum modulation to the original image features, a perturbation constraint loss can optionally be introduced: The model's total loss function is: in, These are the weighting coefficients for the perturbation constraint loss, used to balance classification performance and modulation amplitude constraints. When At that point, the model degenerates into training using only classification loss.
Claims
1. A medical multimodal classification method based on quantum measurement reliability constraint modulation, characterized in that, include: S100: Acquire and preprocess medical images and structured clinical metadata; S200. Extract original image features and clinical metadata features, and group the clinical metadata features; S300: Based on clinical characteristics, grouping is used to generate quantum measurement statistics and quantum characteristics using parameterized quantum circuits; S400. Apply reliability-weighted constraints to quantum characteristics based on quantum measurement statistics. S500. Based on the quantum reliability constraint, quantum features are used to generate quantum modulation parameters, and the original image features are subjected to residual affine modulation using the quantum modulation parameters to obtain modulated image features. S600: Gated residual fusion of modulated image features and original image features is performed through a gating network to obtain fused features; S700: Input the fused features into the classification network to obtain medical classification results.
2. The method according to claim 1, characterized in that, The clinical characteristic-based grouping uses parameterized quantum circuits to generate quantum measurement statistics and quantum characteristics, including: At least one clinical feature is grouped and encoded as input parameters of a parameterized quantum circuit. Quantum state evolution is performed via the parameterized quantum circuit, and multiple quantum measurements are performed on the evolved quantum state to obtain quantum measurement statistics. Quantum features are then generated based on the quantum measurement statistics. The quantum measurement statistics include one or more of the following: measurement results, measurement probability distribution, and measurement expectation value.
3. The method according to claim 2, characterized in that, The generation of quantum measurement statistics and quantum characteristics includes: The clinical feature grouping module divides the clinical metadata features into multiple clinical feature groups according to the medical semantics, field type, and action path corresponding to the clinical metadata features; The grouping features corresponding to each clinical feature group are mapped to grouping quantum coding parameters through a grouping pre-mapping network; The grouped quantum coding parameters are encoded into quantum states using parameterized quantum circuits, and the encoded quantum states are then subjected to parameterized quantum transformations. The quantum state after parameterized quantum transformation is measured and statistically analyzed through a quantum measurement statistical layer; The quantum measurement statistics are mapped into quantum features through a post-mapping network.
4. The method according to claim 2 or 3, characterized in that, The reliability constraint based on quantum measurement statistics includes: generating quantum reliability weights based on quantum measurement statistics, and using the quantum reliability weights to apply weighted constraints to quantum features.
5. The method according to claim 4, characterized in that, The stability index Generate according to the following method: in, R Indicates the number of measurement subsets, Indicates the first r Quantum measurement statistics corresponding to a subset of measurements This represents the mean of the quantum measurement statistics corresponding to each subset of measurements. This indicates a stability index.
6. The method according to claim 5, characterized in that, The process of generating quantum reliability weights based on stability indices and imposing reliability constraints on quantum characteristics includes: generating quantum reliability weights using an exponential decay function. The quantum features are then weighted and constrained based on the quantum reliability weights to obtain the reliability-constrained quantum features: in, Represents quantum reliability weights. Indicates stability index, Indicates the reliability degradation coefficient. Representing quantum characteristics, This represents the quantum characteristics after reliability constraints.
7. The method according to claim 6, characterized in that, The step of generating quantum modulation parameters based on the reliability-constrained quantum features includes: inputting the reliability-constrained quantum features into a quantum modulation parameter generation network to obtain initial quantum modulation parameters; and restricting the value range of the initial modulation parameters through an amplitude constraint function to generate quantum modulation parameters; wherein the quantum modulation parameters include quantum scaling modulation parameters and quantum offset modulation parameters.
8. The method according to claim 7, characterized in that, The process of performing residual affine modulation on the original image features using the quantum modulation parameters to obtain modulated image features includes: performing residual affine modulation on the image features according to the quantum scaling modulation parameters and the quantum offset modulation parameters to obtain modulated image features. in, Represents the features of the modulated image. Represents the features of the original image. Represents the quantum scaling modulation parameters. Represents the quantum offset modulation parameter. This indicates element-wise multiplication.
9. The method according to claim 8, characterized in that, The gated network takes raw image features and clinical metadata features as input, but does not take quantum features, reliability-constrained quantum features, or quantum modulation parameters as input.
10. The method according to claim 9, characterized in that, Based on the fusion gate coefficient, gated residual fusion is performed on the original image features and the modulated image features to obtain the fused features: in, Indicates fusion characteristics, This represents the fusion gating coefficient.
