Patient evaluation method based on NMOSD level model and related device

Abstract

The invention discloses a patient evaluation method based on an NMOSD level model and a related device.The method comprises the steps that multi-dimensional feature data of a patient are collected regularly, and the multi-dimensional feature data comprise vision data, limb data, nervous system symptom data, activity data and self-management behavior data; inputting the multi-dimensional feature data into an NMOSD level model to obtain a self-management capability level of the patient, and dividing the self-management capability of the patient into a predetermined level; according to the preset grade of the patient and the multi-dimensional feature data, self-management suggestions are provided for the patient, and the self-management suggestions comprise medicine management reminding, eyesight protection suggestions, exercise rehabilitation plans, fatigue handling strategies, recurrence risk warning and revival appointment reminding. According to the method, the multi-dimensional data are input into the NMOSD level model, quantitative evaluation of the patient is achieved, and better self-management of the NMOSD patient is helped.

Description

Technical Field

[0001] The present invention relates to the technical field of medical auxiliary diagnosis, and in particular to a patient evaluation method and device based on an NMOSD grade model, and a computing device. Background Art

[0002] NMOSD is a rare autoimmune disease that primarily affects the optic nerves and spinal cord.

[0003] Existing treatments mainly focus on relieving symptoms and preventing relapses, including immunosuppressants, biologics, etc. For the disease management of NMOSD, it currently relies mainly on regular evaluations by doctors and subjective feedback from patients. There is a lack of personalized guidance and proactive intervention for patients' daily self-management, and patients cannot take effective self-management measures in a timely manner. In addition, existing assessment methods only focus on vision, limbs, etc., and it is difficult to comprehensively assess the overall condition of patients. For example, neurological symptoms (fatigue, pain, etc.) have a great impact on the quality of life of patients, but these symptoms are often difficult to be objectively quantified. In addition, the intrinsic connections between vision data, limb data, and neurological symptom data cannot be fully utilized.

[0004] To solve the above problems, the present invention proposes a patient evaluation method based on the NMOSD grade model, which inputs multidimensional data into the NMOSD grade model to achieve quantitative evaluation of patients and help NMOSD patients better manage themselves. Summary of the invention

[0005] In view of the above problems, the present invention provides a patient evaluation method and apparatus, and a computing device based on the NMOSD grade model.

[0006] According to one aspect of the present invention, a patient evaluation method based on the NMOSD grade model is provided, comprising:

[0007] Regularly collect multi-dimensional feature data of the patient, wherein the multi-dimensional feature data includes vision data, limb data, nervous system symptom data, activity data and self-management behavior data;

[0008] Inputting the multidimensional feature data into the NMOSD grade model to obtain the patient's self-management ability grade, and dividing the patient's self-management ability into predetermined grades;

[0009] Based on the patient's predetermined level and the multidimensional feature data, self-management suggestions are provided to the patient, wherein the self-management suggestions include medication management reminders, vision protection suggestions, exercise rehabilitation plans, fatigue coping strategies, relapse risk warnings, and follow-up appointment reminders.

[0010] In an optional manner, the vision data includes visual acuity test results, visual field test results, and eye pain degree scores;

[0011] The limb data include muscle strength test results, walking ability score and balance ability score;

[0012] The neurological symptom data included fatigue level, bladder dysfunction level, pain level, and paresthesia score;

[0013] The activity data includes daily steps, activity time, and sitting time;

[0014] The self-management behavior data included medication adherence scores, frequency of disease relapse symptom monitoring, and frequency of follow-up appointments.

[0015] In an optional manner, the multidimensional feature data also includes patient personal information, wherein the patient personal information includes age, gender, spinal cord involvement length, AQP4 antibody status, serum IL-6 level, and number of previous relapses.

[0016] In an optional manner, the NMOSD hierarchical model includes a dynamic routing module, an embedding layer, a GAT network layer, a Pairwise interaction network layer, a Policy network layer and an output layer;

[0017] Wherein, the dynamic routing module adopts a capsule network to capture the intrinsic connection between the multi-dimensional feature data;

[0018] The GAT network layer learns the relationship and weight between nodes;

[0019] The pairwise interaction network layer constructs the interaction relationship between different modality features by constructing a pairwise interaction matrix;

[0020] The Policy network layer dynamically adjusts the contribution of each network modality feature.

[0021] In an optional manner, the dynamic routing module includes:

[0022] The primary capsule layer is used to convert the initial feature vector into multiple primary capsules through multiple convolutional or fully connected layers, where each primary capsule represents a local feature;

[0023] Multi-head attention routing, which uses the output of each primary capsule as a query and the output of all primary capsules as keys and values ​​to generate multiple different attention heads;

[0024] The GAT network layer includes:

[0025] The node feature initialization layer is used to use the high-level capsules output by the dynamic routing module as the node features of the graph;

[0026] Adaptive adjacency matrix learning layer, used to calculate the similarity between any two node features, and adaptively adjust the similarity threshold and connection weight so that the graph structure represents the relationship between the data;

[0027] Multi-layer GAT layer, used to aggregate neighbor node features according to weights, and multiple GAT layers are superimposed to propagate and aggregate node information in the graph structure;

[0028] The Pairwise interaction network layer includes:

[0029] The modality-specific interaction layer is used to perform Kronecker product tensor operations on low-rank tensors of different modalities to construct interaction tensors between modalities;

[0030] Interaction feature fusion is used to concatenate or weighted sum the interaction features between different modal pairs to obtain fused features;

[0031] The Policy network layer includes:

[0032] The reward function is used to design a reward function based on the patient's self-management ability evaluation result indicators; when the model prediction result meets the expectation, a positive reward is given, otherwise a negative reward is given.

[0033] In an optional manner, the adjustment formula of the similarity threshold is:

[0034]

[0035] in, represents the similarity threshold between node i and node j at the tth iteration; S ij is the similarity between node i and node j; represents the set of neighbor nodes of node i; α is the learning rate parameter; S ik is the similarity between node i and its neighbor node k; k is the set of neighbor nodes of node i An element in .

[0036] In an optional manner, the aggregation formula of the node features in the multi-layer GAT layer is:

[0037]

[0038] in, represents the feature vector of node i at the l+1th layer; represents the set of neighbor nodes of node i; e ij is the attention coefficient between node i and node j; W(l) is the weight matrix of the lth layer; σ is the activation function; Represents the feature vector of node j at layer l.

[0039] In an optional manner, the reward function in the Policy network layer is:

[0040]

[0041] Among them, y is the matching target of the Policy network; is the estimated value generated according to the current strategy; β1 and β2 are weight parameters.

[0042] According to another aspect of the present invention, there is provided a patient evaluation device based on the NMOSD grade model, comprising:

[0043] A data collection module, used to regularly collect multi-dimensional feature data of patients, wherein the multi-dimensional feature data includes vision data, limb data, nervous system symptom data, activity data and self-management behavior data;

[0044] A level assessment module, used for inputting the multidimensional feature data into the NMOSD level model, obtaining the patient's self-management ability level, and dividing the patient's self-management ability into predetermined levels;

[0045] The advice providing module is used to provide self-management advice to the patient based on the patient's predetermined level and the multidimensional feature data, wherein the self-management advice includes medication management reminders, vision protection advice, exercise rehabilitation plans, fatigue coping strategies, recurrence risk warnings and follow-up appointment reminders.

[0046] According to another aspect of the present invention, there is provided a computing device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0047] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned patient evaluation method based on the NMOSD grade model.

[0048] According to the scheme provided by the present invention, multidimensional feature data of patients are collected regularly, wherein the multidimensional feature data includes vision data, limb data, neurological symptom data, activity data and self-management behavior data; the multidimensional feature data is input into the NMOSD grade model to obtain the patient's self-management ability level, and the patient's self-management ability is divided into predetermined levels; according to the patient's predetermined level and the multidimensional feature data, self-management suggestions are provided to the patient, wherein the self-management suggestions include drug management reminders, vision protection suggestions, sports rehabilitation plans, strategies for coping with fatigue, recurrence risk warnings and follow-up appointment reminders. The present invention inputs multidimensional data into the NMOSD grade model to achieve quantitative evaluation of patients and help NMOSD patients better manage themselves.

[0049] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0051] Figure 1 A schematic flow chart of a patient evaluation method based on an NMOSD grade model according to an embodiment of the present invention is shown;

[0052] Figure 2 A schematic diagram of an NMOSD grade model according to an embodiment of the present invention is shown;

[0053] Figure 3 A schematic diagram of a patient evaluation device based on an NMOSD grade model according to an embodiment of the present invention is shown;

[0054] Figure 4 A schematic diagram of the structure of a computing device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0055] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to enable the scope of the present invention to be fully communicated to those skilled in the art.

[0056] Figure 1 FIG. 4 is a flow chart of a patient evaluation method based on an NMOSD grade model according to an embodiment of the present invention. Specifically, Figure 1 As shown, the following steps are included:

[0057] Step S101, regularly collecting multi-dimensional feature data of the patient, wherein the multi-dimensional feature data includes vision data, limb data, nervous system symptom data, activity data and self-management behavior data.

[0058] In this embodiment, the fusion of multimodal data such as visual function, motor function, and patient reports can provide a comprehensive understanding of the clinical phenotype and individual differences of the disease, which is consistent with the bio-psycho-social model of modern medicine and emphasizes the multi-faceted impact of the disease on the patient's physiological, psychological, and social life.

[0059] In an optional manner, the vision data includes visual acuity test results, visual field test results, and eye pain degree scores;

[0060] The limb data include muscle strength test results, walking ability score and balance ability score;

[0061] The neurological symptom data included fatigue level, bladder dysfunction level, pain level, and paresthesia score;

[0062] The activity data includes daily steps, activity time, and sitting time;

[0063] The self-management behavior data included medication adherence scores, frequency of disease relapse symptom monitoring, and frequency of follow-up appointments.

[0064] The multidimensional feature data also includes patient personal information, wherein the patient personal information includes age, gender, spinal cord involvement length, AQP4 antibody status, serum IL-6 level and number of previous relapses.

[0065] As shown in Table 1, the best corrected visual acuity (BCVA) of the patients was measured using a standard visual acuity chart, and the visual field loss and range of the patients were assessed using an automated perimeter. The optic nerve head and retina of the patients were observed using a direct or indirect ophthalmoscope. The thickness of the optic nerve fiber layer (RNFL) was measured by optical coherence tomography (OCT) to assess the degree of damage to the optic nerve. The muscle strength of the patients was assessed using a medical muscle strength score sheet (such as the Modified Medical Research Council Scale). Symptoms such as fatigue, pain, balance disorders, and bladder dysfunction were quantified using standardized scales (such as the Expanded Disability Status Scale (EDSS), Fatigue Severity Scale (FSS), Visual Analog Scale (VAS), etc.). The number of steps, activity duration, sitting time, sleep quality, and other indicators of the patients were monitored by wearable devices (such as sensors such as accelerometers and gyroscopes) or smartphones. The AQP4 antibody level and serum IL-6 level of the patients were regularly tested for inflammatory indicators. The extent and degree of spinal cord inflammation and whether there were brain lesions were assessed based on medical imaging (spinal cord MRI, brain MRI).

[0066] Table 1

[0067]

[0068]

[0069]

[0070]

[0071] Step S102, inputting the multi-dimensional feature data into the NMOSD grade model to obtain the patient's self-management ability grade, and classifying the patient's self-management ability into predetermined grades.

[0072] In the present embodiment, the multidimensional feature data of the patient is input into the trained NMOSD grade model to obtain the patient's self-management ability grade. According to the patient's grade, a personalized intervention plan (drug management, lifestyle guidance, psychological support, rehabilitation training, etc.) is formulated. NMOSD grade assessment is performed based on multidimensional data to reduce the subjective bias of the evaluator and the patient. According to clinical actual conditions and patient needs, the patient's self-management ability is divided into several predetermined levels, for example, a high level (the patient can actively manage his or her disease), a medium level (the patient can basically manage his or her disease), and a low level (the patient is difficult to manage his or her disease independently and needs more support and guidance).

[0073] In an optional manner, the NMOSD hierarchical model includes a dynamic routing module, an embedding layer, a GAT network layer, a Pairwise interaction network layer, a Policy network layer and an output layer;

[0074] Wherein, the dynamic routing module adopts a capsule network to capture the intrinsic connection between the multi-dimensional feature data;

[0075] The GAT network layer learns the relationship and weight between nodes;

[0076] The pairwise interaction network layer constructs the interaction relationship between different modality features by constructing a pairwise interaction matrix;

[0077] The Policy network layer dynamically adjusts the contribution of each network modality feature.

[0078] In this embodiment, capsule networks are used to represent entities and their attributes, and dynamic routing algorithms are used to capture the hierarchical and intrinsic connections between entities in multidimensional feature data. For example, the patient's EDSS score, fatigue level, and anxiety may have complex causal relationships or mutual influences. The GAT learns the strength of association between different features to more accurately assess the impact on self-management ability. For example, if a patient's anxiety has a greater impact on self-management ability, GAT automatically gives it a higher weight. By constructing a pairwise interaction matrix, the interaction between different modal features is explicitly modeled (such as analyzing the interaction between the patient's physiological indicators and psychological indicators). The policy network layer dynamically adjusts the contribution of different modal features according to the patient's specific situation, rather than simply weighting all modalities uniformly. For example, for patients with higher disease activity, physiological indicators are given higher weights; for patients with greater psychological pressure, psychological indicators are given higher weights. In addition, the model architecture uses end-to-end learning to avoid the complex feature engineering and manual parameter adjustment process in traditional methods. Although deep learning models are often regarded as "black boxes", the interpretability of the model is increased through the attention mechanism and pairwise interaction module.

[0079] Specifically, Figure 2 As shown in the figure, the collected multidimensional characteristic data of patients are input into the model, including physiological indicators (EDSS score, MRI lesion load, fatigue level, pain score, neurological deficit assessment, etc.), psychological indicators (anxiety and depression score, coping style assessment, quality of life assessment (SF-36), etc.), behavioral indicators (such as medication adherence score (MARS), lifestyle data (activity, sleep quality, eating habits, etc.), rehabilitation training records, etc.), etc.

[0080] Different types of feature data are mapped to a continuous vector space through the embedding layer, and feature vectors of different dimensions are converted to a unified dimension.

[0081] In the dynamic routing module, the capsule network is used to process the embedded feature vector, and the features are represented in the form of capsules. The dynamic routing algorithm is used to capture the connections between different capsules, identify the hierarchical structure and intrinsic relationship of the features, and output a refined feature representation.

[0082] The feature representation output by the capsule network is regarded as a node in the graph through the GAT network layer. The attention mechanism is used to learn the relationship and weights between different nodes, and the node features are updated according to the learned weights.

[0083] In the Pairwise Interaction Network Layer, a pairwise interaction matrix is ​​constructed to combine the feature vectors of different modalities in pairs to form a pairwise interaction matrix. Through matrix operations and convolution operations, the interaction features between different modalities are extracted, and the interaction features are concatenated with the original features to form a richer feature representation.

[0084] The contribution of different modal features is dynamically adjusted through the Policy Network Layer, and the patient's self-management ability level is predicted by optimizing the objective function.

[0085] The output of the network is mapped to a predetermined self-management capability level using a Softmax or Sigmoid function. In an optional manner, the dynamic routing module includes:

[0086] The primary capsule layer is used to convert the initial feature vector into multiple primary capsules through multiple convolutional or fully connected layers, where each primary capsule represents a local feature;

[0087] Multi-head attention routing, which uses the output of each primary capsule as a query and the output of all primary capsules as keys and values ​​to generate multiple different attention heads;

[0088] The GAT network layer includes:

[0089] The node feature initialization layer is used to use the high-level capsules output by the dynamic routing module as the node features of the graph;

[0090] Adaptive adjacency matrix learning layer, used to calculate the similarity between any two node features, and adaptively adjust the similarity threshold and connection weight so that the graph structure represents the relationship between the data;

[0091] Multi-layer GAT layer, used to aggregate neighbor node features according to weights, and multiple GAT layers are superimposed to propagate and aggregate node information in the graph structure;

[0092] The Pairwise interaction network layer includes:

[0093] The modality-specific interaction layer is used to perform Kronecker product tensor operations on low-rank tensors of different modalities to construct interaction tensors between modalities;

[0094] Interaction feature fusion is used to concatenate or weighted sum the interaction features between different modal pairs to obtain fused features;

[0095] The Policy network layer includes:

[0096] The reward function is used to design a reward function based on the patient's self-management ability evaluation result indicators; when the model prediction result meets the expectation, a positive reward is given, otherwise a negative reward is given.

[0097] In this embodiment, the initial feature vector is converted into multiple primary capsules through a convolution or fully connected layer, each capsule represents a specific local feature, and captures the spatial and local information of the input data. Compared with a simple fully connected layer, the capsule network retains the direction and amplitude information of the local feature through the capsule vector, thereby avoiding the loss of information. The high-level capsule output by the dynamic routing module is directly used as the node feature of the graph to ensure the high-level abstract expression of the input information. By calculating the similarity between node features and adaptively adjusting the similarity threshold and connection weight, the graph structure reflects the relationship between data instead of relying on a predefined structure. By superimposing multiple layers of GAT layers, node information is propagated and aggregated in the graph structure to achieve global information fusion. The interaction relationship between modalities is modeled using a low-rank tensor for Kronecker product operations. The Kronecker product operation introduces nonlinear capture of more complex interaction patterns between modalities. The reward function is directly associated with the ultimate goal of the model (accurately predicting the level of self-management ability) to learn feature representations related to the goal.

[0098] For example, the EDSS score, FSS score, HADS anxiety score, etc. are converted into multiple primary capsules through the convolution layer. For example, one capsule represents the neurological function characteristics related to "moderate disability" and the other represents the fatigue perception characteristics related to "obvious fatigue".

[0099] The multi-head attention mechanism focuses on these primary capsules from different perspectives, e.g., one head may focus on the association between local features of EDSS scores and movement disorders, and another head may focus on the interaction between HADS anxiety scores and fatigue.

[0100] Advanced capsules are used as nodes of the graph to represent various features. The similarity between each feature node is calculated through an adaptive adjacency matrix learning layer (e.g., the "fatigue" node and the "depression" node may have a high similarity). Because feature nodes often coexist in NMOSD, it is possible to learn the connection between fatigue and depression and adaptively adjust the weight of the connection.

[0101] Through multiple GAT layers, information is propagated in the graph structure. For example, nodes related to "movement disorder" may receive information from related nodes such as "fatigue" and "pain", thus forming a more comprehensive representation.

[0102] The Kronecker product operation is performed on the low-rank tensors of the modality-specific interaction layer to generate an interaction tensor that captures the complex interactions between the body and mind. Features such as physiological-psychological interaction and physiological-behavioral interaction are spliced ​​together to form a more complete feature representation.

[0103] If the predicted self-management ability level is correct, a positive reward is given, otherwise a negative reward is given. Through continuous iterative learning, the Policy network adjusts the weights of each mode, so that the model pays more attention to factors that have a greater impact on self-management ability, such as psychological factors, during the prediction process, thereby improving the accuracy of the prediction.

[0104] In an optional manner, the adjustment formula of the similarity threshold is:

[0105]

[0106] in, represents the similarity threshold between node i and node j at the tth iteration; S ij is the similarity between node i and node j; represents the set of neighbor nodes of node i; α is the learning rate parameter; S ik is the similarity between node i and its neighbor node k; k is the set of neighbor nodes of node i An element in .

[0107] In this embodiment, if S ij If the similarity of the two nodes is higher than the average neighbor similarity of node i, the similarity between the two nodes is higher than the average connection of the network they are in, and the threshold between the two nodes is lowered, making it easier to identify them as similar. This makes it easier to identify node pairs with similarity higher than the average level as similar, and makes it more difficult to identify node pairs with similarity lower than the average level as similar. Local similarity analysis is performed to make the threshold adjustment more refined to capture the characteristics of local structures in the network.

[0108] In an optional manner, the aggregation formula of the node features in the multi-layer GAT layer is:

[0109]

[0110] in, represents the feature vector of node i at the l+1th layer; represents the set of neighbor nodes of node i; e ij is the attention coefficient between node i and node j; W (l) is the weight matrix of the lth layer; σ is the activation function; Represents the feature vector of node j at layer l.

[0111] In this embodiment, the attention coefficient determines the importance of each neighbor node j when aggregating the neighbor node features of node i, and then focuses on the neighbors that are more important to node i, rather than treating all neighbors equally, which can capture the complex relationship between nodes in the graph. The model can dynamically adjust the influence of each neighbor on the central node and adapt to the local structure around different nodes.

[0112] In an optional manner, the reward function in the Policy network layer is:

[0113]

[0114] Among them, y is the matching target of the Policy network; is the estimated value generated according to the current strategy; β1 and β2 are weight parameters.

[0115] In this embodiment, the reward function considers the absolute error between the estimated value and the target value and the directional consistency (through the dot product The double optimization of the output accuracy of the Policy network is achieved by measuring the product ratio of the model length and the model length, ensuring that the treatment recommendations are highly consistent with the ideal goals.

[0116] Step S103, providing self-management suggestions to the patient based on the patient's predetermined level and the multidimensional feature data, wherein the self-management suggestions include medication management reminders, vision protection suggestions, exercise rehabilitation plans, fatigue coping strategies, recurrence risk warnings, and follow-up appointment reminders.

[0117] In this embodiment, as shown in Table 2, personalized self-management suggestions are formulated for each patient based on their predetermined level and multi-dimensional feature data.

[0118] Table 2

[0119]

[0120]

[0121] According to the scheme provided by the present invention, multidimensional feature data of patients are collected regularly, wherein the multidimensional feature data includes vision data, limb data, neurological symptom data, activity data and self-management behavior data; the multidimensional feature data is input into the NMOSD grade model to obtain the patient's self-management ability level, and the patient's self-management ability is divided into predetermined levels; according to the patient's predetermined level and the multidimensional feature data, self-management suggestions are provided to the patient, wherein the self-management suggestions include drug management reminders, vision protection suggestions, sports rehabilitation plans, strategies for coping with fatigue, recurrence risk warnings and follow-up appointment reminders. The present invention inputs multidimensional data into the NMOSD grade model to achieve quantitative evaluation of patients and help NMOSD patients better manage themselves.

[0122] Figure 3 The schematic diagram of the framework of the patient evaluation device based on the NMOSD grade model of an embodiment of the present invention is shown. The patient evaluation device based on the NMOSD grade model includes:

[0123] The data collection module 310 is used to regularly collect multi-dimensional characteristic data of the patient, wherein the multi-dimensional characteristic data includes vision data, limb data, nervous system symptom data, activity data and self-management behavior data;

[0124] A level assessment module 320, for inputting the multidimensional feature data into an NMOSD level model, obtaining the patient's self-management ability level, and dividing the patient's self-management ability into predetermined levels;

[0125] The advice providing module 330 is used to provide self-management advice to the patient based on the patient's predetermined level and the multidimensional feature data, wherein the self-management advice includes medication management reminders, vision protection advice, exercise rehabilitation plans, fatigue coping strategies, recurrence risk warnings and follow-up appointment reminders.

[0126] Figure 4 The schematic diagram of the structure of the computing device embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computing device.

[0127] like Figure 4As shown, the computing device may include: a processor 402, a communications interface 404, a memory 406, and a communications bus 408. The processor 402, the communications interface 404, and the memory 406 communicate with each other through the communications bus 408. The communications interface 404 is used to communicate with other devices such as a client or other server network elements. The processor 402 is used to execute a program 410, which may specifically execute the relevant steps in the above-mentioned patient evaluation method embodiment based on the NMOSD grade model.

[0128] Specifically, the program 410 may include a program code, which includes a computer operation instruction. The processor 402 may be a central processing unit CPU, or an application specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiment of the present invention. The one or more processors included in the computing device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0129] The memory 406 is used to store the program 410. The memory 406 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0130] According to the scheme provided by the present invention, multidimensional feature data of patients are collected regularly, wherein the multidimensional feature data includes vision data, limb data, neurological symptom data, activity data and self-management behavior data; the multidimensional feature data is input into the NMOSD grade model to obtain the patient's self-management ability level, and the patient's self-management ability is divided into predetermined levels; according to the patient's predetermined level and the multidimensional feature data, self-management suggestions are provided to the patient, wherein the self-management suggestions include drug management reminders, vision protection suggestions, sports rehabilitation plans, strategies for coping with fatigue, recurrence risk warnings and follow-up appointment reminders. The present invention inputs multidimensional data into the NMOSD grade model to achieve quantitative evaluation of patients and help NMOSD patients better manage themselves.

[0131] Those skilled in the art will appreciate that the modules in the devices in the embodiments can be adaptively changed and set in one or more devices different from the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and in addition they can be divided into multiple submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner can be combined in any combination. Unless otherwise explicitly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) can be replaced by an alternative feature that provides the same, equivalent or similar purpose. In addition, those skilled in the art will appreciate that although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments means being within the scope of the present invention and forming different embodiments. For example, in the following claims, any one of the claimed embodiments can be used in any combination. The present invention can be implemented by hardware including several different elements and by appropriately programmed computers. In a unit claim that lists several means, several of these means may be embodied by the same hardware item. The steps in the above embodiments should not be understood as limiting the execution order unless otherwise specified.

Claims

1. A patient evaluation method based on the NMOSD grade model, characterized in that: include: Regularly collect multi-dimensional feature data of the patient, wherein the multi-dimensional feature data includes vision data, limb data, nervous system symptom data, activity data and self-management behavior data; Inputting the multidimensional feature data into the NMOSD grade model to obtain the patient's self-management ability grade, and dividing the patient's self-management ability into predetermined grades; Based on the patient's predetermined level and the multidimensional feature data, self-management suggestions are provided to the patient, wherein the self-management suggestions include medication management reminders, vision protection suggestions, exercise rehabilitation plans, fatigue coping strategies, relapse risk warnings, and follow-up appointment reminders.

2. The patient evaluation method based on the NMOSD grade model according to claim 1, characterized in that: The vision data include visual acuity test results, visual field test results and eye pain degree scores; The limb data include muscle strength test results, walking ability score and balance ability score; The neurological symptom data included fatigue level, bladder dysfunction level, pain level, and paresthesia score; The activity data includes daily steps, activity time, and sitting time; The self-management behavior data included medication adherence scores, frequency of disease relapse symptom monitoring, and frequency of follow-up appointments.

3. The patient evaluation method based on the NMOSD grade model according to claim 1, characterized in that: The multidimensional feature data also includes patient personal information, wherein the patient personal information includes age, gender, spinal cord involvement length, AQP4 antibody status, serum IL-6 level and number of previous relapses.

4. The patient evaluation method based on the NMOSD grade model according to claim 1, characterized in that: The NMOSD hierarchical model includes a dynamic routing module, an embedding layer, a GAT network layer, a Pairwise interaction network layer, a Policy network layer and an output layer; Wherein, the dynamic routing module adopts a capsule network to capture the intrinsic connection between the multi-dimensional feature data; The GAT network layer learns the relationship and weight between nodes; The pairwise interaction network layer constructs the interaction relationship between different modality features by constructing a pairwise interaction matrix; The Policy network layer dynamically adjusts the contribution of each network modality feature.

5. The patient evaluation method based on the NMOSD grade model according to claim 4, characterized in that: The dynamic routing module includes: The primary capsule layer is used to convert the initial feature vector into multiple primary capsules through multiple convolutional or fully connected layers, where each primary capsule represents a local feature; Multi-head attention routing, which uses the output of each primary capsule as a query and the output of all primary capsules as keys and values ​​to generate multiple different attention heads; The GAT network layer includes: The node feature initialization layer is used to use the high-level capsules output by the dynamic routing module as the node features of the graph; Adaptive adjacency matrix learning layer, used to calculate the similarity between any two node features, and adaptively adjust the similarity threshold and connection weight so that the graph structure represents the relationship between the data; Multi-layer GAT layer, used to aggregate neighbor node features according to weights, and multiple GAT layers are superimposed to propagate and aggregate node information in the graph structure; The Pairwise interaction network layer includes: The modality-specific interaction layer is used to perform Kronecker product tensor operations on low-rank tensors of different modalities to construct interaction tensors between modalities; Interaction feature fusion is used to concatenate or weighted sum the interaction features between different modal pairs to obtain fused features; The Policy network layer includes: The reward function is used to design a reward function based on the patient's self-management ability evaluation result indicators; when the model prediction result meets the expectation, a positive reward is given, otherwise a negative reward is given.

6. The patient evaluation method based on the NMOSD grade model according to claim 5, characterized in that: The adjustment formula of the similarity threshold is: in, represents the similarity threshold between node i and node j at the tth iteration; S ij is the similarity between node i and node j; represents the set of neighbor nodes of node i; α is the learning rate parameter; S ik is the similarity between node i and its neighbor node k; k is the set of neighbor nodes of node i An element in .

7. The patient evaluation method based on the NMOSD grade model according to claim 5, characterized in that: The aggregation formula of node features in the multi-layer GAT layer is: in, represents the feature vector of node i at the l+1th layer; represents the set of neighbor nodes of node i; e ij is the attention coefficient between node i and node j; W (l) is the weight matrix of the lth layer; σ is the activation function; Represents the feature vector of node j at layer l.

8. The patient evaluation method based on the NMOSD grade model according to claim 5, characterized in that: The reward function in the Policy network layer is: Among them, y is the matching target of the Policy network; is the estimated value generated according to the current strategy; β1 and β2 are weight parameters.

9. A patient assessment device based on the NMOSD grade model, characterized in that: include: A data collection module, used to regularly collect multi-dimensional feature data of patients, wherein the multi-dimensional feature data includes vision data, limb data, nervous system symptom data, activity data and self-management behavior data; A level assessment module, used for inputting the multidimensional feature data into the NMOSD level model, obtaining the patient's self-management ability level, and dividing the patient's self-management ability into predetermined levels; The advice providing module is used to provide self-management advice to the patient based on the patient's predetermined level and the multidimensional feature data, wherein the self-management advice includes medication management reminders, vision protection advice, exercise rehabilitation plans, fatigue coping strategies, recurrence risk warnings and follow-up appointment reminders.

10. A computing device comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned patient evaluation method based on the NMOSD grade model.

Images