Method for dynamically predicting bone mass reduction risk of psoriasis patient and public platform

Abstract

According to the method for dynamically predicting the bone mass reduction risk of the psoriasis patient and the public platform, firstly, high-quality data are screened from historical medical records, bone mineral density data and biomarker data of the patient, and multi-modal feature extraction is carried out; high-order interaction of each modal feature is captured through a graph attention network, and multi-modal feature fusion is realized; then, constructing a deep attention neural network, and realizing probabilistic prediction of the bone mass reduction risk of the psoriasis patient; performing visual explanation on an individual prediction result by using the attention weight of the model and the SHAP value, and enhancing clinical availability; then, the network is compressed and optimized through technologies such as model distillation, pruning and quantification, and efficient deployment of the model on edge computing equipment is achieved; finally, continuously optimizing model parameters in combination with actual clinical results and doctor feedback; according to the invention, the accuracy of risk prediction is effectively improved.

Description

Technical field

[0001] The present invention belongs to the technical field of medical model processing, and in particular relates to a method and a public platform for dynamically predicting the risk of bone loss in psoriasis patients. [Background Technology]

[0002] Recent studies have shown that psoriasis, a systemic inflammatory disease, carries a significantly higher risk of bone loss and osteoporosis in patients compared to the general population. The disease also has a complex course and diverse mechanisms, making it difficult to accurately predict individualized risk using a single clinical indicator. Early and accurate identification of bone loss in psoriasis patients and risk intervention remain key areas of focus and challenge in the fields of rheumatology, immunology, bone metabolism, and dermatology.

[0003] Traditional bone mass risk assessment tools are often based on a single modality, ignoring the high-dimensional interactions between multiple data sources, such as patient medical records, biomarkers, medication history, and lifestyle. This results in limited predictive accuracy and generalization. Furthermore, traditional models, which mostly employ linear or shallow structures like logistic regression and decision trees, struggle to effectively capture complex nonlinear and cross-modal interactions. Furthermore, they lack the ability to automatically interpret key risk factors for individual patients, making them inadequate for meeting the demands of modern precision medicine.

[0004] Currently, there is a lack of a full-process, closed-loop, multimodal, explainable, and edge-deployable intelligent platform system for the risk of bone loss in psoriasis patients, both domestically and internationally. There is an urgent need for an innovative technical solution covering the entire chain, including data collection and cleaning, multimodal feature fusion, deep prediction, model interpretation, compression deployment, and feedback optimization, to achieve intelligent, personalized, real-time prediction and clinical decision-making support for the risk of bone loss in psoriasis patients, and to enhance early screening, early prevention, and precise intervention capabilities. The establishment of such a technical platform will not only help improve the risk prediction level of a single disease, but also provide an innovative paradigm with promotional value for chronic disease management and the application of artificial intelligence in precision medicine. [Summary of the invention]

[0005] In view of this, an embodiment of the present invention provides an MRI image processing method for a cochlear implant patient.

[0006] The method for dynamically predicting the risk of osteopenia in patients with psoriasis provided by an embodiment of the present invention comprises:

[0007] S1. Filter high-quality data from each psoriasis patient record, wherein the psoriasis patient record includes historical medical records, bone density data, and biomarker data;

[0008] S2. Extract multimodal features from each of the high-quality data, capture high-order feature interactions through a graph attention network, and then perform multimodal fusion;

[0009] S3. Constructing a deep attention neural network for the risk of osteopenia in patients with psoriasis, and outputting a predicted probability of the risk of osteopenia through the deep attention neural network;

[0010] S4. Build a model based on model attention weights and SHAP values ​​to visualize individual prediction results.

[0011] S5. Distill, prune, and quantize the deep attention neural network and compress it through the edge computing model;

[0012] S6. Collect actual clinical results and doctor feedback data to optimize the parameters of the deep attention neural network.

[0013] According to the above aspects and any possible implementation, an implementation is further provided, wherein S1 specifically includes:

[0014] S11, extracting feature vector X from the historical medical records, bone density data, and biomarker data i =[X ehr ,X bmd ,X bio ];in,

[0015] Perform natural language processing on historical medical records and use medical text embedding models to extract medical record text features X ehr ;

[0016] Extract bone density measurement indicators: lumbar spine bone density L-BMD, femoral neck bone density F-BMD, T-score and Z-score to form the bone density feature vector X bmd =[L-BMD,F-BMD,T-score,Z-score];

[0017] Extract biomarker indicators: inflammatory factors IL-6, TNF-α, CRP and bone metabolism indicators OC, CTX, P1NP to form the bone density feature vector X bmd =[IL-6,TNF-α,CRP,OC,CTX,P1NP];

[0018] S12. Calculate data reconstruction error using the autoencoder model, determine the abnormality scoring threshold, and perform abnormal data screening; wherein,

[0019] If the reconstruction error AS i >T, then data X i Data that are judged to be abnormal are screened out; the abnormal score threshold T = μ(AS) + α·σ(AS), the reconstruction error h i =f enc (X i ), hi is the encoding representation of the data, f enc and f dec are the encoding and decoding functions of the autoencoder, α is the anomaly threshold adjustment parameter, μ(AS) and σ(AS) are the mean and standard deviation of the anomaly score;

[0020] S13. Calculate the integrity score and consistency score of the data after preliminary screening, obtain high-quality data scores through a multi-factor comprehensive scoring model, and screen out high-quality data;

[0021] The multi-factor comprehensive scoring model is expressed as follows: The completeness score is expressed as The consistency score is expressed as n missing is the number of missing features, n total is the total number of features, n inconsistent is the number of features that violate the rule, n checks is the total number of rule checks, AS max is the maximum value of the abnormality score, and ω1, ω2 and ω3 are weight coefficients.

[0022] According to the above aspects and any possible implementation, an implementation is further provided, wherein S2 specifically includes:

[0023] S21. Construct a multimodal feature heterogeneous graph based on medical record text features, bone density features, and biomarker features;

[0024] Construct node set: V=V text ∪V bmd ∪V bio , V text is the feature node set of medical record text, V bmd is the bone density index node set, V bio is the biomarker indicator node set;

[0025] Construct edge set: E={(v i ,r k ,v j )|v i ,v j ∈V,r k ∈R}, including homomodal edges, cross-modal edges and edge types;

[0026] Construct a set of relationship types: R = {r1, r2, ..., v k}, including at least co-occurrence, statistical correlation, pathway regulation, clinical guidance, drug efficacy, and intra-modality similarity;

[0027] Define the modal feature heterogeneous graph as G = (V, E, R);

[0028] S22, based on the graph attention network, learn the high-order feature interaction relationship between nodes and obtain the node embedding feature;

[0029] The feature update formula of the graph attention network layer is: is the feature representation of node i in layer l, W (l) is the weight matrix learned in the lth layer, σ(·) is the activation function, and N(i) is the set of neighbor nodes of node i;

[0030] The attention weight formula is expressed as: a (l) and W (l) are the model trainable parameters, is the contribution of node j to the feature update of node i;

[0031] S23, calculate the feature fusion weights between modalities through cross-modal attention, and output the fused multimodal feature representation; where,

[0032] The multimodal features after fusion are h fusion =Concat(z text ,z bmd ,z bio ,Att text←bmd ,Att text←bio ,Att bmd←bio ); Each modality is pooled to obtain an overall representation: The calculation formula of the attention of modality m to modality n is: W Q , W K and W V is the learnable weight matrix and d is the feature dimension.

[0033] According to the above aspects and any possible implementation, an implementation is further provided, wherein S3 specifically includes:

[0034] S31. Construct a deep attention neural network, wherein the network includes at least two nonlinear hidden layers, and introduces a feature attention weighting mechanism in the hidden layer output feature representation;

[0035] S32. Outputting a continuous probability prediction value of the risk of osteopenia in patients with psoriasis through the deep attention neural network;

[0036] S33. Generate an individualized bone mass loss risk prediction report for psoriasis patients based on the predicted probability value; the report includes the predicted probability of bone mass loss risk, the influencing characteristics of the bone mass loss risk and its contribution, and recommends an individualized prevention or treatment strategy based on the prediction results.

[0037] According to the above aspects and any possible implementation, an implementation is further provided, wherein S31 specifically includes:

[0038] S311. Map the input multimodal fusion feature representation to the first hidden layer to obtain an intermediate representation:

[0039] h (1) =σ(W (1) h fusion +b (1) );

[0040] S312: Map the intermediate representation to the second hidden layer to obtain a higher-order feature representation:

[0041] h (2) =σ(W (2) h (1) +b (2) );

[0042] S313. Calculate the attention weight vector for the feature representation of the second hidden layer through the attention mechanism:

[0043] α=softmax(W attn h (2) +b attn );

[0044] S314. Use the attention weight vector to weight the feature representation of the second hidden layer element by element to obtain a weighted feature representation:

[0045] h attn =αe h (2)

[0046] S315, mapped to the predicted probability value of bone loss risk through the output layer:

[0047]

[0048] Among them, W (1) 、W (2) and W (3) The weight matrix of each layer, b (1) 、b (2) and b (3) is the bias term of the corresponding layer, σ(·) is the nonlinear activation function, W attn and b attn are the attention weights and biases, h (1) and h (2) is the hidden layer feature representation;

[0049] The cross entropy loss function of the deep attention neural network is expressed as:

[0050]

[0051] Among them, y i represents the true label of the i-th sample, represents the predicted probability value of osteopenia risk of the i-th sample, N is the total number of training samples, and θ is the set of model trainable parameters including all weights and bias terms.

[0052] According to the above aspects and any possible implementation, an implementation is further provided, wherein S4 specifically includes:

[0053] S41, extract the attention weight of each input feature of the patient sample during the prediction process;

[0054] S42. Using the Shapley addition interpretation algorithm, the Shapley value of each feature is calculated for the current input features and the predicted results of the patient sample. The calculation formula is:

[0055]

[0056] in, is the Shapley value of the i-th feature for the current sample x, N is the total feature set, M is the total number of features, S is any feature subset that does not contain i, f x is the model prediction output for a specific feature set, S∪{i} is the union of subset S and feature i;

[0057] S43. A feature importance bar chart is plotted with Shapley values ​​on the horizontal axis, showing the quantitative impact of the main features on individual predictions and the positive and negative directions. Also, a one-dimensional heat map is constructed based on the feature-level attention vector α, with the feature names arranged on the axis and the corresponding α values ​​represented by light and dark colors.

[0058] S44. Integrate the graphics generated by S41-S43 and the specified numerical values ​​into the interactive interface for display.

[0059] According to the above aspects and any possible implementation, an implementation is further provided, wherein S5 specifically includes:

[0060] S51. Transfer the knowledge of the original teacher model with higher performance to the student model with lighter structure through knowledge distillation of the trained deep attention neural network.

[0061] S52. Prune the weight parameters, neurons, feature channels or branches in the neural network according to the absolute value of the weight, feature attention weight or structural importance score, and prune the parts below the set threshold; the pruning formula is:

[0062]

[0063] Among them, θ j is the parameter to be pruned, Threshold is the threshold;

[0064] S53. Convert the model weights and activation values ​​of the pruned model from high-precision floating-point numbers to low-bit width using a linear or dynamic quantization method:

[0065]

[0066] Among them, ω is the original weight, ω q is the quantization weight, ω min and ω max is the weight interval, b is the number of quantization bits;

[0067] S54. Export the compressed model after distillation, pruning, and quantization into a format supported by the edge computing device and deploy it to the designated edge platform.

[0068] According to the above aspects and any possible implementation, an implementation is further provided, wherein S51 specifically includes:

[0069] S511. The soft label probability distribution generated by the teacher model is used as the training target of the student model, and the distillation loss function is defined as:

[0070] L distill =KL(softmax(z T / T),softmax(z S / T));

[0071] Among them, z T is the output logits of the teacher model, z S is the output logits of the student model, T is the distillation temperature coefficient, and KL is the Kullback-Leibler divergence;

[0072] S512. The distillation loss is weighted and fused with the cross entropy loss of the student model on the true label to form the total loss function:

[0073] L total =λ1L hard +λ2L distill ;

[0074] Among them, λ1 and λ2 are weighting coefficients;

[0075] S513. Train the student model through backpropagation and gradient descent algorithm.

[0076] According to the above aspects and any possible implementation, an implementation is further provided, wherein S6 specifically includes:

[0077] S61. Collect clinical actual results and doctor feedback data; the clinical actual results are represented by binary labels as The feedback data given by the doctor based on the model prediction results is expressed as F j ={(f i ,v i )}, j = 1, 2, ..., J, f i is the feature that the doctor believes the model underestimates or overestimates, v i is the correction value of the corresponding feature;

[0078] S62. Construct a joint incremental loss function based on the historical training data loss term, the actual clinical result loss term, and the doctor feedback data loss term:

[0079]

[0080] Among them, θ is the parameter set to be optimized for the model, λ real and λ fb is the loss term weight coefficient, y i is the true label in the original training data, is the model’s predicted probability for the data, is the actual clinical diagnosis label collected, N is the sample size of the original training dataset, N r is the sample size of new feedback clinical data, J is, ω j is the feature f in the current model i The weight of Characteristics expected for feedback f i The weight correction value, L base (θ) is the historical training data loss, L real (θ) is the actual clinical result loss, L fb (θ) is the doctor feedback data loss;

[0081] S63. Update parameters using the gradient descent method based on the joint incremental loss function.

[0082] On the other hand, an embodiment of the present invention provides a public platform for dynamically predicting the risk of osteopenia in patients with psoriasis, the public platform comprising:

[0083] Central server, medical terminal and user interaction equipment;

[0084] The central server includes:

[0085] a screening module for screening high-quality data from each psoriasis patient record, wherein the psoriasis patient record includes historical medical records, bone density data, and biomarker data;

[0086] An extraction module is used to extract multimodal features from each of the high-quality data, and to perform multimodal fusion after capturing high-order feature interactions through a graph attention network;

[0087] A prediction module, configured to construct a deep attention neural network for the risk of osteopenia in patients with psoriasis, and output a predicted probability of the risk of osteopenia through the deep attention neural network;

[0088] The visualization module is used to build a model based on the model attention weight and SHAP value to visualize the individual prediction results;

[0089] Compression module, used to distill, prune and quantize deep attention neural networks and compress them through edge computing models;

[0090] The optimization module is used to collect actual clinical results and doctor feedback data to optimize the parameters of the deep attention neural network.

[0091] One of the above technical solutions has the following beneficial effects:

[0092] (1) By innovatively integrating multimodal data such as medical records, bone density, and biomarkers, and adopting graph neural networks and cross-modal attention mechanisms, it is possible to fully explore the complex high-order relationships between features and provide a more accurate and personalized risk probability assessment for each patient, which is superior to traditional single-modality or linear models.

[0093] (2) This technology combines the model's inherent attention weight with the SHAP-based feature attribution method to clearly display the main influencing features and their contributions, providing doctors and patients with a traceable and verifiable decision-making basis, thereby enhancing the trust and usability of the artificial intelligence system.

[0094] (3) Through model compression technologies such as knowledge distillation, pruning and quantization, the number of model parameters and computational complexity are significantly reduced, enabling high-performance prediction models to run efficiently on edge computing platforms such as local hospital servers, mobile terminals or wearable devices, meeting the real-time and energy consumption requirements in actual clinical scenarios.

[0095] (4) The platform can continuously collect real clinical results from patients and physician feedback, and utilize incremental learning and expert guidance to achieve dynamic optimization and self-correction of model parameters, so that the prediction system always maintains a high degree of consistency with the latest clinical practice, improves the long-term stability and adaptability of the model, and greatly improves the early identification and intervention level of complications such as bone loss in high-risk populations such as psoriasis.

Brief Description of the Drawings

[0096] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0097] Figure 1 A schematic flow chart of a method for dynamically predicting the risk of bone loss in psoriasis patients provided by an embodiment of the present invention;

[0098] Figure 2 A schematic block diagram of a public platform central server provided in an embodiment of the present invention;

[0099] Figure 3 A schematic diagram of the structure of the public platform provided by an embodiment of the present invention. [Specific implementation method]

[0100] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0101] Please refer to Figure 1 , which is a flow chart of a method for dynamically predicting the risk of osteopenia in psoriasis patients provided by an embodiment of the present invention. As shown in the figure, the method includes the following steps:

[0102] S1. Filter high-quality data from each psoriasis patient record, wherein the psoriasis patient record includes historical medical records, bone density data, and biomarker data;

[0103] S2. Extract multimodal features from each of the high-quality data, capture high-order feature interactions through a graph attention network, and then perform multimodal fusion;

[0104] S3. Constructing a deep attention neural network for the risk of osteopenia in patients with psoriasis, and outputting a predicted probability of the risk of osteopenia through the deep attention neural network;

[0105] S4. Build a model based on model attention weights and SHAP values ​​to visualize individual prediction results.

[0106] S5. Distill, prune, and quantize the deep attention neural network and compress it through the edge computing model;

[0107] S6. Collect actual clinical results and doctor feedback data to optimize the parameters of the deep attention neural network.

[0108] The purpose of S1 is to accurately extract feature vectors from historical medical records, bone density data, and biomarker data of psoriasis patients, and through innovative anomaly detection and data quality scoring models, automatically screen out high-quality data, providing a solid data foundation for subsequent multimodal fusion prediction. S1 specifically includes:

[0109] S11, extracting feature vector X from the historical medical records, bone density data, and biomarker data i =[X ehr ,X bmd ,X bio ];in,

[0110] Perform natural language processing on historical medical records and use medical text embedding models to extract medical record text features X ehr ;

[0111] Extract bone density measurement indicators: lumbar spine bone density L-BMD, femoral neck bone density F-BMD, T-score and Z-score to form the bone density feature vector X bmd =[L-BMD,F-BMD,T-score,Z-score];

[0112] Extract biomarker indicators: inflammatory factors IL-6, TNF-α, CRP and bone metabolism indicators OC, CTX, P1NP to form the bone density feature vector X bmd =[IL-6,TNF-α,CRP,OC,CTX,P1NP];

[0113] Bone density measurements included lumbar spine bone mineral density (L-BMD), femoral neck bone mineral density (F-BMD), T-score, and Z-score. Biomarkers included inflammatory factors (IL-6, TNF-α, CRP) and bone metabolism markers (OC, CTX, P1NP).

[0114] S12. Calculate data reconstruction error using the autoencoder model, determine the abnormality scoring threshold, and perform abnormal data screening; wherein,

[0115] If the reconstruction error AS i >T, then data X i Data that are judged to be abnormal are screened out; the abnormal score threshold T = μ(AS) + α·σ(AS), the reconstruction error h i =f enc (X i ), h i is the encoding representation of the data, fenc and f dec are the encoding and decoding functions of the autoencoder, α is the anomaly threshold adjustment parameter, μ(AS) and σ(AS) are the mean and standard deviation of the anomaly score;

[0116] The autoencoder model can be trained in an unsupervised manner, and the optimization goal is to minimize the reconstruction error of normal data in the training set.

[0117] S13. Calculate the integrity score and consistency score of the data after preliminary screening, obtain high-quality data scores through a multi-factor comprehensive scoring model, and screen out high-quality data;

[0118] The multi-factor comprehensive scoring model is expressed as follows: The completeness score is expressed as The consistency score is expressed as n missing is the number of missing features, n total is the total number of features, n inconsistent is the number of features that violate the rule, n checks is the total number of rule checks, AS max is the maximum value of the abnormality score, and ω1, ω2 and ω3 are weight coefficients.

[0119] S2 uses innovative graph neural networks (GNNs) and cross-modal attention mechanisms to accurately capture high-order nonlinear interactions between multimodal data (medical records, bone density, and biomarkers) of psoriasis patients, enabling more accurate and personalized risk prediction. S2 specifically includes:

[0120] S21. Construct a multimodal feature heterogeneous graph based on medical record text features, bone density features, and biomarker features;

[0121] Construct node set: V = V text ∪V bmd ∪V bio , V text is the feature node set of medical record text, V bmd is the bone density index node set, V bio is the biomarker indicator node set;

[0122] Construct edge set: E={(v i ,r k ,v j )|v i ,v j ∈V,r k ∈R}, including homomodal edges, cross-modal edges and edge types;

[0123] Construct a set of relationship types: R = {r1, r2, ..., v k}, including at least co-occurrence, statistical correlation, pathway regulation, clinical guidance, drug efficacy, and intra-modality similarity;

[0124] Define the modal feature heterogeneous graph as G = (V, E, R);

[0125] Here are some examples to illustrate homomodal edges, cross-modal edges, and relationship types. Homomodal edges include intra-text edges (Text–Text), intra-bone density edges (BMD–BMD), and intra-biomarker edges (Bio–Bio), such as ("psoriatic arthritis", "methotrexate", co-occurrence), ("T-score", "Z-score", strong correlation), and ("IL-6", "TNF-α", co-regulation of inflammation). Cross-modal edges include Text–BMD, Text–Bio, and BMD–Bio, such as ("psoriasis", "L-BMD", guiding testing), ("chronic inflammation", "IL-6", common pathway), and ("Z-score", "osteocalcin", metabolic correlation). Relationship types include co-occurrence, statistical correlation, pathway regulation, clinical guidance, causality / drug efficacy, and intra-modal similarity.

[0126] S22, based on the graph attention network, learn the high-order feature interaction relationship between nodes and obtain the node embedding feature;

[0127] The feature update formula of the graph attention network layer is: is the feature representation of node i in layer l, W (l) is the weight matrix learned in the lth layer, σ(·) is the activation function, and N(i) is the set of neighbor nodes of node i;

[0128] The attention weight formula is expressed as: a (l) and W (l) are the model trainable parameters, is the contribution of node j to the feature update of node i;

[0129] S23, calculate the feature fusion weights between modalities through cross-modal attention, and output the fused multimodal feature representation; where,

[0130] The multimodal features after fusion are h fusion =Concat(z text ,z bmd ,z bio ,Att text←bmd ,Att text←bio ,Att bmd←bio ); Each modality is pooled to obtain an overall representation: The calculation formula of the attention of modality m to modality n is: W Q, W K and W V is the learnable weight matrix and d is the feature dimension.

[0131] Specifically, S3 includes:

[0132] S31. Construct a deep attention neural network, wherein the network comprises at least two nonlinear hidden layers.

[0133] And introduce the feature attention weighting mechanism into the hidden layer output feature representation;

[0134] S31 specifically includes:

[0135] S311. Map the input multimodal fusion feature representation to the first hidden layer to obtain an intermediate representation:

[0136] h (1) =σ(W (1) h fusion +b (1) );

[0137] S312: Map the intermediate representation to the second hidden layer to obtain a higher-order feature representation:

[0138] h (2) =σ(W (2) h (1) +b (2) );

[0139] S313. Calculate the attention weight vector for the feature representation of the second hidden layer through the attention mechanism:

[0140] α=softmax(W attn h (2) +b attn );

[0141] S314. Use the attention weight vector to weight the feature representation of the second hidden layer element by element to obtain a weighted feature representation:

[0142] h attn =αe h (2)

[0143] S315, mapped to the predicted probability value of bone loss risk through the output layer:

[0144]

[0145] Among them, W (1) 、W (2) and W (3) The weight matrix of each layer, b (1) 、b (2) and b (3)is the bias term of the corresponding layer, σ(·) is the nonlinear activation function, W attn and b attn are the attention weights and biases, h (1) and h (2) is the hidden layer feature representation;

[0146] The cross entropy loss function of the deep attention neural network is expressed as:

[0147]

[0148] Among them, y i represents the true label of the i-th sample, represents the predicted probability value of osteopenia risk of the i-th sample, N is the total number of training samples, and θ is the set of model trainable parameters including all weights and bias terms.

[0149] S32. Outputting a continuous probability prediction value of the risk of osteopenia in patients with psoriasis through the deep attention neural network;

[0150] S33. Generate an individualized bone mass loss risk prediction report for psoriasis patients based on the predicted probability value; the report includes the predicted probability of bone mass loss risk, the influencing characteristics of the bone mass loss risk and its contribution, and recommends an individualized prevention or treatment strategy based on the prediction results.

[0151] Specifically, S4 includes:

[0152] S41, extract the attention weight of each input feature of the patient sample during the prediction process;

[0153] S42. Using the Shapley addition interpretation algorithm, the Shapley value of each feature is calculated for the current input features and the predicted results of the patient sample. The calculation formula is:

[0154]

[0155] in, is the Shapley value of the i-th feature for the current sample x, N is the total feature set, M is the total number of features, S is any feature subset that does not contain i, f x is the model prediction output for a specific feature set, S∪{i} is the union of subset S and feature i;

[0156] S43. A feature importance bar chart is plotted with Shapley values ​​on the horizontal axis, showing the quantitative impact of the main features on individual predictions and the positive and negative directions. Also, a one-dimensional heat map is constructed based on the feature-level attention vector α, with the feature names arranged on the axis and the corresponding α values ​​represented by light and dark colors.

[0157] S44. Integrate the graphics generated by S41-S43 and the specified numerical values ​​into the interactive interface for display.

[0158] Among them, S5 specifically includes:

[0159] S51. Transfer the knowledge of the original teacher model with higher performance to the student model with lighter structure through knowledge distillation of the trained deep attention neural network.

[0160] S51 specifically includes:

[0161] S511. The soft label probability distribution generated by the teacher model is used as the training target of the student model, and the distillation loss function is defined as:

[0162] L distill =KL(softmax(z T / T),softmax(z S / T));

[0163] Among them, z T is the output logits of the teacher model, z S is the output logits of the student model, T is the distillation temperature coefficient, and KL is the Kullback-Leibler divergence;

[0164] S512. The distillation loss is weighted and fused with the cross entropy loss of the student model on the true label to form the total loss function:

[0165] L total =λ1L hard +λ2L distill ;

[0166] Among them, λ1 and λ2 are weighting coefficients;

[0167] S513. Train the student model through backpropagation and gradient descent algorithm.

[0168] S52. Prune the weight parameters, neurons, feature channels or branches in the neural network according to the absolute value of the weight, feature attention weight or structural importance score, and prune the parts below the set threshold; the pruning formula is:

[0169]

[0170] Among them, θ j is the parameter to be pruned, Threshold is the threshold;

[0171] S53. Convert the model weights and activation values ​​of the pruned model from high-precision floating-point numbers to low-bit width using a linear or dynamic quantization method:

[0172]

[0173] Among them, ω is the original weight, ω q is the quantization weight, ω min and ω max is the weight interval, b is the number of quantization bits;

[0174] S54. Export the compressed model after distillation, pruning, and quantization into a format supported by the edge computing device and deploy it to the designated edge platform.

[0175] Among them, S6 specifically includes:

[0176] S61. Collect clinical actual results and doctor feedback data; the clinical actual results are represented by binary labels as The feedback data given by the doctor based on the model prediction results is expressed as F j ={(f i ,v i )}, j = 1, 2, ..., J, f i is the feature that the doctor believes the model underestimates or overestimates, v i is the correction value of the corresponding feature;

[0177] S62. Construct a joint incremental loss function based on the historical training data loss term, the actual clinical result loss term, and the doctor feedback data loss term:

[0178]

[0179] Among them, θ is the parameter set to be optimized for the model, λ real and λ fb is the loss term weight coefficient, y i is the true label in the original training data, is the model’s predicted probability for the data, is the actual clinical diagnosis label collected, N is the sample size of the original training dataset, N r is the sample size of new feedback clinical data, J is, ω j is the feature f in the current model i The weight of Characteristics expected for feedback f i The weight correction value, L base (θ) is the historical training data loss, L real (θ) is the actual clinical result loss, L fb (θ) is the doctor feedback data loss;

[0180] S63. Update parameters using the gradient descent method based on the joint incremental loss function.

[0181] Through the above steps, the present invention achieves the following technical effects:

[0182] (1) By innovatively integrating multimodal data such as medical records, bone density, and biomarkers, and adopting graph neural networks and cross-modal attention mechanisms, it is possible to fully explore the complex high-order relationships between features and provide a more accurate and personalized risk probability assessment for each patient, which is superior to traditional single-modality or linear models.

[0183] (2) This technology combines the model's inherent attention weight with the SHAP-based feature attribution method to clearly display the main influencing features and their contributions, providing doctors and patients with a traceable and verifiable decision-making basis, thereby enhancing the trust and usability of the artificial intelligence system.

[0184] (3) Through model compression technologies such as knowledge distillation, pruning and quantization, the number of model parameters and computational complexity are significantly reduced, enabling high-performance prediction models to run efficiently on edge computing platforms such as local hospital servers, mobile terminals or wearable devices, meeting the real-time and energy consumption requirements in actual clinical scenarios.

[0185] (4) The platform can continuously collect real clinical results from patients and physician feedback, and utilize incremental learning and expert guidance to achieve dynamic optimization and self-correction of model parameters, so that the prediction system always maintains a high degree of consistency with the latest clinical practice, improves the long-term stability and adaptability of the model, and greatly improves the early identification and intervention level of complications such as bone loss in high-risk populations such as psoriasis.

[0186] The embodiments of the present invention further provide device embodiments for implementing the steps and methods in the above method embodiments.

[0187] Please refer to Figure 2 , which is provided by the embodiment of the present invention Figure 2 The public platform central server provided by the embodiment of the present invention includes:

[0188] A screening module 210 is used to screen high-quality data from each psoriasis patient record, wherein the psoriasis patient record includes historical medical records, bone density data, and biomarker data;

[0189] Extraction module 220, for performing multimodal feature extraction on each of the high-quality data, capturing high-order feature interactions through a graph attention network, and performing multimodal fusion;

[0190] Prediction module 230, for constructing a deep attention neural network for the risk of osteopenia in patients with psoriasis, and outputting a predicted probability of the risk of osteopenia through the deep attention neural network;

[0191] A visualization module 240 is used to establish a model based on the model attention weight and SHAP value to visualize the individual prediction results;

[0192] Compression module 250, for distilling, pruning, and quantizing the deep attention neural network and compressing it through the edge computing model;

[0193] The optimization module 260 is used to collect actual clinical results and doctor feedback data to optimize the parameters of the deep attention neural network.

[0194] Since each unit module in this embodiment can execute Figure 1 For the method shown in the embodiment, the part not described in detail in this embodiment can be referred to Figure 1 Related instructions.

[0195] Please refer to Figure 3 , which is a structural diagram of the public platform system provided by an embodiment of the invention.

[0196] include Figure 2 The central server, medical terminals and user interaction devices, and edge inference and acquisition devices are shown. Medical terminals and user interaction devices include but are not limited to doctor workstations, mobile medical tablets, and patient self-query terminals. Edge inference and acquisition devices include but are not limited to departmental edge AI boxes and bedside all-in-one machines.

[0197] At the hardware level, the central server may include a processor and, optionally, an internal bus, a network interface, and memory. The memory may include internal memory, such as high-speed random-access memory (RAM), and non-volatile memory, such as at least one disk drive. Of course, the central server may also include other hardware required for the business.

[0198] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus. The bus can be divided into an address bus, a data bus, a control bus, and the like.

[0199] The memory is used to store programs. Specifically, the program may include program code, which includes computer operating instructions. The memory may include internal memory and non-volatile memory, and provides instructions and data to the processor.

[0200] The steps of the method disclosed in conjunction with the embodiments of the present invention can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules within the decoding processor. The software modules can be located in a storage medium well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the method described above.

[0201] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0202] For the convenience of description, the above device is described as being divided into various units or modules according to their functions. Of course, when implementing the present invention, the functions of each unit or module can be implemented in the same or multiple software and / or hardware.

[0203] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0204] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts 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, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1A device that provides the functions specified in a block or multiple blocks.

[0205] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0206] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0207] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0208] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0209] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0210] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0211] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0212] The present invention may be described in the general context of computer-executable instructions, such as program modules, executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media, including storage devices.

[0213] The various embodiments of the present invention are described in a progressive manner. Similar portions between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiment is generally similar to the method embodiment, so its description is relatively simple. For relevant portions, refer to the description of the method embodiment.

[0214] The foregoing is merely an embodiment of the present invention and is not intended to limit the present invention. It will be apparent to those skilled in the art that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are intended to be included within the scope of the claims of the present invention.

Claims

1. A method for dynamically predicting the risk of osteopenia in patients with psoriasis, characterized in that: The method comprises: S1. Filter high-quality data from each psoriasis patient record, wherein the psoriasis patient record includes historical medical records, bone density data, and biomarker data; S2. Extract multimodal features from each of the high-quality data, capture high-order feature interactions through a graph attention network, and then perform multimodal fusion; S3. Constructing a deep attention neural network for the risk of osteopenia in patients with psoriasis, and outputting a predicted probability of the risk of osteopenia through the deep attention neural network; S4. Build a model based on model attention weights and SHAP values ​​to visualize individual prediction results. S5. Distill, prune, and quantize the deep attention neural network and compress it through the edge computing model; S6. Collect actual clinical results and doctor feedback data to optimize the parameters of the deep attention neural network.

2. The method for dynamically predicting the risk of osteopenia in patients with psoriasis according to claim 1, characterized in that: Said S1 specifically includes: S11, extracting feature vector X from the historical medical records, bone density data, and biomarker data i =[X ehr ,X bmd ,X bio ];in, Perform natural language processing on historical medical records and use medical text embedding models to extract medical record text features X ehr ; Extract bone density measurement indicators: lumbar spine bone density L-BMD, femoral neck bone density F-BMD, T-score and Z-score to form the bone density feature vector X bmd =[L-BMD,F-BMD,T-score,Z-score]; Extract biomarker indicators: inflammatory factors IL-6, TNF-α, CRP and bone metabolism indicators OC, CTX, P1NP to form the bone density feature vector X bmd =[IL-6,TNF-α,CRP,OC,CTX,P1NP]; S12. Calculate data reconstruction error using the autoencoder model, determine the abnormality scoring threshold, and perform abnormal data screening; wherein, If the reconstruction error AS i >T, then data X i Data that are judged to be abnormal are screened out; the abnormal score threshold T = μ(AS) + α·σ(AS), the reconstruction error h i =f enc (X i ), h i is the encoding representation of the data, f enc and f dec are the encoding and decoding functions of the autoencoder, α is the anomaly threshold adjustment parameter, μ(AS) and σ(AS) are the mean and standard deviation of the anomaly score; S13. Calculate the integrity score and consistency score of the data after preliminary screening, obtain high-quality data scores through a multi-factor comprehensive scoring model, and screen out high-quality data; The multi-factor comprehensive scoring model is expressed as follows: The completeness score is expressed as The consistency score is expressed as n missing is the number of missing features, n total is the total number of features, n inconsistent is the number of features that violate the rule, n checks is the total number of rule checks, AS max is the maximum value of the abnormality score, and ω1, ω2 and ω3 are weight coefficients.

3. The method for dynamically predicting the risk of osteopenia in patients with psoriasis according to claim 1, characterized in that: The S2 specifically includes: S21. Construct a multimodal feature heterogeneous graph based on medical record text features, bone density features, and biomarker features; Construct node set: V = V text ∪V bmd ∪V bio , V text is the feature node set of medical record text, V bmd is the bone density index node set, V bio is the biomarker indicator node set; Construct edge set: E={(v i ,r k ,v j )|v i ,v j ∈V,r k ∈R}, including homomodal edges, cross-modal edges and edge types; Construct a set of relationship types: R = {r1, r2, ..., v k }, including at least co-occurrence, statistical correlation, pathway regulation, clinical guidance, drug efficacy, and intra-modality similarity; Define the modal feature heterogeneous graph as G = (V, E, R); S22, based on the graph attention network, learn the high-order feature interaction relationship between nodes and obtain the node embedding feature; The feature update formula of the graph attention network layer is: is the feature representation of node i in layer l, W (l) is the weight matrix learned in the lth layer, σ(·) is the activation function, and N(i) is the set of neighbor nodes of node i; The attention weight formula is expressed as: a (l) and W (l) are the model trainable parameters, is the contribution of node j to the feature update of node i; S23, calculate the feature fusion weights between modalities through cross-modal attention, and output the fused multimodal feature representation; where, The multimodal features after fusion are h fusion =Concat(z text ,z bmd ,z bio ,Att text←bmd ,Att text←bio ,Att bmd←bio ); Each modality is pooled to obtain an overall representation: m∈{text,bmd,bio}; The attention calculation formula of modality m to modality n is: W Q , W K and W V is the learnable weight matrix and d is the feature dimension.

4. The method for dynamically predicting the risk of osteopenia in patients with psoriasis according to claim 3, characterized in that: The S3 specifically includes: S31. Construct a deep attention neural network, wherein the network includes at least two nonlinear hidden layers, and introduces a feature attention weighting mechanism in the hidden layer output feature representation; S32. Outputting a continuous probability prediction value of the risk of osteopenia in patients with psoriasis through the deep attention neural network; S33. Generate an individualized bone mass loss risk prediction report for psoriasis patients based on the predicted probability value; the report includes the predicted probability of bone mass loss risk, the influencing characteristics of the bone mass loss risk and its contribution, and recommends an individualized prevention or treatment strategy based on the prediction results.

5. The method for dynamically predicting the risk of osteopenia in patients with psoriasis according to claim 3, characterized in that: The S31 specifically includes: S311. Map the input multimodal fusion feature representation to the first hidden layer to obtain an intermediate representation: h (1) =σ(W (1) h fusion +b (1) ); S312: Map the intermediate representation to the second hidden layer to obtain a higher-order feature representation: h (2) =σ(W (2) h (1) +b (2) ); S313. Calculate the attention weight vector for the feature representation of the second hidden layer through the attention mechanism: α=softmax(W attn h (2) +b attn ); S314. Use the attention weight vector to weight the feature representation of the second hidden layer element by element to obtain a weighted feature representation: h attn =αe h (2) S315, mapped to the predicted probability value of bone loss risk through the output layer: Among them, W (1) 、W (2) and W (3) The weight matrix of each layer, b (1) 、b (2) and b (3) is the bias term of the corresponding layer, σ(·) is the nonlinear activation function, W attn and b attn are the attention weights and biases, h (1) and h (2) is the hidden layer feature representation; The cross entropy loss function of the deep attention neural network is expressed as: Among them, y i represents the true label of the i-th sample, represents the predicted probability value of osteopenia risk of the i-th sample, N is the total number of training samples, and θ is the set of model trainable parameters including all weights and bias terms.

6. The method for dynamically predicting the risk of osteopenia in patients with psoriasis according to claim 4, characterized in that: The S4 specifically includes: S41, extract the attention weight of each input feature of the patient sample during the prediction process; S42. Using the Shapley addition interpretation algorithm, the Shapley value of each feature is calculated for the current input features and the predicted results of the patient sample. The calculation formula is: in, is the Shapley value of the i-th feature for the current sample x, N is the total feature set, M is the total number of features, S is any feature subset that does not contain i, f x is the model prediction output for a specific feature set, S∪{i} is the union of subset S and feature i; S43. A feature importance bar chart is plotted with Shapley values ​​on the horizontal axis, showing the quantitative impact of the main features on individual predictions and the positive and negative directions. Also, a one-dimensional heat map is constructed based on the feature-level attention vector α, with the feature names arranged on the axis and the corresponding α values ​​represented by light and dark colors. S44. Integrate the graphics generated by S41-S43 and the specified numerical values ​​into the interactive interface for display.

7. The method for dynamically predicting the risk of osteopenia in patients with psoriasis according to claim 1, characterized in that: The S5 specifically includes: S51. Transfer the knowledge of the original teacher model with higher performance to the student model with lighter structure through knowledge distillation of the trained deep attention neural network. S52. Prune the weight parameters, neurons, feature channels or branches in the neural network according to the absolute value of the weight, feature attention weight or structural importance score, and prune the parts below the set threshold; the pruning formula is: Among them, θ j is the parameter to be pruned, Threshold is the threshold; S53. Convert the model weights and activation values ​​of the pruned model from high-precision floating-point numbers to low-bit width using a linear or dynamic quantization method: Among them, ω is the original weight, ω q is the quantization weight, ω min and ω max is the weight interval, b is the number of quantization bits; S54. Export the compressed model after distillation, pruning, and quantization into a format supported by the edge computing device and deploy it to the designated edge platform.

8. The method for dynamically predicting the risk of osteopenia in patients with psoriasis according to claim 1, characterized in that: The S51 specifically includes: S511. The soft label probability distribution generated by the teacher model is used as the training target of the student model, and the distillation loss function is defined as: L distill =KL(softmax(z T / T),softmax(z S / T)); Among them, z T is the output logits of the teacher model, z S is the output logits of the student model, T is the distillation temperature coefficient, and KL is the Kullback-Leibler divergence; S512. The distillation loss is weighted and fused with the cross entropy loss of the student model on the true label to form the total loss function: L total =λ1L hard +λ2L distill ; Among them, λ1 and λ2 are weighting coefficients; S513. Train the student model through backpropagation and gradient descent algorithm.

9. The method for dynamically predicting the risk of osteopenia in patients with psoriasis according to claim 1, characterized in that: The S6 specifically includes: S61. Collect clinical actual results and doctor feedback data; the clinical actual results are represented by binary labels as The feedback data given by the doctor based on the model prediction results is expressed as F j ={(f i ,v i )}, j = 1, 2, ..., J, f i is the feature that the doctor believes the model underestimates or overestimates, v i is the correction value of the corresponding feature; S62. Construct a joint incremental loss function based on the historical training data loss term, the actual clinical result loss term, and the doctor feedback data loss term: Among them, θ is the parameter set to be optimized for the model, λ real and λ fb is the loss term weight coefficient, y i is the true label in the original training data, is the model’s predicted probability for the data, is the actual clinical diagnosis label collected, N is the sample size of the original training dataset, N r is the sample size of new feedback clinical data, J is, ω j is the feature f in the current model i The weight of Characteristics expected for feedback f i The weight correction value, L base (θ) is the historical training data loss, L real (θ) is the actual clinical result loss, L fb (θ) is the doctor feedback data loss; S63. Update parameters using the gradient descent method based on the joint incremental loss function.

10. A public platform for dynamically predicting the risk of osteopenia in psoriasis patients using the method according to any one of claims 1 to 9, characterized in that: The public platform includes: Central server, medical terminal and user interaction equipment; The central server includes: a screening module for screening high-quality data from each psoriasis patient record, wherein the psoriasis patient record includes historical medical records, bone density data, and biomarker data; An extraction module is used to extract multimodal features from each of the high-quality data, and to perform multimodal fusion after capturing high-order feature interactions through a graph attention network; A prediction module, configured to construct a deep attention neural network for the risk of osteopenia in patients with psoriasis, and output a predicted probability of the risk of osteopenia through the deep attention neural network; The visualization module is used to build a model based on the model attention weight and SHAP value to visualize the individual prediction results; Compression module, used to distill, prune and quantize deep attention neural networks and compress them through edge computing models; The optimization module is used to collect actual clinical results and doctor feedback data to optimize the parameters of the deep attention neural network.

Images