Method, device and equipment for predicting postoperative recovery condition of diabetic patient

Abstract

The invention relates to the technical field of medical information, in particular to a diabetic postoperative recovery condition prediction method, device and equipment, and the method comprises the steps: collecting historical specificity data and historical recovery condition data of historical diabetic patients; training the deep neural network model to obtain a recovery condition prediction model; acquiring target specific data of the target diabetic patient; standardized specific data is obtained; performing feature extraction on the standardized specific data to obtain respective feature data of the standardized specific data; fusing the respective feature data of the standardized specific data to form a comprehensive feature vector; on the basis of the comprehensive feature vector and a recovery condition prediction model, the complication occurrence probability and the predicted recovery time of the target diabetic patient are determined, features are extracted from specific data of the diabetic patient, and the features are fused, so that the complication occurrence probability and the predicted recovery time of the diabetic patient are predicted. And the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to a method, device and equipment for predicting the postoperative recovery of diabetic patients. Background Art

[0002] Diabetic patients face higher risks when undergoing surgery due to metabolic disorders and vascular and nervous system lesions. The incidence of postoperative complications is high and the recovery time is long, which poses a huge challenge to the patient's quality of life and medical resources.

[0003] Personalized minimally invasive surgery is an advanced surgical method that aims to reduce surgical trauma and accelerate patient recovery. However, due to the specific characteristics of diabetic patients themselves, the surgical risks are increased and it is impossible to accurately obtain postoperative recovery conditions.

[0004] Therefore, how to accurately predict the postoperative recovery of diabetic patients is a technical problem that needs to be solved urgently. Summary of the invention

[0005] In view of the above problems, the present invention provides a method, device and equipment for predicting the postoperative recovery of diabetic patients that overcome the above problems or at least partially solve the above problems.

[0006] In a first aspect, the present invention provides a method for predicting the postoperative recovery of a diabetic patient, comprising: Collect historical specific data and historical recovery data of historical diabetic patients; Based on the historical specific data and historical recovery data of the historical diabetic patients, a deep neural network model is trained to obtain a recovery prediction model; obtain target-specific data for target diabetic patients; Preprocessing the target-specific data to obtain standardized specific data; Performing feature extraction on the standardized specific data to obtain feature data of each of the standardized specific data; Merging the characteristic data of each of the standardized specific data to form a comprehensive characteristic vector; Based on the comprehensive feature vector and the recovery prediction model, the probability of complication occurrence and the estimated recovery time of the target diabetic patient are determined.

[0007] Preferably, the target-specific data includes: structured data, surgical parameter data, and image data; The structured data include: blood sugar level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and operation blood loss; The surgical parameter data include: surgical type and surgical instrument type; The image data includes: blood vessel images and tissue images.

[0008] Preferably, the target-specific data is preprocessed to obtain standardized specific data, including: The blood glucose level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and surgical blood loss were numerically standardized to obtain the corresponding standardized specific data: ; in, Standardized specific data for any one of blood glucose level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and surgical blood loss, The raw data is any one of blood sugar level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and operation blood loss. is the mean value of any one of the blood sugar level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and surgical blood loss. The standard is any one of blood sugar level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and surgical blood loss; Encode the operation type and the surgical instrument type respectively to obtain the operation type code and the surgical instrument type code; Both the vascular image and the tissue image are adjusted to a preset size, and the pixel values ​​are normalized to obtain standardized image data.

[0009] Preferably, feature extraction is performed on the standardized specific data to obtain feature data of each of the standardized specific data, including: Extracting feature vectors from the standardized specific data of the structured data respectively; Extracting high-order features from the feature vector to obtain feature data of the structured data; Mapping the surgery type code and the surgical instrument type code into low-dimensional dense vectors respectively to obtain a surgery type embedding vector and a surgical instrument type embedding vector; A feature map score and an attention weight are calculated for the standardized image data, and a weighted feature of the image data is obtained based on the feature map score and the attention weight.

[0010] Preferably, the characteristic data of each of the standardized specific data are merged to form a comprehensive characteristic vector, including: Performing linear transformation on the feature data of the structured data, the surgery type embedding vector, the surgical instrument type embedding vector, and the weighted features of the image data, respectively, to obtain a structured feature map, a surgery type embedding map, a surgical instrument type embedding map, and an image feature map; The structured feature map, the surgery type embedding map, the surgical instrument type embedding map and the image feature map are directly concatenated, and high-level features are extracted to form a comprehensive feature vector.

[0011] Preferably, the feature data of the structured data, the surgery type embedding vector, the surgical instrument type embedding vector, and the weighted features of the image data are linearly transformed to obtain a structured feature map, a surgery type embedding map, a surgical instrument type embedding map, and an image feature map, including: ; in, is any one of the feature data of the structured data, the surgery type embedding vector, the surgical instrument type embedding vector, and the weighted feature of the image data, is the hidden layer activation function, is any one of the structured feature map, the surgery type embedding map, the surgical instrument type embedding map, and the image feature map, is the weight matrix before concatenation, is the bias vector before splicing.

[0012] Preferably, the structured feature map, the surgery type embedding map, the surgical instrument type embedding map and the image feature map are directly spliced, and high-level features are extracted to form a comprehensive feature vector, including: ; in, represents a vector concatenation operation, is the structured feature map, embedding a mapping for said surgery type, embedding a mapping for said surgical instrument type, is the image feature map, This is the result after direct splicing; ; in, is the comprehensive feature vector, is the weight matrix after concatenation and is the offset vector after splicing, is the hidden layer activation function.

[0013] Preferably, determining the complication probability and estimated recovery time of the target diabetic patient based on the comprehensive feature vector and the recovery prediction model includes: Inputting the comprehensive feature vector into the recovery situation prediction model, the recovery situation prediction model comprises: an input layer, a hidden layer and an output layer, the hidden layer is a multi-layer fully connected layer; Output target probability of complications in diabetic patients: ; in, is the output of the hidden layer, is the Sigmoid output layer activation function, , is the probability of occurrence of the complication; Output the estimated recovery time for target diabetic patients: ; in, is a linear or nonlinear function, Estimated recovery time.

[0014] In a second aspect, the present invention further provides a device for predicting the postoperative recovery of diabetic patients, comprising: A collection module, used to collect historical specific data and historical recovery data of historical diabetic patients; A model acquisition module, used to train a deep neural network model based on the historical specific data and historical recovery data of the historical diabetic patients to obtain a recovery prediction model; An acquisition module, used for acquiring specific data of a target diabetic patient; A preprocessing module, used for preprocessing the specific data to obtain standardized specific data; A feature extraction module, used to extract features from the standardized specific data to obtain feature data of each of the standardized specific data; A fusion module, used for fusing the characteristic data of each of the standardized specific data to form a comprehensive characteristic vector; A determination module is used to determine the probability of complication occurrence and the estimated recovery time of the target diabetic patient based on the comprehensive feature vector and the recovery prediction model.

[0015] In a third aspect, the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the program.

[0016] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0017] One or more technical solutions in the embodiments of the present invention have at least the following technical effects or advantages: The present invention provides a method for predicting the postoperative recovery of diabetic patients, comprising: collecting historical specific data and historical recovery data of historical diabetic patients; training a deep neural network model based on the historical specific data and historical recovery data of historical diabetic patients to obtain a recovery prediction model; obtaining target specific data of target diabetic patients; preprocessing the target specific data to obtain standardized specific data; extracting features from the standardized specific data to obtain feature data of each of the standardized specific data; fusing the feature data of each of the standardized specific data to form a comprehensive feature vector; based on the comprehensive feature vector and the recovery prediction model, determining the probability of complication occurrence and the expected recovery time of the target diabetic patient, extracting features from the specific data of the diabetic patient, and fusing the features, thereby predicting the probability of complication occurrence and the expected recovery time of the diabetic patient, thereby improving the accuracy of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] 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. Also, throughout the accompanying drawings, the same reference figures are used to represent the same components. In the drawings: Figure 1 A schematic diagram of the steps of a method for predicting the postoperative recovery of a diabetic patient according to an embodiment of the present invention is shown; Figure 2 A schematic diagram of the structure of a device for predicting the postoperative recovery of a diabetic patient according to an embodiment of the present invention is shown; Figure 3 A schematic diagram of the structure of a computer device for implementing a method for predicting the postoperative recovery status of a diabetic patient in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0019] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0020] Embodiment 1 The embodiment of the present invention provides a method for predicting the postoperative recovery of diabetic patients. Figure 1 As shown, including: S101, collecting historical specific data and historical recovery data of historical diabetic patients; S102, training a deep neural network model based on historical specific data and historical recovery data of historical diabetic patients to obtain a recovery prediction model; S103, obtaining target-specific data of the target diabetic patient; S104, preprocessing the target specific data to obtain standardized specific data; S105, extracting features from the standardized specific data to obtain feature data of each of the standardized specific data; S106, fusing the feature data of each of the standardized specific data to form a comprehensive feature vector; S107, based on the comprehensive feature vector and the recovery prediction model, determine the probability of complication occurrence and the estimated recovery time of the target diabetic patient.

[0021] In a specific implementation, S101-S102 is to obtain a recovery status prediction model.

[0022] The recovery prediction model is specifically obtained by training through historical specific data and historical recovery data of historical diabetic patients.

[0023] Specifically, the historical specific data is data related to the historical diabetic patient's illness, including structured data, surgical parameter data, and image data. The structured data includes: historical blood sugar level data, historical glycosylated hemoglobin, historical insulin resistance index, historical complication index, historical operation time, and historical operation blood loss; the surgical parameter data includes: historical operation type and historical surgical instrument type; the image data includes: historical vascular images and historical tissue images.

[0024] Among them, historical complication indicators include the degree of retinopathy, renal function indicators and neuropathy score. Renal function indicators include serum creatinine and urine microalbumin.

[0025] It is necessary to pre-process each historical specific data, including standardization. Then, feature extraction is performed on the standardized historical specific data to obtain historical feature data of each historical specific data. Then, each historical feature data is fused to obtain a historical comprehensive feature vector. Finally, the historical comprehensive feature vector and historical recovery data are input into the deep neural network model for training, thereby obtaining a recovery prediction model.

[0026] In this process, the deep neural network model can specifically adopt CNN, or RNN, etc., which is not limited here. The recovery situation prediction model obtained at this time is only an initial model. Next, the model needs to be verified and optimized.

[0027] For the recovery prediction model, it is necessary to design a loss function, specifically a weighted loss function: the loss function L is defined as the weighted sum of the complication prediction loss and the recovery time prediction loss:

[0028] in, and is the weight coefficient, and , and set according to clinical needs.

[0029] Complications predict losses: Using binary cross entropy loss:

[0030] in, is the number of samples, For the The true labels of samples, Predicted probability of complications for any one training session.

[0031] Recovery time predicted loss: Using mean squared error loss:

[0032] Among them, is the predicted recovery time of any training session, and is the actual recovery time.

[0033] The recovery prediction model is trained in two stages: the first stage: training the structured sub-network, surgical parameter embedding and imaging sub-network separately, and initializing the model parameters. The second stage: jointly training the entire model and optimizing the fusion layer and output layer parameters.

[0034] Next, the algorithm is optimized using the Adam optimizer, and the learning rate is adaptively adjusted according to the training situation.

[0035]

[0036] in, are model parameters, is the gradient of the loss function with respect to the parameters.

[0037] In this way, the parameters of the recovery situation prediction model are continuously optimized to obtain the optimal recovery situation prediction model.

[0038] Next is the model explanation enhancement: Using Grad-CAM (gradient-weighted class activation mapping) technology, heat maps are generated in image data processing to display the key areas that the model focuses on and help doctors understand the decision-making basis of the model.

[0039] Among them, the Grad-CAM calculation formula is:

[0040] in, For the feature map, is the corresponding weight, is the hidden layer activation function.

[0041] Next is the validation of the model: The specific evaluation indicators include: evaluation indicators for complication prediction: accuracy, sensitivity, specificity and area under the ROC curve; evaluation indicators for recovery time prediction: mean square error and mean absolute error.

[0042] Cross-validation: Use k-fold cross-validation to evaluate the generalization ability of the model and prevent overfitting.

[0043] After obtaining the optimized model, the next step is to apply the model, which is specifically presented through S103~S107.

[0044] Before application, feature extraction is performed through S103 to S106.

[0045] Specifically, S103, obtaining target-specific data of the target diabetic patient.

[0046] The target-specific data specifically includes: structured data, surgical parameter data and image data.

[0047] The structured data is the same as the structured data in the history-specific data, including: blood sugar level data , Glycated hemoglobin (HbA1c) , insulin resistance index (HOMA-IR) , complication index, operation time and surgical blood loss Among them, the complication indicators include: the degree of retinopathy , renal function indexes and neuropathy scores . Renal function indicators include: serum creatinine and urine microalbumin .

[0048] Surgical parameter data include: type of surgery and surgical instrument type .

[0049] Image data include: Vascular imaging and tissue imaging .

[0050] Next, S104, various types of data are preprocessed to obtain standardized specific data.

[0051] Preprocessing of structured data: The blood glucose level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and surgical blood loss were numerically standardized to obtain the corresponding annotated specific data: ; in, Standardized specific data for any one of blood glucose level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and surgical blood loss, The raw data is any one of blood sugar level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and operation blood loss. is the mean value of any one of the blood sugar level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and surgical blood loss. It is the standard deviation of any one of the blood glucose level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and surgical blood loss.

[0052] For standardization of surgical parameter data: The operation type and the surgical instrument type are encoded respectively to obtain the operation type code and the surgical instrument type code. Specifically, each operation type is assigned a unique integer code: operation type 1: coded as 0; operation type 2: coded as 1; operation type 3, coded as 2... Each surgical instrument type is assigned a unique integer code: surgical instrument type 1: coded as 0; surgical instrument type 2: coded as 1; surgical instrument type 3: coded as 2... The vascular image and the tissue image are both adjusted to a preset size, and the pixel values ​​are normalized to obtain standardized image data. Specifically, the images are uniformly adjusted to a preset size, and the pixel value of each image is normalized to a range of [0,1].

[0053] After the standardization, S105 is executed to extract features from the standardized specific data to obtain feature data of each of the standardized specific data.

[0054] Specifically, feature data are extracted from the standardized specific data of the structured data respectively; high-order features are extracted from the feature vectors to obtain feature data of the structured data; The surgery type code and the surgical instrument type code are respectively mapped into low-dimensional dense vectors to obtain the surgery type embedding vector and the surgical instrument type embedding vector; The feature map scores and attention weights are calculated for the standardized image data, and based on the feature map scores and attention weights, the weighted features of the image data are obtained.

[0055] The feature extraction of the standardized specific data of structured data is as follows: Input feature vector:

[0056] Use a multi-layer fully connected neural network (MLP) to extract high-level features

[0057] The surgery type code and surgical instrument type code are mapped to low-dimensional dense vectors respectively: Specifically, an embedding layer is used to map the surgery type code into a low-dimensional dense vector:

[0058] in, is the embedding matrix of surgery types, with size , Code the type of surgery, is the total number of surgical types, is the dimension of the embedding vector (adjustable parameter).

[0059] Use an embedding layer to map the surgical instrument type encoding into a low-dimensional dense vector:

[0060] in, is the embedding matrix of surgical instrument types, with size , Code the surgical instrument type, is the total number of surgical instrument types, is the dimension of the embedding vector.

[0061] The feature extraction of standardized image data is as follows: In order to make the model pay more attention to the important areas related to diabetes, the CNN is improved by combining the attention mechanism, including channel attention and spatial attention.

[0062] Channel Attention: Calculate the importance of each channel (feature map) to highlight key features: Channel attention weight

[0063] in, is the feature map tensor, is the global average pooling, It is a multi-layer perceptron.

[0064] Spatial Attention Mechanism: Calculate the importance of each position in the feature map, focusing on key areas: Spatial attention weights

[0065] in, is the convolution operation, is the output layer activation function.

[0066] Through multi-scale feature fusion, features of different scales are extracted, local and global information are integrated, and the model's understanding of complex images is improved.

[0067] Attention mechanism calculation, including feature map score calculation and attention weight calculation.

[0068] Feature map score calculation: For each feature map , calculate the score:

[0069] For the The feature maps are globally averaged pooled. is the weight matrix, is the bias vector, For the feature map scores.

[0070] Attention weight calculation: The above scores are converted into weights through the Softmax function:

[0071] in, is the number of feature maps, is the weight.

[0072] Weighted feature representation:

[0073] To expand the feature map into a one-dimensional vector, is a weighted feature.

[0074] Thus, the weighted features of the vascular image and the weighted features of the tissue image are obtained.

[0075] Among them, the channel attention mechanism uses the attention mechanism to calculate, determine the important feature types, and improve the model's sensitivity to key features; the spatial attention mechanism uses the attention mechanism to calculate, determine the important spatial positions, and improve the model's attention to key areas. Both the channel attention mechanism and the spatial attention mechanism are based on the attention mechanism calculation, and finally the important features are highlighted by calculating the scores and weights.

[0076] Next, S106 is executed to merge the feature data of each of the standardized specific data to form a comprehensive feature vector.

[0077] Specifically, linear transformation is performed on the feature data of the structured data, the surgery type embedding vector, the surgical instrument type embedding vector, and the weighted features of the image data to obtain a structured feature map, a surgery type embedding map, a surgical instrument type embedding map, and an image feature map; The structured feature map, surgery type embedding map, surgical instrument type embedding map and image feature map are directly concatenated, and high-level features are extracted to form a comprehensive feature vector.

[0078] ; in, is any one of the feature data of structured data, the embedding vector of surgical type, the embedding vector of surgical instrument type, and the weighted feature of image data, is the hidden layer activation function, is any one of the structured feature map, the surgery type embedding map, the surgical instrument type embedding map, and the image feature map, Weight matrix before concatenation, is the bias vector before splicing.

[0079] Specifically, the structured feature map:

[0080] Surgery type embedding map:

[0081] Surgical instrument type embedding map:

[0082] Image features Image:

[0083] is the weight matrix corresponding to the feature data of structured data, is the weight matrix corresponding to the surgery type embedding vector, is the weight matrix corresponding to the surgical instrument type embedding vector and, is the weight matrix corresponding to the weighted features of the image data, which are all weight matrices before splicing; is the bias vector corresponding to the feature data of structured data, is the bias vector corresponding to the surgery type embedding vector, is the bias vector corresponding to the surgical instrument type embedding vector, is the bias vector corresponding to the weighted feature of the image data. These are all bias vectors before splicing. is the hidden layer activation function.

[0084] Feature fusion method: Vector concatenation: directly concatenate the feature vectors:

[0085] in, represents a vector concatenation operation, is a structured feature map, Embedding mapping for surgery type, Embedding mapping for surgical instrument types, is the image feature map, is the result after direct concatenation; the dimension of the fused feature vector is the sum of the dimensions of each part.

[0086] Then, high-level features are extracted to form a comprehensive feature vector. Specifically, the fused feature vector is input into the fully connected layer to extract higher-level features:

[0087] in, is the comprehensive feature vector, is the weight matrix after concatenation, is the offset vector after splicing, is the hidden layer activation function.

[0088] After the feature fusion, S107 is executed to determine the complication probability and estimated recovery time of the target diabetic patient based on the comprehensive feature vector and the recovery prediction model, specifically including: The comprehensive feature vector is input into the recovery feature situation prediction model, and the recovery situation prediction model includes: an input layer, a hidden layer and an output layer, and the hidden layer is a multi-layer fully connected layer; Output target probability of complications in diabetic patients: ; in, is the output of the hidden layer, is the Sigmoid output layer activation function, , is the probability of occurrence of the complication; Output the estimated recovery time for target diabetic patients: ; in, is a linear or nonlinear function, Estimated recovery time.

[0089] By using the recovery prediction model of the present invention, the prediction results will be higher by specifically targeting diabetic patients, taking into account their specific physiological and pathological characteristics, as well as the impact of personalized minimally invasive surgery.

[0090] Moreover, by fusing structured data, surgical parameter data, and imaging data, the information of diabetic patients can be fully captured, thus improving the prediction performance.

[0091] By combining the improved CNN with the attention mechanism, we focus on the imaging features related to diabetes, thereby improving the efficiency of data utilization.

[0092] The prediction results and explanatory information of this recovery prediction model can help doctors develop personalized rehabilitation plans, prevent complications in advance, and thus promote rapid recovery of patients.

[0093] The structure of the recovery prediction model is general and can be adjusted and applied to different surgical types and patient groups.

[0094] One or more technical solutions in the embodiments of the present invention have at least the following technical effects or advantages: The present invention provides a method for predicting the postoperative recovery of diabetic patients, comprising: collecting historical specific data and historical recovery data of historical diabetic patients; training a deep neural network model based on the historical specific data and historical recovery data of historical diabetic patients to obtain a recovery prediction model; obtaining target specific data of target diabetic patients; preprocessing the target specific data to obtain standardized specific data; extracting features from the standardized specific data to obtain feature data of each of the standardized specific data; fusing the feature data of each of the standardized specific data to form a comprehensive feature vector; based on the comprehensive feature vector and the recovery prediction model, determining the probability of complication occurrence and the expected recovery time of the target diabetic patient, extracting features from the specific data of the diabetic patient, and fusing the features, thereby predicting the probability of complication occurrence and the expected recovery time of the diabetic patient, thereby improving the accuracy of the prediction.

[0095] Embodiment 2 Based on the same inventive concept, the embodiment of the present invention also provides a device for predicting the postoperative recovery of diabetic patients, such as Figure 2 As shown, including: The collection module 201 is used to collect historical specific data and historical recovery data of historical diabetic patients; A model obtaining module 202 is used to train a deep neural network model based on the historical specific data and historical recovery data of the historical diabetic patients to obtain a recovery prediction model; An acquisition module 203 is used to acquire specific data of a target diabetic patient; A preprocessing module 204, used to preprocess the specific data to obtain standardized specific data; A feature extraction module 205 is used to extract features from the standardized specific data to obtain feature data of each of the standardized specific data; A fusion module 206, for fusing the characteristic data of each of the standardized specific data to form a comprehensive characteristic vector; The determination module 207 is used to determine the probability of complication occurrence and the estimated recovery time of the target diabetic patient based on the comprehensive feature vector and the recovery status prediction model.

[0096] In an optional embodiment, the target-specific data includes: structured data, surgical parameter data, and image data; The structured data include: blood sugar level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and operation blood loss; The surgical parameter data include: surgical type and surgical instrument type; The image data includes: blood vessel images and tissue images.

[0097] In an optional implementation, the preprocessing module 204 is used to: The blood glucose level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and surgical blood loss were numerically standardized to obtain the corresponding standardized specific data: ; in, Standardized specific data for any one of blood glucose level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and surgical blood loss, The raw data is any one of blood sugar level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and operation blood loss. is the mean value of any one of the blood sugar level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and surgical blood loss. The standard deviation of any one of the blood glucose level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time, and surgical blood loss; Encode the operation type and the surgical instrument type respectively to obtain the operation type code and the surgical instrument type code; Both the vascular image and the tissue image are adjusted to a preset size, and the pixel values ​​are normalized to obtain standardized image data.

[0098] In an optional implementation, the feature extraction module 205 is used to: Extracting feature vectors from the standardized specific data of the structured data respectively; Extracting high-order features from the feature vector to obtain feature data of the structured data; Mapping the surgery type code and the surgical instrument type code into low-dimensional dense vectors respectively to obtain a surgery type embedding vector and a surgical instrument type embedding vector; A feature map score and an attention weight are calculated for the standardized image data, and a weighted feature of the image data is obtained based on the feature map score and the attention weight.

[0099] In an optional implementation, the fusion module 206 is configured to: Performing linear transformation on the feature data of the structured data, the surgery type embedding vector, the surgical instrument type embedding vector, and the weighted features of the image data, respectively, to obtain a structured feature map, a surgery type embedding map, a surgical instrument type embedding map, and an image feature map; The structured feature map, the surgery type embedding map, the surgical instrument type embedding map and the image feature map are directly concatenated, and high-level features are extracted to form a comprehensive feature vector.

[0100] In an optional implementation, the fusion module 206 is used to obtain any one of the structured feature map, the surgery type embedding map, the surgical instrument type embedding map, and the image feature map in the following manner: ; in, is any one of the feature data of the structured data, the surgery type embedding vector, the surgical instrument type embedding vector, and the weighted feature of the image data, is the hidden layer activation function, is any one of the structured feature map, the surgery type embedding map, the surgical instrument type embedding map, and the image feature map, is the weight matrix before concatenation, is the bias vector before splicing.

[0101] In an optional implementation, the fusion module 206 is used to form a comprehensive feature vector in the following manner:

[0102] in, represents a vector concatenation operation, is a structured feature map, Embedding mapping for surgery type, Embedding mapping for surgical instrument types, is the image feature map, is the result after direct concatenation; the dimension of the fused feature vector is the sum of the dimensions of each part.

[0103]

[0104] in, is the comprehensive feature vector, is the weight matrix after splicing, is the weight matrix after splicing, is the offset vector after splicing, is the hidden layer activation function.

[0105] In an optional implementation, the determination module 207 is configured to: Inputting the comprehensive feature vector into the recovery situation prediction model, the recovery situation prediction model comprises: an input layer, a hidden layer and an output layer, the hidden layer is a multi-layer fully connected layer; Output target probability of complications in diabetic patients: ; in, is the output of the hidden layer, is the Sigmoid output layer activation function, , is the probability of occurrence of the complication; Output the estimated recovery time for target diabetic patients: ; in, is a linear or nonlinear function, Estimated recovery time.

[0106] Embodiment 3 Based on the same inventive concept, an embodiment of the present invention provides a computer device, such as Figure 3 As shown, it includes a memory 304, a processor 302, and a computer program stored in the memory 304 and executable on the processor 302. When the processor 302 executes the program, the steps of the above-mentioned method for predicting the postoperative recovery of diabetic patients are implemented.

[0107] Among them, Figure 3 In the embodiment of the present invention, a bus architecture (represented by bus 300) is shown, and bus 300 may include any number of interconnected buses and bridges, and bus 300 links various circuits including one or more processors represented by processor 302 and memory represented by memory 304. Bus 300 may also link various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. Bus interface 306 provides an interface between bus 300 and receiver 301 and transmitter 303. Receiver 301 and transmitter 303 may be the same element, namely a transceiver, providing a unit for communicating with various other devices on a transmission medium. Processor 302 is responsible for managing bus 300 and general processing, while memory 304 may be used to store data used by processor 302 when performing operations.

[0108] Embodiment 4 Based on the same inventive concept, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-mentioned method for predicting the postoperative recovery status of diabetic patients.

[0109] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing such systems. In addition, the present invention is not directed to any specific programming language either. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the description of the above specific languages ​​is for disclosing the best mode of the present invention.

[0110] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0111] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, the various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting the following intention: that the claimed invention requires more features than the features explicitly recorded in each embodiment. More specifically, as reflected in each embodiment, the inventive aspects lie in less than all the features of the single embodiment disclosed above. Therefore, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, with each claim itself serving as a separate embodiment of the present invention.

[0112] Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments may be combined into one module or unit or component, and further may be divided into a plurality of submodules or subunits or subcomponents. All features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device so disclosed may be combined in any combination, except that at least some of such features and / or processes or units are mutually exclusive. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature that provides the same, equivalent or similar purpose.

[0113] 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 is meant to be within the scope of the present invention and form different embodiments. For example, in a specific embodiment, any one of the claimed embodiments may be used in any combination.

[0114] The various component embodiments of the present invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or a digital signal processor (DSP) may be used in practice to implement some or all functions of some or all components of a device for predicting the postoperative recovery of a diabetic patient and a computer device according to an embodiment of the present invention. The present invention may also be implemented as a device or device program (e.g., a computer program and a computer program product) for executing part or all of the methods described herein. Such a program for implementing the present invention may be stored on a computer-readable medium, or may be in the form of one or more signals. Such a signal may be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0115] It should be noted that the above embodiments illustrate the present invention rather than limit it, and that those skilled in the art may devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets shall not be construed as a limitation on the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention may be implemented by means of hardware comprising a number of different elements and by means of a suitably programmed computer. In a unit claim enumerating a number of devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc., does not indicate any order. These words may be interpreted as names.

Claims

1. A method for predicting the postoperative recovery of diabetic patients, characterized in that: include: Collect historical specific data and historical recovery data of historical diabetic patients; Based on the historical specific data and historical recovery data of the historical diabetic patients, a deep neural network model is trained to obtain a recovery prediction model; obtain target-specific data for target diabetic patients; Preprocessing the target-specific data to obtain standardized specific data; Performing feature extraction on the standardized specific data to obtain feature data of each of the standardized specific data; Merging the characteristic data of each of the standardized specific data to form a comprehensive characteristic vector; Based on the comprehensive feature vector and the recovery prediction model, the probability of complication occurrence and the estimated recovery time of the target diabetic patient are determined.

2. The method according to claim 1, characterized in that The target-specific data includes: structured data, surgical parameter data, and image data; The structured data include: blood sugar level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and operation blood loss; The surgical parameter data include: surgical type and surgical instrument type; The image data includes: blood vessel images and tissue images.

3. The method according to claim 2, characterized in that Preprocessing the target-specific data to obtain standardized specific data includes: The blood glucose level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and surgical blood loss were numerically standardized to obtain the corresponding standardized specific data: ; in, Standardized specific data for any one of blood glucose level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and surgical blood loss, The raw data is any one of blood sugar level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and operation blood loss. is the mean value of any one of the blood sugar level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time and surgical blood loss. The standard deviation of any one of the blood glucose level data, glycosylated hemoglobin, insulin resistance index, complication index, operation time, and surgical blood loss; Encode the operation type and the surgical instrument type respectively to obtain the operation type code and the surgical instrument type code; The vascular images and tissue images are adjusted to a preset size, and the pixel values ​​are normalized to obtain standardized image data.

4. The method according to claim 3, characterized in that Performing feature extraction on the standardized specific data to obtain feature data of each of the standardized specific data includes: Extracting feature vectors from the standardized specific data of the structured data respectively; Extracting high-order features from the feature vector to obtain feature data of the structured data; Mapping the surgery type code and the surgical instrument type code into low-dimensional dense vectors respectively to obtain a surgery type embedding vector and a surgical instrument type embedding vector; A feature map score and an attention weight are calculated for the standardized image data, and a weighted feature of the image data is obtained based on the feature map score and the attention weight.

5. The method according to claim 4, characterized in that The characteristic data of each of the standardized specific data are merged to form a comprehensive characteristic vector, including: Performing linear transformation on the feature data of the structured data, the surgery type embedding vector, the surgical instrument type embedding vector, and the weighted features of the image data, respectively, to obtain a structured feature map, a surgery type embedding map, a surgical instrument type embedding map, and an image feature map; The structured feature map, the surgery type embedding map, the surgical instrument type embedding map and the image feature map are directly concatenated, and high-level features are extracted to form a comprehensive feature vector.

6. The method according to claim 5, characterized in that The feature data of the structured data, the surgery type embedding vector, the surgical instrument type embedding vector, and the weighted features of the image data are linearly transformed to obtain a structured feature map, a surgery type embedding map, a surgical instrument type embedding map, and an image feature map, including: ; in, is any one of the feature data of the structured data, the surgery type embedding vector, the surgical instrument type embedding vector, and the weighted feature of the image data, is the hidden layer activation function, is any one of the structured feature map, the surgery type embedding map, the surgical instrument type embedding map, and the image feature map, is the weight matrix before concatenation, is the bias vector before splicing.

7. The method according to claim 5, characterized in that The structured feature map, the surgery type embedding map, the surgical instrument type embedding map and the image feature map are directly concatenated, and high-level features are extracted to form a comprehensive feature vector, including: ; in, represents a vector concatenation operation, is the structured feature map, embedding a mapping for said surgery type, embedding a mapping for said surgical instrument type, is the image feature map, This is the result after direct splicing; ; in, is the comprehensive feature vector, is the concatenated weight matrix and is the bias vector after splicing, is the hidden layer activation function.

8. The method according to claim 7, characterized in that Based on the comprehensive feature vector and the recovery prediction model, determining the probability of occurrence of complications and the estimated recovery time of the target diabetic patient, including: Inputting the comprehensive feature vector into the recovery situation prediction model, the recovery situation prediction model comprises: an input layer, a hidden layer and an output layer, the hidden layer is a multi-layer fully connected layer; Output target probability of complications in diabetic patients: ; in, is the output of the hidden layer, is the Sigmoid output layer activation function, , is the probability of occurrence of the complication; Output the estimated recovery time for target diabetic patients: ; in, is a linear or nonlinear function, Estimated recovery time.

9. A device for predicting the postoperative recovery of diabetic patients, characterized in that: include: A collection module, used to collect historical specific data and historical recovery data of historical diabetic patients; A model acquisition module, used to train a deep neural network model based on the historical specific data and historical recovery data of the historical diabetic patients to obtain a recovery prediction model; An acquisition module, used for acquiring specific data of a target diabetic patient; A preprocessing module, used for preprocessing the specific data to obtain standardized specific data; A feature extraction module, used to extract features from the standardized specific data to obtain feature data of each of the standardized specific data; A fusion module, used for fusing the characteristic data of each of the standardized specific data to form a comprehensive characteristic vector; A determination module is used to determine the probability of complication occurrence and the estimated recovery time of the target diabetic patient based on the comprehensive feature vector and the recovery prediction model.

10. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 8 is implemented.

Images