WO2021159761A1

WO2021159761A1 - Pathological data analysis method and apparatus, and computer device and storage medium

Info

Publication number: WO2021159761A1
Application number: PCT/CN2020/125152
Authority: WO
Inventors: 贾文笑
Original assignee: 平安科技（深圳）有限公司
Priority date: 2020-09-09
Filing date: 2020-10-30
Publication date: 2021-08-19
Also published as: CN112037922A

Abstract

A pathological data analysis method and apparatus, and a computer device and a storage medium, relating to the field of digital healthcare. The method comprises: collecting pathological data of a user (S1); performing feature extraction on the pathological data, so as to obtain specified feature data related to the risk of heart failure occurrence (S2); acquiring structured feature data, related to the risk of heart failure occurrence, of the user (S3); performing splicing processing on the specified feature data and the structured feature data, so as to obtain fusion feature data after same is subjected to splicing processing (S4); taking the fusion feature data as an input of a preset attention module, generating, by means of the attention module, attention weights that are in one-to-one correspondence with each fusion feature of the fusion feature data, and performing, according to the attention weights, weighted sum processing on each fusion feature of the fusion feature data, so as to obtain corresponding output results (S5); and inputting the output results into a preset classification module, performing normalization processing on the output results by means of the classification module, and generating a predicted probability of heart failure occurrence corresponding to the user (S6). The method improves the processing efficiency and accuracy of predicting the risk of heart failure occurrence in a user.

Description

Pathological data analysis method, device, computer equipment and storage medium

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on September 9, 2020, with the application number 2020109416268, and the invention title "Pathological data analysis methods, devices, computer equipment, and storage media". The entire content of the application is approved The reference is incorporated in this application.

Technical field

This application relates to the field of digital medical technology, in particular to a pathological data analysis method, device, computer equipment and storage medium.

Background technique

Heart failure (heart failure) referred to as heart failure, refers to the failure of the systolic and/or diastolic function of the heart, the inability to fully discharge the venous return blood volume out of the heart, resulting in blood stasis in the venous system and insufficient blood perfusion in the arterial system. Cardiac circulatory disorder syndrome, which is manifested in clusters of pulmonary congestion and vena cava congestion. Heart failure is not an independent disease, but the final stage of the development of heart disease. Improper treatment can affect health and even threaten life. However, if a greater risk of heart failure can be detected early and effective prevention and treatment measures can be taken in time, it will be of great significance to improve the prognosis and mortality of patients. However, the inventor realizes that the existing screening of people at high risk of heart failure is still based on traditional empirical judgments, and doctors usually use the user's clinical symptoms to predict the corresponding risk of heart failure. The method of predicting the risk of occurrence requires a lot of time to collect and analyze clinical data, time-consuming and labor-intensive, and must rely on the personal experience of the doctor, and the experience level of different doctors often differs greatly, which leads to the occurrence of heart failure. Forecast accuracy is low.

technical problem

The main purpose of this application is to provide an analysis method, device, computer equipment and storage medium for pathological data, which aims to solve the problem that the existing prediction methods for the risk of heart failure are still based on traditional empirical judgments, which are time-consuming, labor-intensive and accurate. Technical problems with low sexuality.

Technical solutions

This application proposes a method for analyzing pathological data. The method includes the steps:

Collect pathological data of users;

Perform feature extraction on the pathological data to obtain specified feature data related to the risk of heart failure, wherein the specified feature data includes physiological signal feature data, vital signs feature data, and case text feature data;

Acquiring structured characteristic data of the user related to the risk of occurrence of heart failure, wherein the structured characteristic data includes laboratory examination characteristic data and demographic characteristic data;

Performing splicing processing on the designated characteristic data and the structured characteristic data to obtain merged characteristic data after splicing;

The fusion feature data is used as the input of the preset attention module, and the attention weight corresponding to each fusion feature in the fusion feature data is generated by the attention module, and the attention weight is adjusted according to the attention weight. Each fusion feature in the fusion feature data is subjected to weighted summation processing to obtain a corresponding output result;

The output result is input to a preset classification module, and the output result is normalized by the classification module to generate a predicted probability of occurrence of heart failure corresponding to the user.

This application also provides a device for predicting the risk of heart failure, including:

Collection module, used to collect pathological data of users;

The extraction module is used for feature extraction of the pathological data to obtain specified feature data related to the risk of heart failure, wherein the specified feature data includes physiological signal feature data, vital signs feature data, and case text feature data;

The first acquisition module is configured to acquire structured characteristic data of the user related to the risk of heart failure, wherein the structured characteristic data includes laboratory examination characteristic data and demographic characteristic data;

A processing module, configured to perform splicing processing on the designated characteristic data and the structured characteristic data to obtain merged characteristic data after splicing;

The first generation module is configured to use the fusion feature data as the input of the preset attention module, and generate the attention weight corresponding to each fusion feature in the fusion feature data through the attention module, and Performing weighted summation processing on each fusion feature in the fusion feature data according to the attention weight to obtain a corresponding output result;

The second generation module is configured to input the output result into a preset classification module, and normalize the output result through the classification module to generate a predicted probability of occurrence of heart failure corresponding to the user.

The present application also provides a computer device, including a memory and a processor, the memory stores a computer program, and the processor implements a pathological data analysis method when the computer program is executed, wherein the pathological data The analysis method includes the following steps:

Collect pathological data of users;

The present application also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, a method for analyzing pathological data is realized, wherein the method for analyzing pathological data includes the following steps:

Collect pathological data of users;

Perform feature extraction on the pathological data to obtain specified feature data related to the risk of heart failure, where the specified feature data includes physiological signal feature data, vital signs feature data, and case text feature data;

Beneficial effect

The pathological data analysis method, device, computer equipment and storage medium provided in this application can intelligently and accurately generate the prediction probability of the user’s heart failure, realize the accurate prediction of the user’s heart failure risk, and effectively improve the prediction of the user’s heart failure. The efficiency of handling the risk of failure.

Description of the drawings

FIG. 1 is a schematic flowchart of a method for locating the root cause of an alarm according to an embodiment of the present application;

FIG. 2 is a schematic structural diagram of an apparatus for locating the root cause of an alarm according to an embodiment of the present application;

Fig. 3 is a schematic structural diagram of a computer device according to an embodiment of the present application.

The best mode of the present invention

It should be understood that the specific embodiments described here are only used to explain the present application, and are not used to limit the present application.

This solution can be applied to the digital medical field in smart cities, thereby promoting the construction of smart cities.

1, the pathological data analysis method of an embodiment of the present application includes:

S1: Collect pathological data of users;

S2: Perform feature extraction on the pathological data to obtain specified feature data related to the risk of heart failure, where the specified feature data includes physiological signal feature data, vital signs feature data, and case text feature data;

S3: Obtain structured characteristic data of the user related to the risk of heart failure, where the structured characteristic data includes laboratory examination characteristic data and demographic characteristic data;

S4: Perform splicing processing on the designated feature data and the structured feature data to obtain merged feature data after the splicing process;

S5: The fusion feature data is used as the input of the preset attention module, and the attention weight corresponding to each fusion feature in the fusion feature data is generated by the attention module, and the attention weight is generated according to the attention Weight performs weighted summation processing on each fusion feature in the fusion feature data to obtain a corresponding output result;

S6: Input the output result to a preset classification module, and normalize the output result through the classification module to generate a predicted probability of occurrence of heart failure corresponding to the user.

As described in the above steps S1 to S6, the execution subject of this method embodiment is a heart failure risk prediction device. In practical applications, the above-mentioned heart failure risk prediction device can be realized by a virtual device, such as software code, or by a physical device written or integrated with relevant execution codes, and can communicate with the user through a keyboard, mouse, remote control, Human-computer interaction is carried out by means of touchpad or voice control equipment. The heart failure risk prediction device in this embodiment can intelligently and accurately generate the user's heart failure prediction probability, realizes accurate prediction of the user's heart failure risk, and effectively improves the processing efficiency for predicting the user's heart failure risk. Specifically, first collect the pathological data of the user. Among them, multiple data sources can be used to collect the user's pathological data during the preset historical time period. The aforementioned data sources are not specifically limited. For example, they can include electronic medical record systems, user's disease shooting or scanning files, etc.; and The aforementioned preset historical time period is not specifically limited, and may be, for example, the past two years. In addition, the above-mentioned pathological data may include at least physiological signal data, vital signs data, case text data and other data. Then, feature extraction is performed on the pathological data to obtain specified feature data related to the risk of heart failure, where the specified feature data includes physiological signal feature data, vital signs feature data, and case text feature data. In addition, different feature extraction methods will be used for pathological data of different modalities. Specifically, convolutional neural networks can be used to extract the above-mentioned physiological signal feature data, recurrent neural networks can be used to extract the above-mentioned vital sign feature data, and Chinese Natural language processing technology extracts case text feature data. In addition, the above-mentioned physiological signal characteristic data may include electrocardiographic characteristic data, etc.; the above-mentioned vital sign data may include blood pressure characteristic data, heart rate characteristic data, and respiratory rate characteristic data, etc.; Related characteristic data such as treatment, family medical history, marital status, etc. And obtain the structured characteristic data of the user related to the risk of heart failure, where the structured characteristic data includes laboratory test characteristic data and demographic characteristic data, and the laboratory test characteristic data may include blood routine and urine routine. The above-mentioned demographic characteristic data may include characteristic data such as age, gender, and disease history. In addition, because the laboratory test feature data and demographic feature data in the structured feature data are information that does not change with time, the structured feature data can be extracted directly without the need for feature extraction processing operations. The above-mentioned pathological data may also include user laboratory examination characteristic data and demographic characteristic data, so that the above-mentioned structured characteristic data can be directly obtained from the above-mentioned pathological data. Then, the specified feature data and the structured feature data are spliced to obtain the merged feature data after splicing. Among them, a staged integrated learning algorithm can be used to realize the data splicing process between the specified feature data and the structured feature data. After the fusion feature data is obtained, the fusion feature data is used as the input of the preset attention module, and the attention weight corresponding to each fusion feature in the fusion feature data is generated through the attention module, and Perform weighted summation processing on each fusion feature in the fusion feature data according to the above attention weight to obtain a corresponding output result. Among them, the process of calculating the attention weight includes: first calculating the attention score of each fusion feature in the fusion feature data, and then using the softmax function to perform numerical conversion on each attention score to generate the attention weight of each fusion feature. Attention weight refers to the weight of the attention degree of each fusion feature, and the sum of all attention weights is 1. Specifically, the attention score of each fusion feature in the fusion feature data can be calculated by the formula e _k,j =relu(W _a h _j +b _a _{), e k,j} are the attention scores, and W _a and b _a is the network parameters that can be learned by the attention layer of the model, h _j is any fusion feature in the fusion feature data, and relu is the activation function. After getting the attention score, pass the formula

Calculate the attention weight corresponding to each fusion feature in the fusion feature data, that is, use the softmax function to _{normalize each e k, j} to assign the corresponding attention weight to each fusion feature, and Obtain the attention distribution with the sum of all attention weights being 1. In addition, the formula

Perform dot multiplication processing, that is, perform weighted summation processing on each fusion feature in the fusion feature data according to the attention weight, and generate the Attention value corresponding to the fusion feature data, that is, the above output result. Finally, when the aforementioned output result is obtained, the aforementioned output result is input to a preset classification module, and the aforementioned output result is normalized by the aforementioned classification module to generate a predicted probability of occurrence of heart failure corresponding to the aforementioned user. Among them, the output result of the attention module can be normalized by the formula O _τ =sigmoid(W _c C+b _c ) to generate the final predicted probability of occurrence of heart failure. O _τ is the predicted probability of heart failure, and _{the range of O τ} is between 0 and 1, W _c and b _c are network parameters that can be learned by the classification module, and C is the output result of the attention module. This solution can be applied to the digital medical field in smart cities, thereby promoting the construction of smart cities. This application collects multiple modal feature data of users related to the risk of heart failure, and performs splicing processing on the multiple modal feature data, and based on the attention module and the classification module on the fusion feature data generated after the splicing process Data analysis and processing can intelligently and accurately generate the user's heart failure prediction probability, realize the accurate prediction of the user's heart failure risk, and effectively improve the processing efficiency of predicting the user's heart failure risk.

Further, the aforementioned attention module is an attention network in the heart failure risk prediction model generated by training, the classification module is a classifier in the heart failure risk prediction model, and the heart failure risk prediction model may further include Feature extraction network for feature extraction. Specifically, the feature extraction network takes pathological data of heart failure patient data as input, and outputs feature data corresponding to the risk of heart failure after feature extraction processing. The attention network uses the feature data of the patients corresponding to the risk of heart failure and the fusion feature data obtained by splicing the structured feature data of the patients corresponding to the risk of heart failure as input, outputs the attention weight, and integrates the features The data and the corresponding attention weight are weighted and summed to obtain the output result, which is the attention value. The classifier takes the attention value output by the attention network as input, normalizes the attention value, and outputs the classification result. Train the initial model by acquiring pathological data of a preset number of patients as a training sample, using the training sample as the input layer in the initial model, and using the truth label corresponding to the training sample as the output layer of the initial model , Obtain the corresponding feature extraction module, attention module and classification module, and finally generate the above-mentioned heart failure risk prediction model. Among them, the training and generation process of the above-mentioned heart failure risk prediction model can refer to the existing model training and generation methods, and will not be repeated here. In addition, the above-mentioned feature extraction module corresponds to the above-mentioned feature extraction network, the above-mentioned attention module corresponds to the above-mentioned attention network, and the above-mentioned classification module corresponds to the above-mentioned classifier. And the aforementioned training samples also include structured feature data corresponding to the risk of heart failure.

Further, in an embodiment of the present application, the above step S2 includes:

S200: Perform feature extraction on the physiological signal data in the pathological data by using a convolutional neural network to obtain physiological signal feature data corresponding to the physiological signal data; and,

S201: Perform feature extraction on vital sign data in the pathological data by using a cyclic neural network to obtain corresponding vital sign feature data; and,

S202: Use Chinese natural language processing technology to extract key features of the case text data in the pathology data to obtain corresponding case text feature data.

As described in the above steps S200 to S202, for the data of different modalities in the pathological data, corresponding deep neural networks will be used to extract the features of the different modal data to improve the accuracy of the feature extraction of the pathological data. . The step of extracting features from the pathological data to obtain specified feature data related to the risk of heart failure may specifically include: using a convolutional neural network to extract features from the physiological signal data in the pathological data to obtain the physiological signal data. Corresponding physiological signal characteristic data. Among them, the physiological signal data includes ECG data. Since the physiological signal data belongs to high-density sampling time-series waveform data, the characteristics of the physiological signal data can be extracted by using a convolutional neural network, and the corresponding physiological signal characteristic data can be output. Specifically, P=CNN (ECG), ECG represents the physiological signal data of the user, such as electrocardiogram data, CNN represents the convolutional neural network, and P represents the output physiological signal characteristic data corresponding to the physiological signal data. And at the same time, the recurrent neural network is used to extract the characteristics of the vital sign data in the above-mentioned pathological data to obtain the corresponding vital sign characteristic data. Among them, the above vital sign data may include blood pressure, body temperature, respiration and other data. Since the vital sign data belongs to low-density sampling of discontinuous waveform data, the above vital sign data can be feature extracted through a cyclic neural network, and the corresponding vital signs can be output. Physical characteristics data. Recurrent neural networks can effectively consider the timing dependence between data. Specifically, Q=RNN (waveform), waveform represents the vital sign data of the user, RNN represents cyclic neural network, which may include LSTM or GRU network structure, for example, and Q represents the output vital sign feature data corresponding to the vital sign data. And at the same time, Chinese natural language processing (NLP) technology is used to extract key features of the case text data in the above pathological data to obtain the corresponding case text feature data. Among them, for text data, advanced NLP technology can be used to automatically extract based on preset keyword information, such as hypertension, diabetes, anti-hypertensive treatment, family history and marital status, etc., through the combination of Embedding and RNN. Key patient characteristics. Specifically, V=Embedding(word), R=RNN(V), word represents the user's case text data, Embedding represents the processing of word embedding technology, and RNN represents cyclic neural network, which may include LSTM or GRU network structure, for example. For example, by splicing the above-mentioned physiological signal feature data P, vital sign feature data Q, and case text feature data R, the corresponding fusion feature data can be obtained as: S=[P,Q,R]. According to the data characteristics of the different modal data in the pathological data, this embodiment will use the corresponding deep neural network to extract the features of the different modal data, so as to accurately and quickly extract the required specifications from the pathological data. The characteristic data enables the subsequent accurate and quick prediction of the risk of the user's heart failure based on the designated characteristic data.

Further, in an embodiment of the present application, the above step S201 includes:

S2010: Perform feature extraction on vital sign data in the pathological data by using the recurrent neural network to obtain first vital sign feature data;

S2011: Determine whether there are missing values in the first vital sign characteristic data;

S2012: If there is a missing value in the first vital sign characteristic data, obtain the data missing position in the first vital sign characteristic data;

S2013: Obtain the last characteristic observation value corresponding to the designated data missing position, and acquire the mean value of the first vital sign characteristic data, wherein the designated data missing position is any data missing position of all the data missing positions ；

S2014: According to the last feature observation value and the average value, call a preset calculation formula to calculate the designated filling value corresponding to the designated data missing position; S2015: Use the designated filling value to delete the designated data Position data filling processing;

S2016: Obtain second vital sign feature data obtained after performing corresponding data filling processing on all missing data positions in the first vital sign feature data;

S2017: Use the second vital sign feature data as the vital sign feature data.

As described in the above steps S2010 to S2017, after the vital sign feature data is extracted by the cyclic neural network, the missing data in the vital sign feature data can be filled in to complete the data processing of the vital sign feature data. . Specifically, the step of using the recurrent neural network to extract features from the vital sign data in the pathological data to obtain the corresponding vital sign feature data includes: first using the recurrent neural network to feature the vital sign data in the pathological data Extract and obtain the first vital sign feature data. Then it is judged whether there are missing values in the first vital sign feature data. If there is a missing value in the first vital sign feature data, the data missing position in the first vital sign feature data is acquired. Then, the last feature observation value corresponding to the designated data missing location is obtained, and the average value of the first vital sign feature data is obtained, where the specified data missing location is any data missing location of all the above data missing locations. Then, according to the above-mentioned last feature observation value and the above-mentioned average value, a preset calculation formula is called to calculate the designated filling value corresponding to the above-mentioned designated data missing position. Specifically, it can be calculated by the formula

Calculate the above specified fill value, where,

Is the missing value that needs to be filled, that is, the specified fill value above,

To the last observed value characteristic corresponding to the position of the designated data deletion, x ^'d is the mean of the data of the first feature vital signs (also referred to as average experience),

It is a mask matrix, indicating whether the current data variable has been observed, if it is observed, the value is 1, and if it is not observed, the value is 0. For example, if at a certain moment, the d-th data variable is observed, then this variable is equal to the observed value at this moment, and if it is not observed, it is indicated as missing data or missing value.

For the designated time attenuation factor corresponding to the designated data missing position, the time attenuation factor can be obtained by the formula Υ _t = exp(-max(0, W _Υ δ _t + b _Υ )), where Υ _t is the relationship with the circulatory nerve The time decay factor corresponding to the network, W _Υ and b _Υ are network parameters that can be learned by the cyclic neural network, δ _t is the time interval between the position of the missing data and the last observation, and Υ _t will eventually be normalized to a value of 0～ Within the range of 1. In addition, the specified fill value corresponding to the missing position of the specified data needs to be balanced between the last feature observation value and the mean value. When calculating the specified fill value by adding the time attenuation factor, if the missing location of the specified data is farther from the last feature observation value, the weight of the mean will be greater; and if the specified data missing location is closer to the last feature observation value, then The weight of the last feature observation will be greater. After the specified filling value is obtained, the specified filling value is used to perform data filling processing on the missing position of the specified data. Then, obtain the second vital sign feature data obtained by performing corresponding data filling processing on all the missing data positions in the first vital sign feature data. Finally, after the second vital sign feature data is generated, the second vital sign feature data is used as the vital sign feature data. In this embodiment, data filling processing is performed on the actual value of the vital sign feature data to realize the data completion processing of the vital sign feature data, and then the user can be processed based on the vital sign feature number after the data completion processing. The prediction of the risk of heart failure effectively improves the accuracy of the subsequent prediction of the probability of occurrence of heart failure.

Further, in an embodiment of the present application, after the above step S5, the method includes:

S500: Obtain the importance coefficients corresponding to each type of modal feature data according to the attention weight according to preset rules, where the modal feature data includes physiological signal feature data, vital sign feature data, and case text Characteristic data, laboratory examination characteristic data, and demographic characteristic data;

S501: Sort all the importance coefficients in descending order of numerical value to obtain a corresponding sorting result;

S502: According to the sorting result, generate an importance prediction report for each type of the modal characteristic data corresponding to the risk of occurrence of heart failure;

S503: Display the importance prediction report.

As described in the above steps S500 to S503, after the attention weight corresponding to each fusion feature in the fusion feature data is generated, the attention weight can be subsequently used to intelligently generate each type of modal feature data corresponding to The importance of the risk of heart failure predicts the outcome. Specifically, the above-mentioned fusion feature data is used as the input of the preset attention module, and the attention weight corresponding to each fusion feature in the above-mentioned fusion feature data is generated by the above-mentioned attention module, and the attention weight is adjusted according to the above-mentioned attention weight. The step of performing weighted summation processing for each fusion feature in the aforementioned fusion feature data to obtain the corresponding output result includes: first obtaining the importance coefficient of each type of modal feature data according to the aforementioned attention weight according to a preset rule, where The aforementioned modal characteristic data includes physiological signal characteristic data, vital sign characteristic data, case text characteristic data, laboratory examination characteristic data, and demographic characteristic data. In addition, the aforementioned preset rule may refer to obtaining the importance coefficient corresponding to each type of modal feature data by calculating the average value of all attention weights corresponding to each type of modal feature data. After the above-mentioned importance coefficients are obtained, all the above-mentioned importance coefficients are sorted in descending order of numerical value to obtain the corresponding sorting result. Then, according to the above sorting results, a prediction report of the importance of each type of modal characteristic data corresponding to the risk of heart failure is generated. Finally, when the above-mentioned importance forecast report is obtained, the above-mentioned importance forecast report will be displayed. In this embodiment, after the importance coefficient corresponding to each type of modal feature data is obtained, the importance of each type of modal feature data is sorted by using the importance coefficient, so that each type of modal feature data can be intuitively and clearly understood. A type of modal feature data and the importance result of the risk of heart failure, and then can quickly screen out the high value that affects the risk of heart failure from all the modal feature data, that is, the target modal feature data with higher importance, and More attention resources can be devoted to the target modal feature data. In addition, if there are certain modal feature data that are extremely low in importance to the risk of heart failure (for example, the importance parameter is less than a preset threshold), the attention of these irrelevant information can be reduced, and even some irrelevant information can be filtered out. Information, which can solve the problem of information overload and improve the generation efficiency and prediction accuracy of the subsequent prediction of the probability of occurrence of heart failure corresponding to the user.

Further, in an embodiment of the present application, the above step S500 includes:

S5000: Filter out the first attention weight corresponding to each of the physiological signal feature data, the second attention weight corresponding to each of the vital sign feature data, and each of the case feature data respectively. The third attention weight of, the fourth attention weight corresponding to each of the laboratory examination characteristic data, and the fifth attention weight corresponding to each of the demographic characteristic data;

S5001: Calculate the first average value of all the first attention weights, the second average value of all the second attention weights, the third average value of all the third attention weights, and all the first attention weights. The fourth average value of four attention weights, and the fifth average value of all the fifth attention weights;

S5002: Use the first average value as the first importance coefficient of the physiological signal characteristic data relative to the risk of heart failure, and use the second average value as the vital sign characteristic data relative to the risk of heart failure. The second importance coefficient, the third average value is used as the third importance coefficient of the case characteristic data relative to the risk of heart failure, and the fourth average value is used as the laboratory examination characteristic data relative to the heart failure. The fourth importance coefficient of the risk of occurrence of heart failure, and the fifth average value is used as the fifth importance coefficient of the demographic characteristic data relative to the risk of occurrence of heart failure.

As described in the foregoing steps S5000 to S5002, the foregoing step of obtaining the importance coefficients corresponding to each type of modal feature data according to the preset rules according to the foregoing attention weight may specifically include: The first attention weight corresponding to the signal feature data, the second attention weight corresponding to each of the above-mentioned vital sign feature data, the third attention weight corresponding to each of the above-mentioned case feature data, and each of the above experiments The fourth attention weight corresponding to the laboratory examination feature data, and the fifth attention weight corresponding to each of the above-mentioned demographic feature data. Then calculate the first average of all the above-mentioned first attention weights, the second average of all the above-mentioned second attention weights, the third average of all the above-mentioned third attention weights, and all the above-mentioned fourth attention weights. The fourth average, and the fifth average of all the above fifth attention weights. Finally, the first average value is used as the first importance coefficient of the physiological signal characteristic data relative to the risk of heart failure, and the second average value is used as the second importance coefficient of the vital sign characteristic data relative to the risk of heart failure. , The above-mentioned third average value is regarded as the third importance coefficient of the above-mentioned case characteristic data relative to the risk of heart failure, and the above-mentioned fourth average value is regarded as the fourth importance coefficient of the above-mentioned laboratory examination characteristic data relative to the risk of heart failure , And the above fifth average value as the fifth importance coefficient of the above demographic characteristic data relative to the risk of heart failure. In this embodiment, by calculating the average value of all the attention weights contained in each type of modal feature data, the importance coefficients corresponding to each type of modal feature data can be calculated intelligently and quickly, which is beneficial to follow-up according to the importance. Parameters to intelligently and quickly obtain the degree of importance of each type of modal characteristic data corresponding to the risk of heart failure, and can quickly filter out the high value of the risk of heart failure from all the modal characteristic data according to the degree of importance , That is, the more important target modal feature data, so that more attention resources can be devoted to the target modal feature data.

Further, in an embodiment of the present application, after the above step S6, the method includes:

S600: Obtain a preset risk threshold;

S601: Determine whether the predicted probability of occurrence of heart failure is greater than the risk threshold;

S602: If the predicted probability of occurrence of heart failure is greater than the risk threshold, determine that the risk of occurrence of heart failure of the user is a high risk level;

S603: If the predicted probability of occurrence of heart failure is not greater than the risk threshold, determine whether the predicted probability of occurrence of heart failure is within a first preset range;

S604: If the predicted probability of occurrence of heart failure is within a first preset range, determine that the risk of occurrence of heart failure of the user is a medium risk level;

S605: If the heart failure risk prediction probability is not within the first preset range, determine that the heart failure risk of the user is a low risk level.

As described in the above steps S600 to S605, after the predicted probability of occurrence of heart failure corresponding to the user is generated, the corresponding risk level can be intelligently generated subsequently according to the predicted probability of occurrence of heart failure. Specifically, after the step of inputting the output result to a preset classification module, and normalizing the output result through the classification module to obtain the predicted probability of occurrence of heart failure corresponding to the user, the method includes: first obtaining the prediction Set the risk threshold. Among them, the above-mentioned risk threshold is not specifically limited, and can be set according to actual needs, for example, it can be set to 0.8. Then determine whether the predicted probability of occurrence of heart failure is greater than the risk threshold. If the predicted probability of occurrence of heart failure is greater than the risk threshold, it is determined that the heart failure occurrence risk of the user is at a high risk level, and the high risk level represents that the heart failure occurrence risk of the user is at least twice the average risk. And if the predicted probability of occurrence of heart failure is not greater than the risk threshold, it is determined whether the predicted probability of occurrence of heart failure is within a first preset range. Among them, the above-mentioned first preset range is not specifically limited, and can be set according to actual needs, for example, it can be set to 0.5-0.8. If the predicted probability of occurrence of heart failure is within the first preset range, it is determined that the risk of occurrence of heart failure of the user is at a medium risk level. The medium risk level means that the user's heart failure risk is slightly higher than the average risk. If the predicted probability of occurrence of heart failure is not within the first preset range, it is determined that the risk of occurrence of heart failure of the user is at a low risk level. The low risk level means that the user's heart failure risk is close to or lower than the average risk. In addition, three colors of red, yellow, and green can be used to visually display the above-mentioned risk levels. Red represents high risk levels, yellow represents medium risk levels, and green represents low risk levels. This embodiment converts the predicted probability of occurrence of heart failure into the corresponding risk level, so that the user can intelligently understand the current risk of heart failure more intuitively, so that the corresponding prevention can be taken intelligently and quickly in the future. Treatment measures.

S610: When the user's heart failure occurrence risk is a high-risk level state or a medium-risk level state, generate early warning information, where the early warning information includes the predicted probability of the occurrence of heart failure and corresponding risk level information;

S611: Obtain recommended information related to heart failure prevention; and,

S612: Obtain the identity information of the user;

S613: According to the identity information, send the warning information and the advice information to the user terminal corresponding to the identity information.

As described in the above steps S610 to S613, after the predicted probability of occurrence of heart failure corresponding to the user is generated, if the risk of occurrence of heart failure of the user is at a high risk level or a medium risk level, the corresponding early warning can be sent to the user intelligently. Information and advice. Specifically, after the above-mentioned step of inputting the above-mentioned output result into a preset classification module, and performing normalization processing on the above-mentioned output result through the above-mentioned classification module to obtain the corresponding heart failure risk prediction probability, it includes: When the occurrence risk is a high-risk level state or a medium-risk level state, early warning information is generated, where the foregoing early warning information includes the predicted probability of occurrence of the heart failure and the corresponding risk level information. Then get advice related to heart failure prevention. Then obtain the above-mentioned user's identity information. Finally, after obtaining the above-mentioned identity information, according to the above-mentioned identity information, the above-mentioned warning information and the above-mentioned suggestion information are sent to the user terminal corresponding to the above-mentioned identity information. In this embodiment, by intelligently sending the user-related early warning information and advice information to the user-related user terminal, it is beneficial for the user to understand the risk of his own heart failure in time, and to provide the user with corresponding heart failure prevention suggestions. Information to improve user experience.

The pathological data analysis method in the embodiment of the present application can also be applied to the blockchain field, such as storing the aforementioned data such as the predicted probability of occurrence of heart failure on the blockchain. By using the blockchain to store and manage the predicted probability of occurrence of heart failure, the security and non-tamperability of the predicted probability of occurrence of heart failure can be effectively guaranteed.

The above-mentioned blockchain is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. Blockchain, essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block. The blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.

The underlying platform of the blockchain can include processing modules such as user management, basic services, smart contracts, and operation monitoring. Among them, the user management module is responsible for the identity information management of all blockchain participants, including the maintenance of public and private key generation (account management), key management, and maintenance of the correspondence between the user’s real identity and the blockchain address (authority management), etc. In the case of authorization, supervise and audit certain real-identity transactions, and provide risk control rule configuration (risk control audit); basic service modules are deployed on all blockchain node devices to verify the validity of business requests, After completing the consensus on the valid request, it is recorded on the storage. For a new business request, the basic service first performs interface adaptation analysis and authentication processing (interface adaptation), and then encrypts the business information through the consensus algorithm (consensus management), After encryption, it is completely and consistently transmitted to the shared ledger (network communication), and recorded and stored; the smart contract module is responsible for contract registration and issuance, contract triggering and contract execution. Developers can define the contract logic through a certain programming language and publish it to On the blockchain (contract registration), according to the logic of the contract terms, call keys or other events to trigger execution, complete the contract logic, and also provide the function of contract upgrade and cancellation; the operation monitoring module is mainly responsible for the deployment of the product release process , Configuration modification, contract settings, cloud adaptation, and visual output of real-time status during product operation, such as: alarms, monitoring network conditions, monitoring node equipment health status, etc.

2, an embodiment of the present application also provides a heart failure risk prediction device, including:

Collection module 1, used to collect pathological data of users;

The extraction module 2 is configured to perform feature extraction on the pathological data to obtain specified feature data related to the risk of heart failure, wherein the specified feature data includes physiological signal feature data, vital signs feature data, and case text feature data;

The first acquisition module 3 is configured to acquire structured characteristic data of the user related to the risk of heart failure, wherein the structured characteristic data includes laboratory examination characteristic data and demographic characteristic data;

The processing module 4 is configured to perform splicing processing on the designated feature data and the structured feature data to obtain merged feature data after the splicing process;

The first generating module 5 is configured to use the fusion feature data as the input of a preset attention module, and generate the attention weight corresponding to each fusion feature in the fusion feature data through the attention module, And performing weighted summation processing on each fusion feature in the fusion feature data according to the attention weight to obtain a corresponding output result;

The second generation module 6 is configured to input the output result into a preset classification module, and normalize the output result through the classification module to generate a predicted probability of occurrence of heart failure corresponding to the user.

In this embodiment, the realization process of the functions and functions of the acquisition module, extraction module, first acquisition module, processing module, first generation module and second generation module in the above-mentioned heart failure risk prediction device is detailed in the above-mentioned pathological data. The analysis method corresponds to the implementation process of steps S1 to S6, which will not be repeated here.

Further, in an embodiment of the present application, the aforementioned extraction module includes:

The first extraction sub-module is configured to use a convolutional neural network to perform feature extraction on the physiological signal data in the pathological data to obtain physiological signal feature data corresponding to the physiological signal data; and,

The second extraction sub-module is used for feature extraction of vital sign data in the pathological data by using a cyclic neural network to obtain corresponding vital sign feature data; and,

The third extraction sub-module is used to extract key features of the case text data in the pathological data by using Chinese natural language processing technology to obtain corresponding case text feature data.

In this embodiment, the realization process of the functions and roles of the first extraction submodule, the second extraction submodule and the third extraction submodule in the above-mentioned heart failure risk prediction device are detailed in the corresponding steps in the above-mentioned pathological data analysis method. The implementation process of S200 to S202 will not be repeated here.

Further, in an embodiment of the present application, the above-mentioned second extraction submodule includes:

An extraction unit, configured to use the cyclic neural network to perform feature extraction on vital sign data in the pathological data to obtain first vital sign feature data;

A judging unit for judging whether there is a missing value in the first vital sign feature data;

The first acquiring unit is configured to, if there is a missing value in the first vital sign characteristic data, acquire the data missing position in the first vital sign characteristic data;

The second acquiring unit is configured to acquire the last feature observation value corresponding to the designated data missing position, and acquire the mean value of the first vital sign feature data, wherein the designated data missing position is the value of all the data missing positions Any data missing position;

The calculation unit is configured to call a preset calculation formula to calculate the designated filling value corresponding to the missing position of the designated data according to the last feature observation value and the average value; the filling unit is configured to use the designated filling value Performing data filling processing on the designated data missing position;

The third acquiring unit is configured to acquire the second vital sign characteristic data obtained after corresponding data filling processing is performed on all missing data positions in the first vital sign characteristic data;

The determining unit is configured to use the second vital sign characteristic data as the vital sign characteristic data.

In this embodiment, the functions and functions of the extraction unit, the judgment unit, the first acquisition unit, the second acquisition unit, the calculation unit, the filling unit, the third acquisition unit, and the determination unit in the above-mentioned heart failure risk prediction device For details, please refer to the implementation process of corresponding steps S2010 to S2017 in the above analysis method of pathological data, which will not be repeated here.

Further, in an embodiment of the present application, the aforementioned pathological data analysis device includes:

The second acquisition module is configured to acquire the importance coefficients corresponding to each type of modal feature data according to the attention weight according to preset rules, wherein the modal feature data includes physiological signal feature data and vital signs Feature data, case text feature data, laboratory examination feature data, and demographic feature data;

The sorting module is used for sorting all the importance coefficients in descending order of numerical value to obtain the corresponding sorting result;

The third generation module is configured to generate the importance prediction report of each type of the modal characteristic data corresponding to the risk of occurrence of heart failure according to the sorting result;

The display module is used to display the importance prediction report.

In this embodiment, the realization process of the functions and functions of the second acquisition module, the sorting module, the third generation module, and the display module in the above-mentioned pathological data analysis device is detailed in the corresponding steps S500 to S503 in the above-mentioned pathological data analysis method. The realization process of this will not be repeated here.

Further, in an embodiment of the present application, the above-mentioned second acquisition module includes:

The screening sub-module is used to screen out the first attention weight corresponding to each of the physiological signal feature data, the second attention weight corresponding to each of the vital sign feature data, and each of the cases A third attention weight corresponding to the characteristic data, a fourth attention weight corresponding to each of the laboratory examination characteristic data, and a fifth attention weight corresponding to each of the demographic characteristic data;

The calculation sub-module is used to calculate the first average value of all the first attention weights, the second average value of all the second attention weights, the third average value of all the third attention weights, A fourth average value of all the fourth attention weights, and a fifth average value of all the fifth attention weights;

The determining sub-module is configured to use the first average value as the first importance coefficient of the physiological signal characteristic data relative to the risk of heart failure, and use the second average value as the vital sign characteristic data relative to the heart failure. The second importance coefficient of the risk of occurrence of heart failure, the third average value is used as the third importance coefficient of the case feature data relative to the risk of heart failure, and the fourth average value is used as the laboratory test feature The fourth importance coefficient of the data relative to the risk of heart failure, and the fifth average value is used as the fifth importance coefficient of the demographic characteristic data relative to the risk of heart failure.

In this embodiment, the functions and functions of the screening sub-module, the calculation sub-module, and the determination sub-module in the above-mentioned pathological data analysis device are detailed in the realization process of corresponding steps S5000 to S5002 in the above-mentioned pathological data analysis method. I won't repeat them here.

The third obtaining module is used to obtain the preset risk threshold;

The first judgment module is used to judge whether the predicted probability of occurrence of heart failure is greater than the risk threshold;

The first determination module is configured to determine that the risk of occurrence of heart failure of the user is a high risk level if the predicted probability of occurrence of heart failure is greater than the risk threshold;

The second judgment module is configured to judge whether the predicted probability of heart failure is within a first preset range if the predicted probability of occurrence of heart failure is not greater than the risk threshold;

The second determination module is configured to determine that the risk of occurrence of heart failure of the user is a medium risk level if the predicted probability of occurrence of heart failure is within a first preset range;

The third determining module is configured to determine that the heart failure risk of the user is a low risk level if the predicted probability of the heart failure risk is not within the first preset range.

In this embodiment, the realization process of the functions and functions of the third acquisition module, the first judgment module, the first judgment module, the second judgment module, the second judgment module, and the third judgment module in the above analysis device for pathological data is specific For details, refer to the implementation process of corresponding steps S600 to S605 in the above analysis method of pathological data, which will not be repeated here.

The third generation module is used to generate early warning information when the user's heart failure risk is in a high-risk state or a medium-risk state, wherein the early warning information includes the predicted probability of the heart failure and the corresponding risk Grade information

The fourth acquisition module is used to acquire recommended information related to the prevention of heart failure; and,

The fifth obtaining module is used to obtain the identity information of the user;

The sending module is configured to send the warning information and the suggestion information to the user terminal corresponding to the identity information according to the identity information.

In this embodiment, the realization process of the functions and roles of the third generation module, the fourth acquisition module, the fifth acquisition module, and the sending module in the above analysis device for pathological data is detailed in the corresponding step S610 in the above analysis method for pathology data. The implementation process to S613 will not be repeated here.

Referring to FIG. 3, an embodiment of the present application also provides a computer device. The computer device may be a server, and its internal structure may be as shown in FIG. 3. The computer equipment includes a processor, a memory, a network interface, a display screen, an input device and a database connected by a system bus. Among them, the processor designed for the computer equipment is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer equipment is used to store data such as designated feature data, structured feature data, fusion feature data, attention weight, output results, and predicted probability of occurrence of heart failure. The network interface of the computer device is used to communicate with an external terminal through a network connection. The display screen of the computer equipment is an indispensable image and text output device in the computer, which is used to convert digital signals into optical signals so that text and graphics can be displayed on the screen of the display screen. The input device of the computer equipment is the main device for information exchange between the computer and the user or other equipment, and is used to transfer data, instructions, and certain flag information to the computer. The computer program is executed by the processor to realize a pathological data analysis method.

The above-mentioned processor executes the steps of the above-mentioned pathological data analysis method:

Collect pathological data of users;

Those skilled in the art can understand that the structure shown in FIG. 3 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the devices and computer equipment to which the solution of the present application is applied.

An embodiment of the present application also provides a computer-readable storage medium. The computer-readable storage medium may be non-volatile or volatile, and has a computer program stored thereon, which is realized when the computer program is executed by a processor. In the method for analyzing pathological data shown in any of the above exemplary embodiments, the method for analyzing pathological data includes the following steps:

Collect pathological data of users;

In summary, the pathological data analysis method, device, computer equipment, and storage medium provided in the embodiments of the present application collect multiple modal characteristic data of users related to the risk of heart failure, and perform the analysis on the multiple modalities. State feature data is spliced, and based on the attention module and classification module, the fusion feature data generated after the splicing process is analyzed and processed, so as to intelligently and accurately generate the predicted probability of the user's heart failure, and realize the risk of the user's heart failure. Accurate prediction of, effectively improve the processing efficiency of predicting the risk of heart failure of users.

A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing relevant hardware through a computer program. The computer program can be stored and a non-volatile computer readable storage. In the medium, when the computer program is executed, it may include the processes of the above-mentioned method embodiments. Wherein, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and/or volatile memory. Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not a limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

The above are only the preferred embodiments of this application, and do not limit the scope of this application. Any equivalent structure or equivalent process transformation made using the content of the specification and drawings of this application, or directly or indirectly applied to other related The technical field is equally included in the scope of patent protection of this application.

Claims

A method for analyzing pathological data, including:

Collect pathological data of users;

Perform feature extraction on the pathological data to obtain specified feature data related to the risk of heart failure, where the specified feature data includes physiological signal feature data, vital signs feature data, and case text feature data;

Acquiring structured characteristic data of the user related to the risk of occurrence of heart failure, wherein the structured characteristic data includes laboratory examination characteristic data and demographic characteristic data;

Performing splicing processing on the designated characteristic data and the structured characteristic data to obtain merged characteristic data after splicing;

The fusion feature data is used as the input of the preset attention module, and the attention weight corresponding to each fusion feature in the fusion feature data is generated by the attention module, and the attention weight is adjusted according to the attention weight. Each fusion feature in the fusion feature data is subjected to weighted summation processing to obtain a corresponding output result;

The output result is input to a preset classification module, and the output result is normalized by the classification module to generate a predicted probability of occurrence of heart failure corresponding to the user.
The method for analyzing pathological data according to claim 1, wherein the feature extraction is performed on the pathological data to obtain designated characteristic data related to the risk of heart failure, wherein the designated characteristic data includes physiological signal characteristic data The steps of vital signs feature data and case text feature data include:

Using a convolutional neural network to perform feature extraction on the physiological signal data in the pathological data to obtain physiological signal feature data corresponding to the physiological signal data; and,

Perform feature extraction on vital sign data in the pathological data by using a recurrent neural network to obtain corresponding vital sign feature data; and,

Using Chinese natural language processing technology to extract key features from the case text data in the pathological data, to obtain corresponding case text feature data.
The method for analyzing pathological data according to claim 2, wherein the step of using a cyclic neural network to perform feature extraction on vital sign data in the pathological data to obtain corresponding vital sign feature data comprises:

Using the recurrent neural network to perform feature extraction on vital sign data in the pathological data to obtain first vital sign feature data;

Judging whether there are missing values in the first vital sign feature data;

If there is a missing value in the first vital sign characteristic data, acquiring the data missing position in the first vital sign characteristic data;

Acquiring the last characteristic observation value corresponding to the designated data missing position, and acquiring the mean value of the first vital sign characteristic data, wherein the designated data missing position is any data missing position of all the data missing positions;

According to the last feature observation value and the average value, call a preset calculation formula to calculate a designated filling value corresponding to the designated data missing position; use the designated filling value to perform data filling on the designated data missing position deal with;

Acquiring second vital sign feature data obtained after performing corresponding data filling processing on all missing positions in the first vital sign feature data;

Use the second vital sign characteristic data as the vital sign characteristic data.
The method for analyzing pathological data according to claim 1, wherein the fusion feature data is used as the input of a preset attention module, and each fusion of the fusion feature data is generated by the attention module The feature has one-to-one correspondence with attention weights, and performing weighted summation processing on each fused feature in the fused feature data according to the attention weight to obtain the corresponding output result, including:

According to the attention weight, the importance coefficient corresponding to each type of modal feature data is obtained according to preset rules, wherein the modal feature data includes physiological signal feature data, vital sign feature data, and case text feature data , Laboratory examination characteristic data and demographic characteristic data;

Sort all the importance coefficients in descending order of numerical value to obtain the corresponding sorting result;

According to the sorting result, generating the importance prediction report of each type of the modal characteristic data corresponding to the risk of occurrence of heart failure;

Show the importance forecast report.
The method for analyzing pathological data according to claim 4, wherein the importance coefficient corresponding to each type of modal characteristic data is obtained according to the attention weight according to a preset rule, wherein the modal The characteristic data includes the steps of physiological signal characteristic data, vital sign characteristic data, case text characteristic data, laboratory examination characteristic data and demographic characteristic data, including:

The first attention weight corresponding to each of the physiological signal feature data, the second attention weight corresponding to each of the vital signs feature data, and the first attention weight corresponding to each of the case feature data are screened out. Three attention weights, a fourth attention weight corresponding to each of the laboratory examination characteristic data, and a fifth attention weight corresponding to each of the demographic characteristic data;

Calculate the first average value of all the first attention weights, the second average value of all the second attention weights, the third average value of all the third attention weights, and all the fourth attention weights. The fourth average value of the force weights, and the fifth average value of all the fifth attention weights;

The first average value is used as the first importance coefficient of the physiological signal characteristic data relative to the risk of heart failure, and the second average value is used as the second importance coefficient of the vital sign characteristic data relative to the risk of heart failure. Importance coefficient, the third average value is used as the third importance coefficient of the case characteristic data relative to the risk of heart failure, and the fourth average value is used as the laboratory examination characteristic data relative to the occurrence of heart failure The fourth importance coefficient of the risk, and the fifth average value is used as the fifth importance coefficient of the demographic characteristic data relative to the risk of heart failure.
The method for analyzing pathological data according to claim 1, wherein the output result is input to a preset classification module, and the output result is normalized by the classification module to obtain the After the user's corresponding steps of predicting the probability of occurrence of heart failure, include:

Obtain the preset risk threshold;

Judging whether the predicted probability of occurrence of heart failure is greater than the risk threshold;

If the predicted probability of occurrence of heart failure is greater than the risk threshold, determining that the risk of occurrence of heart failure of the user is a high risk level;

If the predicted probability of occurrence of heart failure is not greater than the risk threshold, determining whether the predicted probability of occurrence of heart failure is within a first preset range;

If the predicted probability of occurrence of heart failure is within the first preset range, determining that the risk of occurrence of heart failure of the user is a medium risk level;

If the heart failure risk prediction probability is not within the first preset range, it is determined that the heart failure risk of the user is a low risk level.
The method for analyzing pathological data according to claim 6, wherein the output result is input to a preset classification module, and the output result is normalized by the classification module to obtain the corresponding heart After the steps of predicting the probability of failure risk, include:

When the user's heart failure occurrence risk is a high-risk level state or a medium-risk level state, generating early warning information, where the early warning information includes the predicted probability of the occurrence of heart failure and corresponding risk level information;

Obtain advice related to heart failure prevention; and,

Obtaining the identity information of the user;

According to the identity information, the warning information and the suggestion information are sent to the user terminal corresponding to the identity information.
A device for predicting the risk of heart failure, which includes:

Collection module, used to collect pathological data of users;

The extraction module is used for feature extraction of the pathological data to obtain specified feature data related to the risk of heart failure, wherein the specified feature data includes physiological signal feature data, vital signs feature data, and case text feature data;

The first acquisition module is configured to acquire structured characteristic data of the user related to the risk of heart failure, wherein the structured characteristic data includes laboratory examination characteristic data and demographic characteristic data;

A processing module, configured to perform splicing processing on the designated characteristic data and the structured characteristic data to obtain merged characteristic data after splicing;

The first generation module is configured to use the fusion feature data as the input of the preset attention module, and generate the attention weight corresponding to each fusion feature in the fusion feature data through the attention module, and Performing weighted summation processing on each fusion feature in the fusion feature data according to the attention weight to obtain a corresponding output result;

The second generation module is configured to input the output result into a preset classification module, and normalize the output result through the classification module to generate a predicted probability of occurrence of heart failure corresponding to the user.
A computer device includes a memory and a processor, and a computer program is stored in the memory, wherein the processor implements a pathological data analysis method when the computer program is executed by the processor:

Wherein, the analysis method of the pathological data includes:

Collect pathological data of users;

Perform feature extraction on the pathological data to obtain specified feature data related to the risk of heart failure, where the specified feature data includes physiological signal feature data, vital signs feature data, and case text feature data;

Acquiring structured characteristic data of the user related to the risk of occurrence of heart failure, wherein the structured characteristic data includes laboratory examination characteristic data and demographic characteristic data;

Performing splicing processing on the designated characteristic data and the structured characteristic data to obtain merged characteristic data after splicing;

The fusion feature data is used as the input of the preset attention module, and the attention weight corresponding to each fusion feature in the fusion feature data is generated by the attention module, and the attention weight is adjusted according to the attention weight. Each fusion feature in the fusion feature data is subjected to weighted summation processing to obtain a corresponding output result;

The output result is input to a preset classification module, and the output result is normalized by the classification module to generate a predicted probability of occurrence of heart failure corresponding to the user.
The computer device according to claim 9, wherein the characteristic extraction of the pathological data is performed to obtain designated characteristic data related to the risk of occurrence of heart failure, wherein the designated characteristic data includes physiological signal characteristic data and vital signs The steps of feature data and case text feature data include:

Using a convolutional neural network to perform feature extraction on the physiological signal data in the pathological data to obtain physiological signal feature data corresponding to the physiological signal data; and,

Perform feature extraction on vital sign data in the pathological data by using a recurrent neural network to obtain corresponding vital sign feature data; and,

Using Chinese natural language processing technology to extract key features from the case text data in the pathological data, to obtain corresponding case text feature data.
The computer device according to claim 10, wherein the step of using a cyclic neural network to extract features of vital signs data in the pathological data to obtain corresponding feature data of vital signs comprises:

Using the recurrent neural network to perform feature extraction on vital sign data in the pathological data to obtain first vital sign feature data;

Judging whether there are missing values in the first vital sign feature data;

If there is a missing value in the first vital sign characteristic data, acquiring the data missing position in the first vital sign characteristic data;

Acquiring the last characteristic observation value corresponding to the designated data missing position, and acquiring the mean value of the first vital sign characteristic data, wherein the designated data missing position is any data missing position of all the data missing positions;

According to the last feature observation value and the average value, call a preset calculation formula to calculate a designated filling value corresponding to the designated data missing position; use the designated filling value to perform data filling on the designated data missing position deal with;

Acquiring second vital sign feature data obtained after performing corresponding data filling processing on all missing positions in the first vital sign feature data;

Use the second vital sign characteristic data as the vital sign characteristic data.
The computer device according to claim 9, wherein the fusion feature data is used as the input of a preset attention module, and each fusion feature in the fusion feature data is generated by the attention module one by one. Corresponding attention weight, and performing weighted summation processing on each fusion feature in the fusion feature data according to the attention weight to obtain the corresponding output result, including:

According to the attention weight, the importance coefficient corresponding to each type of modal feature data is obtained according to preset rules, wherein the modal feature data includes physiological signal feature data, vital sign feature data, and case text feature data , Laboratory examination characteristic data and demographic characteristic data;

Sort all the importance coefficients in descending order of numerical value to obtain the corresponding sorting result;

According to the sorting result, generating the importance prediction report of each type of the modal characteristic data corresponding to the risk of occurrence of heart failure;

Show the importance forecast report.
The computer device according to claim 12, wherein the importance coefficient corresponding to each type of modal characteristic data is obtained according to the attention weight according to a preset rule, wherein the modal characteristic data includes The steps of physiological signal characteristic data, vital signs characteristic data, case text characteristic data, laboratory examination characteristic data, and demographic characteristic data include:

The first attention weight corresponding to each of the physiological signal feature data, the second attention weight corresponding to each of the vital sign feature data, and the first attention weight corresponding to each of the case feature data are screened out. Three attention weights, a fourth attention weight corresponding to each of the laboratory examination characteristic data, and a fifth attention weight corresponding to each of the demographic characteristic data;

Calculate the first average value of all the first attention weights, the second average value of all the second attention weights, the third average value of all the third attention weights, and all the fourth attention weights. The fourth average value of the force weights, and the fifth average value of all the fifth attention weights;

The first average value is used as the first importance coefficient of the physiological signal characteristic data relative to the risk of heart failure, and the second average value is used as the second vital sign characteristic data relative to the risk of heart failure. Importance coefficient, the third average value is used as the third importance coefficient of the case characteristic data relative to the risk of heart failure, and the fourth average value is used as the laboratory examination characteristic data relative to the occurrence of heart failure The fourth importance coefficient of the risk, and the fifth average value is used as the fifth importance coefficient of the demographic characteristic data relative to the risk of heart failure.
9. The computer device according to claim 9, wherein the output result is input to a preset classification module, and the output result is normalized by the classification module to obtain the corresponding After the steps of predicting the probability of occurrence of heart failure, include:

Obtain the preset risk threshold;

Judging whether the predicted probability of occurrence of heart failure is greater than the risk threshold;

If the predicted probability of occurrence of heart failure is greater than the risk threshold, determining that the risk of occurrence of heart failure of the user is a high risk level;

If the predicted probability of occurrence of heart failure is not greater than the risk threshold, determining whether the predicted probability of occurrence of heart failure is within a first preset range;

If the predicted probability of occurrence of heart failure is within the first preset range, determining that the risk of occurrence of heart failure of the user is a medium risk level;

If the heart failure risk prediction probability is not within the first preset range, it is determined that the heart failure risk of the user is a low risk level.
14. The computer device according to claim 14, wherein the output result is input to a preset classification module, and the output result is normalized by the classification module to obtain a corresponding heart failure risk prediction After the probabilistic steps, include:

When the user's heart failure occurrence risk is a high-risk level state or a medium-risk level state, generating early warning information, where the early warning information includes the predicted probability of the occurrence of heart failure and corresponding risk level information;

Obtain advice related to heart failure prevention; and,

Obtaining the identity information of the user;

According to the identity information, the warning information and the suggestion information are sent to the user terminal corresponding to the identity information.
A computer-readable storage medium has a computer program stored thereon, wherein when the computer program is executed by a processor, a method for analyzing pathological data is realized, wherein the method for analyzing pathological data includes the following steps:

Collect pathological data of users;

Perform feature extraction on the pathological data to obtain specified feature data related to the risk of heart failure, where the specified feature data includes physiological signal feature data, vital signs feature data, and case text feature data;

Acquiring structured characteristic data of the user related to the risk of occurrence of heart failure, wherein the structured characteristic data includes laboratory examination characteristic data and demographic characteristic data;

Performing splicing processing on the designated characteristic data and the structured characteristic data to obtain merged characteristic data after splicing;

The fusion feature data is used as the input of the preset attention module, and the attention weight corresponding to each fusion feature in the fusion feature data is generated by the attention module, and the attention weight is adjusted according to the attention weight. Each fusion feature in the fusion feature data is subjected to weighted summation processing to obtain a corresponding output result;

The output result is input to a preset classification module, and the output result is normalized by the classification module to generate a predicted probability of occurrence of heart failure corresponding to the user.
The computer-readable storage medium according to claim 16, wherein the characteristic extraction of the pathological data is performed to obtain designated characteristic data related to the risk of heart failure, wherein the designated characteristic data includes physiological signal characteristic data The steps of vital signs feature data and case text feature data include:

Using a convolutional neural network to perform feature extraction on the physiological signal data in the pathological data to obtain physiological signal feature data corresponding to the physiological signal data; and,

Perform feature extraction on vital sign data in the pathological data by using a recurrent neural network to obtain corresponding vital sign feature data; and,

Using Chinese natural language processing technology to extract key features from the case text data in the pathological data, to obtain corresponding case text feature data.
18. The computer-readable storage medium according to claim 17, wherein the step of using a cyclic neural network to extract features of vital signs in the pathological data to obtain corresponding feature data of vital signs comprises:

Using the recurrent neural network to perform feature extraction on vital sign data in the pathological data to obtain first vital sign feature data;

Judging whether there are missing values in the first vital sign feature data;

If there is a missing value in the first vital sign characteristic data, acquiring the data missing position in the first vital sign characteristic data;

Acquiring the last characteristic observation value corresponding to the designated data missing position, and acquiring the mean value of the first vital sign characteristic data, wherein the designated data missing position is any data missing position of all the data missing positions;

According to the last feature observation value and the average value, call a preset calculation formula to calculate a designated filling value corresponding to the designated data missing position; use the designated filling value to perform data filling on the designated data missing position deal with;

Acquiring second vital sign feature data obtained after performing corresponding data filling processing on all missing positions in the first vital sign feature data;

Use the second vital sign characteristic data as the vital sign characteristic data.
The computer-readable storage medium according to claim 16, wherein the fusion feature data is used as the input of a preset attention module, and each fusion of the fusion feature data is generated by the attention module The feature has one-to-one correspondence with attention weights, and performing weighted summation processing on each fused feature in the fused feature data according to the attention weight to obtain the corresponding output result, including:

According to the attention weight, the importance coefficient corresponding to each type of modal feature data is obtained according to preset rules, wherein the modal feature data includes physiological signal feature data, vital sign feature data, and case text feature data , Laboratory examination characteristic data and demographic characteristic data;

Sort all the importance coefficients in descending order of numerical value to obtain the corresponding sorting result;

According to the sorting result, generating the importance prediction report of each type of the modal characteristic data corresponding to the risk of occurrence of heart failure;

Show the importance forecast report.
The computer-readable storage medium according to claim 16, wherein said inputting said output result to a preset classification module, and performing normalization processing on said output result through said classification module to obtain After the user's corresponding steps of predicting the probability of occurrence of heart failure, include:

Obtain the preset risk threshold;

Judging whether the predicted probability of occurrence of heart failure is greater than the risk threshold;

If the predicted probability of occurrence of heart failure is greater than the risk threshold, determining that the risk of occurrence of heart failure of the user is a high risk level;

If the predicted probability of occurrence of heart failure is not greater than the risk threshold, determining whether the predicted probability of occurrence of heart failure is within a first preset range;

If the predicted probability of occurrence of heart failure is within the first preset range, determining that the risk of occurrence of heart failure of the user is a medium risk level;

If the heart failure risk prediction probability is not within the first preset range, it is determined that the heart failure risk of the user is a low risk level.