Medical data monitoring and early warning method and system based on data processing
Through medical data monitoring and early warning methods based on data processing, real-time physiological medical data and pre-training models, combined with context dependencies, the problem of insufficient basis for disease assessment in the existing technology is solved, and timely and accurate early warning of the onset of the disease is achieved.
Patent Information
- Application Number
- CN202411457978.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-17
- Publication Date
- 2025-02-11
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the prior art, the basis for evaluating the development of patients' disease is low, real-time and objectivity, which leads to insufficient reliability and timeliness of the assessment, and the inability to avoid further disease development in a timely and accurate manner.
Provide a medical data monitoring and early warning method based on data processing, by receiving real-time physiological medical data streams from wearable devices, extracting the physiological medical data feature sequences and the feature sequences after slices associated with multiple consecutive time periods, and using pre-trained Comprehensive target disease warning model, combined with context dependencies, outputs the probability of target disease onset, and issues early warning to medical terminals when the preset probability threshold is reached.
It improves the real-time and reliability of early warning of target disease attacks, so that medical staff can timely and accurately formulate medical plans to avoid the further development of the disease.
Smart Images
Figure CN119418923A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of medical technology, and in particular to a medical data monitoring and early warning method and system based on data processing. Background Art
[0002] At present, for some specific diseases (such as neurological diseases, brain diseases, and heart diseases), when patients are hospitalized, doctors can measure the patient's medical data (such as pulse data, heart rate data, brain waves, and blood pressure data) at regular intervals (such as one day). In turn, doctors can understand the progression of the patient's condition and make timely medical plans to avoid further progression of the disease.
[0003] However, the above-mentioned evaluation basis for evaluating the progression of the patient's condition is low in completeness, timeliness, and objectivity, resulting in low reliability and timeliness in evaluating the progression of the patient's condition, which may result in a medical plan that cannot timely and accurately avoid further progression of the disease. Summary of the invention
[0004] The present application provides a medical data monitoring and early warning method and system based on data processing, which is used to solve the problem that the basis for evaluating the progression of a patient's condition in the prior art is of low integrity, real-timeness and objectivity, resulting in low reliability and timeliness in evaluating the progression of a patient's condition.
[0005] In a first aspect, the present application provides a medical data monitoring and early warning method based on data processing, which is applied to a medical server. The method provided by the present application includes:
[0006] When a start instruction is obtained from the medical terminal, a physiological medical data stream of the patient collected in real time from the wearable device is received, where the physiological medical data stream includes at least a pulse beat data stream, a heart rate data stream, an electroencephalogram data stream, and a blood pressure data stream;
[0007] Input the physiological and medical data feature sequences associated with the M consecutive time periods before the day, the physiological and medical data feature sequences obtained from the Nth slices corresponding to the previous M consecutive days, the physiological and medical data feature sequences obtained from the Nth slices within the day, the first context dependency in time series, and the second context dependency in time series into the pre-trained comprehensive target disease warning model to output a first probability of the target disease occurring in the patient after a preset second time period, wherein the comprehensive target disease warning model is trained based on a plurality of first training samples, wherein each first training sample includes the physiological and medical data feature sequences associated with the M consecutive time periods before the day in history, the physiological and medical data feature sequences obtained from the Nth slices corresponding to the previous M consecutive days in history, the physiological and medical data feature sequences obtained from the Nth slices within the day in history, the historical first context dependency in time series, the historical second context dependency in time series, and the first probability of the target disease occurring in the patient after the corresponding second preset time period in history, and the second preset time period is greater than the preset first time period;
[0008] When the first probability is greater than a set probability threshold, disease warning information is sent to the medical terminal.
[0009] In a possible implementation, after outputting the first probability that the target disease will occur in the patient after the preset second time period, the method provided by the present application further includes:
[0010] Inputting physiological and medical data feature sequences associated with the M consecutive time periods before the current day, physiological and medical data feature sequences obtained from the Nth slice of the current day, and the first context dependency relationship in time series into a pre-trained first target disease warning model to output a second probability of the target disease occurring in the patient after a preset second time period; wherein the first target disease warning model is obtained by training based on a plurality of second training samples, each second training sample comprising the physiological and medical data feature sequences associated with the M consecutive time periods before the current day, physiological and medical data feature sequences obtained from the Nth slice of the current day, and the first context dependency relationship in time series;
[0011] Inputting the second context dependency relationship in time series between the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day into the pre-trained second target disease warning model to output a third probability that the target disease will occur in the patient after a preset second time period, wherein the second target disease warning model is trained based on a plurality of third training samples, each of which includes the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days, the physiological and medical data feature sequence obtained from the Nth slice on the current day, and the second context dependency relationship in the historical time series;
[0012] The first probability is modified according to the second probability and the third probability.
[0013] In a possible implementation, extracting a first context dependency relationship in time sequence between a physiological and medical data feature sequence associated with M consecutive time periods before the day and a physiological and medical data feature sequence obtained by the Nth slicing, including: extracting a first context dependency relationship in time sequence between a physiological and medical data feature sequence associated with M consecutive time periods before the day and a physiological and medical data feature sequence obtained by the Nth slicing according to a deep confidence DBN network;
[0014] Determine the second context dependency relationship in time sequence between the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day, including: determining the second context dependency relationship in time sequence between the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day according to the deep confidence DBN network.
[0015] In a second aspect, the present application provides a medical data monitoring and early warning system based on data processing, which is applied to a medical server.
[0016] The present application provides a medical data monitoring and early warning method and system based on data processing, which can receive the physiological medical data stream of the patient collected in real time from the wearable device. Since the physiological medical data stream is real-time and will not be interrupted, the real-time and reliability of the subsequent early warning of the onset of the target disease can be improved;
[0017] Furthermore, within the same day, the physiological and medical data stream is sliced every preset first time period to obtain a physiological and medical data feature sequence associated with multiple consecutive time periods; within the same day, when the physiological and medical data stream is sliced for the Nth time, the physiological and medical data feature sequence associated with the M consecutive time periods before the same day is extracted, where N is an integer greater than or equal to 3, M is an integer greater than or equal to 2, and M is less than N.
[0018] It can be understood that the changes in the physiological and medical data feature sequence in multiple adjacent time periods can be a good basis for judging the development of the disease, and can improve the real-time and reliability of the subsequent warning of the onset of the target disease. Therefore, the physiological and medical data feature sequence associated with the previous M consecutive time periods within the day and the physiological and medical data feature sequence obtained by the Nth slice can be extracted, and the first context dependency in time sequence;
[0019] It can be understood that the changes in the characteristic sequence of physiological and medical data in the same time period can also be a good basis for judging the development of the disease, which can improve the real-time and reliability of subsequent warnings for the onset of the target disease.
[0020] Therefore, the second context dependency relationship in time sequence between the physiological and medical data feature sequence obtained from the Nth slices corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day can be determined;
[0021] The physiological and medical data feature sequences associated with the M consecutive time periods before the day, the physiological and medical data feature sequences obtained from the Nth slices corresponding to the previous M consecutive days, the physiological and medical data feature sequences obtained from the Nth slices within the day, the first context dependency in time sequence, and the second context dependency in time sequence are input into the pre-trained comprehensive target disease warning model to output the first probability of the target disease attacking the patient after the preset second time length. In this way, the reliability and real-time performance of the first probability of the target disease attacking the patient after the preset second time length are also high. When the first probability is greater than the set probability threshold, the disease warning information is sent to the medical terminal. In this way, medical staff can make timely and accurate medical plans to avoid further development of the disease. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0023] Figure 1 A flowchart of a medical data monitoring and early warning method based on data processing provided in an embodiment of the present application;
[0024] Figure 2 A functional module block diagram of a medical data monitoring and early warning system based on data processing provided in an embodiment of the present application. DETAILED DESCRIPTION
[0025] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments made by ordinary technicians in this field under the enlightenment of the embodiments belong to the scope of protection of the present application.
[0026] The Long Short-Term Memory (LSTM) network is a time-recurrent neural network that is specially designed to solve the long-term dependency problem of ordinary RNNs (recurrent neural networks). Due to its unique design structure, LSTM is suitable for processing and predicting important events with very long intervals and delays in time series. The cleverness of the LSTM network is that by adding input gates, forget gates and output gates, and designing the weight coefficients between connections, the LSTM network can accumulate long-term connections between nodes that are far away, thereby achieving long-term memory of data.
[0027] DBN network: Deep Belief Networks (DBN) is a neural network model in deep learning methods. DBN network has multiple hidden layers. The hidden layer of DBN is connected to the next hidden layer, and the last hidden layer is connected to the visible layer, but the units in the same layer are not connected; multiple hidden layers can perform higher-level correlation operations on data, which also makes the hidden layers before and after training have more order. After the previous hidden layer is fully trained, the next hidden layer can be trained again with the training data obtained from the previous hidden layer. Repeat this process continuously to finally obtain trained data.
[0028] See also Figure 1 The embodiment of the present application provides a medical data monitoring and early warning method based on data processing, which is applied to a medical server. The method provided by the embodiment of the present application includes:
[0029] S101: When a start instruction is obtained from a medical terminal, a physiological medical data stream of a patient collected in real time from a wearable device is received, wherein the physiological medical data stream includes at least a pulse data stream, a heart rate data stream, an electroencephalogram data stream, and a blood pressure data stream.
[0030] Among them, the wearable components may include smart bracelets and smart headgear.
[0031] S102: within the day, slicing the physiological and medical data stream at intervals of a preset first time length to obtain physiological and medical data feature sequences associated with a plurality of continuous time periods.
[0032] Exemplarily, the preset first duration may be 5 minutes, 10 minutes, or 20 minutes, etc., which is not limited here.
[0033] S103: When slicing the physiological and medical data stream for the Nth time within the same day, extract the physiological and medical data feature sequences associated with the M consecutive time periods before the same day.
[0034] Wherein, N is an integer greater than or equal to 3, M is an integer greater than or equal to 2, and M is less than N.
[0035] For example, when the physiological and medical data stream is sliced for the tenth time within the day, the physiological and medical data feature sequences associated with the five consecutive time periods before the day are extracted. The length of the time period is equal to the preset first time length.
[0036] S104: extracting the first context dependency in time sequence between the physiological and medical data feature sequences associated with the M consecutive time periods before the current day and the physiological and medical data feature sequences obtained by the Nth slicing.
[0037] For example, when the time period is 9:00-9:10 on September 6, and M=5, the M consecutive time periods before that day can be understood as: 8:50-9:00 on September 6, 8:40-8:50 on September 6, 8:30-9:40 on September 6, 8:20-8:30 on September 6, and 8:10-8:20 on September 6.
[0038] For example, the specific implementation of S104 includes but is not limited to the following two methods:
[0039] The first type: The server is configured with a long short-term memory model, which includes two groups of long short-term memory network layers and a random inactivation processing layer and a feature connection layer connected in sequence.
[0040] According to the context dependency in time sequence of the physiological and medical data feature sequences associated with the M consecutive time periods before the day and the physiological and medical data feature sequences obtained by the Nth slicing extracted by the first group of long short-term memory network layers; then, the physiological and medical data feature sequences associated with the M consecutive time periods before the day and the physiological and medical data feature sequences obtained by the Nth slicing with the context dependency are randomly reset to zero by the random inactivation processing layer; the physiological and medical data feature sequences associated with the M consecutive time periods before the day and the physiological and medical data feature sequences obtained by the Nth slicing after the processing are output to the next group of long short-term memory network layers connected in sequence; The network layer and the random inactivation processing layer are used to sequentially perform the same processing on the physiological and medical data feature sequence associated with the M consecutive time periods before the day and the physiological and medical data feature sequence obtained by the Nth slicing by the long short-term memory network layer and the random inactivation processing layer connected in sequence with the first group, so as to obtain the context dependency in time sequence based on the physiological and medical data feature sequence associated with the M consecutive time periods before the day and the physiological and medical data feature sequence obtained by the Nth slicing extracted by the first group of long short-term memory network layers; the previous context dependency and the following context dependency respectively output by the two long short-term memory network layers are connected through the feature connection layer to obtain the first context dependency.
[0041] The second type: extracting the physiological and medical data feature sequence associated with the M consecutive time periods before the day and the physiological and medical data feature sequence obtained by the Nth slice based on the deep confidence DBN network, and the first context dependency in time sequence.
[0042] S105: Determine a second context dependency relationship in time sequence between the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day.
[0043] For example, if today is September 6 and M=5, the previous M consecutive days are September 5, September 4, September 3, September 2, and September 1.
[0044] For example, the specific implementation of S105 includes but is not limited to the following two methods:
[0045] The first method: extracting the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day through the first group of long short-term memory network layers, and the context dependency in time sequence; then performing random zeroing processing on some feature values in the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day with the context dependency through the random inactivation processing layer; outputting the processed physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day to the next group of sequentially connected The long short-term memory network layer and the random inactivation processing layer are used to sequentially perform the same processing on the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day by the long short-term memory network layer and the random inactivation processing layer which are sequentially connected to the first group, so as to obtain the context dependency relationship in time sequence between the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days extracted by the first group of long short-term memory network layers and the physiological and medical data feature sequence obtained from the Nth slice on the current day; the previous context dependency relationship and the following context dependency relationship respectively output by the two long short-term memory network layers are connected through the feature connection layer to obtain a second context dependency relationship.
[0046] As mentioned above, LSTM is a time-recurrent neural network, which is composed of memory blocks that are expanded over time. In the structure of LSTM, the hidden layer at each moment contains multiple memory blocks (usually one memory block), each memory block contains multiple historical information records, and each historical information record contains an interval and three gates. LSTM can process multiple physiological and medical data feature sequences in a sequence. The memory block in LSTM can obtain the relationship between the current physiological and medical data feature sequence and the previous physiological and medical data feature sequence through the interaction of the input gate, the forget gate, and the output gate, so as to achieve the purpose of absorbing information, updating the state and outputting information. Therefore, the function of LSTM is to capture the dependency between physiological and medical data feature sequences. By connecting the output of LSTM at each time step with the fully connected network distributed over time, the dependency between physiological and medical data feature sequences can be better captured, and the output of LSTM at each time step can be used as the feature of the current physiological and medical data feature sequence, including contextual dependencies. Dropout is a method for optimizing artificial neural networks with deep structures. In the learning process, some weights or outputs of the hidden layer are randomly reset to zero to reduce the mutual dependence between nodes, thereby realizing the regularization of the neural network, reducing its structural risk, and preventing or reducing overfitting. For the same physiological and medical data feature sequence, an output will be obtained through the first group of long short-term memory network layers and random dropout processing layers connected in sequence, and then the obtained output will be passed through the second group of long short-term memory network layers and random dropout processing layers connected in sequence to obtain another output, and then the two obtained outputs will be connected into a vector. Therefore, for the same physiological and medical data feature sequence, the first group of long short-term memory network layers and random dropout processing layers connected in sequence can capture the dependency relationship between it and the physiological and medical data feature sequence on its left, and the second group of long short-term memory network layers and random dropout processing layers connected in sequence can capture the dependency relationship between it and the physiological and medical data feature sequence on its right. Therefore, the vector connected by the above two outputs can better capture the contextual dependency of the current physiological and medical data feature sequence, so as to better represent the current physiological and medical data feature sequence.
[0047] The second type: determining the second context dependency in time sequence between the physiological and medical data feature sequence obtained from the Nth slices corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the same day based on the deep confidence DBN network.
[0048] S106: Input the physiological and medical data feature sequence associated with the M consecutive time periods before the day, the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days, the physiological and medical data feature sequence obtained from the Nth slice within the day, the first context dependency in time sequence, and the second context dependency in time sequence into the pre-trained comprehensive target disease early warning model to output a first probability of the target disease occurring in the patient after a preset second time length.
[0049] Exemplarily, the preset second duration may be 1 hour, 2 hours, or 4 hours, etc.
[0050] Among them, the comprehensive target disease warning model is obtained by training based on multiple first training samples, wherein each first training sample includes the physiological and medical data feature sequence associated with M consecutive time periods before the historical day, the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days in history, the physiological and medical data feature sequence obtained from the Nth slice on the historical day, the historical first context dependency in time series, the historical second context dependency in time series, and the first probability of the target disease occurring in the patient after the corresponding historical second preset time length, and the second preset time length is greater than the preset first time length.
[0051] S107: When the first probability is greater than a set probability threshold, disease warning information is sent to the medical terminal.
[0052] To sum up, the medical data monitoring and early warning method based on data processing provided in the embodiment of the present application can receive the physiological medical data stream of the patient collected in real time from the wearable device. Since the physiological medical data stream is real-time and will not be interrupted, the real-time and reliability of subsequent early warning of the onset of the target disease can be improved.
[0053] Furthermore, within the same day, the physiological and medical data stream is sliced every preset first time period to obtain a physiological and medical data feature sequence associated with multiple consecutive time periods; within the same day, when the physiological and medical data stream is sliced for the Nth time, the physiological and medical data feature sequence associated with the M consecutive time periods before the same day is extracted, where N is an integer greater than or equal to 3, M is an integer greater than or equal to 2, and M is less than N.
[0054] It can be understood that the changes in the physiological and medical data feature sequence in multiple adjacent time periods can be a good basis for judging the development of the disease, and can improve the real-time and reliability of the subsequent warning of the onset of the target disease. Therefore, the physiological and medical data feature sequence associated with the previous M consecutive time periods within the day and the physiological and medical data feature sequence obtained by the Nth slice can be extracted, and the first context dependency in time sequence;
[0055] It can be understood that the changes in the characteristic sequence of physiological and medical data in the same time period can also be a good basis for judging the development of the disease, which can improve the real-time and reliability of subsequent warnings for the onset of the target disease.
[0056] Therefore, the second context dependency relationship in time sequence between the physiological and medical data feature sequence obtained from the Nth slices corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day can be determined;
[0057] The physiological and medical data feature sequences associated with the M consecutive time periods before the day, the physiological and medical data feature sequences obtained from the Nth slices corresponding to the previous M consecutive days, the physiological and medical data feature sequences obtained from the Nth slices within the day, the first context dependency in time sequence, and the second context dependency in time sequence are input into the pre-trained comprehensive target disease warning model to output the first probability of the target disease attacking the patient after the preset second time length. In this way, the reliability and real-time performance of the first probability of the target disease attacking the patient after the preset second time length are also high. When the first probability is greater than the set probability threshold, the disease warning information is sent to the medical terminal. In this way, medical staff can make timely and accurate medical plans to avoid further development of the disease.
[0058] In addition, after S106, the method provided in the embodiment of the present application further includes:
[0059] Step 1: Input the physiological and medical data feature sequence associated with the M consecutive time periods before the day, the physiological and medical data feature sequence obtained from the Nth slice on the day, and the first context dependency in time series into the pre-trained first target disease warning model to output the second probability of the target disease occurring in the patient after a preset second time period.
[0060] Among them, the first target disease warning model is trained based on multiple second training samples, each second training sample includes a physiological and medical data feature sequence associated with M consecutive time periods before the current day, a physiological and medical data feature sequence obtained from the Nth slice on the current day, and the first context dependency relationship in the history in time series.
[0061] Step 2: Input the second context dependency in time sequence between the physiological and medical data feature sequence obtained from the Nth slices corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day into the pre-trained second target disease warning model to output a third probability of the target disease occurring in the patient after a preset second time period.
[0062] Among them, the second target disease warning model is trained based on multiple third training samples, each third training sample includes the physiological and medical data feature sequence obtained from the Nth slice corresponding to the M consecutive days before the history, the physiological and medical data feature sequence obtained from the Nth slice on the historical day, and the second context dependency relationship on the historical time series.
[0063] Step 3: Modify the first probability according to the second probability and the third probability.
[0064] Specifically, according to the formula P 11 =K 1 P 1 +K 2 P 2 +K 3 P 3 , corrected first probability, where K 1 Greater than K 2 and K 3 , and K 1 +K 2 +K 3 =1,P 1 is the first probability before correction, P 2 is the second probability, P 3 is the third probability, P 11 is the corrected first probability, K 1 , K 2 , and K 3 are weighted coefficients. In this way, a more accurate first probability can be obtained. For example, K 1 =0.6, K 2 and K 3 Both are equal to 0.4.
[0065] In addition, the medical data monitoring and early warning system based on data processing provided in the embodiment of the present application is applied to a medical server. Figure 2 As shown, the system includes a data receiving unit, a data slicing unit, a sequence extraction unit, a dependency extraction unit, a disease warning unit, and a prompt information output unit.
[0066] The data receiving unit is used to receive the physiological medical data stream of the patient collected in real time from the wearable device when obtaining the start-up instruction from the medical terminal. The physiological medical data stream includes at least pulse beat data stream, heart rate data stream, brain wave data stream and blood pressure data stream.
[0067] The data slicing unit is used to slice the physiological and medical data stream at intervals of a preset first time length within a day to obtain physiological and medical data feature sequences associated with multiple continuous time periods.
[0068] The sequence extraction unit is used to extract the physiological and medical data feature sequences associated with M consecutive time periods before the same day when slicing the physiological and medical data stream for the Nth time within the same day. Wherein N is an integer greater than or equal to 3, M is an integer greater than or equal to 2, and M is less than N.
[0069] The dependency extraction unit is used to extract the first context dependency in time sequence between the physiological and medical data feature sequences associated with the M consecutive time periods before the day and the physiological and medical data feature sequences obtained by the Nth slice; and determine the second context dependency in time sequence between the physiological and medical data feature sequences obtained by the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequences obtained by the Nth slice on the day.
[0070] The disease warning unit is used to input the physiological and medical data feature sequences associated with the M consecutive time periods before the day, the physiological and medical data feature sequences obtained from the Nth slices corresponding to the previous M consecutive days, the physiological and medical data feature sequences obtained from the Nth slices within the day, the first context dependency in time sequence, and the second context dependency in time sequence into the pre-trained comprehensive target disease warning model to output a first probability of the target disease occurring in the patient after a preset second time length.
[0071] Among them, the comprehensive target disease warning model is obtained by training based on multiple first training samples, wherein each first training sample includes the physiological and medical data feature sequence associated with M consecutive time periods before the historical day, the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days in history, the physiological and medical data feature sequence obtained from the Nth slice on the historical day, the historical first context dependency in time series, the historical second context dependency in time series, and the first probability of the target disease occurring in the patient after the corresponding historical second preset time length, and the second preset time length is greater than the preset first time length.
[0072] The prompt information output unit is used to send disease warning information to the medical terminal when the first probability is greater than a set probability threshold.
[0073] In a possible implementation, the system provided by the present application further includes:
[0074] A probability correction unit is used to input the physiological and medical data feature sequence associated with the M consecutive time periods before the current day, the physiological and medical data feature sequence obtained by the Nth slice of the current day, and the first context dependency relationship in time series into the pre-trained first target disease warning model to output a second probability that the target disease will occur in the patient after a preset second time period; wherein the first target disease warning model is obtained by training based on a plurality of second training samples, each second training sample comprising the physiological and medical data feature sequence associated with the M consecutive time periods before the current day, the physiological and medical data feature sequence obtained by the Nth slice of the current day, and the first context dependency relationship in time series;
[0075] Inputting the second context dependency relationship in time series between the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day into the pre-trained second target disease warning model to output a third probability that the target disease will occur in the patient after a preset second time period, wherein the second target disease warning model is trained based on a plurality of third training samples, each of which includes the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days, the physiological and medical data feature sequence obtained from the Nth slice on the current day, and the second context dependency relationship in the historical time series;
[0076] The first probability is modified according to the second probability and the third probability.
[0077] In a possible implementation manner, the data correction unit is specifically configured to: 11 =K 1 P 1 +K 2 P 2 +K 3 P 3 , corrected first probability, where K 1 Greater than K 2 and K 3 , and K 1 +K 2 +K 3 =1,P 1 is the first probability before correction, P 2 is the second probability, P 3 is the third probability, P 11 is the corrected first probability, K 1 , K 2 , and K 3 are all weighting coefficients.
[0078] In a possible implementation, the server is configured with a long short-term memory model, which includes two groups of long short-term memory network layers and a random inactivation processing layer, a feature connection layer, and a dependency extraction unit, which is specifically used to extract the physiological and medical data feature sequence associated with the M consecutive time periods before the day and the physiological and medical data feature sequence obtained by the Nth slice according to the temporal context dependency through the first group of long short-term memory network layers; and then perform random zeroing processing on some feature values in the physiological and medical data feature sequence associated with the M consecutive time periods before the day and the physiological and medical data feature sequence obtained by the Nth slice that have the context dependency through the random inactivation processing layer; Output the processed physiological and medical data feature sequence associated with the M consecutive time periods before the day and the physiological and medical data feature sequence obtained by the Nth slicing to the next group of sequentially connected long short-term memory network layers and random inactivation processing layers, so as to sequentially perform the same processing as the first group of sequentially connected long short-term memory network layers and random inactivation processing layers on the above-mentioned physiological and medical data feature sequence associated with the M consecutive time periods before the day and the physiological and medical data feature sequence obtained by the Nth slicing, and obtain the physiological and medical data feature sequence associated with the M consecutive time periods before the day and the physiological and medical data feature sequence obtained by the Nth slicing extracted by the first group of sequentially connected long short-term memory network layers, and the following context dependency in time sequence is obtained. relationship; connect the context dependency and the context dependency respectively output by the two long short-term memory network layers through the feature connection layer to obtain the first context dependency; extract the physiological and medical data feature sequence obtained by the first group of long short-term memory network layers from the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained by the Nth slice on the same day, and the context dependency in time sequence; then perform random zeroing processing on some feature values in the physiological and medical data feature sequence obtained by the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained by the Nth slice on the same day with the context dependency relationship; output the processed physiological and medical data feature sequence corresponding to the previous M consecutive days respectively. The physiological and medical data feature sequence obtained from the Nth slice and the physiological and medical data feature sequence obtained from the Nth slice on the same day are transmitted to the next group of sequentially connected long short-term memory network layers and random inactivation processing layers, so as to sequentially perform the same processing as the first group of sequentially connected long short-term memory network layers and random inactivation processing layers on the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the same day, so as to obtain the context dependency relationship in time sequence between the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days extracted by the first group of long short-term memory network layers and the physiological and medical data feature sequence obtained from the Nth slice on the same day;The previous context dependency and the following context dependency outputted by the two long short-term memory network layers are connected through the feature connection layer to obtain the second context dependency. ;
[0079] In one possible implementation, the dependency extraction unit is specifically used to extract the first context dependency in time sequence between the physiological and medical data feature sequences associated with the M consecutive time periods before the day and the physiological and medical data feature sequences obtained by the Nth slice according to the deep confidence DBN network; and to determine the second context dependency in time sequence between the physiological and medical data feature sequences obtained by the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequences obtained by the Nth slice on the day according to the deep confidence DBN network.
[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.
Claims
1. A medical data monitoring and early warning method based on data processing, characterized in that: Applied to a medical server, the method comprises: When a start instruction is obtained from the medical terminal, a physiological medical data stream of the patient collected in real time from the wearable device is received, wherein the physiological medical data stream includes at least a pulse beat data stream, a heart rate data stream, an electroencephalogram data stream, and a blood pressure data stream; Input the physiological and medical data feature sequences associated with the M consecutive time periods before the day, the physiological and medical data feature sequences obtained from the Nth slices corresponding to the previous M consecutive days, the physiological and medical data feature sequences obtained from the Nth slices within the day, the first context dependency in time series, and the second context dependency in time series into a pre-trained comprehensive target disease warning model to output a first probability of the target disease occurring in the patient after a preset second time period, wherein the comprehensive target disease warning model is trained based on a plurality of first training samples, wherein each of the first training samples includes the physiological and medical data feature sequences associated with the M consecutive time periods before the day in history, the physiological and medical data feature sequences obtained from the Nth slices corresponding to the previous M consecutive days in history, the physiological and medical data feature sequences obtained from the Nth slices within the day in history, the historical first context dependency in time series, the historical second context dependency in time series, and the first probability of the target disease occurring in the patient after a corresponding second preset time period in history, and the second preset time period is greater than the preset first time period; When the first probability is greater than a set probability threshold, disease warning information is sent to the medical terminal.
2. The method according to claim 1, characterized in that After outputting the first probability that the target disease will occur in the patient after the preset second time period, the method further includes: The physiological and medical data feature sequence associated with the M consecutive time periods before the current day, the physiological and medical data feature sequence obtained from the Nth slice of the current day, and the first context dependency relationship in time series are input into the pre-trained first target disease warning model to output a second probability of the target disease occurring in the patient after a preset second time period; wherein the first target disease warning model is trained based on multiple second training samples, and each of the second training samples includes the physiological and medical data feature sequence associated with the M consecutive time periods before the current day, the physiological and medical data feature sequence obtained from the Nth slice of the current day, and the first context dependency relationship in time series.
3. The method according to claim 2, characterized in that After outputting the first probability that the target disease will occur in the patient after the preset second time period, the method further includes: Inputting the second context dependency relationship in time series between the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day into the pre-trained second target disease warning model to output a third probability that the target disease will occur in the patient after a preset second time period, wherein the second target disease warning model is trained based on a plurality of third training samples, each of the third training samples including the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days, the physiological and medical data feature sequence obtained from the Nth slice on the current day, and the second context dependency relationship in the historical time series; The first probability is modified according to the second probability and the third probability.
4. The method according to claim 1, characterized in that: The first context dependency relationship in time sequence between the physiological and medical data feature sequence associated with the M consecutive time periods before the day and the physiological and medical data feature sequence obtained by the Nth slicing is extracted, including: extracting the physiological and medical data feature sequence associated with the M consecutive time periods before the day and the physiological and medical data feature sequence obtained by the Nth slicing according to the deep confidence DBN network, and the first context dependency relationship in time sequence.
5. The method according to claim 4, characterized in that The second context dependency relationship in time sequence between the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day is determined, including: determining the second context dependency relationship in time sequence between the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day according to the deep confidence DBN network.
6. A medical data monitoring and early warning system based on data processing, characterized in that: Applied to a medical server, the system comprises: A data receiving unit, configured to receive a physiological medical data stream of a patient collected in real time from a wearable device when a start instruction from a medical terminal is obtained, wherein the physiological medical data stream includes at least a pulse beat data stream, a heart rate data stream, an electroencephalogram data stream, and a blood pressure data stream; A symptom warning unit, used for inputting physiological and medical data feature sequences associated with the M consecutive time periods before the day, physiological and medical data feature sequences obtained from the Nth slices corresponding to the previous M consecutive days, physiological and medical data feature sequences obtained from the Nth slices within the day, the first context dependency in time series, and the second context dependency in time series into a pre-trained comprehensive target disease warning model to output a first probability of the target disease attacking the patient after a preset second time period, wherein the comprehensive target disease warning model is trained based on a plurality of first training samples, wherein each of the first training samples includes physiological and medical data feature sequences associated with the M consecutive time periods before the day in history, physiological and medical data feature sequences obtained from the Nth slices corresponding to the previous M consecutive days in history, physiological and medical data feature sequences obtained from the Nth slices within the day in history, the historical first context dependency in time series, the historical second context dependency in time series, and the first probability of the target disease attacking the patient after the corresponding second preset time period in history, and the second preset time period is greater than the preset first time period; A prompt information output unit is used to send disease warning information to the medical terminal when the first probability is greater than a set probability threshold.
7. The system according to claim 6, characterized in that The system further comprises: A probability correction unit is used to input the physiological and medical data feature sequence associated with the M consecutive time periods before the current day, the physiological and medical data feature sequence obtained from the Nth slice of the current day, and the first context dependency relationship in time series into a pre-trained first target disease warning model to output a second probability that the target disease will occur in the patient after a preset second time period; wherein the first target disease warning model is trained based on multiple second training samples, and each of the second training samples includes the physiological and medical data feature sequence associated with the M consecutive time periods before the current day, the physiological and medical data feature sequence obtained from the Nth slice of the current day, and the first context dependency relationship in time series.
8. The system according to claim 6, characterized in that Inputting the second context dependency relationship in time series between the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequence obtained from the Nth slice on the current day into the pre-trained second target disease warning model to output a third probability that the target disease will occur in the patient after a preset second time period, wherein the second target disease warning model is trained based on a plurality of third training samples, each of the third training samples including the physiological and medical data feature sequence obtained from the Nth slice corresponding to the previous M consecutive days, the physiological and medical data feature sequence obtained from the Nth slice on the current day, and the second context dependency relationship in the historical time series; The first probability is modified according to the second probability and the third probability.
9. The system according to claim 6, characterized in that The dependency extraction unit is specifically used to extract the first context dependency in time sequence between the physiological and medical data feature sequences associated with the M consecutive time periods before the day and the physiological and medical data feature sequences obtained by the Nth slice according to the deep confidence DBN network; and to determine the second context dependency in time sequence between the physiological and medical data feature sequences obtained by the Nth slice corresponding to the previous M consecutive days and the physiological and medical data feature sequences obtained by the Nth slice on the day according to the deep confidence DBN network.