Risk reminding method and device, equipment and medium
By using timing test results and classification models in CDSS to generate transient probability values, the problem that CDSS is difficult to distinguish between emergency and non-emergency risk reminders is solved, and doctors' work efficiency and decision-making quality are improved.
Patent Information
- Application Number
- CN202411476257.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-22
- Publication Date
- 2025-01-10
AI Technical Summary
When the existing clinical decision support system (CDSS) processes a large amount of medical information, it is difficult to quickly distinguish between emergency and non-emergency risk reminders, resulting in the omission or delay of crisis alerts, and reduces the work efficiency and decision quality of doctors.
By obtaining the timing test results, input them into the pre-trained classification model, a transient probability value is generated, and a risk reminder is triggered based on the preset probability threshold, filtering out non-emergency and non-important risk reminders.
It effectively reduces the work burden of doctors, improves work efficiency and decision-making quality, and ensures timely attention to real emergencies.
Smart Images
Figure CN119296782A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of medical technology, and in particular to a risk reminder method, device, equipment and medium. Background Art
[0002] "Transient" refers to the phenomenon that during medical tests, certain test indicators (such as hemoglobin, blood sugar, proteinuria, etc.) temporarily deviate from the normal range, but then return to the normal range on their own. This "transient" change reflects the patient's physiological fluctuations, mild stress or temporary pathological state, which has little impact on the patient's health, but has a greater impact on the early prevention of the disease, the base adjustment of the treatment plan and the patient's emotional stability.
[0003] In related technologies, in order to accurately identify the "transient" and "persistent" nature of test indicators, the Clinical Decision Support System (CDSS) usually issues risk alerts for abnormal test indicators based on preset normal and critical values. Figure 1 As shown, when the test value of the test indicator is greater than the normal value and less than the critical value, CDSS will trigger a risk reminder to prompt the doctor to monitor the patient's test indicators.
[0004] However, doctors need to process a large amount of medical information in their daily work, and CDSS risk alerts undoubtedly increase this burden. When doctors are faced with a large number of risk alerts, it is difficult to quickly distinguish which risk alerts are truly urgent and important, which can easily lead to the omission of crisis alerts or delay in handling, thereby reducing work efficiency and decision-making quality. Summary of the invention
[0005] Based on the above problems, the present application provides a risk reminder method, device, equipment and medium, which can filter out a large number of non-urgent and non-important risk reminders, reduce the burden on doctors, and improve doctors' work efficiency and decision-making quality.
[0006] The embodiments of the present application disclose the following technical solutions:
[0007] In a first aspect, an embodiment of the present application discloses a risk reminder method, the method comprising:
[0008] Acquire a time series test result, wherein the time series test result includes a sequence of multiple test values for a single test indicator arranged in chronological order;
[0009] Inputting the time series test result into a classification model so that the classification model outputs a transient probability value corresponding to the time series test result;
[0010] If the transient probability value is higher than the preset probability threshold corresponding to the inspection indicator, a risk reminder is triggered.
[0011] Optionally, the classification model is constructed as follows:
[0012] Acquire a test result set for a single test indicator, the test result set comprising historical test results of a plurality of patients, the historical test results comprising a sequence of multiple test values for the test indicator arranged in chronological order;
[0013] According to the test result set, target patients whose test values at the Mth and M+Nth time points are both abnormal values are screened out, wherein M and N are both positive integers;
[0014] Based on the historical test results of the target patient, model training is performed using a machine learning algorithm to obtain a classification model.
[0015] Optionally, the method of performing model training by a machine learning algorithm based on the historical test results of the target patient to obtain a classification model includes:
[0016] Obtaining case information of the target patient;
[0017] Extracting case segmentation information from the case information of the target patient through natural language processing technology;
[0018] Based on the historical test results of the target patient and the case segmentation information, model training is performed through a machine learning algorithm to obtain a classification model.
[0019] Optionally, the classification model includes an input layer, a long short-term memory network (LSTM) layer, a fully connected layer, and an output layer;
[0020] Inputting the time series test result into a classification model so that the classification model outputs a transient probability value corresponding to the time series test result, including:
[0021] The timing test result is input into the input layer, and the input layer passes the timing test result to the LSTM layer, so that the LSTM layer obtains the timing feature of the timing test result, and passes the timing feature to the fully connected layer, so that the fully connected layer determines a transient probability value corresponding to the timing test result according to the timing feature, and passes the transient probability value to the output layer, so that the output layer outputs the transient probability value.
[0022] Optionally, if the transient probability value is higher than a preset probability threshold corresponding to the inspection indicator, triggering a risk reminder includes:
[0023] If the transient probability value is higher than the preset probability threshold corresponding to the inspection indicator, and the difference between the transient probability value and the preset probability threshold is higher than the preset difference threshold, a first level risk reminder is triggered.
[0024] Optionally, if the transient probability value is higher than a preset probability threshold corresponding to the inspection indicator, triggering a risk reminder includes:
[0025] Obtaining a dynamic change ratio of the timing test result;
[0026] If the transient probability value is higher than the preset probability threshold corresponding to the inspection indicator, and the dynamic change ratio is greater than the preset ratio threshold, a first level risk reminder is triggered.
[0027] In a second aspect, an embodiment of the present application discloses a risk reminder device, the device comprising: a result acquisition module, a probability output module and a risk reminder module;
[0028] The result acquisition module is used to acquire a time series test result, wherein the time series test result includes a sequence of multiple test values for a single test indicator arranged in chronological order;
[0029] The probability output module is used to input the time series test result into the classification model, so that the classification model outputs a transient probability value corresponding to the time series test result;
[0030] The risk reminder module is used to trigger a risk reminder if the transient probability value is higher than a preset probability threshold corresponding to the inspection indicator.
[0031] Optionally, the classification model construction device is specifically as follows: a set acquisition module, a patient screening module and a model training module;
[0032] The set acquisition module is used to acquire a test result set for a single test indicator, wherein the test result set includes historical test results of multiple patients, and the historical test results include a sequence of multiple test values for the test indicator arranged in chronological order;
[0033] The patient screening module is used to screen out target patients whose test values at the Mth and M+Nth time points are both abnormal values according to the test result set, wherein M and N are both positive integers;
[0034] The model training module is used to perform model training through a machine learning algorithm according to the historical test results of the target patient to obtain a classification model.
[0035] Optionally, the model training module is specifically used to: obtain the case information of the target patient; extract the case segmentation information in the case information of the target patient through natural language processing technology; perform model training through a machine learning algorithm based on the historical test results of the target patient and the case segmentation information to obtain a classification model.
[0036] Optionally, the classification model includes an input layer, a long short-term memory network LSTM layer, a fully connected layer and an output layer; the probability output module is specifically used to: input the timing test result into the input layer, and the input layer passes the timing test result to the LSTM layer, so that the LSTM layer obtains the timing characteristics of the timing test result, and passes the timing characteristics to the fully connected layer, so that the fully connected layer determines a transient probability value corresponding to the timing test result according to the timing characteristics, and passes the transient probability value to the output layer, so that the output layer outputs the transient probability value.
[0037] Optionally, the risk reminder module is specifically used to trigger a first level risk reminder if the transient probability value is higher than a preset probability threshold corresponding to the inspection indicator, and the difference between the transient probability value and the preset probability threshold is higher than a preset difference threshold.
[0038] Optionally, the risk reminder module is specifically used to: obtain the dynamic change ratio of the timing test result; if the transient probability value is higher than the preset probability threshold corresponding to the test index, and the dynamic change ratio is greater than the preset ratio threshold, then trigger the first level risk reminder.
[0039] In a third aspect, an embodiment of the present application discloses a risk reminder device, including: a memory and a processor;
[0040] The memory is used to store programs;
[0041] The processor is used to execute the program to implement the various steps of the method described in the first aspect.
[0042] In a fourth aspect, an embodiment of the present application discloses a computer storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.
[0043] The embodiments of the present application provide a risk reminder method, device, equipment and medium. The risk reminder method first captures the dynamic changes of the test value of the test indicator over time by acquiring the time series test results (i.e., multiple test values for a single test indicator arranged in chronological order), and then inputs the time series test results into a pre-trained classification model to obtain a transient probability value of the time series test results. Finally, by setting a preset probability threshold corresponding to the test indicator, the risk reminder is triggered only when the transient probability value output by the classification model is higher than the threshold. In this way, a large number of non-urgent and non-important risk reminders are effectively filtered out, reducing the burden on doctors and improving their work efficiency and decision-making quality. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0045] Figure 1 A schematic diagram of a risk reminder method;
[0046] Figure 2 A flow chart of a risk reminder method provided in an embodiment of the present application;
[0047] Figure 3 A flowchart of another risk reminder method provided in an embodiment of the present application;
[0048] Figure 4 A schematic diagram of a risk reminder method provided in an embodiment of the present application;
[0049] Figure 5 A schematic diagram of a risk reminder device provided in an embodiment of the present application;
[0050] Figure 6 A schematic diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0051] As described above, in order to accurately identify the "transient" and "persistent" nature of test indicators, CDSS usually issues risk alerts for abnormal test values based on preset normal and critical values. Figure 1, which is a schematic diagram of a risk reminder method. When the test value of the test indicator is greater than the normal value and less than the critical value, CDSS will trigger a risk reminder to remind the doctor that the patient's test indicator needs to be monitored. When the test value of the test indicator is greater than the critical value, a critical alert will be issued to the doctor through a specific management process (such as a critical value management system).
[0052] However, on the one hand, doctors need to deal with a large amount of medical information in their daily work, and CDSS risk alerts undoubtedly increase this burden. When doctors are faced with a large number of risk alerts, it is difficult to quickly distinguish which risk alerts are truly urgent and important, which can easily lead to the omission of crisis alerts or delayed processing, which not only increases the workload of doctors, but also reduces their work efficiency and decision-making quality.
[0053] On the other hand, since the risk reminder logic of CDSS is relatively simple, that is, as long as the test value of the test indicator is greater than the normal value and less than the critical value, CDSS will trigger a risk reminder, which makes it easy to generate false reminders for mild, non-pathological test values. Frequent false reminders not only waste medical resources, but may also cause doctors to lose trust in risk reminders. When doctors are accustomed to ignoring a large number of insignificant risk reminders, truly important risk reminders may also be missed.
[0054] On the other hand, medical decision-making is a complex process that requires comprehensive consideration of the patient's medical history, symptoms, signs, test results, and other aspects. However, CDSS can often only provide alerts based on a single test value, lacking in-depth analysis and comprehensive judgment of the patient's overall clinical condition. This limitation makes it difficult for CDSS to truly assist doctors in making complex clinical decisions.
[0055] After research, the inventors proposed a risk reminder method, device, system and medium. The risk reminder method first captures the dynamic changes of the test values of the test indicators over time by obtaining the time series test results (i.e., multiple test values for a single test indicator arranged in chronological order), and then inputs the time series test results into a pre-trained classification model to obtain a transient probability value of the time series test results. Finally, by setting a preset probability threshold corresponding to the test indicator, the risk reminder is triggered when the transient probability value output by the classification model is higher than the threshold. In this way, a large number of non-urgent and non-important risk reminders are effectively filtered out, reducing the burden on doctors and improving doctors' work efficiency and decision-making quality. At the same time, the technical solution also provides more scientific and reasonable support for early prevention of diseases, base adjustment of treatment plans and emotional stability of patients.
[0056] In order to enable those skilled in the art to better understand the solution of the present application, 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 only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0057] See also Figure 2 , which is a flow chart of a risk reminder method provided in an embodiment of the present application. The method includes:
[0058] S201: Acquire a time series test result, where the time series test result includes a sequence of multiple test values for a single test indicator arranged in chronological order.
[0059] A test index refers to an index that can quantitatively represent a specific physiological, pathological or environmental state. For example, in the medical field, the test index includes but is not limited to hemoglobin concentration (used to reflect anemia), blood sugar level (used for diabetes monitoring), blood amylase (used to evaluate pancreatic function), urine protein (used for kidney disease screening), serum creatinine (used for kidney function evaluation), etc.
[0060] Time series testing refers to measuring a test indicator multiple times at different time points to obtain a sequence of test values that change over time. This test method aims to reveal the dynamic changes of test indicators over time and then analyze possible physiological mechanisms, disease processes or treatment effects.
[0061] Specifically, first, determine the test indicators that need to be monitored and the time points for monitoring (such as hourly, daily, weekly, monthly, etc.). Then, take samples of patients or samples at the specified time points. After each measurement, record the measurement value and its corresponding time point. Finally, sort the collected measurement values in chronological order to form the time series test results.
[0062] In some specific implementations, the timing test results of the user's physiological indicators can be obtained through wearable devices such as electronic watches. Among them, the physiological indicators include at least one of heart rate, blood pressure, blood oxygen saturation and skin temperature. This application does not limit the specific acquisition method.
[0063] S202: Inputting the time series test result into the classification model, so that the classification model outputs a transient probability value corresponding to the time series test result.
[0064] In the risk reminder method of the embodiment of the present application, the classification model is constructed based on a machine learning algorithm, such as logistic regression, random forest, gradient boosting machines (GBM) or a deep learning model.
[0065] The output of the classification model is a probability value, which represents the probability of transient occurrence of the test indicator corresponding to the time series test result. Here, "transient" usually refers to a test indicator that becomes abnormal in a short period of time, but then quickly returns to normal, without long-term or continuous changes. The generation of the probability of transient occurrence depends on a large number of features and rules that the classification model deeply learns and masters during the training phase. These features and rules may include, but are not limited to: the trend of test values changing over time (such as rising, falling, fluctuating, etc.), the periodic relationship between test values, the frequency of occurrence of specific test value combinations, etc.
[0066] S203: If the transient probability value is higher than the preset probability threshold corresponding to the inspection indicator, a risk reminder is triggered.
[0067] When the time series test result of a test indicator outputs a high transient probability through the classification model (for example, the transient probability is higher than the preset probability threshold), CDSS can issue an early warning or reminder to remind relevant personnel to pay attention to potential unstable factors or risks so that preventive or intervention measures can be taken in time to ensure the health and safety of patients. Among them, the preset probability threshold is a value set according to actual needs, which is used to determine whether the transient probability value has reached a level that requires vigilance.
[0068] It is understandable that the risk reminder method can also be applied to other fields that require monitoring and early warning, such as environmental monitoring, industrial safety, etc. In some examples, in terms of environmental monitoring, the risk reminder method can obtain the time series test results of environmental indicators such as air quality, water quality, and noise, and use similar classification models to evaluate the probability of transient abnormalities (such as sudden pollution events) in these indicators. Once a high risk is predicted, the corresponding environmental risk reminder can be triggered, prompting relevant departments to take prompt action to protect the ecological environment and public health. In other examples, in the field of industrial safety, the risk reminder method can monitor the time series data of various process parameters (such as temperature, pressure, flow, etc.) on the production line in real time, and the classification model can identify potential safety hazards, such as precursors to equipment failures or abnormal fluctuations during operation. Once a high-risk situation is identified, a safety warning can be triggered immediately to help companies and employees take preventive measures in a timely manner to avoid accidents and ensure production safety. It should be noted that this application does not limit specific application scenarios.
[0069] In summary, an embodiment of the present application provides a risk reminder method, which first captures the dynamic changes of the test indicators over time by acquiring the time series test results (i.e., multiple test values for a single test indicator arranged in chronological order), and then inputs the time series test results into a pre-trained classification model to obtain a transient probability value of the time series test results. Finally, by setting a reasonable preset probability threshold, the risk reminder is triggered only when the transient probability value output by the classification model is higher than the threshold. In this way, a large number of non-urgent and non-important risk reminders are effectively filtered out, reducing the burden on doctors and improving their work efficiency and decision-making quality.
[0070] See also Figure 3 , which is a flow chart of another risk reminder method provided in an embodiment of the present application. The method includes:
[0071] S301: Construct a test result set for a single test indicator.
[0072] First, obtain a test result set for a single test indicator. Among them, the test result set includes historical test results of multiple patients, and the historical test results include a sequence of multiple test values for a single test indicator arranged in chronological order. The test result set can provide a basis for subsequent model training and analysis. See Table 1, which is a schematic table of a test result set provided in an embodiment of the present application (taking blood sugar as an example in Table 1).
[0073] Table 1
[0074]
[0075] It should be noted that the above test result set can be obtained from a hospital information system (HIS) or an electronic medical record system (EMR), or the patient's historical test results can be directly obtained through a wearable device. This application does not limit this.
[0076] S302: According to the test result set, target patients whose test values at the Mth and M+Nth time points are both abnormal values are screened out, where M and N are both positive integers.
[0077] In a specific implementation, after obtaining the test result set, it is necessary to identify the index value corresponding to each patient to see whether there is a situation where the test result is normal first and then abnormal. It should be noted that the abnormal test result here refers to any abnormal test result after the normal test result, and this application does not limit the number and order of abnormal test results.
[0078] Exemplarily, taking blood sugar as an example, in the test result set, the blood sugar level of patient D at 10:00 on January 1 was 5.5, the blood sugar level at 12:00 on January 1 was 9.5, and the blood sugar level at 16:00 on January 1 was 5.6. Blood sugar levels of 5.5 and 5.6 are normal, and blood sugar level of 9.5 is abnormal. Therefore, it is determined that the patient D had a normal test result first and then an abnormal test result.
[0079] Subsequently, based on the historical test results of patients who first had normal test results and then had abnormal test results, it is determined whether the test values at the Mth time point and the M+Nth time point are both abnormal (i.e., whether it is transient). Specifically, if at least one test result remains abnormal among all the test results after the abnormal test result, it is classified into the non-transient abnormal group; if all the test results after the abnormal test result return to normal, it is classified into the transient abnormal group.
[0080] For example, taking serum creatinine as an example, if the serum creatinine of patient B is abnormal for the first time, and all subsequent test results are still abnormal, the patient is classified into the non-transient abnormal group. If the serum creatinine of the patient is abnormal for the first time, and the subsequent test results return to normal, the patient is classified into the transient abnormal group.
[0081] Subsequently, patients in the non-transient abnormal group are screened as target patients, and the target patients are target patients whose test values at the Mth and M+Nth time points are both abnormal values.
[0082] It should be noted that when judging the persistence of anomalies, a certain limit on the number of tests should be considered. Specifically, within a period of time, if the index value of a certain test indicator is abnormal only a few times, and remains normal for most of the other time, then such anomalies are likely to be accidental fluctuations rather than true persistent anomalies. Therefore, by setting a threshold for the number of tests, truly persistent phenomena can be further screened out, thereby improving the accuracy and reliability of risk alerts. This application does not limit this.
[0083] S303: For each test indicator, in the corresponding test result set, select target patients whose test values at the Mth and M+Nth time points are both abnormal values.
[0084] In order to comprehensively analyze the characteristics of different test indicators, it is necessary to repeat the above steps S301-S302 for each test indicator, establish their own test result sets, and screen out target patients whose test values at the Mth and M+Nth time points are both abnormal values.
[0085] S304: Based on the historical test results of the target patient, a model is trained using a machine learning algorithm to obtain a classification model.
[0086] In some specific implementations, the classification model can be obtained by directly training the model through a machine learning algorithm based on the historical test results of the target patient.
[0087] In another specific implementation, in order to enhance the prediction ability of the classification model, the basic information and case information of the target patient can be obtained first, and then the classification model can be obtained by training the model through a machine learning algorithm based on the basic information, case information and historical test results of the target patient. The basic information refers to the patient's age, gender, medical history and other information.
[0088] It should be noted that the numerical information in the basic information (such as age, etc.) can be directly used for the training of the classification model without additional processing, while the categorical information in the basic information (such as gender, etc.) needs to be encoded and converted to obtain an embedded vector before it can be used for the training of the classification model. Exemplarily, encoding can be performed by One-Hot Encoding and Embedding Vectors methods. This application does not limit the specific encoding method.
[0089] In addition, for case information, it is necessary to first use natural language processing technology to extract the case segmentation information in the case information, and then encode the case segmentation information to obtain an embedded vector before it can be used for classification model training. Among them, case segmentation information refers to the information (such as symptom description, diagnosis results, treatment plan, etc.) further extracted using pre-trained models such as BERT or natural language processing technology after the patient's medical record text is segmented.
[0090] Specifically, the method of obtaining the classification model by training the model through a machine learning algorithm based on the embedding vectors of the above basic information, the embedding vectors of the case segmentation information, and the historical test results can be as follows:
[0091] In the risk reminder method provided in the embodiment of the present application, the classification model includes an input layer, a long short-term memory network (Long Short Term Memory, LSTM) layer, a fully connected layer and an output layer, which aims to accurately judge the abnormal persistence of the test results.
[0092] First, in order to use the embedding vector of basic information, the embedding vector of case segmentation information and the historical test results as the input of the classification model, the above three can be combined.
[0093] In one example, the three can be directly concatenated and merged, that is, the dimensions of the three vectors are directly added (d1+d2+d3). This method retains all information, but may increase the complexity of the classification model.
[0094] In another example, a weighted summation method may be used to combine the three, where weights are assigned to different vectors and added according to the weights to balance the importance of each vector in the classification model.
[0095] It should be noted that, since the sequence lengths of the historical test results of different patients may be different, the historical test results of different patients at each time point can be sorted into sequences of fixed length, and then the above three can be merged. Exemplarily, this can be achieved through a padding or truncating strategy. Among them, the padding strategy is to add a specific padding value (such as 0 or a negative value) at the end of a shorter sequence to ensure that the length of all sequences is consistent. The truncation strategy is to directly delete sequences that exceed the fixed length. This application does not limit the specific sorting method.
[0096] It should be noted that if the amount of data used to train the model (i.e., the embedding vector of basic information, the embedding vector of case segmentation information, and historical test results) is small, it is easy to have overfitting problems. In the risk reminder method provided in the embodiment of the present application, a part of the neurons in the classification model can be randomly discarded by the Dropout technique to prevent overfitting, so as to improve the generalization ability of the classification model. In addition, the early stopping method can also be used to prevent overfitting. This application does not limit the method for preventing overfitting.
[0097] The LSTM layer in the classification model is responsible for capturing the temporal features in the time series data by processing the merged vectors. Each LSTM unit in the LSTM layer includes a forget gate, an input gate, and an output gate, which control the inflow, outflow, and retention of input information. Through its internal forget gate, input gate, and output gate structure, LSTM is able to selectively retain important information and forget unnecessary information.
[0098] It should be noted that in the risk reminder method provided in the embodiment of the present application, the number of LSTM units in the LSTM layer also needs to be considered. More LSTM units can usually capture more complex time series features, but may also lead to increased model training time and the risk of overfitting. Therefore, it is necessary to adjust the number of LSTM units according to the requirements of the task and the complexity of the data to balance the ability to capture time series features and the complexity of the classification model.
[0099] The time series features extracted by the LSTM layer are then sent to the fully connected layer for further integration and classification, so as to obtain the transient probability value corresponding to the time series test result. The fully connected layer introduces nonlinear factors through activation functions (such as ReLU, sigmoid, etc.), maps high-dimensional features to low-dimensional space, and sets the number of output units and activation functions according to task requirements. Exemplarily, in a binary classification problem, a sigmoid activation function and a single output unit can be used to predict the probability of belonging to a certain category; in a multi-classification problem, a softmax activation function and multiple output units can be used to predict the probability of belonging to each category. Since the classification model involved in this application provides a binary classification method, that is, the time series test result of the classification input model is "transient" or "non-transient", a binary cross entropy loss function (Binary Cross-EntropyLoss) can be used as the loss function of the classification model.
[0100] In the training process of the classification model, the update of the model parameters depends on the core mechanism of the backpropagation algorithm. This algorithm accurately calculates the gradient of the loss function to the model parameters and adjusts the parameters in the opposite direction of the gradient to drive the loss function to evolve in the direction of minimization.
[0101] In order to further optimize the training efficiency and effect, the present application can also use an adaptive learning rate optimizer such as the Adam optimizer, which can automatically adjust the learning rate of each parameter during the training process. By calculating the first-order moment estimate and the second-order moment estimate of the gradient, the Adam optimizer can implement different learning rate adjustment strategies for different parameters, thereby accelerating the convergence speed of the model and improving its training effect.
[0102] It should be noted that when the classification model is built and put into use, the accuracy, recall and F1 score of the classification model can also be obtained. Among them, the accuracy refers to the proportion of the classification model correctly classified in all categories. The recall rate measures the ability of the classification model to identify "transient abnormalities". The F1 score is the harmonic mean of the accuracy and recall rate, which can balance the above indicators. In practical applications, the training data can be input into the classification model to obtain the training results of the training data. Subsequently, the training results are compared with the actual results of the training data to obtain the accuracy, recall and F1 score of the classification model, which helps to understand the performance of the classification model in identifying "transient abnormalities" and "non-transient abnormalities". In addition, when the classification model is built and put into use, the confusion matrix (Confusion Matrix) of the classification model can also be obtained. The confusion matrix includes true positives (the number of samples that are actually "transient abnormalities" and the model predicts that they are also "transient abnormalities"), false positives (the number of samples that are actually "non-transient abnormalities" but the model incorrectly predicts that they are "transient abnormalities"), true negatives (the number of samples that are actually "non-transient abnormalities" and the model predicts that they are also "non-transient abnormalities"), and false negatives (the number of samples that are actually "transient abnormalities" but the model incorrectly predicts that they are "non-transient abnormalities"). The above confusion matrix can help understand the performance of the classification model in identifying "transient abnormalities" and "non-transient abnormalities".
[0103] S305: Inputting the time series test results to be classified into the classification model, so that the classification model outputs a transient probability value corresponding to the time series test results.
[0104] Specifically, the classification model includes an input layer, an LSTM layer, a fully connected layer and an output layer. Therefore, the timing test result is input into the input layer, and the input layer passes the timing test result to the LSTM layer, so that the LSTM layer obtains the timing features of the timing test result, and passes the timing features to the fully connected layer, so that the fully connected layer determines the transient probability value corresponding to the timing test result according to the timing features, and passes the transient probability value to the output layer, so that the output layer outputs the transient probability value.
[0105] S306: If the transient probability value is higher than the preset probability threshold corresponding to the inspection indicator, a risk reminder is triggered.
[0106] See also Figure 4, this figure is a schematic diagram of a risk warning method provided in an embodiment of the present application. After the time series test results are input into the classification model and a transient probability value is obtained, it can be determined whether the transient probability value is greater than the preset probability threshold. If so, CDSS issues a risk reminder. If not, CDSS does not issue a risk reminder. Exemplarily, the above-mentioned preset probability threshold can be 0.5. When the transient probability value is greater than 0.5, it is judged as a "transient abnormality" and a risk reminder needs to be triggered; on the contrary, when the transient probability value is less than or equal to 0.5, it is judged as a "non-transient abnormality" and no risk reminder needs to be triggered. This application does not limit the specific preset probability threshold.
[0107] In some specific implementations, after the classification model analyzes the test index and outputs its corresponding transient probability value, it can also determine whether the transient probability value is higher than the preset probability threshold corresponding to the test index, and whether the difference between this transient probability value and the preset probability threshold is also higher than the preset difference threshold. If so, the first level of risk reminder is triggered. The first level of risk reminder includes but is not limited to sending alarm information to relevant personnel, initiating emergency plans, or conducting further in-depth analysis. In this way, it can be ensured that the first level of risk reminder is triggered only when the transient probability value is significantly higher than the preset probability threshold, thereby reducing the possibility of false alarms due to slight excesses, avoiding unnecessary panic or overreaction, and also ensuring timely attention to truly high-risk cases.
[0108] In another specific implementation, after the classification model analyzes the test indicator and outputs the corresponding transient probability value, it can also be determined whether the transient probability value is higher than a preset probability threshold corresponding to the test indicator, and whether the dynamic change ratio of the timing test result is greater than the preset ratio threshold. Among them, the dynamic change ratio is greater than the preset ratio threshold, indicating that the state or event represented by the test indicator is undergoing significant changes, which may be a signal of increased risk. If so, the first level of risk reminder is triggered. In this way, significant changes in risk status can be discovered and responded to in a timely manner, which helps to take preventive measures before the risk increases.
[0109] In summary, an embodiment of the present application provides a risk reminder method, which first captures the dynamic changes of the test indicators over time by acquiring the time series test results (i.e., multiple test values for a single test indicator arranged in chronological order), and then inputs the time series test results into a pre-trained classification model to obtain a transient probability value of the time series test results. Finally, by setting a reasonable preset probability threshold, the risk reminder is triggered only when the transient probability value output by the classification model is higher than the threshold. In this way, a large number of non-urgent and non-important risk reminders are effectively filtered out, reducing the burden on doctors and improving their work efficiency and decision-making quality.
[0110] See also Figure 5 , which is a schematic diagram of a risk warning device provided in an embodiment of the present application. The risk warning device 500 includes: a result acquisition module 501, a probability output module 502 and a risk reminder module 503.
[0111] The result acquisition module 501 is used to acquire the time series test result, which includes a sequence of multiple test values for a single test indicator arranged in chronological order;
[0112] A probability output module 502, used to input the time series test result into the classification model, so that the classification model outputs a transient probability value corresponding to the time series test result;
[0113] The risk reminder module 503 is used to trigger a risk reminder if the transient probability value is higher than a preset probability threshold corresponding to the inspection indicator.
[0114] In some specific implementations, the classification model construction device is specifically as follows: a set acquisition module, a patient screening module, and a model training module;
[0115] A set acquisition module, used to acquire a test result set for a single test indicator, the test result set including historical test results of multiple patients, the historical test results including multiple test values for the test indicator arranged in chronological order;
[0116] A patient screening module is used to screen out target patients whose test values at the Mth and M+Nth time points are both abnormal values in the test result set, where M and N are both positive integers;
[0117] The model training module is used to train the model through a machine learning algorithm based on the historical test results of the target patients to obtain a classification model.
[0118] In some specific implementations, the model training module is specifically used to: obtain the case information of the target patient; extract the case segmentation information in the case information of the target patient through natural language processing technology; and perform model training through a machine learning algorithm based on the historical test results and case segmentation information of the target patient to obtain a classification model.
[0119] In some specific implementations, the risk reminder module 503 is specifically used to: if the transient probability value is higher than the preset probability threshold corresponding to the inspection index, and the difference between the transient probability value and the preset probability threshold is higher than the preset difference threshold, then trigger the first level risk reminder.
[0120] In some specific implementations, the risk alert module 503 is specifically used to: obtain the dynamic change ratio of the timing test results; if the transient probability value is higher than the preset probability threshold corresponding to the test index, and the dynamic change ratio is greater than the preset ratio threshold, then trigger the first level risk alert.
[0121] In summary, an embodiment of the present application provides a risk reminder device, which first captures the dynamic changes of the test indicators over time by acquiring the time series test results (i.e., multiple test values for a single test indicator arranged in chronological order), and then inputs the time series test results into a pre-trained classification model to obtain a transient probability value of the time series test results. Finally, by setting a reasonable preset probability threshold, the risk reminder is triggered only when the transient probability value output by the classification model is higher than the threshold. In this way, a large number of non-urgent and non-important risk reminders are effectively filtered out, reducing the burden on doctors and improving their work efficiency and decision-making quality.
[0122] See also Figure 6 , which is a schematic diagram of a computer device provided in an embodiment of the present application.
[0123] like Figure 6 As shown, the computer device 01 is in the form of a general computing device. The components of the computer device 01 may include but are not limited to: one or more processors or processing units 03, a system memory 08, and a bus 04 connecting different system components (including the system memory 08 and the processing unit 03).
[0124] Bus 04 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor or a local bus using any of a variety of bus structures. For example, these architectures include but are not limited to Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus and Peripheral Component Interconnect (PCI) bus.
[0125] The computer device 01 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computer device 01, including volatile and non-volatile media, removable and non-removable media.
[0126] The system memory 08 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 09 and / or cache memory 10. The computer device 01 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 11 may be used to read and write non-removable, non-volatile magnetic media ( Figure 6 not shown, usually called a "hard drive"). Although Figure 6 Not shown, a removable non-volatile disk (e.g., a "floppy disk") may be provided.
[0127] The memory 08 may include a magnetic disk drive that reads and writes to a removable non-volatile optical disk (such as a CD-ROM, a DVD-ROM or other optical media). In these cases, each drive may be connected to the bus 04 via one or more data medium interfaces. The memory 08 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present invention.
[0128] A program / utility 12 having a set (at least one) of program modules 13 may be stored, for example, in the memory 08, such program modules 13 including but not limited to an operating system, one or more application programs, other program modules, and program data, each of which or some combination thereof may include an implementation of a network environment. The program modules 13 generally perform the functions and / or methods of the embodiments described herein.
[0129] The computer device 01 may also communicate with one or more external devices 02 (e.g., keyboard, pointing device, display 07, etc.), one or more devices that enable a user to interact with the computer device 01, and / or any device that enables the computer device 01 to communicate with one or more other computing devices (e.g., network card, modem, etc.). Such communication may be performed via input / output (I / O) interface 06. Furthermore, the computer device 01 may also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN) and / or public network, such as the Internet) via network adapter 05. Figure 6 As shown, the network adapter 05 communicates with other modules of the computer device 01 via the bus 04. It should be understood that although Figure 6Not shown, other hardware and / or software modules may be used in conjunction with computer device 01, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.
[0130] The processor unit 03 executes various functional applications and data processing by running programs stored in the system memory 08, such as implementing the risk reminder method provided in the embodiment of the present application.
[0131] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components indicated as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.
[0132] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. A risk reminder method, characterized in that: The method comprises: Acquire a time series test result, wherein the time series test result includes a sequence of multiple test values for a single test indicator arranged in chronological order; Inputting the time series test result into a classification model so that the classification model outputs a transient probability value corresponding to the time series test result; If the transient probability value is higher than the preset probability threshold corresponding to the inspection indicator, a risk reminder is triggered.
2. The method according to claim 1, characterized in that The construction method of the classification model is as follows: Acquire a test result set for a single test indicator, the test result set comprising historical test results of a plurality of patients, the historical test results comprising a sequence of multiple test values for the test indicator arranged in chronological order; According to the test result set, target patients whose test values at the Mth and M+Nth time points are both abnormal values are screened out, wherein M and N are both positive integers; Based on the historical test results of the target patient, model training is performed using a machine learning algorithm to obtain a classification model.
3. The method according to claim 2, characterized in that The method of training a model by a machine learning algorithm based on the historical test results of the target patient to obtain a classification model includes: Obtaining case information of the target patient; Extracting case segmentation information from the case information of the target patient through natural language processing technology; Based on the historical test results of the target patient and the case segmentation information, model training is performed through a machine learning algorithm to obtain a classification model.
4. The method according to claim 1, characterized in that: The classification model includes an input layer, a long short-term memory network LSTM layer, a fully connected layer and an output layer; Inputting the time series test result into a classification model so that the classification model outputs a transient probability value corresponding to the time series test result, including: The timing test result is input into the input layer, and the input layer passes the timing test result to the LSTM layer, so that the LSTM layer obtains the timing feature of the timing test result, and passes the timing feature to the fully connected layer, so that the fully connected layer determines a transient probability value corresponding to the timing test result according to the timing feature, and passes the transient probability value to the output layer, so that the output layer outputs the transient probability value.
5. The method according to claim 1, characterized in that If the transient probability value is higher than the preset probability threshold corresponding to the inspection indicator, a risk reminder is triggered, including: If the transient probability value is higher than the preset probability threshold corresponding to the inspection indicator, and the difference between the transient probability value and the preset probability threshold is higher than the preset difference threshold, a first level risk reminder is triggered.
6. The method according to claim 1, characterized in that If the transient probability value is higher than the preset probability threshold corresponding to the inspection indicator, a risk reminder is triggered, including: Obtaining a dynamic change ratio of the timing test result; If the transient probability value is higher than the preset probability threshold corresponding to the inspection indicator, and the dynamic change ratio is greater than the preset ratio threshold, a first level risk reminder is triggered.
7. A risk reminder device, characterized in that: The device comprises: a result acquisition module, a probability output module and a risk reminder module; The result acquisition module is used to acquire a time series test result, wherein the time series test result includes a sequence of multiple test values for a single test indicator arranged in chronological order; The probability output module is used to input the time series test result into the classification model, so that the classification model outputs a transient probability value corresponding to the time series test result; The risk reminder module is used to trigger a risk reminder if the transient probability value is higher than a preset probability threshold corresponding to the inspection indicator.
8. The device according to claim 7, wherein the classification model construction device is specifically as follows: a set acquisition module, a patient screening module and a model training module; The set acquisition module is used to acquire a test result set for a single test indicator, wherein the test result set includes historical test results of multiple patients, and the historical test results include a sequence of multiple test values for the test indicator arranged in chronological order; The patient screening module is used to screen out target patients whose test values at the Mth and M+Nth time points are both abnormal values according to the test result set, wherein: M and N are both positive integers; The model training module is used to perform model training through a machine learning algorithm according to the historical test results of the target patient to obtain a classification model.
9. A risk reminder device, characterized in that: include: Memory and processor; The memory is used to store programs; The processor is used to execute the program to implement each step of the method according to any one of claims 1 to 6.
10. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the method according to any one of claims 1 to 6 is implemented.