WO2021103624A1 - 一种脓毒血症的预警装置、设备及存储介质 - Google Patents

一种脓毒血症的预警装置、设备及存储介质 Download PDF

Info

Publication number
WO2021103624A1
WO2021103624A1 PCT/CN2020/105398 CN2020105398W WO2021103624A1 WO 2021103624 A1 WO2021103624 A1 WO 2021103624A1 CN 2020105398 W CN2020105398 W CN 2020105398W WO 2021103624 A1 WO2021103624 A1 WO 2021103624A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
index data
disease
module
sepsis
Prior art date
Application number
PCT/CN2020/105398
Other languages
English (en)
French (fr)
Inventor
何国平
王旭英
李炳强
何婷
董驰
Original Assignee
医惠科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 医惠科技有限公司 filed Critical 医惠科技有限公司
Publication of WO2021103624A1 publication Critical patent/WO2021103624A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients

Definitions

  • the present invention relates to the field of medical equipment, in particular to an early warning device, equipment and computer readable storage medium for sepsis.
  • the prior art calculates the average/maximum/minimum value of the disease data corresponding to the missing value in the historical index data, and supplements the missing value accordingly, and then performs early warning detection. In fact, it is precisely because of abnormal changes in the target user's index data that the user may be ill. Therefore, the method of supplementing the missing values in the current index data in the prior art will make the prediction result inaccurate.
  • an early warning device for sepsis including:
  • the training module is used to train a prediction model by using sample index data and deviation labels with sepsis labels; wherein, each of the sample index data includes multiple types of disease data; the deviation label is the The label corresponding to the deviation direction of the disease data from the standard value;
  • a calculation module for calculating the correlation values of each type of the disease data with sepsis, and calculating the time threshold corresponding to each type of the disease data
  • the acquisition module is used to acquire the current index data and historical index data of the target user
  • the interpolation module is used to determine the type of disease data corresponding to the missing value in the current index data, and select target disease data that meets the time threshold requirement from the historical index data according to the principle of proximity as the missing value , Get updated current indicator data;
  • a prediction module configured to input respective deviation labels of the current disease data in the updated current index data and the updated current index data into the prediction model to obtain a prediction result
  • the display is used to display the prediction result.
  • the training module specifically includes:
  • the interpolation sub-module is used to interpolate the missing values in the sample index data according to preset rules
  • the calculation sub-module is used to calculate the difference between each of the disease data and the corresponding standard value, and determine the corresponding deviation label according to the difference;
  • the input sub-module is used to select a corresponding type of machine learning model according to the number of the disease data in the sample index data, and input the sample index data and each corresponding deviation label into the machine learning model;
  • the training sub-module is used to train the machine learning model by using the sample index data and the deviation label to obtain the prediction model.
  • the acquisition sub-module specifically includes:
  • the first collection unit is used to extract structured disease data in the LIS library and the nursing system;
  • the second collection unit is used to collect unstructured disease data in the case book by using natural language processing technology
  • the setting unit is configured to obtain the sample index data provided with the sepsis label according to the structured disease data, the unstructured disease data, and the diagnosis result of the medical record.
  • it further includes:
  • the cleaning unit is configured to perform data cleaning on each of the sample index data provided with the sepsis label.
  • it further includes:
  • a receiving module for receiving the first index data of the sepsis label marked by a professional
  • the update module is used to input the first indicator data and the corresponding first deviation label into the prediction model for training, so as to update the prediction model.
  • the interpolation module specifically includes:
  • the sorting sub-module is used to sort the historical index data according to a preset time sequence
  • the determining sub-module is used to determine the type of disease data corresponding to the missing value in the current indicator data
  • the selection sub-module is used to select target disease data that meets the time threshold requirement from the arranged historical index data as the missing value according to the principle of proximity to obtain updated current index data.
  • it further includes:
  • the alerter is used to send out corresponding prompt information when the prediction result is that the target user is ill.
  • the present invention also provides an early warning device for sepsis, including:
  • Memory used to store computer programs
  • the processor is used to implement the following steps when executing the computer program:
  • a prediction model is trained by using sample index data with sepsis labels and deviation labels; wherein, each of the sample index data includes multiple types of disease data; the deviation labels are the disease data and the standard value The label corresponding to the deviation direction;
  • the prediction result is displayed.
  • the present invention also provides a computer-readable storage medium having a computer program stored on the computer-readable storage medium, and when the computer program is executed by a processor, the following steps are implemented:
  • a prediction model is trained by using sample index data with sepsis labels and deviation labels; wherein, each of the sample index data includes multiple types of disease data; the deviation labels are the disease data and the standard value The label corresponding to the deviation direction;
  • the prediction result is displayed.
  • the present invention provides an early warning device for sepsis.
  • the device calculates the correlation value of each type of disease data with sepsis through a calculation module, and calculates the correlation value of each type of disease data.
  • the disease data corresponds to the time threshold; then the interpolation module determines the type of disease data corresponding to the missing value in the current indicator data, and selects the target disease data that meets the time threshold requirement from the historical indicator data as the missing value according to the principle of proximity.
  • the device can not only supplement the missing values of the current index data, but also select the target disease data to meet the time threshold requirement according to the principle of proximity, which can be closer to the current actual situation of the target user, and can more accurately perform sepsis on the target user. Early warning detection of disease.
  • the device uses the updated current index data and the corresponding deviation direction of each current disease data in the updated current index data to predict. By combining multiple forms of data for prediction, the accuracy of the prediction result can be further improved.
  • the present invention also provides an early warning device for sepsis and a computer-readable storage medium, both of which have the above-mentioned beneficial effects.
  • Figure 1 is a flow chart of an early warning device for sepsis provided by an embodiment of the present invention
  • Fig. 2 is a structural diagram of an early warning device for sepsis provided by an embodiment of the present invention.
  • Fig. 1 is a structural diagram of an early warning device for sepsis provided by an embodiment of the present invention. As shown in Fig. 1, an early warning device for sepsis includes:
  • the training module 10 is used to train a prediction model using sample index data and deviation labels with sepsis labels; wherein, each sample index data includes multiple types of disease data; the deviation labels are disease data and standard values The label corresponding to the deviation direction;
  • the calculation module 20 is used to calculate the correlation value of each type of disease data with sepsis, and calculate the time threshold corresponding to each type of disease data;
  • the obtaining module 30 is used to obtain current index data and historical index data of the target user;
  • the interpolation module 40 is used to determine the type of disease data corresponding to the missing value in the current index data, and select the target disease data that meets the time threshold requirement from the historical index data according to the principle of proximity as the missing value to obtain the updated current index data ;
  • the prediction module 50 is used to input the respective deviation labels of the updated current index data and the current disease data in the updated current index data into the prediction model to obtain the prediction result;
  • the display 60 is used to display the prediction result.
  • the training module 10 needs to obtain sample index data, and then use the sample index data and deviation labels to train a prediction model, so that the prediction model can be subsequently used to predict sepsis for the target user based on the current index data of the target user.
  • sample index data refers to index data with sepsis labels, and each sample index data includes multiple types of disease data.
  • Disease data includes inspection data and inspection data. Inspection data is data obtained through laboratory experiments, such as the number of red blood cells, white blood cells, etc.; inspection data is data obtained directly through observation or inspection equipment, such as body temperature, heartbeat frequency, etc.
  • One sample index data includes multiple disease data, and this embodiment does not limit the data type of the disease data specifically included in the sample index data. It is understandable that in actual operation, the more types of disease data associated with sepsis in the sample index data, the more quickly and accurately the early warning can be carried out; and the larger the amount of sample index data, the more training The forecasting model will also be more accurate.
  • the deviation label is the label corresponding to the deviation direction of the disease data from the standard value.
  • the disease data is greater than the standard value, or the disease data is less than the standard value, or the disease data is equal to the standard value.
  • the sample index data includes multiple types of disease data, it is necessary to use the calculation module 20 to calculate the correlation value of each type of disease data with sepsis, and the correlation value represents each type of disease data. The effect on whether you suffer from sepsis, and calculate the time threshold corresponding to each type of disease data.
  • the time threshold refers to the effective duration of the disease data, which is used to restrict the retrospective data from the historical indicator data as a missing The time when the value of the disease data.
  • the process of calculating the correlation value between each type of disease data and sepsis includes two steps:
  • the first step Determine the traceability time window for each type of disease data.
  • the second step determine the timeliness of each type of disease data.
  • the retrospective time window is used to slide on the time axis of the sample index data to calculate the average value of each disease data in each retrospective time window, and the calculated average Correlation analysis is performed between the value and the target average value of the target time window when the target user is determined to be sepsis. Specifically, whether the retrospective time window is correlated with the target time window is determined according to whether the calculated correlation value is greater than 0.6. The window before the first occurrence is less than 0.6 is regarded as the longest window of the final retrospective. If all are less than 0.6, the first time window is defaulted as the retrospective window.
  • the acquisition module 30 is used to obtain the current index data and historical index data of the target user; it is understandable that the current index data refers to the index data obtained at the current time, and the historical index Data refers to the indicator data obtained before the current time.
  • the interpolation module 40 is used to first determine the type of disease data corresponding to the missing value in the current index data, and then select the target disease data that meets the time threshold requirement from the historical index data as the missing value according to the principle of proximity to obtain the updated current index Data; that is to say, the updated current indicator data refers to the indicator data after the current indicator data with missing values is interpolated based on the historical indicator data.
  • T time the time threshold corresponding to disease data of type A is t1
  • T time the benchmark to search from historical disease data of the same type
  • the target disease data is determined from the historical index data as the missing value of this type of disease data at time T; the disease data beyond t1 time is invalid.
  • the same type of disease data cannot be regarded as missing values.
  • the prediction module 50 is configured to input the respective deviation labels of the current disease data in the updated current index data and the updated current index data into the prediction model to obtain a prediction result.
  • the display 60 is used to display the prediction result obtained by the prediction module 50, so that the medical staff can intuitively obtain the prediction result through the display. Specifically, by displaying the prediction result on the display, the prediction result can be highlighted as high-risk and suspected patients. In addition, abnormal disease data can be further highlighted; the target user's personal information, laboratory examination information, medical record information, nursing records, comparison of updated current index data and standard values can also be further displayed. It should be noted that setting a display method in advance to display the corresponding prediction result using the display is a technical content known to those skilled in the art, and this embodiment does not limit this.
  • the embodiment of the present invention provides an early warning device for sepsis.
  • the device calculates the correlation value of each type of disease data with sepsis through a calculation module, and calculates the correlation value of each type of disease data.
  • the type of disease data corresponds to the time threshold; then after the current index data and historical index data of the target user are acquired through the acquisition module, the interpolation module determines the type of disease data corresponding to the missing value in the current index data, according to the nearest
  • the principle is to select the target disease data that meets the time threshold requirement from the historical index data as the missing value, and obtain the updated current index data; then, the updated current index data and the updated current index data are respectively corresponding to the current disease data.
  • the deviation direction is input into the prediction model, and the prediction result is obtained. Therefore, the device can not only supplement the missing values of the current index data, but also select the target disease data to meet the time threshold requirement according to the principle of proximity, which can be closer to the current actual situation of the target user, and can more accurately perform sepsis on the target user. Early warning detection of disease.
  • the device uses the updated current index data and the corresponding deviation direction of each current disease data in the updated current index data to predict. By combining multiple forms of data for prediction, the accuracy of the prediction result can be further improved.
  • the training module specifically includes:
  • the obtaining sub-module is used to obtain the sample index data with sepsis label
  • the interpolation sub-module is used to interpolate the missing values in the sample index data according to preset rules
  • the calculation sub-module is used to calculate the difference between each disease data and the corresponding standard value, and determine the corresponding deviation label according to the difference;
  • the input sub-module is used to select the corresponding type of machine learning model according to the number of disease data in the sample index data, and input the sample index data and each corresponding deviation label into the machine learning model;
  • the training sub-module is used to train the machine learning model using sample index data and deviation labels to obtain a prediction model.
  • the calculation sub-module calculates the difference between each disease data and the corresponding standard value, and then determines the difference according to the calculated difference, including the difference is positive, the difference is negative, or the difference is 0.
  • the corresponding deviation label does not limit the specific types of deviation labels. For example, deviation labels of "0", “1” and “2” can be used to indicate three different deviation directions; or "+” and "-” can be used. The deviation labels with "0” respectively indicate three different deviation directions and so on.
  • machine learning models such as vector machine SVM, logistic regression, XGboost, etc.
  • the training dimensions of different machine learning models are different, that is, the number of different machine learning models according to the type of disease data
  • the corresponding training method is different, and the prediction model obtained by the corresponding training is also different. Therefore, in actual operation, it is necessary to use the input submodule to select the corresponding type of machine according to the number of disease data in the sample index data Learn the model, and input the sample index data and the corresponding deviation label into the selected machine learning model, and then train the input sample index data and deviation label through the training sub-module to obtain a prediction model.
  • the input sub-module selects the corresponding machine learning model according to the number of disease data types in the sample index data, and then uses the corresponding machine learning model to train the sample index data and the deviation label, so that the trained prediction The model is more accurate.
  • the acquisition submodule specifically includes:
  • the first collection unit is used to extract structured disease data in the LIS library and the nursing system;
  • the second collection unit is used to collect unstructured disease data in the case book by using natural language processing technology
  • the setting unit is used to obtain the sample index data with sepsis label set according to the structured disease data, the unstructured disease data and the diagnosis result of the medical record book.
  • the sample index data includes structured disease data and unstructured disease data.
  • the LIS library (LIS, Laboratory Information Management System) is extracted through the first collection unit, which is a set of experiments designed specifically for the hospital laboratory.
  • Room Information Management System) and Nursing System is the use of information technology, computer technology and network communication technology to collect, store, process, transmit, and query nursing management and business technical information to improve the quality of nursing management information.
  • the system is an important sub-system of the hospital information system) structured disease data, and uses natural language processing technology to collect unstructured disease data in the medical record through the second collection unit; then through the setting unit according to the structured disease data And unstructured disease data and the diagnosis result of the medical record book to obtain the sample index data with sepsis label, that is, according to the diagnosis result of the medical record book (whether you have sepsis), it includes structured disease data and The disease data of the unstructured disease data sets the corresponding label as the sample index data.
  • the acquisition sub-module provided in this embodiment is obtained by extracting structured disease data in the LIS library and nursing system through the first acquisition unit and the second acquisition unit using natural language processing technology to acquire unstructured disease data in the medical record.
  • the sample index data is obtained, so the data sources of the sample index data are more extensive.
  • this embodiment further includes:
  • the cleaning unit is used to perform data cleaning on the index data of each sample with a sepsis label.
  • the cleaning unit is further used to perform data cleaning on each sample index data.
  • data cleaning refers to discovering and correcting identifiable errors in the sample index data, including checking data consistency, processing invalid values, etc., and regularizing the sample index data according to the corresponding machine learning model, so that the machine learning model can be used to correct
  • the sample index data is used for training to improve the convenience and accuracy of training.
  • this embodiment further illustrates and optimizes the technical solution. Specifically, this embodiment further includes:
  • the receiving module is used to receive the first index data marked with sepsis labels by professionals;
  • the setting module is used to set the corresponding first deviation label according to the deviation direction of each disease data in the first indicator data from the standard value;
  • the update module is used to input the first indicator data and the corresponding first deviation label into the prediction model for training, so as to update the prediction model.
  • it further includes a receiving module for receiving the first index data marked with a sepsis label by a professional, and the first index data is also the index data marked with a sepsis label by a professional;
  • the setting module obtains the corresponding first deviation label according to the deviation direction of each first disease data in the first indicator data from the standard value, and then inputs the first indicator data into the prediction model for training through the update module to update Forecast model.
  • the first index data is more accurate index data with sepsis labels than the sample index data; therefore, through the update module Inputting the first indicator data and the first deviation label into the prediction model for training can continuously update the prediction model, making the prediction of the prediction model more accurate.
  • the interpolation module specifically includes:
  • the sorting sub-module is used to arrange the historical indicator data in a preset time sequence
  • the determination sub-module is used to determine the type of disease data corresponding to the missing value in the current indicator data
  • the selection sub-module is used to select the target disease data that meets the time threshold requirement from the arranged historical index data as the missing value according to the principle of proximity, and obtain the updated current index data.
  • the interpolation module specifically includes a sorting sub-module for arranging the historical index data in a preset time sequence, that is, the historical index data is arranged in chronological order, from recent time to long-term time or from Arrange the long-term time to the recent time to get the time axis; then determine the type of disease data corresponding to the missing value in the current indicator data through the determining sub-module; then select from the arranged historical indicator data according to the principle of proximity through the selection sub-module The target disease data that meets the time threshold requirement is obtained, and the target disease data is taken as a missing value to obtain updated current index data.
  • a sorting sub-module for arranging the historical index data in a preset time sequence, that is, the historical index data is arranged in chronological order, from recent time to long-term time or from Arrange the long-term time to the recent time to get the time axis; then determine the type of disease data corresponding to the missing value in the current indicator data through the determining sub-module; then select from the
  • the historical index data is sorted in chronological order through the sorting submodule, which can make it easier for the selection submodule to select the target disease data.
  • this embodiment further illustrates and optimizes the technical solution. Specifically, this embodiment further includes:
  • the alerter is used to send out corresponding prompt information when the predicted result is that the target user is ill.
  • this embodiment further includes a warning device.
  • the warning device sends out corresponding prompt information to remind the medical staff of the target. The abnormal situation of the user.
  • the warning device in this embodiment may be specifically a buzzer or an indicator light, and the sound frequency of the buzzer or the light emission frequency of the indicator light is used to set the corresponding prompt information to achieve the prompt effect. It is understandable that the buzzer and indicator light, as commonly used prompting devices, can not only achieve the prompting effect simply and intuitively, but also are cheap.
  • the prompt can be made more timely and intuitively.
  • the embodiment of the early warning device for sepsis provided by the present invention has been described in detail above.
  • the present invention also provides a early warning device for sepsis and a computer-readable storage medium corresponding to the device. Since the embodiments of the device and computer-readable storage medium part correspond to the embodiments of the device part, please refer to the description of the embodiment of the device part for the embodiments of the device and computer-readable storage medium part, which will not be repeated here.
  • Fig. 2 is a structural diagram of an early warning device for sepsis provided by an embodiment of the present invention. As shown in Fig. 2, an early warning device for sepsis includes:
  • the memory 21 is used to store computer programs
  • each sample index data includes multiple types of disease data;
  • the deviation label is the label corresponding to the deviation direction of the disease data from the standard value ;
  • the forecast result is displayed.
  • the early warning device for sepsis provided by the embodiment of the present invention has the beneficial effects of the foregoing early warning device for sepsis.
  • the present invention also provides a computer-readable storage medium with a computer program stored on the computer-readable storage medium, which is characterized in that, when the computer program is executed by a processor, the following steps are implemented:
  • each sample index data includes multiple types of disease data;
  • the deviation label is the label corresponding to the deviation direction of the disease data from the standard value ;
  • the forecast result is displayed.
  • the computer-readable storage medium provided by the embodiment of the present invention has the beneficial effects of the above-mentioned early warning device for sepsis.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Data Mining & Analysis (AREA)
  • Biomedical Technology (AREA)
  • Primary Health Care (AREA)
  • Pathology (AREA)
  • Epidemiology (AREA)
  • General Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

一种脓毒血症的预警装置、设备及介质，包括：训练模块，用于利用设置有脓毒血症标签的样本指标数据和偏差标签训练出预测模型；计算模块，用于计算各类型的疾病数据分别与脓毒血症的相关性值，计算出各类型的疾病数据分别对应的时间阈值；获取模块，用于获取目标用户的当前指标数据和历史指标数据；插值模块，用于确定出当前指标数据中的缺失值对应的疾病数据的类型，按照就近原则从历史指标数据中选择出满足时间阈值要求的目标疾病数据作为缺失值，得到更新的当前指标数据；预测模块，用于将更新的当前指标数据和更新的当前指标数据中各当前疾病数据的分别对应的偏差标签输入至预测模型中，得出预测结果；显示器，用于显示预测结果。

Description

一种脓毒血症的预警装置、设备及存储介质
本申请要求于2019年11月27日提交中国专利局、申请号为201911184380.8、发明名称为“一种脓毒血症的预警装置、设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及医学设备领域，特别涉及一种脓毒血症的预警装置、设备及计算机可读存储介质。
背景技术
随着科技的发展，人工智能技术已经上升至国家战略发展层面，医疗辅助诊断是人工智能的一个重要的应用和研究方向。目前，为了更便捷地辅助诊断用户是否患有脓毒血症，现有技术提供了一种脓毒血症的预警装置。通过将目标用户的当前指标数据输入至预先训练出的预测模型中，便能得出对应的预测结果。但是，在实际操作中，目标用户的当前指标数据往往存在缺失的情况，而在这种情况下，若直接将存在缺失的当前指标数据输入至预测模型，将无法预测或者使得预测结果不准确。因此，现有技术通过计算出历史指标数据中与缺失值对应的疾病数据的平均值/最大值/最小值，并据此补充缺失值，再进行预警检测。而实际上，正是因为目标用户的指标数据发生了异常变化，才表示用户可能出现患病的情况。因此，按照现有技术中补充当前指标数据中的缺失值的方式，将使得预测结果不准确。
因此，如何提高脓毒血症预警装置的准确度，是本领域技术人员目前需要解决的技术问题。
发明内容
有鉴于此，本发明的目的在于提供一种脓毒血症的预警装置，能够提高脓毒血症预警装置预测的准确度；本发明的另一目的是提供一种脓毒血症的预警装置设备及计算机可读存储介质，均具有上述有益效果。
为解决上述技术问题，本发明提供一种脓毒血症的预警装置，包括：
训练模块，用于利用设置有脓毒血症标签的样本指标数据和偏差标签训练出预测模型；其中，每一个所述样本指标数据中包括多个类型的疾病数据；所述偏差标签为所述疾病数据与标准值的偏离方向对应的标签；
计算模块，用于计算各类型的所述疾病数据分别与脓毒血症的相关性值，并计算出各类型的所述疾病数据分别对应的时间阈值；
获取模块，用于获取目标用户的当前指标数据和历史指标数据；
插值模块，用于确定出所述当前指标数据中的缺失值对应的疾病数据的类型，按照就近原则从所述历史指标数据中选择出满足所述时间阈值要求的目标疾病数据作为所述缺失值，得到更新的当前指标数据；
预测模块，用于将所述更新的当前指标数据和所述更新的当前指标数据中各所述当前疾病数据的分别对应的偏差标签输入至所述预测模型中，得出预测结果；
显示器，用于显示所述预测结果。
优选地，所述训练模块具体包括：
获取子模块，用于获取设置有所述脓毒血症标签的所述样本指标数据；
插值子模块，用于按照预设规则对所述样本指标数据中的缺失值进行插补；
计算子模块，用于计算各所述疾病数据分别与对应的标准值的差值，并根据所述差值确定出对应的偏差标签；
输入子模块，用于根据所述样本指标数据中所述疾病数据的数量选择对应类型的机器学习模型，并将所述样本指标数据和各对应的偏离标签输入至所述机器学习模型中；
训练子模块，用于利用所述样本指标数据和所述偏离标签对所述机器学习模型进行训练，得到所述预测模型。
优选地，所述获取子模块具体包括：
第一采集单元，用于抽取LIS库和护理系统中的结构化疾病数据；
第二采集单元，用于利用自然语言处理技术采集所述病例本中的非结构化的疾病数据；
设置单元，用于根据所述结构化疾病数据和所述非结构化疾病数据以及病历本的诊断结果得出所述设置有所述脓毒血症标签的所述样本指标数据。
优选地，进一步包括：
清洗单元，用于对各所述设置有所述脓毒血症标签的所述样本指标数据进行数据清洗。
优选地，进一步包括：
接收模块，用于接收由专业人员标注所述脓毒血症标签的第一指标数据；
更新模块，用于将所述第一指标数据和对应的第一偏差标签输入至所述预测模型中进行训练，以更新所述预测模型。
优选地，所述插值模块具体包括：
排序子模块，用于将各所述历史指标数据按照预设时间顺序排列；
确定子模块，用于确定出所述当前指标数据中的缺失值对应的疾病数据的类型；
选择子模块，用于按照就近原则从排列的所述历史指标数据中选择出满足所述时间阈值要求的目标疾病数据作为所述缺失值，得到更新的当前指标数据。
优选地，进一步包括：
警示器，用于在所述预测结果为所述目标用户患病时，发出对应的提示信息。
为解决上述技术问题，本发明还提供一种脓毒血症的预警设备，包括：
存储器，用于存储计算机程序；
处理器，用于执行所述计算机程序时，实现如下步骤：
利用设置有脓毒血症标签的样本指标数据和偏差标签训练出预测模型；其中，每一个所述样本指标数据中包括多个类型的疾病数据；所述偏差标签为所述疾病数据与标准值的偏离方向对应的标签；
计算各类型的所述疾病数据分别与脓毒血症的相关性值，并计算出各类型的所述疾病数据分别对应的时间阈值；
获取目标用户的当前指标数据和历史指标数据；
确定出所述当前指标数据中的缺失值对应的疾病数据的类型，按照就近原则从所述历史指标数据中选择出满足所述时间阈值要求的目标疾病数据作为所述缺失值，得到更新的当前指标数据；
将所述更新的当前指标数据和所述更新的当前指标数据中各所述当前疾病数据的分别对应的偏差标签输入至所述预测模型中，得出预测结果；
显示所述预测结果。
为解决上述技术问题，本发明还提供一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机程序，所述计算机程序被处理器执行时，实现如下步骤：
利用设置有脓毒血症标签的样本指标数据和偏差标签训练出预测模型；其中，每一个所述样本指标数据中包括多个类型的疾病数据；所述偏差标签为所述疾病数据与标准值的偏离方向对应的标签；
计算各类型的所述疾病数据分别与脓毒血症的相关性值，并计算出各类型的所述疾病数据分别对应的时间阈值；
获取目标用户的当前指标数据和历史指标数据；
确定出所述当前指标数据中的缺失值对应的疾病数据的类型，按照就近原则从所述历史指标数据中选择出满足所述时间阈值要求的目标疾病数据作为所述缺失值，得到更新的当前指标数据；
将所述更新的当前指标数据和所述更新的当前指标数据中各所述当前疾病数据的分别对应的偏差标签输入至所述预测模型中，得出预测结果；
显示所述预测结果。
本发明提供的一种脓毒血症的预警装置，相较于现有技术，本装置是通过计算模块计算各类型的疾病数据分别与脓毒血症的相关性值，并计算出各类型的疾病数据分别对应的时间阈值；然后通过插值模块确定出当前指标数据中的缺失值对应的疾病数据的类型，按照就近原则从历史指标数据中选择出满足时间阈值要求的目标疾病数据作为缺失值，得到更新的当前指标数据；再通过将更新的当前指标数据和更新的当前指标数据中各当前疾病数据的分别对应的偏离方向输入至预测模型中，得出预测结果。因 此，本装置不仅能够补充当前指标数据的缺失值，而且按照就近原则选择的满足时间阈值要求目标疾病数据，能够更接近目标用户当前的实际情况，从而能够更准确地对目标用户进行脓毒血症的预警检测。另外，本装置是利用更新的当前指标数据和更新的当前指标数据中各当前疾病数据的分别对应的偏离方向进行预测，通过综合多种形式的数据进行预测，能够进一步提高预测结果的准确度。
为解决上述技术问题，本发明还提供了一种脓毒血症的预警设备及计算机可读存储介质，均具有上述有益效果。
附图说明
为了更清楚地说明本发明实施例或现有技术的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单的介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据提供的附图获得其他的附图。
图1为本发明实施例提供的一种脓毒血症的预警装置的流程图；
图2为本发明实施例提供的一种脓毒血症的预警设备的结构图。
具体实施方式
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
本发明实施例的核心是提供一种脓毒血症的预警装置，能够提高脓毒血症预警装置预测的准确度；本发明的另一核心是提供一种脓毒血症的预警设备及计算机可读存储介质，均具有上述有益效果。
为了使本领域技术人员更好地理解本发明方案，下面结合附图和具体实施方式对本发明作进一步的详细说明。
图1为本发明实施例提供的一种脓毒血症的预警装置的结构图，如图1所示，一种脓毒血症的预警装置包括：
训练模块10，用于利用设置有脓毒血症标签的样本指标数据和偏差标签训练出预测模型；其中，每一个样本指标数据中包括多个类型的疾病数据；偏差标签为疾病数据与标准值的偏离方向对应的标签；
计算模块20，用于计算各类型的疾病数据分别与脓毒血症的相关性值，并计算出各类型的疾病数据分别对应的时间阈值；
获取模块30，用于获取目标用户的当前指标数据和历史指标数据；
插值模块40，用于确定出当前指标数据中的缺失值对应的疾病数据的类型，按照就近原则从历史指标数据中选择出满足时间阈值要求的目标疾病数据作为缺失值，得到更新的当前指标数据；
预测模块50，用于将更新的当前指标数据和更新的当前指标数据中各当前疾病数据的分别对应的偏差标签输入至预测模型中，得出预测结果；
显示器60，用于显示预测结果。
首先，训练模块10需要获取样本指标数据，然后利用样本指标数据和偏差标签训练出预测模型，以便后续利用预测模型根据目标用户的当前指标数据对目标用户进行脓毒血症的预测。
需要说明的是，样本指标数据指的是设置有脓毒血症标签的指标数据，并且每一个样本指标数据中包括多个类型的疾病数据。疾病数据包括检验数据和检查数据，检验数据为通过实验室实验得出的数据，如红细胞数量、白细胞数量等；检查数据为直接通过观察或者检查仪器得出的数据，如体温、心跳频率等。一个样本指标数据中包括多个疾病数据，本实施例对样本指标数据中具体包括的疾病数据的数据类型不做限定。可以理解的是，在实际操作中，样本指标数据中与脓毒血症相关联的疾病数据的类型越多，越能够快速准确地进行预警；并且，样本指标数据的数据量越大，训练出的预测模型也将更加准确。
需要说明的是，偏差标签为疾病数据与标准值的偏离方向对应的标签，例如，疾病数据大于标准值，或者疾病数据小于标准值，或者疾病数据等 于标准值，这三种情况分别对应不同的偏离方向：正偏、反偏或者没有偏离，将不同的偏离方向分别设置为对应的偏离标签。
需要说明的是，由于样本指标数据中包括多种类型的疾病数据，因此需要利用计算模块20计算各类型的疾病数据分别与脓毒血症的相关性值，相关性值表示各类型的疾病数据对是否患脓毒血症的影响效果，并计算出各类型的疾病数据分别对应的时间阈值，时间阈值指的是疾病数据对应的有效时长，用于约束从历史指标数据中追溯得出作为缺失值的疾病数据时的时间。
具体的，计算各类型的疾病数据分别与脓毒血症的相关性值的过程，包括两个步骤：
第一步：确定各类型的疾病数据的追溯时间窗。
具体的，首先获取每个样本指标数据中，在预设时间段内各不同类型的疾病数据出现的第一时间间隔，将第一时间间隔的70％作为第二时间间隔；
然后计算各样本指标数据中，同一类型的疾病数据的第二时间间隔的平均值，得到追溯时间窗，或者在计算出平均值，得到第三时间间隔后，再将第三时间间隔的70％设置为追溯时间窗。
第二步：确定各类型的疾病数据的时效性。
根据各类型的疾病数据分别对应的追溯时间窗，利用追溯时间窗，在样本指标数据的时间轴上滑动，计算每个追溯时间窗中的各疾病数据的平均值，并将各计算出的平均值与目标用户确定为脓毒血症时的目标时间窗的目标平均值进行相关性分析，具体是根据计算出的相关性值是否大于0.6来确定追溯时间窗与目标时间窗之间是否相关。在第一次出现小于0.6之前的窗口作为最终追溯的最长窗口，若都小于0.6默认第一个时间窗作为追溯的窗口。
在对目标用户进行脓毒血症的预警时，利用获取模块30获取目标用户的当前指标数据和历史指标数据；可以理解的是，当前指标数据指的是当前时间获取到的指标数据，历史指标数据指的是在当前时间之前获取到的指标数据。
插值模块40用于先确定出当前指标数据中的缺失值对应的疾病数据的类型，然后按照就近原则从历史指标数据中选择出满足时间阈值要求的目标疾病数据作为缺失值，得到更新的当前指标数据；也就是说，更新的当前指标数据指的是依据历史指标数据对存有缺失值的当前指标数据进行插值处理之后的指标数据。
具体的，例如，假设类型为A的疾病数据对应的时间阈值为t1，那么要获取该类疾病数据在T时间的缺失值，则以T时间为基准，从历史的同类型的疾病数据中查找，只有在t1时间内的疾病数据才是有效数据，因此按照就近原则从历史指标数据中确定出目标疾病数据，作为该类疾病数据在T时间的缺失值；超过t1时间的疾病数据已失效，即使是同类型的疾病数据，也不能作为缺失值。
预测模块50用于将更新的当前指标数据和更新的当前指标数据中各当前疾病数据的分别对应的偏差标签输入至预测模型中，得出预测结果。
显示器60用于显示通过预测模块50得出的预测结果，以便医护人员能够通过显示器直观地获取到预测结果。具体的，通过显示器显示预测结果的方式，可以将预测结果为高危和疑似患者进行高亮显示。另外，可以进一步将异常的疾病数据进行高亮显示；还可以进一步显示目标用户的个人信息、实验室检查信息、病历信息、护理记录、更新的当前指标数据与标准值的对比情况等。需要说明的是，预先设置显示方法，以利用显示器显示对应的预测结果，是本领域技术人员所公知的技术内容，本实施例对此不做限定。
本发明实施例提供的一种脓毒血症的预警装置，相较于现有技术，本装置是通过计算模块计算各类型的疾病数据分别与脓毒血症的相关性值，并计算出各类型的疾病数据分别对应的时间阈值；然后在通过获取模块获取到目标用户的当前指标数据和历史指标数据之后，通过插值模块确定出当前指标数据中的缺失值对应的疾病数据的类型，按照就近原则从历史指标数据中选择出满足时间阈值要求的目标疾病数据作为缺失值，得到更新的当前指标数据；再通过将更新的当前指标数据和更新的当前指标数据中各当前疾病数据的分别对应的偏离方向输入至预测模型中，得出预测结果。 因此，本装置不仅能够补充当前指标数据的缺失值，而且按照就近原则选择的满足时间阈值要求目标疾病数据，能够更接近目标用户当前的实际情况，从而能够更准确地对目标用户进行脓毒血症的预警检测。另外，本装置是利用更新的当前指标数据和更新的当前指标数据中各当前疾病数据的分别对应的偏离方向进行预测，通过综合多种形式的数据进行预测，能够进一步提高预测结果的准确度。
在上述实施例的基础上，本实施例对技术方案作了进一步的说明和优化，具体的，本实施例中，训练模块具体包括：
获取子模块，用于获取设置有脓毒血症标签的样本指标数据；
插值子模块，用于按照预设规则对样本指标数据中的缺失值进行插补；
计算子模块，用于计算各疾病数据分别与对应的标准值的差值，并根据差值确定出对应的偏差标签；
输入子模块，用于根据样本指标数据中疾病数据的数量选择对应类型的机器学习模型，并将样本指标数据和各对应的偏离标签输入至机器学习模型中；
训练子模块，用于利用样本指标数据和偏离标签对机器学习模型进行训练，得到预测模型。
需要说明的是，训练模块训练得出预测模型的过程中，首先需要利用获取子模块获取设置有脓毒血症标签的样本指标数据，然后利用插值子模块按照预设规则对样本指标数据中的缺失值进行插补，需要说明的是，插值子模块对样本指标数据进行插值的过程，与对目标用户的当前指标数据进行插值的操作方法一致，因此可以参考上一实施例的具体描述，此处不做赘述。
具体的，通过计算子模块计算各疾病数据分别与对应的标准值的差值，然后根据计算出的差值，包括差值为正，差值为负或差值为0这三种情况确定出对应的偏差标签。本实施例对偏差标签的具体类型不做限定，例如，可以用“0”、“1”和“2”的偏差标签分别表示三种不同的偏差方向；或者可以用“+”、“-”和“0”的偏差标签分别表示三种不同的偏差方向等。
可以理解的是，机器学习模型多种多样，如向量机SVM、逻辑斯蒂回归、XGboost等，不同的机器学习模型训练的维度是不同的，即不同的机器学习模型根据疾病数据的类型的数量的不同，对应的训练的方式是不同的，对应训练得出的预测模型也是不同的，因此，在实际操作中，需要利用输入子模块根据样本指标数据中的疾病数据的数量选择对应类型的机器学习模型，并将样本指标数据和对应的偏离标签输入至选择的机器学习模型中，再通过训练子模块根据输入的样本指标数据和偏离标签进行训练，得出预测模型。
可见，本实施例通过输入子模块根据样本指标数据中的疾病数据的类型的数量选择对应的机器学习模型，再利用对应的机器学习模型对样本指标数据和偏离标签进行训练，使得训练出的预测模型更加准确。
在上述实施例的基础上，本实施例对技术方案作了进一步的说明和优化，具体的，本实施例中，获取子模块具体包括：
第一采集单元，用于抽取LIS库和护理系统中的结构化疾病数据；
第二采集单元，用于利用自然语言处理技术采集病例本中的非结构化的疾病数据；
设置单元，用于根据结构化疾病数据和非结构化疾病数据以及病历本的诊断结果得出设置有脓毒血症标签的样本指标数据。
需要说明的是，样本指标数据包括结构化疾病数据和非结构化疾病数据，本实施例通过第一采集单元抽取LIS库(LIS，Laboratory Information Management System，是专为医院检验科设计的一套实验室信息管理系统)和护理系统(护理信息系统是利用信息技术、计算机技术和网络通信技术对护理管理和业务技术信息进行采集、存储、处理、传输、查询，以提高护理管理质量为目的的信息系统，是医院信息系统的一个重要子系统)中的结构化疾病数据，并通过第二采集单元利用自然语言处理技术采集病历本中的非结构化疾病数据；然后通过设置单元根据结构化疾病数据和非结构化疾病数据以及病历本的诊断结果得出设置有脓毒血症标签的样本指标数据，即，根据病历本的诊断结果(是否患有脓毒血症)为包括结构化疾 病数据和非结构化疾病数据的疾病数据设置对应的标签，以作为样本指标数据。
因此，本实施例提供的获取子模块，是通过第一采集单元抽取LIS库和护理系统中的结构化疾病数据和第二采集单元利用自然语言处理技术采集病历本中的非结构化疾病数据得出样本指标数据，因此样本指标数据的数据来源更加广泛。
作为优选的实施方式，本实施例进一步包括：
清洗单元，用于对各设置有脓毒血症标签的样本指标数据进行数据清洗。
也就是说，在得出设置有脓毒血症标签的样本指标数据之后，进一步利用清洗单元对各样本指标数据进行数据清洗。具体的，数据清洗指的是发现并纠正样本指标数据中可识别的错误，包括检查数据一致性，处理无效值等，将样本指标数据按照对应的机器学习模型进行规整，便于利用机器学习模型对样本指标数据进行训练，提高训练的便捷度和准确度。
在上述实施例的基础上，本实施例对技术方案作了进一步的说明和优化，具体的，本实施例进一步包括：
接收模块，用于接收由专业人员标注脓毒血症标签的第一指标数据；
设置模块，用于根据第一指标数据中的各疾病数据与标准值的偏差方向分别设置对应的第一偏差标签；
更新模块，用于将第一指标数据和对应的第一偏差标签输入至预测模型中进行训练，以更新预测模型。
在本实施例中，进一步包括用于接收由专业人员标注脓毒血症标签的第一指标数据的接收模块，第一指标数据也即由专业人员标注脓毒血症标签的指标数据；然后通过设置模块根据第一指标数据中的各第一疾病数据分别与标准值的偏差方向，得出对应的第一偏差标签，再通过更新模块将第一指标数据输入至预测模型中进行训练，以更新预测模型。
可以理解的是，由于第一指标数据的标签是由专业人员标注的，因此 第一指标数据是相较于样本指标数据更为准确的带有脓毒血症标签的指标数据；因此通过更新模块将第一指标数据和第一偏差标签输入至预测模型中进行训练，能够不断地更新预测模型，使得预测模型的预测更加准确。
在上述实施例的基础上，本实施例对技术方案作了进一步的说明和优化，具体的，本实施例中，插值模块具体包括：
排序子模块，用于将各历史指标数据按照预设时间顺序排列；
确定子模块，用于确定出当前指标数据中的缺失值对应的疾病数据的类型；
选择子模块，用于按照就近原则从排列的历史指标数据中选择出满足时间阈值要求的目标疾病数据作为缺失值，得到更新的当前指标数据。
具体的，本实施例中，插值模块具体包括用于将各历史指标数据按照预设时间顺序排列的排序子模块，即，将历史指标数据按照时间先后顺序，由近期时间至远期时间或者由远期时间至近期时间排列，得出时间轴；然后通过确定子模块确定出当前指标数据中的缺失值对应的疾病数据的类型；再通过选择子模块按照就近原则从排列的历史指标数据中选择出满足时间阈值要求的目标疾病数据，并将该目标疾病数据作为缺失值，得到更新的当前指标数据。
本实施例通过排序子模块将历史指标数据按照时间顺序进行排序，能够更便于选择子模块选择出目标疾病数据。
在上述实施例的基础上，本实施例对技术方案作了进一步的说明和优化，具体的，本实施例进一步包括：
警示器，用于在预测结果为目标用户患病时，发出对应的提示信息。
具体的，本实施例进一步包括警示器，当预测模块对目标用户的当前指标数据进行预测得出的预测结果为目标用户患病时，则通过警示器发出对应的提示信息，以提示医护人员目标用户的异常情况。
需要说明的是，本实施例中的警示器可以具体为蜂鸣器或者指示灯，利用蜂鸣器的发声频率或指示灯的发光频率设置对应的提示信息，以起到 提示效果。可以理解的是，蜂鸣器和指示灯作为常用的提示装置，不仅能够简单直观地达到提示的效果，而且价格便宜。
本实施例通过进一步利用警示器在预测结果为目标用户患病时，发出对应的提示信息，能够更加及时直观地进行提示。
上文对于本发明提供的一种脓毒血症的预警装置的实施例进行了详细的描述，本发明还提供了一种与该装置对应的脓毒血症的预警设备及计算机可读存储介质，由于设备及计算机可读存储介质部分的实施例与装置部分的实施例相互照应，因此设备及计算机可读存储介质部分的实施例请参见装置部分的实施例的描述，这里暂不赘述。
图2为本发明实施例提供的一种脓毒血症的预警设备的结构图，如图2所示，一种脓毒血症的预警设备包括：
存储器21，用于存储计算机程序；
处理器22，用于执行存储器中存储的计算机程序时，实现如下步骤：
利用设置有脓毒血症标签的样本指标数据和偏差标签训练出预测模型；其中，每一个样本指标数据中包括多个类型的疾病数据；偏差标签为疾病数据与标准值的偏离方向对应的标签；
计算各类型的疾病数据分别与脓毒血症的相关性值，并计算出各类型的疾病数据分别对应的时间阈值；
获取目标用户的当前指标数据和历史指标数据；
确定出当前指标数据中的缺失值对应的疾病数据的类型，按照就近原则从历史指标数据中选择出满足时间阈值要求的目标疾病数据作为缺失值，得到更新的当前指标数据；
将更新的当前指标数据和更新的当前指标数据中各当前疾病数据的分别对应的偏差标签输入至预测模型中，得出预测结果；
显示预测结果。
本发明实施例提供的脓毒血症的预警设备，具有上述脓毒血症的预警装置的有益效果。
为解决上述技术问题，本发明还提供一种计算机可读存储介质，计算机可读存储介质上存储有计算机程序，其特征在于，计算机程序被处理器执行时，实现如下步骤：
利用设置有脓毒血症标签的样本指标数据和偏差标签训练出预测模型；其中，每一个样本指标数据中包括多个类型的疾病数据；偏差标签为疾病数据与标准值的偏离方向对应的标签；
计算各类型的疾病数据分别与脓毒血症的相关性值，并计算出各类型的疾病数据分别对应的时间阈值；
获取目标用户的当前指标数据和历史指标数据；
确定出当前指标数据中的缺失值对应的疾病数据的类型，按照就近原则从历史指标数据中选择出满足时间阈值要求的目标疾病数据作为缺失值，得到更新的当前指标数据；
将更新的当前指标数据和更新的当前指标数据中各当前疾病数据的分别对应的偏差标签输入至预测模型中，得出预测结果；
显示预测结果。
本发明实施例提供的计算机可读存储介质，具有上述脓毒血症的预警装置的有益效果。
以上对本发明所提供的脓毒血症的预警装置、设备及计算机可读存储介质进行了详细介绍。本文中应用了具体实施例对本发明的原理及实施方式进行了阐述，以上实施例的说明只是用于帮助理解本发明的方法及其核心思想。应当指出，对于本技术领域的普通技术人员来说，在不脱离本发明原理的前提下，还可以对本发明进行若干改进和修饰，这些改进和修饰也落入本发明权利要求的保护范围内。
说明书中各个实施例采用递进的方式描述，每个实施例重点说明的都是与其他实施例的不同之处，各个实施例之间相同相似部分互相参见即可。对于实施例公开的装置而言，由于其与实施例公开的方法相对应，所以描述的比较简单，相关之处参见方法部分说明即可。
专业人员还可以进一步意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、计算机软件或者二者的结合来实现，为了清楚地说明硬件和软件的可互换性，在上述说明中已经按照功能一般性地描述了各示例的组成及步骤。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本发明的范围。

Claims (9)

  1. 一种脓毒血症的预警装置，其特征在于，包括：
    训练模块，用于利用设置有脓毒血症标签的样本指标数据和偏差标签训练出预测模型；其中，每一个所述样本指标数据中包括多个类型的疾病数据；所述偏差标签为所述疾病数据与标准值的偏离方向对应的标签；
    计算模块，用于计算各类型的所述疾病数据分别与脓毒血症的相关性值，并计算出各类型的所述疾病数据分别对应的时间阈值；
    获取模块，用于获取目标用户的当前指标数据和历史指标数据；
    插值模块，用于确定出所述当前指标数据中的缺失值对应的疾病数据的类型，按照就近原则从所述历史指标数据中选择出满足所述时间阈值要求的目标疾病数据作为所述缺失值，得到更新的当前指标数据；
    预测模块，用于将所述更新的当前指标数据和所述更新的当前指标数据中各所述当前疾病数据的分别对应的偏差标签输入至所述预测模型中，得出预测结果；
    显示器，用于显示所述预测结果。
  2. 根据权利要求1所述的装置，其特征在于，所述训练模块具体包括：
    获取子模块，用于获取设置有所述脓毒血症标签的所述样本指标数据；
    插值子模块，用于按照预设规则对所述样本指标数据中的缺失值进行插补；
    计算子模块，用于计算各所述疾病数据分别与对应的标准值的差值，并根据所述差值确定出对应的偏差标签；
    输入子模块，用于根据所述样本指标数据中所述疾病数据的数量选择对应类型的机器学习模型，并将所述样本指标数据和各对应的偏离标签输入至所述机器学习模型中；
    训练子模块，用于利用所述样本指标数据和所述偏离标签对所述机器学习模型进行训练，得到所述预测模型。
  3. 根据权利要求2所述的装置，其特征在于，所述获取子模块具体包括：
    第一采集单元，用于抽取LIS库和护理系统中的结构化疾病数据；
    第二采集单元，用于利用自然语言处理技术采集所述病例本中的非结构化的疾病数据；
    设置单元，用于根据所述结构化疾病数据和所述非结构化疾病数据以及病历本的诊断结果得出所述设置有所述脓毒血症标签的所述样本指标数据。
  4. 根据权利要求3所述的装置，其特征在于，进一步包括：
    清洗单元，用于对各所述设置有所述脓毒血症标签的所述样本指标数据进行数据清洗。
  5. 根据权利要求2所述的装置，其特征在于，进一步包括：
    接收模块，用于接收由专业人员标注所述脓毒血症标签的第一指标数据；
    更新模块，用于将所述第一指标数据和对应的第一偏差标签输入至所述预测模型中进行训练，以更新所述预测模型。
  6. 根据权利要求1所述的装置，其特征在于，所述插值模块具体包括：
    排序子模块，用于将各所述历史指标数据按照预设时间顺序排列；
    确定子模块，用于确定出所述当前指标数据中的缺失值对应的疾病数据的类型；
    选择子模块，用于按照就近原则从排列的所述历史指标数据中选择出满足所述时间阈值要求的目标疾病数据作为所述缺失值，得到更新的当前指标数据。
  7. 根据权利要求1至6任一项所述的装置，其特征在于，进一步包括：
    警示器，用于在所述预测结果为所述目标用户患病时，发出对应的提示信息。
  8. 一种脓毒血症的预警设备，其特征在于，包括：
    存储器，用于存储计算机程序；
    处理器，用于执行所述存储器中存储的计算机程序时，实现如下步骤：
    利用设置有脓毒血症标签的样本指标数据和偏差标签训练出预测模型；其中，每一个所述样本指标数据中包括多个类型的疾病数据；所述偏差标签为所述疾病数据与标准值的偏离方向对应的标签；
    计算各类型的所述疾病数据分别与脓毒血症的相关性值，并计算出各类型的所述疾病数据分别对应的时间阈值；
    获取目标用户的当前指标数据和历史指标数据；
    确定出所述当前指标数据中的缺失值对应的疾病数据的类型，按照就近原则从所述历史指标数据中选择出满足所述时间阈值要求的目标疾病数据作为所述缺失值，得到更新的当前指标数据；
    将所述更新的当前指标数据和所述更新的当前指标数据中各所述当前疾病数据的分别对应的偏差标签输入至所述预测模型中，得出预测结果；
    显示所述预测结果。
  9. 一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机程序，其特征在于，所述计算机程序被处理器执行时，实现如下步骤：
    利用设置有脓毒血症标签的样本指标数据和偏差标签训练出预测模型；其中，每一个所述样本指标数据中包括多个类型的疾病数据；所述偏差标签为所述疾病数据与标准值的偏离方向对应的标签；
    计算各类型的所述疾病数据分别与脓毒血症的相关性值，并计算出各类型的所述疾病数据分别对应的时间阈值；
    获取目标用户的当前指标数据和历史指标数据；
    确定出所述当前指标数据中的缺失值对应的疾病数据的类型，按照就近原则从所述历史指标数据中选择出满足所述时间阈值要求的目标疾病数据作为所述缺失值，得到更新的当前指标数据；
    将所述更新的当前指标数据和所述更新的当前指标数据中各所述当前疾病数据的分别对应的偏差标签输入至所述预测模型中，得出预测结果；
    显示所述预测结果。
PCT/CN2020/105398 2019-11-27 2020-07-29 一种脓毒血症的预警装置、设备及存储介质 WO2021103624A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911184380.8A CN110993096B (zh) 2019-11-27 2019-11-27 一种脓毒血症的预警装置、设备及存储介质
CN201911184380.8 2019-11-27

Publications (1)

Publication Number Publication Date
WO2021103624A1 true WO2021103624A1 (zh) 2021-06-03

Family

ID=70087537

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/105398 WO2021103624A1 (zh) 2019-11-27 2020-07-29 一种脓毒血症的预警装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN110993096B (zh)
WO (1) WO2021103624A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110993096B (zh) * 2019-11-27 2021-04-09 医惠科技有限公司 一种脓毒血症的预警装置、设备及存储介质
CN111939018A (zh) * 2020-08-19 2020-11-17 郑州铁路职业技术学院 一种内科病人用腿部理疗装置
CN113035352B (zh) * 2021-04-27 2022-06-21 河南科技大学 基于bp神经网络的糖尿病视网膜病变预警方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109902881A (zh) * 2019-03-19 2019-06-18 武汉乐易创想科技有限公司 基于多元统计分析和lstm融合的pm2.5浓度预测方法
CN110123274A (zh) * 2019-04-29 2019-08-16 上海电气集团股份有限公司 一种脓毒血症的监测系统
CN110123277A (zh) * 2019-05-17 2019-08-16 上海电气集团股份有限公司 一种脓毒血症的数据处理系统
CN110147367A (zh) * 2019-05-14 2019-08-20 中国科学院深圳先进技术研究院 一种温度缺失数据填补方法、系统及电子设备
CN110993096A (zh) * 2019-11-27 2020-04-10 医惠科技有限公司 一种脓毒血症的预警装置、设备及存储介质

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
ES2656150T3 (es) * 2011-11-14 2018-02-23 Universitätsklinikum Jena Diagnóstico de la sepsis y del síndrome de respuesta inflamatoria sistémica
CN107491638A (zh) * 2017-07-28 2017-12-19 深圳和而泰智能控制股份有限公司 一种基于深度学习模型的icu用户病情预测方法及终端设备
CN109785976B (zh) * 2018-12-11 2022-08-23 青岛智能产业技术研究院 一种基于Soft-Voting的痛风病分期预测系统
CN109711606A (zh) * 2018-12-13 2019-05-03 平安医疗健康管理股份有限公司 一种基于模型的数据预测方法及装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109902881A (zh) * 2019-03-19 2019-06-18 武汉乐易创想科技有限公司 基于多元统计分析和lstm融合的pm2.5浓度预测方法
CN110123274A (zh) * 2019-04-29 2019-08-16 上海电气集团股份有限公司 一种脓毒血症的监测系统
CN110147367A (zh) * 2019-05-14 2019-08-20 中国科学院深圳先进技术研究院 一种温度缺失数据填补方法、系统及电子设备
CN110123277A (zh) * 2019-05-17 2019-08-16 上海电气集团股份有限公司 一种脓毒血症的数据处理系统
CN110993096A (zh) * 2019-11-27 2020-04-10 医惠科技有限公司 一种脓毒血症的预警装置、设备及存储介质

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
AGOR JOSEPH; ÖZALTIN OSMAN Y.; IVY JULIE S.; CAPAN MUGE; ARNOLD RYAN; ROMERO SANTIAGO: "The value of missing information in severity of illness score development", JOURNAL OF BIOMEDICAL INFORMATICS, vol. 97, 1 September 2019 (2019-09-01), pages 1 - 15, XP085809698, ISSN: 1532-0464, DOI: 10.1016/j.jbi.2019.103255 *
SINGH JANMAJAY; OSHIRO KENTARO; KRISHNAN RAGHAVA; SATO MASAHIRO; OHKUMA TOMOKO; KATO NORIJI: "Utilizing Informative Missingness for Early Prediction of Sepsis", 2019 COMPUTING IN CARDIOLOGY (CINC), 8 September 2019 (2019-09-08), pages 1 - 4, XP033720777, DOI: 10.23919/CinC49843.2019.9005809 *

Also Published As

Publication number Publication date
CN110993096A (zh) 2020-04-10
CN110993096B (zh) 2021-04-09

Similar Documents

Publication Publication Date Title
WO2021103624A1 (zh) 一种脓毒血症的预警装置、设备及存储介质
EP3739596B1 (en) Clinical predictive analytics system
US11923094B2 (en) Monitoring predictive models
US20190287661A1 (en) Related systems and method for correlating medical data and diagnostic and health treatment follow-up conditions of patients monitored in real-time
JP6706627B2 (ja) 集中治療室における臨床値の自動化された分析及びリスク通知のためのシステム
US20200058408A1 (en) Systems, methods, and apparatus for linking family electronic medical records and prediction of medical conditions and health management
CN112786203A (zh) 一种机器学习糖尿病视网膜病变发病风险预测方法及应用
Al-Mualemi et al. A deep learning-based sepsis estimation scheme
CN112970070A (zh) 用于健康护理提供者辅助系统的方法和系统
CN110911011B (zh) 一种脓毒血症的预警装置、设备及存储介质
CN110473636B (zh) 基于深度学习的智能医嘱推荐方法及系统
CN116092655B (zh) 一种基于大数据的医院绩效管理方法及系统
WO2021103623A1 (zh) 一种脓毒血症的预警装置、设备及存储介质
US20230060794A1 (en) Diagnostic Tool
CN110555472A (zh) 基于主动学习的心电信号数据标注方法
Xie et al. Continuous-valued annotations aggregation for heart rate detection
CN112669973B (zh) 基于大数据深度学习的疾病协同递进预测方法和机器人
CN107644147A (zh) 从多个结果确定充分原因的系统和方法
Jeong et al. A method for identifying temporal progress of chronic disease using chronological clustering
CN118398150B (zh) 基于大数据的计量数据采集方法及系统
Duggan et al. Gamified Crowdsourcing as a Novel Approach to Lung Ultrasound Dataset Labeling
US20220319650A1 (en) Method and System for Providing Information About a State of Health of a Patient
Jeong et al. Temporal progress model of metabolic syndrome for clinical decision support system
US20230207125A1 (en) Diagnosis-adaptive patient acuity monitoring
Garmendia Bergés Interactive modelling and prognosis of a COVID-19 hospitalized patient via multistate models

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20892577

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20892577

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 20892577

Country of ref document: EP

Kind code of ref document: A1