CN117894466B - Continuous acute nephritis risk prediction method and system based on recurrent neural network - Google Patents

Continuous acute nephritis risk prediction method and system based on recurrent neural network Download PDF

Info

Publication number
CN117894466B
CN117894466B CN202410295446.5A CN202410295446A CN117894466B CN 117894466 B CN117894466 B CN 117894466B CN 202410295446 A CN202410295446 A CN 202410295446A CN 117894466 B CN117894466 B CN 117894466B
Authority
CN
China
Prior art keywords
time
data
patient
value
features
Prior art date
Legal status (The legal status 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 status listed.)
Active
Application number
CN202410295446.5A
Other languages
Chinese (zh)
Other versions
CN117894466A (en
Inventor
田一昕
张瑜
何佳凌
成欣
李天贵
彭丽媛
郝鹏飞
游潮
方芳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
West China Hospital of Sichuan University
Original Assignee
West China Hospital of Sichuan University
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 West China Hospital of Sichuan University filed Critical West China Hospital of Sichuan University
Priority to CN202410295446.5A priority Critical patent/CN117894466B/en
Publication of CN117894466A publication Critical patent/CN117894466A/en
Application granted granted Critical
Publication of CN117894466B publication Critical patent/CN117894466B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • 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
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Primary Health Care (AREA)
  • Biomedical Technology (AREA)
  • Epidemiology (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • Biophysics (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

The invention relates to the technical field of medicine, and particularly discloses a continuous prediction method and a continuous prediction system for acute nephritis risk based on a cyclic neural network, wherein the method comprises the following steps: acquiring information of a to-be-tested person comprising basic clinical information, nursing record information, hospital check information and hospital order information; generating time sequence data of the to-be-detected person based on the to-be-detected person information and the preset time axis information; and inputting the time sequence data into a preset cyclic neural network model, and continuously predicting the risk of the acute nephritis of the tested person in real time. Therefore, not only can the accuracy of acute nephritis prediction of a craniotome be improved, but also continuous real-time prediction can be realized, and the practicability of prediction is improved.

Description

Continuous acute nephritis risk prediction method and system based on recurrent neural network
Technical Field
The invention relates to the technical field of medicine, in particular to a continuous prediction method and system for acute nephritis risk based on a recurrent neural network.
Background
Acute kidney injury such as acute nephritis is a common hospital acquired disease that manifests as reduced renal function for a short period of time during hospital. According to previous literature reports, 30% to 40% of acute kidney injury occurs in the perioperative period in all newly diagnosed cases of acute kidney injury in the hospital. Whereas acute kidney injury during the perioperative period can lead to a variety of complications including metabolic acidosis, high potassium, uremia and altered fluid balance, and often leads to an increase in mortality from both short and long term causes of the patient, closely related to the final poor prognosis.
Craniotomy is the most common procedure in neurosurgical clinical practice. Compared with other operations, the craniotomy perioperative medication and the patient management have certain uniqueness. The usual perioperative drugs for neurosurgery such as mannitol, furosemide, hypertonic saline solution, vancomycin, etc. are all reported to have nephrotoxicity. According to the findings of Kovacheva VP et al in a retrospective study involving two-center 1656 craniotomy patients in 2016, the incidence of acute kidney injury was about 1% -10% among craniotomy patients, and the 30-day mortality of craniotomy patients with acute kidney injury was significantly increased. It is therefore highly necessary to predict acute kidney injury in patients undergoing craniotomy.
In the prior art, the risk of acute kidney injury of a person to be tested is generally predicted through a general acute kidney injury model based on pre-admission data of the person to be tested, but the prediction effect is poor.
Disclosure of Invention
Therefore, the invention aims to provide a continuous acute nephritis risk prediction method and system based on a circulatory neural network, so as to solve the problem that accurate acute nephritis risk prediction cannot be performed on craniotomy patients at present.
In order to achieve the above purpose, the invention adopts the following technical scheme:
In a first aspect, an embodiment of the present application provides a continuous prediction method for acute nephritis risk based on a recurrent neural network, including:
Obtaining information of a to-be-tested person, wherein the information of the to-be-tested person comprises basic clinical information, nursing record information, hospital check information and hospital doctor advice information;
generating time sequence data of the to-be-detected person based on the to-be-detected person information and preset time axis information;
And inputting the time sequence data into a preset cyclic neural network model, and continuously predicting the risk of the acute nephritis of the to-be-detected person in real time.
Further, the generating process of the preset cyclic neural network model includes:
acquiring historical data of a plurality of patients;
formatting and digitizing the historical data to obtain data characteristics corresponding to each patient;
generating, for each patient, a time axis data characteristic of the patient based on a first preset length of time and the data characteristic of the patient; wherein each of the timeline data features includes a plurality of values distributed over a plurality of time nodes;
For each patient, marking the acute nephritis risk of the time axis data characteristic of the patient by taking a time node as a unit to obtain a label result of the patient;
For each patient, performing sliding window processing on the time axis data characteristics based on a second preset time length to generate sliding window data;
training and verifying a pre-constructed basic cyclic neural network model based on the sliding window data and the corresponding labels to obtain the pre-set cyclic neural network model.
Further, before generating the time axis data characteristic of each patient based on the first preset time length and the data characteristic of the patient, the method further comprises: screening the data features to determine the data features corresponding to the target features;
The target features include a fixed feature and a timing feature;
The fixed features are features that do not change over time during patient hospitalization, including basic clinical features; the basic clinical features include: age, sex, smoking history, drinking history, hypertension history, diabetes history, ASA score, hospitalized GCS score, chronic renal disease history, chronic obstructive pulmonary history, coronary heart disease history, duration of surgery, blood loss from surgery, nosocomial infection, surgical incision infection, deep vein infection, and epilepsy;
The timing characteristic is a characteristic that changes over time during patient hospitalization, comprising: care record features, hospital check features, and hospital medication features;
The care record features include: body temperature, respiratory rate, urine volume, heart rate, systolic pressure, diastolic pressure, pulse, pain score, oxygen saturation, and 24h intake; the hospital check feature includes: sodium, potassium, blood oxygen, serum cystatin, blood protein, AST/ALT, blood glucose, creatine kinase, lactate dehydrogenase, hydrocarbon butyrate dehydrogenase, urea, cholesterol, creatinine, red blood cell count, hemoglobin, white blood cell count, neutrophil absolute, lymphocyte absolute, monocyte absolute, ph, platelet count, and blood calcium; the hospital administration features include: vancomycin, meropenem, cephalosporins, beta-lactams, aminoglycosides, quinolones, amphotericin B, fluconazole, hypertonic salt solution, mannitol, furosemide, glycerofructose, spironolactone, desmopressin, iomeprol, iodixanol, gadolinium contrast agent, iohexol, baterin, human serum albumin, nonsteroidal species, and ACEIARB species;
The generating, for each patient, a time axis data characteristic of the patient based on a first preset length of time and the data characteristic of the patient, comprising:
For each patient, generating a time axis data feature of the patient corresponding to each target feature based on a first preset time length and the data features of the patient.
Further, the method comprises the steps of,
For the data features corresponding to the fixed features, generating a time axis data feature of the patient corresponding to each target feature based on a first preset time length and the data features of the patient, including:
Determining each time node of the time axis data feature based on the first preset time length;
Filling the numerical values in the data features into all time nodes of the time axis data features corresponding to the fixed features, and generating the time axis data features of the patient, which are respectively corresponding to each fixed feature one by one;
For the feature data corresponding to the time sequence feature, the generating a time axis data feature of the patient corresponding to each target feature based on a first preset time length and the data feature of the patient includes:
Determining each time node of the time axis data feature based on the first preset time length;
Calculating a time of value in each of the data features based on the patient admission time;
And filling the numerical values in the data features into corresponding time nodes of time axis data features of corresponding time sequence features based on the time of the numerical values in each data feature, and generating the time axis data features of the patient, which are respectively in one-to-one correspondence with each time sequence feature.
Further, the method further comprises the following steps:
Performing continuous value processing on the time axis data characteristics corresponding to the state type characteristics, wherein the state type characteristics comprise all the hospital check characteristics and all nursing record characteristics except the 24h input;
the continuation value processing includes:
Aiming at the time axis data feature with the value of the first time node not being 0, taking the value of the last time node with the value not being 0 in the time axis data feature as the value of the current time node with the value being 0 in the time axis data feature;
Taking the global non-0 median value as the value of the first bit time node in the time axis data characteristic aiming at the time axis data characteristic with the value of 0 of the first bit time node; and taking the value of the time node with the last value not being 0 in the time axis data characteristic as the value of the time node with the current value being 0 in the time axis data characteristic.
Further, the method further comprises the following steps:
Adding a time node with a second preset time length to each time axis data characteristic;
wherein, in the added time node, the same value of the original time axis data characteristic is taken for the value corresponding to the basic clinical characteristic; for the numerical value corresponding to the state type characteristic, taking the numerical value of the first time node in the original time axis data characteristic; and taking 0 for the numerical value corresponding to the hospital administration characteristic.
Further, the marking the acute nephritis risk of the time axis data feature of each patient by taking a time node as a unit comprises:
generating a grading judgment standard based on a preset grading judgment index;
and extracting the numerical value corresponding to the grading judgment index from all time nodes in the time axis data characteristics of the patient, and marking the acute nephritis risk of the data of each time node in the time axis data characteristics of the patient by taking the time node as a unit based on the grading judgment standard and the model target prediction grade.
Further, the grading decision index includes creatinine;
the generating the grading judgment standard based on the grading judgment index comprises the following steps: defining the creatinine test value of the current time node as 1.5-1.9 times of the baseline value as grade 1; defining the creatinine test value of the current time node as 2.0-2.9 times of the baseline value as grade 2; defining the creatinine test value of the current time node as 3.0 times of the baseline value as grade 3; wherein the baseline value comprises a creatinine test value for a first time node of a patient time axis data characteristic;
The step of extracting the numerical value corresponding to the grading judgment index in all time nodes in the time axis data characteristics of the patient, and carrying out acute nephritis risk marking on the data of each time node in the time axis data characteristics of the patient by taking the time node as a unit based on the grading judgment standard and the model target prediction grade comprises the following steps:
extracting a creatinine test value of a first time node in the time axis data characteristics of the patient, and multiplying the creatinine test value by a target multiple to generate grading judgment standard information, wherein the target multiple is a multiple corresponding to the model target prediction grade in the grading judgment standard;
Extracting creatinine test values of all time nodes of the patient to generate a creatinine array;
And converting the creatinine array into a tag array based on the grading judgment standard information, and taking the tag array as the tag result.
Further, the method comprises the steps of,
The corresponding label of the sliding window data is a label of a time node which is positioned behind the time node corresponding to the sliding window data and is separated from the time node corresponding to the sliding window data by a second target time length in the same time axis data characteristic;
Wherein the second target time length is determined based on a predicted time length of the network model.
In a second aspect, an embodiment of the present application provides a continuous acute nephritis risk prediction system based on a recurrent neural network, including:
The system comprises an acquisition module, a control module and a control module, wherein the acquisition module is used for acquiring information of a to-be-tested person, and the information of the to-be-tested person comprises basic clinical information, nursing record information, hospital check information and hospital doctor advice information;
the processing module is used for generating time sequence data of the to-be-detected person based on the to-be-detected person information and the preset time axis information;
and the prediction module is used for continuously predicting the risk of the acute nephritis of the to-be-detected person in real time based on the time sequence data.
The invention provides a continuous prediction method and a continuous prediction system for acute nephritis risk based on a recurrent neural network, wherein the method comprises the following steps: acquiring information of a to-be-tested person comprising basic clinical information, nursing record information, hospital check information and hospital order information; generating time sequence data of the to-be-detected person based on the to-be-detected person information and the preset time axis information; and inputting the time sequence data into a preset cyclic neural network model, and continuously predicting the risk of the acute nephritis of the tested person in real time. Therefore, not only can the accuracy of acute nephritis prediction of a craniotome be improved, but also continuous real-time prediction can be realized, and the practicability of prediction is improved.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic flow chart of a continuous prediction method for acute nephritis risk based on a recurrent neural network according to an embodiment of the present invention;
Fig. 2 is a schematic flow chart of a prediction model generated in a continuous prediction method of acute nephritis risk based on a recurrent neural network according to an embodiment of the present invention;
fig. 3 is a schematic diagram of training and prediction of a prediction model in a prediction model generated in a continuous prediction method of acute nephritis risk based on a recurrent neural network according to an embodiment of the present invention;
fig. 4 is a prediction effect diagram of a continuous prediction method for acute nephritis risk based on a recurrent neural network according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a continuous acute nephritis risk prediction device based on a recurrent neural network according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be described in detail below. It will be apparent that the described embodiments are only some, but not all, embodiments of the invention. All other embodiments, based on the examples herein, which are within the scope of the invention as defined by the claims, will be within the scope of the invention as defined by the claims.
Method embodiment:
Fig. 1 is a schematic flow chart of a continuous prediction method for acute nephritis risk based on a recurrent neural network according to an embodiment of the present invention, referring to fig. 1, the present embodiment may include the following steps:
S101, obtaining information of a person to be tested.
Specifically, the information of the to-be-tested person can comprise basic clinical information, nursing record information, hospital check information, hospital order information (such as a hospital administration condition) and the like.
S102, generating time sequence data of the testee based on the information of the testee and the preset time axis information.
Specifically, the time series data for the subject, that is, the data having the time series may be generated based on the preset time axis information (the preset time length and the like as mentioned below) and the subject information, for example, the time series data based on the admission time of the subject as the start time.
S103, inputting the time sequence data into a preset cyclic neural network model, and continuously predicting the risk of the acute nephritis of the person to be tested in real time.
Specifically, the preset cyclic neural network model is generated based on patient data training in a preset database and is used for predicting the risk of acute nephritis of the person to be tested at the target moment based on time sequence data of the person to be tested. The time sequence data with different time sequences are input into a preset cyclic neural network model, so that the risk of acute nephritis of a person to be tested in corresponding time (for example, after 2 days) can be obtained.
The application provides a continuous prediction method for acute nephritis risk based on a recurrent neural network, which comprises the following steps: acquiring information of a to-be-tested person comprising basic clinical information, nursing record information, hospital check information and hospital order information; generating time sequence data of the to-be-detected person based on the to-be-detected person information and the preset time axis information; and inputting the time sequence data into a preset cyclic neural network model, and continuously predicting the risk of the acute nephritis of the tested person in real time. Therefore, not only can the accuracy of acute nephritis prediction of a craniotome be improved, but also continuous real-time prediction can be realized, and the practicability of prediction is improved.
Fig. 2 is a schematic flow chart of generating a prediction model (i.e., a cyclic neural network model for performing prediction) in a cyclic neural network-based continuous prediction method for acute nephritis risk according to an embodiment of the present application, as shown in fig. 2, in an embodiment of the present application, a flow of generating the cyclic neural network model may at least include the following flows:
S201, historical data of a plurality of patients are acquired.
S202, formatting and digitizing historical data to obtain data characteristics corresponding to each patient.
Specifically, the historical data of all patients in a certain time period can be obtained from a preset hospital or medical institution, for example, through a China Western medicine institute big data center, and all patient group data of the China Western medicine institute neurosurgery center, which is subjected to the period selection craniotomy, are collected from 1 month in 2011 to 3 months in 2021. The data (based on the Chinese western medicine institute ethics No. 20211701) namely the historical data can comprise basic clinical information (such as gender, age, smoking and drinking history, hypertension history and the like) of each patient and all indexes of the patient in-hospital examination; all hospital orders for patients; all nursing records of patients in hospital. And then carrying out formatting and digitizing processing on a large amount of obtained data to finally obtain the data forming an editable form of the computer language.
The data characteristics for each patient were obtained after the above operations. Wherein the data features comprise fixed features that do not change over time during the hospital, such as features relating to the underlying clinic; and features that change over time during the hospital, such as all with respect to the order of the hospital (medication), all with respect to the check-up index at the hospital, and all with respect to the three parts of the care record at the hospital, and the specific values of each feature, i.e. feature entry, of the features of the time are time stamped with an accuracy of hours, such as blood glucose after a patient is admitted: first day 4.5, second day 5.5, etc.
In some embodiments of the present application, when the above data features are initially processed and included in the data, for features related to basic clinics, a floating point number may be used to represent a continuous variable (composed of a plurality of values, such as continuous blood glucose values or blood pressure values, etc.), and 0/1, etc., to represent a classification variable (indicating the presence/absence of smoking history, etc.).
In addition, in actual hospital administration, there may be a difference in the record of dosage forms (such as ml/mg/bottle) in the items ascribed to the same characteristics, and in the present application, the dosage form units are collectively converted into mg for inclusion. All continuous variables were outliers removed in 99.9% bins.
In some embodiments of the present application, after the data features described above are obtained, all the data may be combined to form a feature array, as for the large data center of the Huaxi hospital mentioned in the above embodiments, a total of 10,943,528 feature entries may be obtained for subsequent processing (where the feature array for each patient may also be sorted).
S203, generating a time axis data characteristic of each patient based on the first preset time length and the data characteristic of the patient.
Wherein each time axis data feature comprises a plurality of values distributed over a plurality of time nodes, e.g. one time axis data comprises 39 values, located at 39 time nodes, respectively.
First, before generating the time axis data features, a sampling and missing value step may be performed, specifically: for the missing values of the feature entries (the feature entries should theoretically have corresponding values under each data feature, but the information integrity of different patients is not the same, so that a certain data feature of a patient may lack a specific value under the data feature name determined by integrating all patient information), the missing values of the classification variables are directly removed, and for the missing values of the continuous variables, the global non-zero median value of the data feature of all patients is filled. And calculating the time of each characteristic item with the sampling precision of hours according to the admission time of each patient.
On this basis, a time axis data feature is generated.
First, a time axis data feature is generated based on the feature data and a first preset time length.
Specifically, for example, a 39-day (i.e., a first preset time length) time axis (including 39 time nodes from day 1 to day 39 after admission) is generated with a day precision. For a patient's data feature, for a fixed-value data feature, the patient generates a timeline data feature with the same value at all 39 timestamp locations (one-to-one correspondence with the 39 time nodes) that incorporate the feature. For a data feature with continuous values, traversing the values of the data feature of the patient every day by a program, and recording a time stamp of the current day as a unique value of the data feature if the patient exists on the current day; if the value of the data characteristic (unverified or not dosed, etc.) does not appear on the same day, the time stamp of the day is filled with 0; if a plurality of values exist in the data feature of the patient on the same day, the last value is taken as the recorded value of the current time stamp of the data feature, and a time axis data feature is generated.
In some embodiments, to increase the efficiency of the subsequent processing, the above-described processing may be performed on each patient, generating a new data array, i.e., a time axis data feature array, for the subsequent processing.
Further, in some embodiments of the present application, the data in the time axis data feature array may be ejected and filtered based on a preset ejection standard and target feature information or a manual command.
For example, patient data may be screened according to drainage criteria, such as: patients less than 18 years old are excluded; emergency craniotomy patients resulting from cerebral hemorrhage, subarachnoid hemorrhage and trauma were excluded, and a total of 26133 patients were eventually included in subsequent data processing in the above-mentioned large data center.
And determining to incorporate the features potentially related to the occurrence of acute kidney injury, namely target features, through comprehensive discussion evaluation of early research experience and neurosurgery specialists, and screening all data features of the patient, wherein 73 features of the target features are finally incorporated, and the method specifically comprises the following steps: fixed features (corresponding to the fixed features in the data features) and timing features (corresponding to the features having timing in the data features).
Wherein the fixed characteristic is a characteristic that does not change over time during patient hospitalization, including a basic clinical characteristic; the basic clinical features include: age, sex, smoking history, drinking history, hypertension history, diabetes history, ASA score, hospitalized GCS score, chronic renal disease history, chronic obstructive pulmonary history, coronary heart disease history, duration of surgery, blood loss from surgery, nosocomial infection, surgical incision infection, deep vein infection, and epilepsy; the time sequence features are features that change over time during patient hospitalization, including: care record features, hospital check features, and hospital medication features; the care record features include: body temperature, respiratory rate, urine volume, heart rate, systolic pressure, diastolic pressure, pulse, pain score, oxygen saturation, and 24h intake; the hospital check features include: sodium, potassium, blood oxygen, serum cystatin, blood protein, AST/ALT, blood glucose, creatine kinase, lactate dehydrogenase, hydrocarbon butyrate dehydrogenase, urea, cholesterol, creatinine, red blood cell count, hemoglobin, white blood cell count, neutrophil absolute, lymphocyte absolute, monocyte absolute, ph, platelet count, and blood calcium; the hospital administration features include: vancomycin, meropenem, cephalosporins, beta-lactams, aminoglycosides, quinolones, amphotericin B, fluconazole, hypertonic salt solution, mannitol, furosemide, glycerofructose, spironolactone, desmopressin, iomeprol, iodixanol, gadolinium contrast agent, iohexol, baterin, human serum albumin, nonsteroidal species, and ACEIARB species.
It should be noted that, in the present application, after the above data missing value processing and the time axis data feature processing, and filtering and screening, each patient has 73 time axis data features corresponding to the above 73 item target features one by one, and each time axis data feature has 39 time nodes for placing specific values, and in the above specific data embodiment, the total processing inclusion data point is 87,754,614 (each data point corresponds to one of the above mentioned values) for all patients.
In other embodiments of the present application, the information of the target features may be determined first, and then the numerical value corresponding to each target feature may be obtained, so as to obtain the data feature of each patient corresponding to each target feature, and further obtain the time axis data feature corresponding to each target feature.
The method specifically comprises the following steps: acquiring data features corresponding to the fixed features, and then determining each time node of the time axis data features based on a first preset time length for 39 days; and filling the numerical values in the data features into all time nodes of the time axis data features corresponding to the fixed features, and generating the time axis data features of the patient corresponding to each fixed feature respectively. Acquiring data characteristics corresponding to the time sequence characteristics, and determining each time node of the time axis data characteristics based on a first preset time length; and calculating a time for the value in each data feature based on the patient admission time; and then, based on the time of the numerical value in each data feature, filling the numerical value in the data feature into the corresponding time node of the time axis data feature of the corresponding time sequence feature, and generating the time axis data feature of the patient corresponding to each time sequence feature. The specific filling principle is the same as that in the above embodiment, and a detailed description thereof will be omitted.
Further, in some embodiments of the present application, when generating the prediction model, the method may further include: and performing continuous value processing on the time axis data characteristics corresponding to the state class characteristics. Wherein the status class features include all of the hospital check features described above, and all of the care record features except for the 24h entry described above.
The continuation value processing specifically comprises the following steps: aiming at the time axis data feature with the value of the first time node not being 0, taking the value of the last time node with the value not being 0 in the time axis data feature as the value of the current time node with the value being 0 in the time axis data feature; and regarding the time axis data feature with the value of the first bit time node being 0, taking the global non-0 median value as the value of the first bit time node in the time axis data feature (namely, the time axis data feature with the value of the first bit time node being 0); and taking the time node with the last value not being 0 in the time axis data characteristic as the time node with the current value being 0 in the time axis data characteristic.
For example: traversing each time stamp value of each feature of each patient by using a preset program for the time axis data feature of the corresponding state class, and continuing the value of t n to tn+1 if the value of the time stamp t n is greater than 0 and the value of the time stamp t n+1 is equal to zero; if the first bit timestamp t 0 of a time axis data feature has a value of zero, then the global non-zero median is padded. If the values of all time stamps of a certain time axis data feature are 0, the global non-zero median is filled.
And in order to consider the data of the patient before admission and adapt to the subsequent generation of window data, a time node of a second preset time length can be added to each obtained time axis data characteristic; wherein, in the added time node, the same value of the original time axis data characteristic is taken for the value corresponding to the basic clinical characteristic; for the numerical value of the corresponding state class feature, taking the numerical value of the first time node in the original time axis data feature; for the values corresponding to the hospital medication profile, 0 was taken.
For example, 7 time stamps (for characterization as a pre-patient admission status, in preparation for a sliding window procedure, the length of the bit may be adjusted according to the period of the recurrent neural network that requires backtracking) are appended to the front of each timeline data feature, extending the time stamp of 39d to 46d. In the filled 7-bit time stamp, the part corresponding to the basic clinical feature is the original value in the original time axis data feature, the part corresponding to the state type feature is the value of the time stamp of the first day (t 0) in the original time axis data feature (null value does not exist in the first day after the continuous value processing), and the part corresponding to the application type feature is 0.
S204, for each patient, marking the acute nephritis risk of the time axis data characteristics of the patient by taking the time node as a unit, and obtaining the label result of the patient.
Specifically, for example, for one patient, it may include: firstly, generating a grading judgment standard based on a preset grading judgment index; and then extracting the numerical values corresponding to the grading judgment indexes in all time nodes in the time axis data characteristics of the patient, and carrying out acute nephritis risk marking on the data of each time node in the time axis data characteristics of the patient by taking the time node as a unit based on the grading judgment standard and the model target prediction grade.
For example, the grading criterion may select creatinine. At this time, in the above embodiment, the generation of the classification judgment criterion based on the classification judgment index may specifically include: defining the creatinine test value of the current time node as 1.5-1.9 times of the baseline value as grade 1; defining the creatinine test value of the current time node as 2.0-2.9 times of the baseline value as grade 2; defining the creatinine test value of the current time node as 3.0 times of the baseline value as grade 3; wherein the baseline value comprises a creatinine test value for a first time node of the patient.
On the basis, the creatinine test value of the current time stamp can be extracted for each time axis characteristic data of a patient, compared with the creatinine test value of the first day time stamp in the time axis characteristic data of the patient, a label of whether the current time stamp has acute nephritis or not is obtained, and the marking of the data of each time stamp in all the time axis characteristic data of all the patients is completed in the same way. Alternatively, after obtaining the creatinine test value of the first day timestamp in the patient time axis feature data, the grading determination standard information may be generated, for example, the creatinine test value of the current timestamp may be evaluated as several grades in the grading determination standard, or whether the creatinine test value belongs to one grade, and further marking is performed.
In some embodiments of the present application, to improve the marking efficiency, the marking may be performed by using an array manner based on the above-described time axis data feature array for all patients, for example:
After determining the grading judgment standard, traversing each patient by using a preset program, taking the t 0 time stamp creatinine value (no null value exists on the first day after continuous value processing) of each patient, and multiplying the value by the corresponding multiple to generate the grading judgment standard array. Then, taking a creatinine array of a full time axis (39 d) of each patient, and converting the creatinine array into a tag array according to the AKI (for example, a model predicts AKI 2 grade risk) predicted as required, and taking the tag array as a tag result. Wherein each timestamp indicates whether AKI occurs or not by 1/0.
Thus, after the marking is completed, the queue including the time axis feature data and the marking is randomly divided into a training set and a test set in a ratio of 7:3, and the training set (specifically, 18293 patient data) is used for training model parameters, preferred model structures, super parameters and the like. The test set (7840 patients' data) was used to detect model generalization performance.
S205, carrying out sliding window processing on the time axis data characteristics based on a second preset time length for each patient to generate sliding window data.
S206, training and verifying the pre-constructed basic cyclic neural network model based on the sliding window data and the corresponding labels to obtain a preset cyclic neural network model.
Specifically, in some examples of the present application, based on the training set and the test set obtained as described above, an input basic cyclic neural network model may be obtained through a sliding window principle, and training and verifying the input basic cyclic neural network model, thereby obtaining a final cyclic neural network model.
Specifically, in some embodiments of the present application, the sliding window program may take 7 days as the window width. Each time stamp bit is traversed for each patient, and the time axis data feature array corresponding to each patient (consisting of all time axis data features for that patient) is segmented into 39 window feature arrays (i.e., window data, such as X t0-Xt7,Xt1-Xt8,Xt2-Xt9,…,Xtn-Xtn+7) of 7 days in width. The label of each window feature array depends on the given prediction time window, i.e. the prediction duration of the prediction model, for example, if the model is to predict the probability of acute kidney injury occurring within 48 hours (the second target time length is 48 hours at this time, it will be understood that in practical applications, both 24 hours/48 hours/72 hours can be set), then the label value of the Xt n window feature array is the or value of the label values in the next two timestamps of the time axis data feature of the patient (Yt n+1|Ytn+2).
Then based on the data, a feature tensor slice is generated. Specifically, the "from_ tensor _slots" function module TensorFlow may be used to convert the processed training set and test set into tensor slices. Wherein, for the specific data example, the value of "batch" is set to 390, the dimension of the training set feature matrix is (713427,7,73), the dimension of the testing set feature matrix is (305760,7,73) and the dimension after slicing is (390,7,73); training set label dimensions (713427), test set label dimensions (305760), post-slicing label dimensions (390). And inputting the generated data array into a basic cyclic network model for processing.
In some embodiments of the application, the underlying recurrent network model may be as shown in FIG. 3, and a multi-layer long-short term memory recurrent neural network (LSTM) may be constructed using TensorFlow deep learning framework. The specific model structure is as follows: three LSTM (node number is 64,64,32 respectively) layers and one full-connection layer are connected in series, and a sigmoid function is used as a final prediction function. Each LSTM layer includes a Dropout layer and an L2 Regularization layer to prevent model overfitting, and the specific model structure and prediction flow are shown in fig. 3.
It should be noted that, in the embodiment of the present application, for the initial parameters and the super parameters of the model, specific steps are as follows: the initial parameters are random arrays, and the initial super parameters are as follows: the learning rate adopts a step-by-step subtraction method (0.1-0.001); the optimizer employs adaptive moment estimation (Adam); the loss function adopts binary cross entropy; the Dropout rate and the L2 Regularization rate are respectively 0.5 and 0.001; because of the imbalance in the ratio of positive to negative tags (AKI incidence of about 3% in this dataset), the positive/negative tag weight was set to 5:1. the above super parameters are determined by multiple pre-experimental grid searches, and the final parameters of the model are determined according to the lowest Loss function value (Loss) of multiple training-internal verification tests.
And in the model training process, determining the optimal super parameters by using a grid search method, and storing an optimal effect model after each model training iteration is performed for 100 times. And storing the Loss value, accuracy and AUC list of the model stored in each iteration, and selecting an optimal model after all training is finished.
And when the model test set is verified, inputting the test set into the obtained optimal model, and calculating accuracy and AUC. Samples of the AKI pre-warning that were successful 48 hours earlier are shown in fig. 4, for example.
In the continuous prediction method of acute nephritis risk based on the recurrent neural network, provided by the application, the prediction model systematically incorporates risk factors related to acute kidney injury in patients receiving craniotomy. The included risk factors (i.e., the target features described above) are comprehensive and include the contents of the admitted clinical features and dynamic checkups and perioperative drug management. Compared with the existing prediction system for acute kidney injury of craniotomy patients, the prediction system can only predict according to the admission characteristics, can only predict the total probability of acute kidney injury during hospital, omits the data characteristics of a large number of patients during hospital, and improves the practicability. The model is a deep learning model which is built based on real-time changing test indexes, medicine taking conditions and the like during patient admission and is used for continuously predicting the risk of acute kidney injury, and the model can consider the difference between a medicine treatment strategy and other diseases of patients receiving craniotomy, so that the prediction effect is better.
According to the continuous prediction method for acute nephritis risk based on the recurrent neural network, provided by the application, through a series of tendency score proportioning queue researches, the risk factors of acute renal injury of patients receiving craniotomy are systematically explored, and the method mainly comprises three aspects: 1) Admission clinical features and inspection; 2) Performing perioperative medication management and checking indexes; 3) The state and the test index in the operation. On the basis of systematically evaluating risk factors of acute kidney injury, an online prediction model is built, so that the risk of acute kidney injury in the hospital can be predicted through clinical characteristics of admission and inspection and examination, and the method is used by a doctor in charge of the hospital online, and has wide usability. Moreover, the risk of acute kidney injury to the patient can be predicted continuously in real time based on dynamic information through the patient during the hospital for the ongoing changes in the patient's condition and the impact of treatment during the hospital.
System embodiment:
Based on the same inventive concept, the embodiment of the application also provides a continuous acute nephritis risk prediction system based on a recurrent neural network, as shown in fig. 5, the system at least can comprise:
The obtaining module 51 is configured to obtain information of a to-be-tested person, where the information of the to-be-tested person includes basic clinical information, nursing record information, hospital check information, and hospital order information.
The processing module 52 is configured to generate time series data of the tester based on the tester information and the preset time axis information.
The prediction module 53, such as the prediction model mentioned in the above method embodiment, is configured to continuously predict, in real time, the risk of occurrence of acute nephritis in the subject based on the time series data.
The specific manner in which the various modules perform the operations in the apparatus of the above embodiments have been described in detail in connection with the embodiments of the method, and will not be described in detail herein.
It is to be understood that the same or similar parts in the above embodiments may be referred to each other, and that in some embodiments, the same or similar parts in other embodiments may be referred to.
It should be noted that in the description of the present invention, the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance. Furthermore, in the description of the present invention, unless otherwise indicated, the meaning of "plurality" means at least two.
Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or steps of the process, and further implementations are included within the scope of the preferred embodiment of the present invention in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the present invention.
It is to be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, may be implemented using any one or combination of the following techniques, as is well known in the art: discrete logic circuits having logic gates for implementing logic functions on data signals, application specific integrated circuits having suitable combinational logic gates, programmable Gate Arrays (PGAs), field Programmable Gate Arrays (FPGAs), and the like.
Those of ordinary skill in the art will appreciate that all or a portion of the steps carried out in the method of the above-described embodiments may be implemented by a program to instruct related hardware, where the program may be stored in a computer readable storage medium, and where the program, when executed, includes one or a combination of the steps of the method embodiments.
In addition, each functional unit in the embodiments of the present invention may be integrated in one processing module, or each unit may exist alone physically, or two or more units may be integrated in one module. The integrated modules may be implemented in hardware or in software functional modules. The integrated modules may also be stored in a computer readable storage medium if implemented in the form of software functional modules and sold or used as a stand-alone product.
The above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, or the like.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the invention, and that variations, modifications, alternatives and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the invention.

Claims (1)

1. A continuous prediction method for acute nephritis risk based on a recurrent neural network is characterized by comprising the following steps:
Obtaining information of a to-be-tested person, wherein the information of the to-be-tested person comprises basic clinical information, nursing record information, hospital check information and hospital doctor advice information;
generating time sequence data of the to-be-detected person based on the to-be-detected person information and preset time axis information;
inputting the time sequence data into a preset cyclic neural network model, and continuously predicting the risk of the acute nephritis of the to-be-detected person in real time;
The generation process of the preset cyclic neural network model comprises the following steps:
acquiring historical data of a plurality of patients;
formatting and digitizing the historical data to obtain data characteristics corresponding to each patient;
Screening the data features to determine the data features corresponding to the target features;
The target features include a fixed feature and a timing feature;
The fixed features are features that do not change over time during patient hospitalization, including basic clinical features; the basic clinical features include: age, sex, smoking history, drinking history, hypertension history, diabetes history, ASA score, hospitalized GCS score, chronic renal disease history, chronic obstructive pulmonary history, coronary heart disease history, duration of surgery, blood loss from surgery, nosocomial infection, surgical incision infection, deep vein infection, and epilepsy;
The timing characteristic is a characteristic that changes over time during patient hospitalization, comprising: care record features, hospital check features, and hospital medication features;
The care record features include: body temperature, respiratory rate, urine volume, heart rate, systolic pressure, diastolic pressure, pulse, pain score, oxygen saturation, and 24h intake; the hospital check feature includes: sodium, potassium, blood oxygen, serum cystatin, blood protein, AST/ALT, blood glucose, creatine kinase, lactate dehydrogenase, hydrocarbon butyrate dehydrogenase, urea, cholesterol, creatinine, red blood cell count, hemoglobin, white blood cell count, neutrophil absolute, lymphocyte absolute, monocyte absolute, ph, platelet count, and blood calcium; the hospital administration features include: vancomycin, meropenem, cephalosporins, beta-lactams, aminoglycosides, quinolones, amphotericin B, fluconazole, hypertonic salt solutions, mannitol, furosemide, glycerofructose, spironolactone, desmopressin, iomeprol, iodixanol, gadolinium contrast agents, iohexol, baterin, human serum albumin, nonsteroidal species, ACEI species and ARBs;
Generating, for each patient, a time axis data characteristic of the patient based on a first preset length of time and the data characteristic of the patient, comprising: generating, for each patient, a time axis data feature of the patient corresponding to each of the target features, respectively, based on a first preset length of time and the data features of the patient; wherein each of the timeline data features includes a plurality of values distributed over a plurality of time nodes;
For the data features corresponding to the fixed features, generating a time axis data feature of the patient corresponding to each target feature based on a first preset time length and the data features of the patient, including:
Determining each time node of the time axis data feature based on the first preset time length;
Filling the numerical values in the data features into all time nodes of the time axis data features corresponding to the fixed features, and generating the time axis data features of the patient, which are respectively corresponding to each fixed feature one by one;
For the feature data corresponding to the time sequence feature, the generating a time axis data feature of the patient corresponding to each target feature based on a first preset time length and the data feature of the patient includes:
Determining each time node of the time axis data feature based on the first preset time length;
Calculating a time of value in each of the data features based on the patient admission time;
Filling the numerical values in the data features into corresponding time nodes of time axis data features of corresponding time sequence features based on the time of the numerical values in each data feature, and generating time axis data features of the patient corresponding to each time sequence feature one by one;
For each patient, marking the acute nephritis risk of the time axis data characteristic of the patient by taking a time node as a unit to obtain a label result of the patient;
For each patient, performing sliding window processing on the time axis data characteristics based on a second preset time length to generate sliding window data;
training and verifying a pre-constructed basic cyclic neural network model based on the sliding window data and the corresponding labels to obtain the preset cyclic neural network model;
Further comprises:
Performing continuous value processing on the time axis data characteristics corresponding to the state type characteristics, wherein the state type characteristics comprise all the hospital check characteristics and all nursing record characteristics except the 24h input;
the continuation value processing includes:
Aiming at the time axis data feature with the value of the first time node not being 0, taking the value of the last time node with the value not being 0 in the time axis data feature as the value of the current time node with the value being 0 in the time axis data feature;
Taking the global non-0 median value as the value of the first bit time node in the time axis data characteristic aiming at the time axis data characteristic with the value of 0 of the first bit time node; and taking the value of the time node with the last value not being 0 in the time axis data characteristic as the value of the time node with the current value being 0 in the time axis data characteristic;
Further comprises:
Adding a time node with a second preset time length to each time axis data characteristic;
wherein, in the added time node, the same value of the original time axis data characteristic is taken for the value corresponding to the basic clinical characteristic; for the numerical value corresponding to the state type characteristic, taking the numerical value of the first time node in the original time axis data characteristic; taking 0 for the numerical value corresponding to the hospital medication characteristic;
Wherein said for each patient, in time node units, labeling the acute nephritis risk for the time axis data characteristic of that patient, comprises:
generating a grading judgment standard based on a preset grading judgment index; the grading judgment index comprises creatinine; the generating the grading judgment standard based on the grading judgment index comprises the following steps: defining the creatinine test value of the current time node as 1.5-1.9 times of the baseline value as grade 1; defining the creatinine test value of the current time node as 2.0-2.9 times of the baseline value as grade 2; defining the creatinine test value of the current time node as 3.0 times of the baseline value as grade 3; wherein the baseline value comprises a creatinine test value for a first time node of a patient time axis data characteristic;
Extracting values corresponding to the grading judgment indexes from all time nodes in the time axis data characteristics of the patient, and carrying out acute nephritis risk marking on data of each time node in the time axis data characteristics of the patient by taking the time node as a unit based on the grading judgment standards and model target prediction grades, wherein the method comprises the following steps: extracting a creatinine test value of a first time node in the time axis data characteristics of the patient, and multiplying the creatinine test value by a target multiple to generate grading judgment standard information, wherein the target multiple is a multiple corresponding to the model target prediction grade in the grading judgment standard; extracting creatinine test values of all time nodes of the patient to generate a creatinine array; converting the creatinine array into a tag array based on the grading judgment standard information, and taking the tag array as the tag result;
The corresponding label of the sliding window data is a label of a time node which is positioned behind the time node corresponding to the sliding window data and is separated from the time node corresponding to the sliding window data by a second target time length in the same time axis data characteristic;
Wherein the second target time length is determined based on a predicted time length of the network model.
CN202410295446.5A 2024-03-15 2024-03-15 Continuous acute nephritis risk prediction method and system based on recurrent neural network Active CN117894466B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202410295446.5A CN117894466B (en) 2024-03-15 2024-03-15 Continuous acute nephritis risk prediction method and system based on recurrent neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202410295446.5A CN117894466B (en) 2024-03-15 2024-03-15 Continuous acute nephritis risk prediction method and system based on recurrent neural network

Publications (2)

Publication Number Publication Date
CN117894466A CN117894466A (en) 2024-04-16
CN117894466B true CN117894466B (en) 2024-05-17

Family

ID=90652139

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202410295446.5A Active CN117894466B (en) 2024-03-15 2024-03-15 Continuous acute nephritis risk prediction method and system based on recurrent neural network

Country Status (1)

Country Link
CN (1) CN117894466B (en)

Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109872819A (en) * 2019-01-30 2019-06-11 杭州脉兴医疗科技有限公司 A kind of acute kidney injury incidence rate forecasting system based on Intensive Care Therapy detection
CN110459328A (en) * 2019-07-05 2019-11-15 梁俊 A kind of Clinical Decision Support Systems for assessing sudden cardiac arrest
CN112837819A (en) * 2021-01-20 2021-05-25 尹丽君 Method for establishing acute kidney injury prediction model after coronary artery bypass grafting
KR20210152047A (en) * 2020-06-05 2021-12-14 인하대학교 산학협력단 Apparatus for predicting acute kidney injury using machine learning based on artificial intelligence and method thereof
KR20220132397A (en) * 2021-03-23 2022-09-30 서울대학교병원 System and method for supporting clinical decision for acute kidney injury using recurrent neural network model
CN115171881A (en) * 2022-07-08 2022-10-11 陕西霖桥实业有限公司 Acute kidney injury prediction method based on full-connection feature fusion convolution neural network
CN115410716A (en) * 2022-08-17 2022-11-29 四川大学华西医院 Method and system for establishing neural function prognosis prediction model of cerebral hemorrhage patient
CN115762783A (en) * 2022-11-22 2023-03-07 深圳大学 Acute kidney injury prediction system
CN116030976A (en) * 2022-12-20 2023-04-28 浙江大学 Risk prediction system for critical patients suffering from acute kidney injury
CN116913502A (en) * 2023-05-15 2023-10-20 上海交通大学医学院附属第九人民医院 Method for predicting GO disease risk in GD patient based on time sequence data
CN117153380A (en) * 2023-11-01 2023-12-01 北京大学第一医院 Method, system and equipment for predicting postoperative acute kidney injury of non-cardiac surgery patient
CN117558464A (en) * 2024-01-12 2024-02-13 四川大学华西医院 Method for constructing ADR prediction model of elderly patient, prediction system and storage medium

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020102435A1 (en) * 2018-11-13 2020-05-22 Google Llc Prediction of future adverse health events using neural networks by pre-processing input sequences to include presence features
EP3909054A4 (en) * 2019-01-11 2022-10-26 Quadrus Medical Technologies, Inc. Systems and methods for assessing and evaluating renal health diagnosis, staging, and therapy recommendation
KR20240023874A (en) * 2022-08-16 2024-02-23 가톨릭대학교 산학협력단 Method for predicting the occurrence of postoperative acute kidney injury and system thereof

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109872819A (en) * 2019-01-30 2019-06-11 杭州脉兴医疗科技有限公司 A kind of acute kidney injury incidence rate forecasting system based on Intensive Care Therapy detection
CN110459328A (en) * 2019-07-05 2019-11-15 梁俊 A kind of Clinical Decision Support Systems for assessing sudden cardiac arrest
KR20210152047A (en) * 2020-06-05 2021-12-14 인하대학교 산학협력단 Apparatus for predicting acute kidney injury using machine learning based on artificial intelligence and method thereof
CN112837819A (en) * 2021-01-20 2021-05-25 尹丽君 Method for establishing acute kidney injury prediction model after coronary artery bypass grafting
KR20220132397A (en) * 2021-03-23 2022-09-30 서울대학교병원 System and method for supporting clinical decision for acute kidney injury using recurrent neural network model
CN115171881A (en) * 2022-07-08 2022-10-11 陕西霖桥实业有限公司 Acute kidney injury prediction method based on full-connection feature fusion convolution neural network
CN115410716A (en) * 2022-08-17 2022-11-29 四川大学华西医院 Method and system for establishing neural function prognosis prediction model of cerebral hemorrhage patient
CN115762783A (en) * 2022-11-22 2023-03-07 深圳大学 Acute kidney injury prediction system
CN116030976A (en) * 2022-12-20 2023-04-28 浙江大学 Risk prediction system for critical patients suffering from acute kidney injury
CN116913502A (en) * 2023-05-15 2023-10-20 上海交通大学医学院附属第九人民医院 Method for predicting GO disease risk in GD patient based on time sequence data
CN117153380A (en) * 2023-11-01 2023-12-01 北京大学第一医院 Method, system and equipment for predicting postoperative acute kidney injury of non-cardiac surgery patient
CN117558464A (en) * 2024-01-12 2024-02-13 四川大学华西医院 Method for constructing ADR prediction model of elderly patient, prediction system and storage medium

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
A dual-channel deep learning approach to continuous prediction of acute kidney injury in the intensive care unit;Xuetong Kong，等;2023 IEEE International Conference on Bioinformatics and Biomedicine;20240118;4320-4327 *
住院患者急性肾损伤的风险预测与关键影响因素分析;张钦，等;中国生物医学工程学报;20191220;第38卷(第06期);702-710 *
基于多尺度卷积和长短期记忆神经网络的临床危重症疾病预警模型;王天罡，等;医学信息学杂志;20200425;第41卷(第04期);41-45 *
快速序贯器官衰竭评估评分筛检颅内感染致脓毒症的效力评价;王浩宇，等;西部医学;20161220;第28卷(第12期);1664-1668 *
神经电生理监测在重型颅脑损伤患者生存预后相关因素中的应用;陈钫，等;中国医学物理学杂志;20230925;第40卷(第09期);1151-1155 *
颅内动脉瘤再破裂风险评估和防治方法;伍聪，等;华西医学;20121214;第27卷(第12期);1906-1909 *

Also Published As

Publication number Publication date
CN117894466A (en) 2024-04-16

Similar Documents

Publication Publication Date Title
Yuan et al. The development an artificial intelligence algorithm for early sepsis diagnosis in the intensive care unit
US20220254493A1 (en) Chronic disease prediction system based on multi-task learning model
Hunter et al. Definition of osteoarthritis on MRI: results of a Delphi exercise
Basu How to conduct meta-analysis: a basic tutorial
Abd-Elrazek et al. Predicting length of stay in hospitals intensive care unit using general admission features
García-Gallo et al. A machine learning-based model for 1-year mortality prediction in patients admitted to an Intensive Care Unit with a diagnosis of sepsis
CN114023441A (en) Severe AKI early risk assessment model and device based on interpretable machine learning model and development method thereof
CN112183572B (en) Method and device for generating prediction model for predicting severity of pneumonia
JP7404581B1 (en) Chronic nephropathy subtype mining system based on self-supervised graph clustering
CN113057587A (en) Disease early warning method and device, electronic equipment and storage medium
CN113128654A (en) Improved random forest model for coronary heart disease pre-diagnosis and pre-diagnosis system thereof
Mohammadi et al. Learning to identify patients at risk of uncontrolled hypertension using electronic health records data
Qin et al. Improving early sepsis prediction with multi modal learning
CN116580847B (en) Method and system for predicting prognosis of septic shock
CN116884631B (en) Comprehensive liver failure prediction and treatment reference system based on AI and similar patient analysis
CN117894466B (en) Continuous acute nephritis risk prediction method and system based on recurrent neural network
CN115954102A (en) Artificial joint prosthesis peripheral infection diagnosis model and diagnosis system
CN113782197B (en) New coronary pneumonia patient outcome prediction method based on interpretable machine learning algorithm
CN115602317A (en) Seven-point blood glucose prediction model construction method and device based on deep learning
Feng Support vector machine for stroke risk prediction
Zhao et al. Development and application of an intensive care medical data set for deep learning
Chan et al. Investigation of diabetic microvascular complications using data mining techniques
Panigrahy et al. Predictive Modelling of Diabetes Complications: Insights from Binary Classifier on Chronic Diabetic Mellitus
Kennedy Machine Learning in Thoracic Aortic Aneurysm Risk Assessment
Chen et al. Preoperative risk prediction of heart failure with numerical and textual attributes

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant