CN114937486A - Construction method and application of IDH prediction and intervention measure recommendation multitask model - Google Patents
Construction method and application of IDH prediction and intervention measure recommendation multitask model Download PDFInfo
- Publication number
- CN114937486A CN114937486A CN202210711223.3A CN202210711223A CN114937486A CN 114937486 A CN114937486 A CN 114937486A CN 202210711223 A CN202210711223 A CN 202210711223A CN 114937486 A CN114937486 A CN 114937486A
- Authority
- CN
- China
- Prior art keywords
- idh
- data
- layer
- task
- output
- 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.)
- Granted
Links
- 238000010276 construction Methods 0.000 title abstract description 4
- 208000020372 Infective dermatitis associated with HTLV-1 Diseases 0.000 claims abstract description 221
- 210000002569 neuron Anatomy 0.000 claims abstract description 87
- 238000001631 haemodialysis Methods 0.000 claims abstract description 80
- 230000000322 hemodialysis Effects 0.000 claims abstract description 80
- 238000012360 testing method Methods 0.000 claims abstract description 61
- 238000000034 method Methods 0.000 claims abstract description 43
- 238000012549 training Methods 0.000 claims abstract description 31
- 238000011282 treatment Methods 0.000 claims abstract description 31
- 230000008569 process Effects 0.000 claims abstract description 23
- 208000001953 Hypotension Diseases 0.000 claims abstract description 8
- 208000012866 low blood pressure Diseases 0.000 claims abstract description 5
- 238000000502 dialysis Methods 0.000 claims description 109
- 210000004369 blood Anatomy 0.000 claims description 57
- 239000008280 blood Substances 0.000 claims description 57
- 238000000108 ultra-filtration Methods 0.000 claims description 31
- 230000006870 function Effects 0.000 claims description 29
- 210000002700 urine Anatomy 0.000 claims description 28
- 238000001802 infusion Methods 0.000 claims description 25
- 238000011038 discontinuous diafiltration by volume reduction Methods 0.000 claims description 21
- 230000035515 penetration Effects 0.000 claims description 21
- 230000003827 upregulation Effects 0.000 claims description 20
- 238000004364 calculation method Methods 0.000 claims description 19
- 230000035488 systolic blood pressure Effects 0.000 claims description 18
- DDRJAANPRJIHGJ-UHFFFAOYSA-N creatinine Chemical compound CN1CC(=O)NC1=N DDRJAANPRJIHGJ-UHFFFAOYSA-N 0.000 claims description 16
- FAPWRFPIFSIZLT-UHFFFAOYSA-M Sodium chloride Chemical compound [Na+].[Cl-] FAPWRFPIFSIZLT-UHFFFAOYSA-M 0.000 claims description 15
- 238000003491 array Methods 0.000 claims description 15
- 238000012423 maintenance Methods 0.000 claims description 12
- 238000012544 monitoring process Methods 0.000 claims description 10
- 239000002504 physiological saline solution Substances 0.000 claims description 10
- 102000009027 Albumins Human genes 0.000 claims description 8
- 108010088751 Albumins Proteins 0.000 claims description 8
- BHPQYMZQTOCNFJ-UHFFFAOYSA-N Calcium cation Chemical compound [Ca+2] BHPQYMZQTOCNFJ-UHFFFAOYSA-N 0.000 claims description 8
- NPYPAHLBTDXSSS-UHFFFAOYSA-N Potassium ion Chemical compound [K+] NPYPAHLBTDXSSS-UHFFFAOYSA-N 0.000 claims description 8
- FKNQFGJONOIPTF-UHFFFAOYSA-N Sodium cation Chemical compound [Na+] FKNQFGJONOIPTF-UHFFFAOYSA-N 0.000 claims description 8
- 230000017531 blood circulation Effects 0.000 claims description 8
- 230000037396 body weight Effects 0.000 claims description 8
- 229910001424 calcium ion Inorganic materials 0.000 claims description 8
- HVYWMOMLDIMFJA-DPAQBDIFSA-N cholesterol Chemical compound C1C=C2C[C@@H](O)CC[C@]2(C)[C@@H]2[C@@H]1[C@@H]1CC[C@H]([C@H](C)CCCC(C)C)[C@@]1(C)CC2 HVYWMOMLDIMFJA-DPAQBDIFSA-N 0.000 claims description 8
- 229940109239 creatinine Drugs 0.000 claims description 8
- 238000003745 diagnosis Methods 0.000 claims description 8
- 230000035487 diastolic blood pressure Effects 0.000 claims description 8
- 210000003743 erythrocyte Anatomy 0.000 claims description 8
- 210000000265 leukocyte Anatomy 0.000 claims description 8
- 229910001414 potassium ion Inorganic materials 0.000 claims description 8
- 229910001415 sodium ion Inorganic materials 0.000 claims description 8
- 230000036387 respiratory rate Effects 0.000 claims description 7
- 206010020772 Hypertension Diseases 0.000 claims description 6
- 238000007689 inspection Methods 0.000 claims description 6
- WQZGKKKJIJFFOK-GASJEMHNSA-N Glucose Natural products OC[C@H]1OC(O)[C@H](O)[C@@H](O)[C@@H]1O WQZGKKKJIJFFOK-GASJEMHNSA-N 0.000 claims description 5
- 230000004872 arterial blood pressure Effects 0.000 claims description 5
- 206010012601 diabetes mellitus Diseases 0.000 claims description 5
- 239000008103 glucose Substances 0.000 claims description 5
- OYPRJOBELJOOCE-UHFFFAOYSA-N Calcium Chemical compound [Ca] OYPRJOBELJOOCE-UHFFFAOYSA-N 0.000 claims description 4
- 102000001554 Hemoglobins Human genes 0.000 claims description 4
- 108010054147 Hemoglobins Proteins 0.000 claims description 4
- DGAQECJNVWCQMB-PUAWFVPOSA-M Ilexoside XXIX Chemical compound C[C@@H]1CC[C@@]2(CC[C@@]3(C(=CC[C@H]4[C@]3(CC[C@@H]5[C@@]4(CC[C@@H](C5(C)C)OS(=O)(=O)[O-])C)C)[C@@H]2[C@]1(C)O)C)C(=O)O[C@H]6[C@@H]([C@H]([C@@H]([C@H](O6)CO)O)O)O.[Na+] DGAQECJNVWCQMB-PUAWFVPOSA-M 0.000 claims description 4
- OAICVXFJPJFONN-UHFFFAOYSA-N Phosphorus Chemical compound [P] OAICVXFJPJFONN-UHFFFAOYSA-N 0.000 claims description 4
- ZLMJMSJWJFRBEC-UHFFFAOYSA-N Potassium Chemical compound [K] ZLMJMSJWJFRBEC-UHFFFAOYSA-N 0.000 claims description 4
- LEHOTFFKMJEONL-UHFFFAOYSA-N Uric Acid Chemical compound N1C(=O)NC(=O)C2=C1NC(=O)N2 LEHOTFFKMJEONL-UHFFFAOYSA-N 0.000 claims description 4
- TVWHNULVHGKJHS-UHFFFAOYSA-N Uric acid Natural products N1C(=O)NC(=O)C2NC(=O)NC21 TVWHNULVHGKJHS-UHFFFAOYSA-N 0.000 claims description 4
- PNNCWTXUWKENPE-UHFFFAOYSA-N [N].NC(N)=O Chemical compound [N].NC(N)=O PNNCWTXUWKENPE-UHFFFAOYSA-N 0.000 claims description 4
- 230000009471 action Effects 0.000 claims description 4
- 239000003146 anticoagulant agent Substances 0.000 claims description 4
- 229940127219 anticoagulant drug Drugs 0.000 claims description 4
- 238000004820 blood count Methods 0.000 claims description 4
- 230000036760 body temperature Effects 0.000 claims description 4
- 239000011575 calcium Substances 0.000 claims description 4
- 229910052791 calcium Inorganic materials 0.000 claims description 4
- 239000011574 phosphorus Substances 0.000 claims description 4
- 229910052698 phosphorus Inorganic materials 0.000 claims description 4
- 239000011591 potassium Substances 0.000 claims description 4
- 229910052700 potassium Inorganic materials 0.000 claims description 4
- 102000004169 proteins and genes Human genes 0.000 claims description 4
- 108090000623 proteins and genes Proteins 0.000 claims description 4
- 239000011734 sodium Substances 0.000 claims description 4
- 229910052708 sodium Inorganic materials 0.000 claims description 4
- 239000000725 suspension Substances 0.000 claims description 4
- UFTFJSFQGQCHQW-UHFFFAOYSA-N triformin Chemical compound O=COCC(OC=O)COC=O UFTFJSFQGQCHQW-UHFFFAOYSA-N 0.000 claims description 4
- 229940116269 uric acid Drugs 0.000 claims description 4
- 230000004584 weight gain Effects 0.000 claims description 4
- 235000019786 weight gain Nutrition 0.000 claims description 4
- VEXZGXHMUGYJMC-UHFFFAOYSA-M Chloride anion Chemical compound [Cl-] VEXZGXHMUGYJMC-UHFFFAOYSA-M 0.000 claims description 3
- 235000012000 cholesterol Nutrition 0.000 claims description 3
- 230000000694 effects Effects 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000002615 hemofiltration Methods 0.000 claims description 3
- 238000000465 moulding Methods 0.000 claims description 3
- 239000011780 sodium chloride Substances 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 2
- 238000001035 drying Methods 0.000 claims 1
- 238000004393 prognosis Methods 0.000 abstract description 3
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 230000001154 acute effect Effects 0.000 description 5
- 208000019901 Anxiety disease Diseases 0.000 description 4
- 206010008479 Chest Pain Diseases 0.000 description 4
- 206010009866 Cold sweat Diseases 0.000 description 4
- 206010013954 Dysphoria Diseases 0.000 description 4
- 208000000059 Dyspnea Diseases 0.000 description 4
- 206010013975 Dyspnoeas Diseases 0.000 description 4
- 208000007101 Muscle Cramp Diseases 0.000 description 4
- 206010028813 Nausea Diseases 0.000 description 4
- 208000005392 Spasm Diseases 0.000 description 4
- 208000003443 Unconsciousness Diseases 0.000 description 4
- 206010047700 Vomiting Diseases 0.000 description 4
- 206010000059 abdominal discomfort Diseases 0.000 description 4
- 230000036506 anxiety Effects 0.000 description 4
- 230000007211 cardiovascular event Effects 0.000 description 4
- 208000002173 dizziness Diseases 0.000 description 4
- 230000008693 nausea Effects 0.000 description 4
- 230000001052 transient effect Effects 0.000 description 4
- 230000008673 vomiting Effects 0.000 description 4
- 208000037157 Azotemia Diseases 0.000 description 3
- 241001282135 Poromitra oscitans Species 0.000 description 3
- 208000012886 Vertigo Diseases 0.000 description 3
- 206010048232 Yawning Diseases 0.000 description 3
- 230000036543 hypotension Effects 0.000 description 3
- 208000009852 uremia Diseases 0.000 description 3
- 231100000889 vertigo Toxicity 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 230000036772 blood pressure Effects 0.000 description 2
- 239000006185 dispersion Substances 0.000 description 2
- 239000003792 electrolyte Substances 0.000 description 2
- 239000012510 hollow fiber Substances 0.000 description 2
- 238000001727 in vivo Methods 0.000 description 2
- 210000004303 peritoneum Anatomy 0.000 description 2
- 230000001105 regulatory effect Effects 0.000 description 2
- 208000009304 Acute Kidney Injury Diseases 0.000 description 1
- 208000012639 Balance disease Diseases 0.000 description 1
- ZAMOUSCENKQFHK-UHFFFAOYSA-N Chlorine atom Chemical compound [Cl] ZAMOUSCENKQFHK-UHFFFAOYSA-N 0.000 description 1
- 206010018367 Glomerulonephritis chronic Diseases 0.000 description 1
- 208000003351 Melanosis Diseases 0.000 description 1
- 208000033626 Renal failure acute Diseases 0.000 description 1
- 201000011040 acute kidney failure Diseases 0.000 description 1
- 208000012998 acute renal failure Diseases 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 239000000460 chlorine Substances 0.000 description 1
- 229910052801 chlorine Inorganic materials 0.000 description 1
- 208000020832 chronic kidney disease Diseases 0.000 description 1
- 208000022831 chronic renal failure syndrome Diseases 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000000385 dialysis solution Substances 0.000 description 1
- 239000008151 electrolyte solution Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 210000003734 kidney Anatomy 0.000 description 1
- 210000004379 membrane Anatomy 0.000 description 1
- 239000012528 membrane Substances 0.000 description 1
- 230000002503 metabolic effect Effects 0.000 description 1
- 239000002207 metabolite Substances 0.000 description 1
- 210000003200 peritoneal cavity Anatomy 0.000 description 1
- 231100000614 poison Toxicity 0.000 description 1
- 238000010837 poor prognosis Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000036391 respiratory frequency Effects 0.000 description 1
- 238000001179 sorption measurement Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 239000003440 toxic substance Substances 0.000 description 1
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
- G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
- G16H20/40—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to mechanical, radiation or invasive therapies, e.g. surgery, laser therapy, dialysis or acupuncture
-
- 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
-
- 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
- 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
- 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)
- Medical Informatics (AREA)
- Public Health (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Pathology (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Urology & Nephrology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Surgery (AREA)
- External Artificial Organs (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The invention discloses a construction method and application of an IDH prediction and intervention measure recommendation multitask model, belonging to the field of artificial intelligence of the medical industry and comprising the following steps: s1, collecting time-invariant data and time-invariant data before hemodialysis to form input data; s2 setting labels for each piece of input data, wherein each label corresponds to a learning task; s3, constructing a mid-low blood pressure prediction and intervention measure recommendation multi-task model, wherein the model comprises an auxiliary task X, an auxiliary task Y and a main task Z; the model architecture of each task is divided into an input layer, a hidden layer and an output layer, and each layer is composed of a plurality of neurons; s4, constructing training set data and test set data to train the model; s5, verifying the multitask model; the invention reduces the risk of IDH of the patient by means of advance prediction and advance intervention, thereby achieving the purposes of preventing and treating IDH and improving the prognosis of MHD patients in the MHD treatment process.
Description
Technical Field
The invention relates to a construction method of a medical prediction model, belongs to the field of artificial intelligence in the medical industry, and particularly relates to a method for constructing a maintenance hemodialysis mid-low blood pressure prediction and intervention measure recommendation model by adopting a multi-task learning method and application.
Background
The maintenance hemodialysis treatment is a transitional method for prolonging the life of patients with uremia, which means to save the life of patients by hemodialysis or peritoneal dialysis. Patients undergoing maintenance hemodialysis not only include uremia due to the development of chronic nephritis, but also other uremia due to diabetes and hypertension are common causes of patient undergoing maintenance dialysis. Maintenance dialysis treatments are divided into two categories, hemodialysis and peritoneal dialysis:
hemodialysis (HD) is one of the alternative treatments for the kidney of patients with acute and chronic renal failure. The method comprises the steps of draining blood in vivo to the outside of a human body, enabling the blood and electrolyte solution (dialysate) with similar body concentration to be inside and outside one hollow fiber through a dialyzer consisting of a plurality of hollow fibers, and performing substance exchange through dispersion, ultrafiltration, adsorption and convection principles to remove metabolic wastes in the human body and maintain electrolyte and acid-base balance; at the same time, the excess water in the body is removed, and the purified blood is returned to the body.
Peritoneal Dialysis (PD) is to regularly and regularly inject prepared dialysate into the peritoneal cavity of a patient through a catheter by utilizing the characteristic that the peritoneum is used as a semi-permeable membrane through the action of gravity, and due to the concentration gradient difference of solutes on two sides of the peritoneum, the solute on one side with high concentration moves to one side with low concentration (dispersion action); the water moves from the hypotonic side to the hypertonic side (osmosis). The peritoneal dialysis solution is continuously replaced to achieve the purposes of removing in vivo metabolites and toxic substances and correcting water and electrolyte balance disorder.
In dialysis, when the systolic pressure of a patient is reduced by more than or equal to 20mmHg or the mean arterial pressure is reduced by more than or equal to 10mmHg, the patient can have dizziness, vertigo, dysphoria, anxiety, pale complexion, yawning, nausea, vomiting, chest distress, increased heart rate, abdominal discomfort and cold sweat, and serious patients can have dyspnea, melanism, muscle spasm and even transient consciousness loss, and acute cardiovascular events can be caused and the death risk can be increased when the serious patients are serious. Research shows that the incidence rate of IDH in MHD treatment patients in China is about 39%, and the frequent occurrence of IDH is an important factor for poor prognosis of MHD patients, so that the prevention and treatment of IDH in the MHD treatment process are very important.
Disclosure of Invention
In order to solve the technical problem, the invention provides a method for constructing a maintenance hemodialysis hypotension prediction and intervention measure recommendation multitask model, which comprises the following steps of:
s1 collecting input data; collecting time-invariant data before hemodialysis and time-variant data at each moment during hemodialysis as input data; each time-invariant data and each time-variant data jointly form an input data;
s2, setting a label for each piece of input data; setting 3 labels for each piece of input data, namely label A, label B and label C; each label corresponds to a learning task;
s3, constructing a mid-low blood pressure prediction and intervention measure recommendation multitask model; the multi-task model comprises an auxiliary task X, an auxiliary task Y and a main task Z; the model architecture of each task is divided into an input layer, a hidden layer and an output layer, and each layer is composed of a plurality of neurons; each neuron in the hidden layer and the output layer has a weight, and the weight is obtained in the process of model training; each neuron of the output layer has different weight calculation on the data output by the hidden layer, the final result is output, and the number of neurons of the output layer is determined by the number of label categories;
s4, training a model; constructing training set data and test set data from the input data collected in the step S1, constructing a loss function, inputting the training set data into a model, and training the model;
s5, verifying the multitask model; and the multitask model calculates the test set data obtained in the step S4 to obtain results of three tasks of the test set data, respectively calculates the accuracy, the recall rate and the accuracy rate of each task, evaluates the effect of the multitask model according to the three indexes, and obtains a qualified IDH prediction and intervention measure recommendation multitask model when the accuracy, the recall rate and the accuracy rate of each task reach preset values.
Further, in the above-mentioned case,
the time invariant data at step S1 includes: the measured physical sign data before hemodialysis of the patient, personal information, medical history records, inspection results, historical dialysis records and current hemodialysis prescriptions; wherein,
the measurement sign data comprises: body weight before penetration, heart rate before penetration, body weight before penetration, pulse before penetration, systolic pressure before penetration, diastolic pressure before penetration, and respiratory rate before penetration;
the personal information includes: age, sex, height, age under dialysis;
the medical history record comprises: history of hypertension, diabetes;
the inspection result comprises: blood leukocyte count, hemoglobin, red blood cell count, blood glucose, blood albumin, blood total cholesterol, blood triglyceride, blood creatinine, blood uric acid, blood urea nitrogen, blood potassium, blood sodium, blood calcium, blood phosphorus, blood chloride, urine leukocyte, urine protein, urine erythrocyte, urine creatinine, urine occult blood, urine microalbumin, urine albumin;
the historical dialysis records include: weight gain during dialysis, length of last dialysis, last off-machine weight, frequency of IDH occurring in nearly seven days, and frequency of IDH occurring in nearly thirty days;
when the hemodialysis prescription includes: dialysis mode, anticoagulant dosage, ultrafiltration volume, dialysis duration, dialysate potassium ion concentration, dialysate calcium ion concentration, dialysate sodium ion concentration, dialysate conductivity, and blood flow volume;
the time-varying data includes: during the hemodialysis of a patient, monitoring physical signs, dialysis treatment parameters and dialysis machine parameters at fixed time intervals; wherein,
the interval monitoring sign data comprises: current body temperature, current heart rate, current systolic pressure, current diastolic pressure, current pulse and current respiratory rate;
dialysis treatment parameters include: current ultrafiltration, current dialysis duration, current dialysate potassium ion concentration, current dialysate calcium ion concentration, current dialysate sodium ion concentration, current dialysate conductivity, and current blood flow;
dialysis machine parameters include: current arterial pressure, current venous pressure, current trans-molding pressure, current amount of hemofiltration.
Further, step S2 includes the following sub-steps:
s21, setting the diagnostic criteria for IDH:
firstly, the intervention measures of IDH are met, and the systolic pressure is reduced by more than 20mmHg compared with the systolic pressure before permeation;
no dry measure is taken, but the systolic pressure is less than 90 mmHg;
s22, setting a label A, a label B and a label C for each piece of input data according to the IDH diagnosis standard;
the label A is: whether IDH occurs in the dialysis or not;
label B is: whether IDH occurs at the next moment;
label C is: intervention measures of IDH at the next moment; wherein the intervention measure of IDH comprises the following steps: no intervention, suspending ultrafiltration, reducing volume, normal saline infusion, and increasing conductivity by 5 categories.
Further, step S22 includes the following sub-steps:
s221, collecting IDH intervention data in dialysis, wherein the IDH intervention data is an intervention measure of the input data in the dialysis process; the intervention measures of the IDH intervention data are also as follows: 5 categories of no intervention, suspension of ultrafiltration, volume reduction, normal saline infusion and conductivity up-regulation;
s222, setting a label A; judging whether each moment is IDH according to the diagnostic standard of IDH for the input data of all moments of the dialysis;
s223, setting a label B; judging whether each moment is IDH according to the IDH diagnosis standard for the input data of all moments of the dialysis; the label B at the previous moment of the moment which is judged to be the IDH is marked as the next moment to generate the IDH, and the label B at the rest moments are marked as the next moment to not generate the IDH;
s224, setting a label C; and setting an input data acquisition label C according to the intervention measure data of the IDH.
Further, in step S3, it is predicted whether IDH occurs in the present dialysis as an auxiliary task X; predicting the intervention measure of the IDH at the next moment as an auxiliary task Y; predicting whether IDH occurs at the next moment to be a main task Z;
the model architecture of the auxiliary task X is divided into an input layer X1, a hidden layer X2 and an output layer X3, each layer is composed of a plurality of neurons, each neuron in the hidden layer X2 and the output layer X3 has a weight, and the weight is obtained in the model training process;
1. the auxiliary task X model comprises the following structure:
(1) an input layer: the data input by the input layer X1 of the auxiliary task X is time-invariant data before hemodialysis, and the number of the input layer X1 neurons is the number of the time-invariant data before hemodialysis;
(2) hiding the layer: each neuron of the hidden layer X2 has different weight calculation on the data input by the input layer X1, so that the prediction of a certain task label is favored; in the invention, the weight value of the hidden layer X2 of the auxiliary task X is biased to predict whether IDH occurs in the dialysis;
setting:
the output array of the input layer X1 of the auxiliary task X is X1 i I is an output arrayThe number of the medium numerical values, i, is 1-n;
the neuron of the hidden layer X2 comprises a plurality of weight parameter arrays W i The number of the weight parameters W is i, and i is 1-n;
the number of the weight parameters W is equal to the number i of the numerical values in the output array;
if the output data array of the hidden layer X2 is X2i, then:
(3) an output layer: one neuron of the output layer X3 is provided, 0 is that IDH does not occur in the dialysis, and 1 is that IDH occurs in the dialysis; the neurons of the output layer X3 perform different weight calculation on the data output by the hidden layer, and output the final result;
setting:
the neurons of the output layer X3 include several weight parameter arrays W j The number of the weight parameters W is j, and j is 1-n;
weight parameter array W j Number and weight parameter array W i Equal number, i.e. i = j
Then, output = Sigmoid (of output layer X3)
)；
The model of auxiliary task Y includes the following structure:
the model architecture of the auxiliary task Y is divided into an input layer Y1, a hidden layer Y2 and an output layer Y3, each layer is composed of a plurality of neurons, each neuron in the hidden layer Y2 and the output layer Y3 has a weight, and the weight is obtained in the model training process;
(1) an input layer: the data input to the input layer Y1 of the auxiliary task Y is the same as the data input to the input layer of the main task Z1, and is time-invariant data before hemodialysis and time-variant data at each time during hemodialysis, and the number of neurons in the input layer Y1 is the sum of the number of time-invariant data before hemodialysis and the number of time-variant data at each time during hemodialysis;
(2) hiding the layer: each neuron of the hidden layer Y2 has different weight calculation on the data input by the input layer Y1, so that the prediction of a certain task label is favored;
setting:
the output array of the input layer Y1 of the auxiliary task Y is Y1 i I is the number of numerical values in the output array, and i is 1-n;
the neuron of the hidden layer Y2 comprises a plurality of weight parameter arrays W i The number of the weight parameters W is i, and i is 1-n;
the number of the weight parameters W is equal to the number i of the numerical values in the output array;
let the output data array of the hidden layer Y2 be Y2 i Then:
in the invention, the weight value of the hidden layer Y2 of the auxiliary task Y is biased to predict the intervention measure of the next moment IDH;
(3) an output layer: five output layer Y3 neurons correspond to the intervention measures of IDH at the next moment such as no intervention, suspended ultrafiltration, volume reduction, normal saline infusion, conductivity up-regulation and the like respectively;
the neurons of the output layer Y3 have different weight calculations on the data output by the hidden layer, and output the final result;
setting:
the neurons of the output layer Y3 comprise a plurality of weight parameter arrays W j The number of the weight parameters W is j, and j is 1-n;
weight parameter array W j Number and weight parameter array W i Equal number, i.e. i = j
Then, output = Sigmoid (of output layer Y3)
)；
3. The main task Z model comprises the following structure:
the model architecture of the main task Z is divided into an input layer Z1, a hidden layer Z2 and an output layer Z3, each layer is composed of a plurality of neurons, each neuron in the hidden layer Z2 and the output layer Z3 has a weight, and the weight is obtained in the model training process;
(1) an input layer: the data input to the input layer of the main task Z1 is the same as the data input to the input layer Y1 of the support task Y, and is time-invariant data before hemodialysis and time-variant data at each time during hemodialysis, and the number of neurons in the input layer Z1 is the sum of the number of time-invariant data before hemodialysis and the number of time-variant data at each time during hemodialysis;
(2) hiding the layer: each neuron of the hidden layer Z2 has different weight calculation on the data input by the input layer Z1, so that the prediction of a certain task label is favored, and in the invention, the weight value of the hidden layer Z2 of the main task Z is favored to predict whether IDH occurs at the next moment;
(3) an output layer: the output layer Z3 neurons of the main task Z have one, which is: the next moment does not generate IDH, which is indicated by '0', or the next moment generates IDH, which is indicated by '1';
taking the output array of the hidden layer X2 of the auxiliary task X, the output array of the hidden layer Y2 of the auxiliary task Y and the output array of the hidden layer Z2 of the main task Z as the input of the output layer Z3 of the main task Z;
setting:
the output array of the hidden layer X2 of the auxiliary task X is X2 i I is the number of numerical values in the output array, and i is 1-n;
output array Y2 of hidden layer Y2 of auxiliary task Y i I is the number of numerical values in the output array, and i is 1-n;
the output array of the hidden layer Z2 with the main task Z is Z2 i I is the number of numerical values in the output array, and i is 1-n;
the numbers of numerical values in an output array of a hidden layer X2 of an auxiliary task X, an output array of a hidden layer Y2 of the auxiliary task Y and an output array of a hidden layer Z2 of a main task Z are the same, and are i, wherein i is 1-n;
the neurons of the main task Z output layer Z3 comprise a plurality of weight parameter arrays W i The number of the weight parameters W is i, and i is 1-n;
the number of the weight parameters W is equal to the number i of the numerical values in the output array;
array the output as X2 i And an output array Y2 i Output array Z2 i Adding the values in the same sequence to obtain the input array Q of the output layer Z3 of the main task Z i ；
The output of the primary task Z = Sigmoid ().
Further, step S4 includes the following sub-steps:
s41, constructing training set data and test set data;
sorting the input data collected in the step S1 according to dialysis time, wherein the first 80% is taken as training set data, and the second 20% is taken as test set data;
s42, constructing a loss function;
firstly, constructing a loss function by adopting cross entropy for an auxiliary task X, an auxiliary task Y and a main task Z;
then, constructing a model total loss function, wherein the model total loss function is the weighted sum of the loss functions of an auxiliary task X, an auxiliary task Y and a main task Z; wherein,
the weight ratio of the cross entropy loss function of the main task Z, the auxiliary task X and the auxiliary task Y is 2: 1: 1;
s43, calculating the weight parameters of each corresponding neuron when the loss function takes the minimum value by adopting a gradient descent method for the training set data;
the neurons were those in the model constructed in S3.
Further, in step S5:
the accuracy of the auxiliary task A is the ratio of the number of IDH occurrences and IDH non-occurrences to the total number of test data, which are correctly predicted by a model in the test set data;
the recall rate of the auxiliary task A is the ratio of the number of IDH (inverse discrete cosine transform) occurrences correctly predicted by the model in the test set data to the total number of IDH occurrences in the test set data;
the accuracy rate of the auxiliary task A is the ratio of the number of the IDH correctly predicted by the model in the test set data to the total number of the IDH predicted by the model;
the accuracy of the auxiliary task B is the ratio of the sum of the number of model correct prediction non-intervention, suspended ultrafiltration, reduced volume, normal saline infusion and up-regulated conductivity to the total number of the test data in the test set data;
the recall rate of the auxiliary task B is the ratio of the number of model correctly predicting the total number of non-intervention, suspended ultrafiltration, volume reduction, physiological saline infusion and up-regulation conductivity in the test data to the total number of non-intervention, suspended ultrafiltration, volume reduction, physiological saline infusion and the total number of up-regulation conductivity in the test data;
the precision rate of the auxiliary task B is the ratio of the number of model correctly predicting the total number of non-intervention, suspended ultrafiltration, volume reduction, physiological saline infusion and up-regulation conductivity in the test set data to the total number of model correctly predicting the non-intervention, the total number of suspended ultrafiltration, the total number of volume reduction, the total number of physiological saline infusion and the total number of up-regulation conductivity;
the accuracy of the main task C is the ratio of the number of the IDH occurring at the next moment and the number of the IDH not occurring at the next moment in the test set data correctly predicted by the model to the total number of the test data;
the recall rate of the main task C is the ratio of the number of IDH occurrences at the next moment to the total number of IDH occurrences at the next moment in the test set data, which is correctly predicted by the model in the test set data;
the accuracy rate of the main task C is the ratio of the number of IDH generated at the next moment when the model correctly predicts the IDH in the test set data to the total number of IDH generated at the next moment when the model predicts the IDH.
Furthermore, the invention also provides an application of the IDH prediction and intervention measure recommendation multitask model constructed by the method of the invention in predicting IDH in maintenance hemodialysis treatment.
Has the advantages that: the invention recommends a multitask model based on a multitask learning method through the low and middle blood pressure prediction and intervention measures, and reduces the risk of IDH occurrence of a patient in a mode of prediction in advance and intervention in advance, thereby achieving the purposes of preventing and treating IDH and improving the prognosis of MHD patients in the MHD treatment process, and specifically comprises the following steps:
on one hand, whether IDH occurs in the dialysis can be predicted before hemodialysis of the patient, so that attention of medical care to the patient who possibly has IDH is improved.
On the other hand, the model of the invention can also be used for predicting whether IDH occurs at the next monitoring moment according to the physical signs and dialysis treatment parameters monitored during hemodialysis in the dialysis process.
Thirdly, the model of the invention can provide a high-quality IDH intervention scheme for medical staff and provide a high-quality treatment scheme recommendation, namely a high-quality IDH intervention measure recommendation for IDH occurring in hemodialysis, thereby solving the problems that the medical staff selects an improper treatment scheme when IDH occurs in hemodialysis, or cannot provide a treatment scheme in a short time, or the inexperienced medical staff does not know how to treat the patient, so that the patient has dizziness, dysphoria, anxiety, pale complexion, malade, nausea, vomiting, chest distress, rapid heart rate increase, abdominal discomfort and cold sweat, and severe patients can have dyspnea, blackness, muscle spasm and even transient consciousness loss, and acute cardiovascular events or even death can be caused in severe cases.
Fourthly, the output of the hidden layer of the auxiliary task is fused with the output of the hidden layer of the main task, and the fused result is used as the input of the output layer of the main task, so that the weight sharing of the main task and the auxiliary task is realized, and the accuracy of predicting whether IDH occurs at the next moment by the main task can be effectively improved by adopting the model of the invention.
Fifthly, the model can simultaneously realize the prediction of a plurality of tasks, and saves the storage cost and the maintenance cost of the model, the time cost and the calculation cost of the model prediction.
Drawings
FIG. 1 is a schematic structural diagram of a multi-task model for predicting and recommending intervention measures based on hypertension and hypotension.
Detailed Description
Example 1: the invention provides a method for constructing a maintenance hemodialysis hypotension prediction and intervention measure recommendation multitask model and application thereof, comprising the following steps of:
s1, collecting input data; collecting time-invariant data before hemodialysis and time-variant data at each moment during hemodialysis as input data; each time-invariant data and each time-variant data jointly form an input data;
the input data is at least 1;
(1) the time-invariant data includes: the measured physical sign data before hemodialysis, personal information, medical history records, inspection results, historical dialysis records and current hemodialysis prescriptions of the patient;
wherein ,
measuring the physical sign data comprises: body weight before penetration, heart rate before penetration, body weight before penetration, pulse before penetration, systolic pressure before penetration, diastolic pressure before penetration, and respiratory rate before penetration;
the personal information includes: age, sex, height, age under dialysis;
the medical history record comprises: history of hypertension, diabetes;
the inspection result comprises: blood leukocyte count, hemoglobin, red blood cell count, blood glucose, blood albumin, blood total cholesterol, blood triglyceride, blood creatinine, blood uric acid, blood urea nitrogen, blood potassium, blood sodium, blood calcium, blood phosphorus, blood chloride, urine leukocyte, urine protein, urine erythrocyte, urine creatinine, urine occult blood, urine microalbumin, urine albumin;
the historical dialysis records include: weight gain during dialysis, length of last dialysis, last off-machine weight, frequency of IDH occurring in nearly seven days, and frequency of IDH occurring in nearly thirty days;
when the hemodialysis prescription includes: dialysis mode, anticoagulant dosage, ultrafiltration volume, dialysis duration, dialysate potassium ion concentration, dialysate calcium ion concentration, dialysate sodium ion concentration, dialysate conductivity, and blood flow.
(2) The time-varying data includes: during the hemodialysis of a patient, monitoring physical signs, dialysis treatment parameters and dialysis machine parameters at fixed time intervals; wherein,
the interval monitoring sign data comprises: current body temperature, current heart rate, current systolic pressure, current diastolic pressure, current pulse and current respiratory rate;
dialysis treatment parameters include: current ultrafiltered, current dialysis duration, current dialysate potassium ion concentration, current dialysate calcium ion concentration, current dialysate sodium ion concentration, current dialysate conductivity, and current blood flow.
Dialysis machine parameters include: current arterial pressure, current venous pressure, current trans-molding pressure, current amount of hemofiltration.
S2, setting a label for each piece of input data; setting 3 labels for each piece of input data, namely label A, label B and label C; each label corresponds to a learning task;
the label A is: whether IDH occurs in the dialysis or not;
label B is: whether IDH occurs at the next moment;
label C is: intervention measures of IDH at the next moment; wherein the intervention measures of IDH include: 5 categories of no intervention, suspension of ultrafiltration, volume reduction, normal saline infusion and conductivity up-regulation;
s21, setting the diagnostic criteria for IDH:
firstly, the intervention measures of IDH are met, and the systolic pressure is reduced by more than 20mmHg compared with the systolic pressure before permeation;
no dry measure is taken, but the systolic pressure is less than 90 mmHg;
s22, setting a label A, a label B and a label C for each piece of input data according to the IDH diagnosis standard;
specifically, the method comprises the following substeps:
s221, collecting IDH intervention data in dialysis, wherein the IDH intervention data is an intervention measure of the input data in the dialysis process; the intervention measures of the IDH intervention data are also as follows: 5 categories of no intervention, suspension of ultrafiltration, volume reduction, normal saline infusion and conductivity up-regulation;
each piece of input data corresponds to one piece of IDH intervention data; the input data, the IDH intervention data, the label A, the label B and the label C form a complete data;
s222, setting a label A; judging whether each moment is IDH according to the IDH diagnosis standard for the input data of all moments of the dialysis;
the input data of all moments of the dialysis comprise N input data, wherein N is at least 1;
specifically, each moment of the dialysis corresponds to one input data, and N moments of the dialysis correspond to N input data; the time refers to a time point of a fixed time interval, and the fixed time interval comprises time values such as hours, minutes, seconds or milliseconds;
if IDH occurs at one moment, marking labels A of input data at all the moments of the dialysis as IDH, namely marking labels A of N pieces of input data of the dialysis as IDH;
if no IDH occurs at each moment, marking the label A of the input data of all the moments of the dialysis as the IDH not occurring; marking labels A of N input data of the dialysis as non-occurrence IDH;
further, the label A is represented by 0 or 1, 0 is no IDH and 1 is IDH; label B is represented by 0 or 1, 0 being the next time when no IDH occurs, and 1 being the next time when an IDH occurs.
S223, setting a label B; judging whether each moment is IDH according to the diagnostic standard of IDH for the input data of all moments of the dialysis; the label B at the previous moment of the moment which is judged to be the IDH is marked as the next moment to generate the IDH, and the label B at the rest moments are marked as the next moment to not generate the IDH;
similarly, the input data of the dialysis at all times comprises N pieces of input data, wherein N is at least 1;
specifically, each moment of the dialysis corresponds to one input data, and N moments of the dialysis correspond to N input data; the time refers to a time point of a fixed time interval, and the fixed time interval comprises hours, minutes, seconds, milliseconds and the like;
for example:
example 1: taking N =5 as an example, that is, the number of pieces of input data is set to 5; when the input data for generating IDH is N =4 as judged by the diagnostic criteria for IDH, label B in the input data of N =3 is labeled as: IDH occurs at the next moment; labels B for the remaining input data are labeled: no IDH occurs at the next moment;
example 2: taking N =5 as an example, that is, the number of pieces of input data is set to 5; setting input data for generating the IDH as N =4, and N =5, as judged by the diagnostic criteria of the IDH;
label B in the input data of N =3 and the input data of N =4 is labeled as: IDH occurs at the next moment; labels B for the remaining input data are labeled: no IDH occurs at the next moment;
the invention judges whether each moment is IDH according to the IDH diagnosis standard for the input data of all moments of the dialysis; the label B at the previous moment of the moment which is judged to be IDH is marked as the next moment when IDH occurs, and the label B at the rest moments are marked as the next moment when IDH does not occur, so that the problem that whether IDH occurs at the next moment in the dialysis process or not is solved, and the problem that the IDH risk is predicted in real time in the hemodialysis process is solved.
S224, setting a label C; setting an input data acquisition label C according to the intervention measure data of the IDH;
further, label C is represented by a one-hot code, five digits are used to encode five interventions; example (c):
10000 means no intervention;
01000 denotes pausing ultrafiltration;
00100 denotes volume reduction;
00010 denotes saline infusion;
00001 represents the upturn conductivity.
Further, in the above-mentioned case,
firstly, the next input data of the input data with the label B marked as 1 of each input data is judged according to the intervention measure of the IDH, then the label C of the input data with the label B marked as 1 is marked as the single-hot code of the corresponding intervention measure, and the labels C at the other moments are marked as the single-hot codes without intervention.
For example:
m pieces of data are set, and M = 6;
and marking the label B in the 4 th input data as 1, judging which intervention measure the 5 th input data belongs to according to the intervention measure of the IDH, marking the label C of the 4 th input data as the single-hot code of the corresponding intervention measure, and marking the labels C at the rest moments as the single-hot codes without intervention.
The input data acquisition label C is set according to the intervention measure data of the IDH, so that a high-quality IDH intervention scheme can be provided for medical staff, and a treatment scheme is provided for IDH occurring in hemodialysis, so that the problems that the medical staff selects an inappropriate treatment scheme or cannot give a treatment scheme in a short time or the inexperienced medical staff does not know how to treat the patient, such as dizziness, vertigo, dysphoria, anxiety, pale complexion, yawning, nausea, vomiting, chest distress, rapid heart rate increase, abdominal discomfort and cold sweat of the patient, and severe patients can have dyspnea, blackness, muscle spasm and even transient consciousness loss, and acute cardiovascular events and even death can be caused in severe cases are solved.
Meanwhile, the invention solves the problem of providing an IDH intervention scheme with high prediction quality by firstly judging which intervention measure belongs to next input data of the input data with the label B marked as 1 according to the intervention measure of the IDH and then marking the label C of the input data with the label B marked as 1 as the corresponding intervention measure in a single-hot coding mode.
S3, constructing a mid-low blood pressure prediction and intervention measure recommendation multitask model;
the multi-task model comprises an auxiliary task X, an auxiliary task Y and a main task Z;
as shown in fig. 1, the model architecture of each task is divided into an input layer, a hidden layer and an output layer, and each layer is composed of a plurality of neurons; each neuron in the hidden layer and the output layer has a weight, and the weight is obtained in the process of model training; each neuron of the output layer has different weight calculation on the data output by the hidden layer, the final result is output, and the number of neurons of the output layer is determined by the number of label categories;
setting:
predicting whether IDH occurs in the dialysis or not as an auxiliary task X;
predicting the intervention measure of the IDH at the next moment as an auxiliary task Y;
predicting whether IDH occurs at the next moment to be a main task Z;
then:
1. the model architecture of the auxiliary task X is divided into an input layer X1, a hidden layer X2 and an output layer X3, each layer is composed of a plurality of neurons, each neuron in the hidden layer X2 and the output layer X3 has a weight, and the weight is obtained in the model training process;
(1) an input layer: the data input by the input layer X1 of the auxiliary task X is time-invariant data before hemodialysis, and the number of the input layer X1 neurons is the number of the time-invariant data before hemodialysis;
(2) hiding the layer: each neuron of the hidden layer X2 has different weight calculation on the data input by the input layer X1, so that the prediction of a certain task label is more favored; in the invention, the weight value of the hidden layer X2 of the auxiliary task X is biased to predict whether IDH occurs in the dialysis;
setting:
the output array of the input layer X1 of the auxiliary task X is X1 i I is the number of numerical values in the output array, and i is 1-n;
the neuron of the hidden layer X2 comprises a plurality of weight parameter arrays W i The number of the weight parameters W is i, and i is 1-n;
the number of the weight parameters W is equal to the number i of the numerical values in the output array;
if the output data array of the hidden layer X2 is X2i, then:
(3) an output layer: one neuron of the output layer X3 is provided, 0 is that IDH does not occur in the dialysis, and 1 is that IDH occurs in the dialysis; the neurons of the output layer X3 perform different weight calculation on the data output by the hidden layer, and output the final result;
setting:
the neurons of the output layer X3 include several weight parameter arrays W j The number of the weight parameters W is j, and j is 1-n;
weight parameter array W j Number and weight parameter array W i Equal in number, i.e. i = j
Then, the output of output layer X3 = Sigmoid (/) (
)。
2. The model architecture of the auxiliary task Y is divided into an input layer Y1, a hidden layer Y2 and an output layer Y3, each layer is composed of a plurality of neurons, each neuron in the hidden layer Y2 and the output layer Y3 has a weight, and the weight is obtained in the model training process;
(1) an input layer: the data input to the input layer Y1 of the auxiliary task Y is the same as the data input to the input layer of the main task Z1, and is time-invariant data before hemodialysis and time-variant data at each time during hemodialysis, and the number of neurons in the input layer Y1 is the sum of the number of time-invariant data before hemodialysis and the number of time-variant data at each time during hemodialysis;
(2) hiding the layer: each neuron of the hidden layer Y2 has different weight calculation on the data input by the input layer Y1, so that the prediction of a certain task label is favored;
setting:
the output array of the input layer Y1 of the auxiliary task Y is Y1 i I is the number of numerical values in the output array, and i is 1-n;
the neuron of the hidden layer Y2 comprises a plurality of weight parameter arrays W i The number of the weight parameters W is i, and i is 1-n;
the number of the weight parameters W is equal to the number i of the numerical values in the output array;
let the output data array of the hidden layer Y2 be Y2 i Then:
in the invention, the weight value of the hidden layer Y2 of the auxiliary task Y is biased to predict the intervention measure of the next moment IDH;
(3) an output layer: five output layer Y3 neurons correspond to the intervention measures of IDH at the next moment such as no intervention, suspended ultrafiltration, volume reduction, normal saline infusion, conductivity up-regulation and the like respectively;
the neurons of the output layer Y3 have different weight calculations on the data output by the hidden layer, and output the final result;
setting:
the neurons of the output layer Y3 include several weight parameter arrays W j The number of the weight parameters W is j, and j is 1-n;
weight parameter array W j Number and weight parameter array W i Equal in number, i.e. i = j
Then, output = Sigmoid (of output layer Y3)
)。
3. The model architecture of the main task Z is divided into an input layer Z1, a hidden layer Z2 and an output layer Z3, each layer is composed of a plurality of neurons, each neuron in the hidden layer Z2 and the output layer Z3 has a weight, and the weight is obtained in the model training process;
(1) an input layer: the data inputted to the input layer of the main task Z1 is time-invariant data before hemodialysis and time-variant data at each time during hemodialysis, the same as the data inputted to the input layer Y1 of the support task Y, and the number of neurons in the input layer Z1 is the sum of the number of time-invariant data before hemodialysis and the number of time-variant data at each time during hemodialysis.
(2) Hiding the layer: each neuron of the hidden layer Z2 has different weight calculation on the data input by the input layer Z1, so that the prediction of a certain task label is favored, and in the invention, the weight value of the hidden layer Z2 of the main task Z is favored to predict whether IDH occurs at the next moment;
(3) an output layer: the output layer Z3 neurons of the main task Z have one, which is: the next time when IDH does not occur is represented by '0', or the next time when IDH occurs is represented by '1';
as shown in figure 1 of the drawings, in which,
taking the output array of the hidden layer X2 of the auxiliary task X, the output array of the hidden layer Y2 of the auxiliary task Y and the output array of the hidden layer Z2 of the main task Z as the input of the output layer Z3 of the main task Z;
setting:
the output array of the hidden layer X2 of the auxiliary task X is X2 i I is the number of numerical values in the output array, and i is 1-n;
output array Y2 of hidden layer Y2 of auxiliary task Y i I is the number of numerical values in the output array, and i is 1-n;
the output array of the hidden layer Z2 with the main task Z is Z2 i I is the number of numerical values in the output array, and i is 1-n;
the numbers of numerical values in an output array of a hidden layer X2 of the auxiliary task X, an output array of a hidden layer Y2 of the auxiliary task Y and an output array of a hidden layer Z2 of the main task Z are the same, and are i, i is 1-n;
the neurons of the main task Z output layer Z3 comprise a plurality of weight parameter arrays W i The number of the weight parameters W is i, and i is 1-n;
the number of the weight parameters W is equal to the number i of the numerical values in the output array;
array the output as X2 i And an output array Y2 i Output array Z2 i Adding the values in the same sequence to obtain the input array Q of the output layer Z3 of the main task Z i ；
Output of primary task Z = Sigmoid(s) ((s))
)；
The Sigmoid function is also called a Logistic function and is also called an S-shaped growth curve, is used for hidden layer neuron output, has a value range of (0,1), can map a real number to an interval of (0,1) and can be used for classification, and the Sigmoid function as an activation function has the advantages of smoothness and easiness in derivation. The invention only adopts Sigmoid as the activation function of the neural network, and the principle of the Sigmoid function is not described in detail here.
The invention fuses the output of the hidden layer of the auxiliary task and the output of the hidden layer of the main task, and the fused result is used as the input of the output layer of the main task, thereby realizing the weight sharing of the main task and the auxiliary task and effectively improving the accuracy of the main task for predicting whether IDH occurs at the next moment.
S4, training a model;
s41, constructing training set data and test set data;
sorting the input data collected in the step S1 according to dialysis time, taking the first 80% as training set data and the last 20% as test set data;
s42, constructing a loss function;
firstly, constructing a loss function by adopting cross entropy for an auxiliary task X, an auxiliary task Y and a main task Z;
then, constructing a model total loss function, wherein the model total loss function is the weighted sum of the loss functions of an auxiliary task X, an auxiliary task Y and a main task Z; wherein,
the weight ratio of the cross entropy loss function of the main task Z, the auxiliary task X and the auxiliary task Y is 2: 1: 1.
s43, calculating the weight parameters of each corresponding neuron when the loss function takes the minimum value by adopting a gradient descent method for the training set data;
the neurons were those in the model constructed in S3.
S5, verifying the multitask model;
and the multitask model calculates the test set data obtained in the step S4 to obtain results of three tasks of the test set data, calculates the accuracy rate, the recall rate and the accuracy rate of each task respectively, and evaluates the effect of the multitask model according to the three indexes.
The accuracy of the auxiliary task A is the ratio of the number of IDH occurrences and IDH non-occurrences to the total number of test data, which are correctly predicted by a model in the test set data;
the recall rate of the auxiliary task A is the ratio of the number of IDH (inverse discrete cosine transform) occurrences correctly predicted by the model in the test set data to the total number of IDH occurrences in the test set data;
the accuracy rate of the auxiliary task A is the ratio of the number of IDH (inverse discrete cosine transform) generated by correctly predicting the IDH by the model in the test set data to the total number of IDH generated by predicting the IDH by the model.
The accuracy of the auxiliary task B is the ratio of the sum of the quantity of model correct prediction non-intervention, suspended ultrafiltration, reduced volume, normal saline infusion and up-regulated conductivity to the total quantity of the test data in the test set data;
the recall rate of the auxiliary task B is the ratio of the number of model correctly predicting the total number of non-intervention, suspended ultrafiltration, volume reduction, physiological saline infusion and up-regulation conductivity in the test data to the total number of non-intervention, suspended ultrafiltration, volume reduction, physiological saline infusion and up-regulation conductivity in the test data;
the precision rate of the auxiliary task B is the ratio of the number of model correctly predicting the total number of non-intervention, suspended ultrafiltration, volume reduction, physiological saline infusion and up-regulation conductivity in the test set data to the total number of model correctly predicting the non-intervention, the total number of suspended ultrafiltration, the total number of volume reduction, the total number of physiological saline infusion and the total number of up-regulation conductivity.
The accuracy of the main task C is the ratio of the number of IDH occurring at the next moment and IDH not occurring at the next moment to the total number of the test data, which is correctly predicted by the model in the test set data;
the recall rate of the main task C is the ratio of the number of IDH occurrences at the next moment to the total number of IDH occurrences at the next moment in the test set data, which are correctly predicted by the model in the test set data;
the accuracy rate of the main task C is the ratio of the number of IDH generated at the next moment when the model correctly predicts the IDH in the test set data to the total number of IDH generated at the next moment when the model predicts the IDH.
The invention recommends a multitask model based on a multitask learning method through the low and middle blood pressure prediction and intervention measures, and reduces the risk of IDH occurrence of a patient in a mode of prediction in advance and intervention in advance, thereby achieving the purposes of preventing and treating IDH and improving the prognosis of MHD patients in the MHD treatment process, and specifically comprises the following steps:
on one hand, whether IDH occurs in the dialysis can be predicted before hemodialysis of the patient, so that attention of medical care to the patient who possibly has IDH is improved.
On the other hand, the model of the invention can also be used for predicting whether IDH occurs at the next monitoring moment according to the physical signs and dialysis treatment parameters monitored during hemodialysis in the dialysis process. Thirdly, by adopting the model of the invention, a high-quality IDH intervention scheme can be provided for medical care personnel, and a high-quality treatment scheme recommendation, namely a high-quality IDH intervention measure recommendation, is provided for the medical care personnel when IDH occurs in hemodialysis, so that the problems that the medical care personnel select an improper treatment scheme or cannot give a treatment scheme in a short time or the inexperienced medical care personnel do not know how to treat the patient, such as dizziness, vertigo, dysphoria, anxiety, pale complexion, yawning, nausea, vomiting, chest distress, rapid heart rate increase, abdominal discomfort and cold sweat, severe patients can have dyspnea, blackness, muscle spasm and even transient consciousness loss, and acute cardiovascular events and even death can be caused in severe cases are solved.
Fourthly, the output of the hidden layer of the auxiliary task is fused with the output of the hidden layer of the main task, and the fused result is used as the input of the output layer of the main task, so that the weight sharing of the main task and the auxiliary task is realized, and the accuracy of predicting whether IDH occurs at the next moment by the main task can be effectively improved by adopting the model of the invention.
Example 2: the embodiment provides an application of an IDH prediction and intervention recommendation multitask model constructed by the method of the invention in predicting IDH in maintenance hemodialysis treatment.
Before the patient starts hemodialysis, the personal information, medical history record, test examination result, historical dialysis record and current hemodialysis prescription of the patient are called from the system, and the physical sign data of the patient are measured to obtain time-invariant data before hemodialysis: body weight before dialysis, heart rate before dialysis, body weight before dialysis, pulse before dialysis, systolic blood pressure before dialysis, diastolic blood pressure before dialysis, respiratory frequency before dialysis, age before dialysis, sex before dialysis, height before dialysis, history of hypertension, history of diabetes, blood leukocyte count, hemoglobin, red blood cell count, blood glucose, blood albumin, total blood cholesterol, blood triglyceride, blood creatinine, blood uric acid, blood urea nitrogen, blood potassium, blood sodium, blood calcium, blood phosphorus, blood chlorine, urine leukocyte, urine protein, urine erythrocyte, urine creatinine, urine occult blood, urine microalbumin, urine albumin, dialysis-interval weight gain, last dialysis time, last machine weight, IDH number of times of last seven days, IDH number of times of last thirty days, dialysis mode, anticoagulant, dose, ultrafiltration amount, dialysis time, potassium ion concentration of dialysate, calcium ion concentration of dialysate, sodium ion concentration of dialysate, and blood glucose concentration of dialysate, Conductivity and blood flow of dialysate;
the time-invariant data before hemodialysis is input into an IDH prediction and intervention measure recommendation multitask model, and the model can predict the IDH risk of the current dialysis before hemodialysis of a patient is started.
During the maintenance hemodialysis treatment of a patient, reading interval monitoring signs, dialysis treatment parameters and dialysis machine parameters obtained by continuous fixed time interval monitoring in a hemodialysis machine, and obtaining time-varying data at each moment during the hemodialysis: current body temperature, current heart rate, current systolic pressure, current diastolic pressure, current pulse, current respiratory rate, current ultrafiltered, current dialysis duration, current dialysate potassium ion concentration, current dialysate calcium ion concentration, current dialysate sodium ion concentration, current dialysate electrical conductivity, current blood flow, current arterial pressure, current venous pressure, current trans-module pressure, current blood filtration volume.
The time-invariant data before hemodialysis and the time-variant data at each moment during hemodialysis are input into an IDH prediction and intervention measure recommendation multitask model, and the model can predict whether IDH occurs at the next moment during the current dialysis of a patient and intervention measure recommendation of the IDH at the next moment.
Claims (8)
- A method for constructing an IDH prediction and intervention measure recommendation multitask model is characterized by comprising the following steps:s1 collecting input data; collecting time-invariant data before hemodialysis and time-variant data at each moment during hemodialysis as input data; each time-invariant data and each time-variant data jointly form an input data;s2, setting a label for each piece of input data; setting 3 labels for each piece of input data, namely label A, label B and label C; each label corresponds to a learning task;s3, constructing a mid-low blood pressure prediction and intervention measure recommendation multitask model; the multi-task model comprises an auxiliary task X, an auxiliary task Y and a main task Z; the model architecture of each task is divided into an input layer, a hidden layer and an output layer, and each layer is composed of a plurality of neurons; each neuron in the hidden layer and the output layer has a weight, and the weight is obtained in the process of model training; each neuron of the output layer has different weight calculation on the data output by the hidden layer, the final result is output, and the number of neurons of the output layer is determined by the number of label categories;s4, training the model; constructing training set data and test set data from the input data collected in the step S1, constructing a loss function, inputting the training set data into a model, and training the model;s5, verifying the multitask model; and the multitask model calculates the test set data obtained in the step S4 to obtain results of three tasks of the test set data, respectively calculates the accuracy, the recall rate and the accuracy rate of each task, evaluates the effect of the multitask model according to the three indexes, and obtains a qualified IDH prediction and intervention measure recommendation multitask model when the accuracy, the recall rate and the accuracy rate of each task reach preset values.
- 2. The method for constructing an IDH prediction and intervention recommendation multitask model according to claim 1, wherein:the time invariant data at step S1 includes: the measured physical sign data before hemodialysis, personal information, medical history records, inspection results, historical dialysis records and current hemodialysis prescriptions of the patient; wherein,the measurement sign data comprises: body weight before penetration, heart rate before penetration, body weight before penetration, pulse before penetration, systolic pressure before penetration, diastolic pressure before penetration, and respiratory rate before penetration;the personal information includes: age, sex, height, age under dialysis;the medical history record comprises: history of hypertension, diabetes;the inspection result comprises: blood leukocyte count, hemoglobin, red blood cell count, blood glucose, blood albumin, blood total cholesterol, blood triglyceride, blood creatinine, blood uric acid, blood urea nitrogen, blood potassium, blood sodium, blood calcium, blood phosphorus, blood chloride, urine leukocyte, urine protein, urine erythrocyte, urine creatinine, urine occult blood, urine microalbumin, urine albumin;the historical dialysis records include: weight gain during dialysis, last dialysis time, last off-machine weight, frequency of IDH occurring in nearly seven days, and frequency of IDH occurring in nearly thirty days;when the hemodialysis prescription includes: dialysis mode, anticoagulant dosage, ultrafiltration volume, dialysis duration, dialysate potassium ion concentration, dialysate calcium ion concentration, dialysate sodium ion concentration, dialysate conductivity, and blood flow volume;the time-varying data includes: during the hemodialysis of a patient, monitoring physical signs, dialysis treatment parameters and dialysis machine parameters at fixed time intervals; wherein,the interval monitoring sign data comprises: current body temperature, current heart rate, current systolic pressure, current diastolic pressure, current pulse and current respiratory rate;dialysis treatment parameters include: current ultrafiltered, current dialysis duration, current dialysate potassium ion concentration, current dialysate calcium ion concentration, current dialysate sodium ion concentration, current dialysate conductivity, and current blood flow;dialysis machine parameters include: current arterial pressure, current venous pressure, current trans-molding pressure, current amount of hemofiltration.
- 3. The method for constructing an IDH prediction and intervention recommendation multitask model according to claim 1, wherein: step S2 includes the following sub-steps:s21, setting the diagnostic criteria for IDH:firstly, the intervention measures of IDH are met, and the systolic pressure is reduced by more than 20mmHg compared with the systolic pressure before permeation;② no drying measure is taken, but the systolic pressure is less than 90 mmHg;s22, setting a label A, a label B and a label C for each piece of input data according to the IDH diagnosis standard;the label A is: whether IDH occurs in the dialysis or not;label B is: whether IDH occurs at the next moment;label C is: intervention measures of IDH at the next moment; wherein the intervention measures of IDH include: no intervention, suspending ultrafiltration, reducing volume, normal saline infusion, and increasing conductivity by 5 categories.
- 4. The method for constructing an IDH prediction and intervention recommendation multitask model according to claim 3, wherein: step S22 includes the following sub-steps:s221, collecting IDH intervention data in dialysis, wherein the IDH intervention data is an intervention measure of the input data in the dialysis process; the intervention measures of the IDH intervention data are also as follows: 5 categories of no intervention, suspension of ultrafiltration, volume reduction, normal saline infusion and conductivity up-regulation;s222, setting a label A; judging whether each moment is IDH according to the IDH diagnosis standard for the input data of all moments of the dialysis;s223, setting a label B; judging whether each moment is IDH according to the IDH diagnosis standard for the input data of all moments of the dialysis; the label B at the previous moment of the moment which is judged to be the IDH is marked as the next moment to generate the IDH, and the label B at the rest moments are marked as the next moment to not generate the IDH;s224, setting a label C; and setting an input data acquisition label C according to the intervention measure data of the IDH.
- 5. The method for constructing an IDH prediction and intervention measure recommendation multitasking model according to claim 1, characterized by: step S3, predicting whether IDH occurs in the dialysis as an auxiliary task X; predicting the intervention measure of the IDH at the next moment as an auxiliary task Y; predicting whether IDH occurs at the next moment to be a main task Z;the model architecture of the auxiliary task X is divided into an input layer X1, a hidden layer X2 and an output layer X3, each layer is composed of a plurality of neurons, each neuron in the hidden layer X2 and the output layer X3 has a weight, and the weight is obtained in the model training process;5.1 the auxiliary task X model comprises the following structure:(1) an input layer: the data input by the input layer X1 of the auxiliary task X is time-invariant data before hemodialysis, and the number of the input layer X1 neurons is the number of the time-invariant data before hemodialysis;(2) hiding the layer: each neuron of the hidden layer X2 has different weight calculation on the data input by the input layer X1, so that the prediction of a certain task label is more favored; in the invention, the weight value of the hidden layer X2 of the auxiliary task X is biased to predict whether IDH occurs in the dialysis;setting:the output array of the input layer X1 of the auxiliary task X is X1 i I is the number of numerical values in the output array, and i is 1-n;the neuron of the hidden layer X2 comprises a plurality of weight parameter arrays W i The number of the weight parameters W is i, and i is 1-n;the number of the weight parameters W is equal to the number i of the numerical values in the output array;if the output data array of the hidden layer X2 is X2i, then:；
(3) an output layer: one neuron of the output layer X3 is provided, 0 is that IDH does not occur in the dialysis, and 1 is that IDH occurs in the dialysis; the neurons of the output layer X3 have different weight calculations for the data output by the hidden layer, and output the final result;setting:the neurons of the output layer X3 comprise a plurality of weight parameter arrays W j The number of the weight parameters W is j, and j is 1-n;weight parameter array W j Number and weight parameter array W i Equal in number, i.e. i = jThen, output = Sigmoid (of output layer X3))；
5.2 model of auxiliary task Y comprises the following structure:the model architecture of the auxiliary task Y is divided into an input layer Y1, a hidden layer Y2 and an output layer Y3, each layer is composed of a plurality of neurons, each neuron in the hidden layer Y2 and the output layer Y3 has a weight, and the weight is obtained in the model training process;(1) an input layer: the data input by the input layer Y1 of the auxiliary task Y is the same as the data input by the input layer of the main task Z1, and is time-invariant data before hemodialysis and time-variant data at each time during hemodialysis, and the number of neurons in the input layer Y1 is the sum of the number of time-invariant data before hemodialysis and the number of time-variant data at each time during hemodialysis;(2) hiding the layer: each neuron of the hidden layer Y2 has different weight calculation on the data input by the input layer Y1, so that the prediction of a certain task label is favored;setting:the output array of the input layer Y1 of the auxiliary task Y is Y1 i I is the number of numerical values in the output array, and i is 1-n;the neuron of the hidden layer Y2 comprises a plurality of weight parameter arrays W i The number of the weight parameters W is i, and i is 1-n;the number of the weight parameters W is equal to the number i of the numerical values in the output array;let the output data array of the hidden layer Y2 be Y2 i And then:；
in the invention, the weight value of the hidden layer Y2 of the auxiliary task Y is biased to predict the intervention measure of the next moment IDH;(3) an output layer: five output layer Y3 neurons correspond to the intervention measures of IDH at the next moment such as no intervention, suspended ultrafiltration, volume reduction, normal saline infusion, conductivity up-regulation and the like respectively;the neurons of the output layer Y3 have different weight calculations on the data output by the hidden layer, and output the final result;setting:the neurons of the output layer Y3 comprise a plurality of weight parameter arrays W j The number of the weight parameters W is j, and j is 1-n;weight parameter array W j Number and weight parameter array W i Equal in number, i.e. i = jThen, output = Sigmoid (of output layer Y3))；
5.3 the main task Z model includes the following structure:the model architecture of the main task Z is divided into an input layer Z1, a hidden layer Z2 and an output layer Z3, each layer is composed of a plurality of neurons, each neuron in the hidden layer Z2 and the output layer Z3 has a weight, and the weight is obtained in the model training process;(1) an input layer: the data input to the input layer of the main task Z1 is the same as the data input to the input layer Y1 of the support task Y, and is time-invariant data before hemodialysis and time-variant data at each time during hemodialysis, and the number of neurons in the input layer Z1 is the sum of the number of time-invariant data before hemodialysis and the number of time-variant data at each time during hemodialysis;(2) hiding the layer: each neuron of the hidden layer Z2 has different weight calculation on the data input by the input layer Z1, so that the prediction of a certain task label is favored, and in the invention, the weight value of the hidden layer Z2 of the main task Z is favored to predict whether IDH occurs at the next moment;(3) and (3) an output layer: the output layer Z3 neurons of the main task Z have one, which is: the next time when IDH does not occur is represented by '0', or the next time when IDH occurs is represented by '1';taking the output array of the hidden layer X2 of the auxiliary task X, the output array of the hidden layer Y2 of the auxiliary task Y and the output array of the hidden layer Z2 of the main task Z as the input of the output layer Z3 of the main task Z;setting:the output array of the hidden layer X2 of the auxiliary task X is X2 i I is the number of numerical values in the output array, and i is 1-n;output array Y2 of hidden layer Y2 of auxiliary task Y i I is the number of numerical values in the output array, and i is 1-n;the output array of the hidden layer Z2 with the main task Z is Z2 i I is the number of numerical values in the output array, and i is 1-n;the numbers of numerical values in an output array of a hidden layer X2 of the auxiliary task X, an output array of a hidden layer Y2 of the auxiliary task Y and an output array of a hidden layer Z2 of the main task Z are the same, and are i, i is 1-n;the neurons of the main task Z output layer Z3 comprise a plurality of weight parameter arrays W i The number of the weight parameters W is i, and i is 1-n;the number of the weight parameters W is equal to the number i of the numerical values in the output array;array the output as X2 i And an output array Y2 i Output array Z2 i Adding the values in the same sequence to obtain the input array Q of the output layer Z3 of the main task Z i ；Output of primary task Z = Sigmoid (/) ()。
- 6. The method for constructing the IDH prediction and intervention recommendation multitask model according to claim 1, wherein the step S4 comprises the following sub-steps:s41, constructing training set data and test set data;sorting the input data collected in the step S1 according to dialysis time, wherein the first 80% is taken as training set data, and the second 20% is taken as test set data;s42, constructing a loss function;firstly, constructing a loss function by adopting cross entropy for an auxiliary task X, an auxiliary task Y and a main task Z;then, constructing a model total loss function, wherein the model total loss function is the weighted sum of the loss functions of an auxiliary task X, an auxiliary task Y and a main task Z; wherein,the cross entropy loss function weight ratio of the main task Z to the auxiliary tasks X and Y is 2: 1: 1;s43, calculating the weight parameters of each corresponding neuron when the loss function takes the minimum value by adopting a gradient descent method for the training set data;the neurons were neurons in the model constructed in S3.
- 7. The method for constructing the IDH prediction and intervention measure recommendation multitask model according to claim 1, wherein in the step S5:the accuracy of the auxiliary task A is the ratio of the number of IDH occurrence and IDH non-occurrence in the test set data and the total number of the test data in the correct prediction of the model;the recall rate of the auxiliary task A is the ratio of the number of IDH (inverse discrete cosine transform) occurrences correctly predicted by the model in the test set data to the total number of IDH occurrences in the test set data;the accuracy rate of the auxiliary task A is the ratio of the number of the IDH correctly predicted by the model in the test set data to the total number of the IDH predicted by the model;the accuracy of the auxiliary task B is the ratio of the sum of the quantity of model correct prediction without intervention, suspended ultrafiltration, volume reduction, normal saline infusion and conductivity up-regulation in the test set data to the total quantity of the test data;the recall rate of the auxiliary task B is the ratio of the number of model correctly predicting the total number of non-intervention, ultrafiltration pause, volume reduction, saline infusion and conductivity up-regulation in the test set data to the total number of non-intervention, ultrafiltration pause, volume reduction, saline infusion and conductivity up-regulation in the test set data respectively;the precision rate of the auxiliary task B is the ratio of the number of model correctly predicting the total number of non-intervention, suspended ultrafiltration, volume reduction, physiological saline infusion and up-regulation conductivity in the test set data to the total number of model correctly predicting the non-intervention, the total number of suspended ultrafiltration, the total number of volume reduction, the total number of physiological saline infusion and the total number of up-regulation conductivity;the accuracy of the main task C is the ratio of the number of the IDH occurring at the next moment and the number of the IDH not occurring at the next moment in the test set data correctly predicted by the model to the total number of the test data;the recall rate of the main task C is the ratio of the number of IDH occurrences at the next moment to the total number of IDH occurrences at the next moment in the test set data, which is correctly predicted by the model in the test set data;the accuracy rate of the main task C is the ratio of the number of IDH generated at the next moment when the model correctly predicts the IDH in the test set data to the total number of IDH generated at the next moment when the model predicts the IDH.
- 8. The method for constructing an IDH prediction and intervention action recommendation multitask model according to claim 1, wherein the IDH prediction and intervention action recommendation multitask model constructed by the method of claim 1 is used for predicting IDH in maintenance hemodialysis treatment.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210711223.3A CN114937486B (en) | 2022-06-22 | 2022-06-22 | IDH prediction and intervention measure recommendation multi-task model construction method and application |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210711223.3A CN114937486B (en) | 2022-06-22 | 2022-06-22 | IDH prediction and intervention measure recommendation multi-task model construction method and application |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114937486A true CN114937486A (en) | 2022-08-23 |
| CN114937486B CN114937486B (en) | 2023-09-26 |
Family
ID=82868450
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210711223.3A Active CN114937486B (en) | 2022-06-22 | 2022-06-22 | IDH prediction and intervention measure recommendation multi-task model construction method and application |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114937486B (en) |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110224565A1 (en) * | 2010-03-15 | 2011-09-15 | Singapore Health Services Pte Ltd. | Method of predicting acute cardiopulmonary events and survivability of a patient |
| US20150045713A1 (en) * | 2013-08-07 | 2015-02-12 | B. Braun Avitum Ag | Device and method for predicting intradialytic parameters |
| CN110507296A (en) * | 2019-08-12 | 2019-11-29 | 重庆大学 | A kind of acute hypotension mixing method for early warning based on LSTM network |
| CN111028913A (en) * | 2019-11-29 | 2020-04-17 | 北京工业大学 | Hemodialysis treatment scheme aid decision-making method |
| CN111460157A (en) * | 2020-04-01 | 2020-07-28 | 哈尔滨理工大学 | Cyclic convolution multitask learning method for multi-field text classification |
| CN111939353A (en) * | 2019-05-14 | 2020-11-17 | 吴元昊 | Construction method of prediction model of hypotensive event in hemodialysis |
| CN112107752A (en) * | 2019-06-20 | 2020-12-22 | 纬创资通股份有限公司 | Blood pressure prediction method and electronic device using same |
| US20210193317A1 (en) * | 2019-12-20 | 2021-06-24 | Fresenius Medical Care Holdings, Inc. | Real-time intradialytic hypotension prediction |
| CN113270192A (en) * | 2021-04-19 | 2021-08-17 | 浙江大学 | Hemodialysis risk prediction system based on combination of dynamic and static data and depth self-encoder |
| CN114091460A (en) * | 2021-11-24 | 2022-02-25 | 长沙理工大学 | Multitask Chinese entity naming identification method |
| CN114186711A (en) * | 2021-10-27 | 2022-03-15 | 中山大学 | Industrial raw material consumption prediction method based on multitask time sequence learning |
-
2022
- 2022-06-22 CN CN202210711223.3A patent/CN114937486B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110224565A1 (en) * | 2010-03-15 | 2011-09-15 | Singapore Health Services Pte Ltd. | Method of predicting acute cardiopulmonary events and survivability of a patient |
| US20150045713A1 (en) * | 2013-08-07 | 2015-02-12 | B. Braun Avitum Ag | Device and method for predicting intradialytic parameters |
| CN111939353A (en) * | 2019-05-14 | 2020-11-17 | 吴元昊 | Construction method of prediction model of hypotensive event in hemodialysis |
| CN112107752A (en) * | 2019-06-20 | 2020-12-22 | 纬创资通股份有限公司 | Blood pressure prediction method and electronic device using same |
| CN110507296A (en) * | 2019-08-12 | 2019-11-29 | 重庆大学 | A kind of acute hypotension mixing method for early warning based on LSTM network |
| CN111028913A (en) * | 2019-11-29 | 2020-04-17 | 北京工业大学 | Hemodialysis treatment scheme aid decision-making method |
| US20210193317A1 (en) * | 2019-12-20 | 2021-06-24 | Fresenius Medical Care Holdings, Inc. | Real-time intradialytic hypotension prediction |
| CN111460157A (en) * | 2020-04-01 | 2020-07-28 | 哈尔滨理工大学 | Cyclic convolution multitask learning method for multi-field text classification |
| CN113270192A (en) * | 2021-04-19 | 2021-08-17 | 浙江大学 | Hemodialysis risk prediction system based on combination of dynamic and static data and depth self-encoder |
| CN114186711A (en) * | 2021-10-27 | 2022-03-15 | 中山大学 | Industrial raw material consumption prediction method based on multitask time sequence learning |
| CN114091460A (en) * | 2021-11-24 | 2022-02-25 | 长沙理工大学 | Multitask Chinese entity naming identification method |
Non-Patent Citations (2)
| Title |
|---|
| 李国;袁闻;王怀超;: "面向不完备数据的民航旅客流失预测模型", 计算机工程与设计, no. 10 * |
| 李戈;靳立忠;: "基于学习向量量化网络建立糖尿病并发症预测模型", 中国自然医学杂志, no. 04 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114937486B (en) | 2023-09-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US8246567B2 (en) | Method of monitoring hypertensive haemodialysis patients | |
| US8303532B2 (en) | Pertioneal membrane function test method, peritoneal membrane function test apparatus and peritoneal membrane function test program | |
| US7788038B2 (en) | Biological information and blood treating device information control system, biological information and blood treating device information control device, and biological information and blood treating device information control method | |
| US8088094B2 (en) | Planning method and apparatus for peritoneal dialysis and hemodialysis hybrid remedy | |
| Madihally | Principles of biomedical engineering | |
| US8449496B2 (en) | Method for testing peritoneum function and a peritoneal dialysis planning apparatus | |
| US20180043076A1 (en) | Intersession adaptive peritoneal dialysis fluid removal for multiple session optimization | |
| Maheshwari et al. | A regional blood flow model for β 2-microglobulin kinetics and for simulating intra-dialytic exercise effect | |
| CN113744832B (en) | Intelligent decision-making and quality control system for continuous kidney substitution therapy | |
| CN114937486B (en) | IDH prediction and intervention measure recommendation multi-task model construction method and application | |
| Marano et al. | A new approach to bicarbonate addition during hemodialysis: Testing model predictions in a patient cohort | |
| CN108986920B (en) | Dialysis adequacy protein decomposition rate evaluation method and system | |
| CN110730989B (en) | Personalized renal failure chronic care system and method | |
| RU2693241C1 (en) | Method of treating renal insufficiency in domestic animals by applying haemodialysis procedure | |
| US20240177866A1 (en) | Kidney health monitoring system | |
| CN117672486A (en) | Estimation system and method for target ultrafiltration volume in hemodialysis | |
| CN115775633B (en) | Mortality real-time prediction method and system based on multidimensional data and electronic equipment | |
| Ghista | Biomedical Engineering of Pancreatic, Pulmonary, and Renal Systems, and Applications to Medicine | |
| Azar | A novel system for haemodialysis efficiency monitoring | |
| Bazaev et al. | In vitro and in vivo trials of wearable artificial kidney | |
| Tayebi et al. | The effect of hypertonic glucose infusion on dialysis adequacy in non-diabetic hemodialysis patients | |
| Kim et al. | The Prediction of Dry Weight for Chronic Hemodialysis Athletes Using a Machine Learning Approach: Sports Health Implications | |
| CN113241174A (en) | Checking system for CRRT treatment parameters of blood purification patient | |
| Xiang et al. | Causal Convolution-Transformer for Time Series Forecasting of Dry Weight in Hemodialysis Patients | |
| Mitra | The pathophysiology of fluid removal during haemodialysis: implications for blood volume monitoring and determination of dry weight |
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 | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: The construction method and application of a multi task model for IDH prediction and intervention recommendation Granted publication date: 20230926 Pledgee: Nanjing Bank Co.,Ltd. Nanjing North Branch Pledgor: Shentai Health Technology (Nanjing) Co.,Ltd. Registration number: Y2024980005062 |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right |