System for early monitoring and prevention of low blood pressure in hemodialysis

Abstract

The application relates to a system for early monitoring and prevention of low blood pressure in hemodialysis, which comprises a dialysis device for performing hemodialysis on a user and collecting blood-related information of the user during the dialysis process; a liquid supplementing device for adding liquid supplement for the user into the dialysis device and capable of adjusting the temperature and output flow of the liquid supplement; a camera device for acquiring images of the user; a memory and a processor, the memory being used for storing image analysis and control programs and liquid supplement adjustment programs, the processor being used for loading the stored image analysis and control programs to detect whether the user has a trembling symptom, loading the liquid supplement adjustment programs to generate the temperature and output flow of the liquid supplement, and adjusting the temperature and output flow of the liquid supplement in real time based on the detected trembling symptom of the user. The application has the advantage that parameters of the dialysis device and the liquid supplementing device can be calculated and adjusted according to real-time conditions of the user.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical devices, in particular to a low blood pressure early monitoring and prevention system in hemodialysis. BACKGROUND

[0002] Dialysis technology has a wide range of applications in the medical field, used to treat various kidney diseases. During dialysis, the patient's blood is cleaned through a dialyzer, removing excess waste and water from the blood. However, the dialysis process itself is not completely risk-free. Low blood pressure is a common complication that patients may encounter during dialysis, which can be caused by factors such as dialysate temperature, infusion rate and volume.

[0003] During dialysis, it is very important to maintain stable blood pressure. In order to prevent the occurrence of low blood pressure, medical personnel will try to adjust the infusion rate and temperature, as these two factors can affect blood flow rate and blood vessel dilation, thereby affecting blood pressure. For example, rapid infusion can cause blood pressure to rise, while excessively high infusion temperature can cause blood vessels to dilate excessively, leading to a drop in blood pressure. While an excessively low infusion temperature can cause patients to experience shivering and other adverse reactions.

[0004] However, for each patient, the appropriate infusion rate and temperature are not the same, which needs to be adjusted according to the individual differences of the patient. In reality, medical personnel often need to rely on experience to determine the appropriate infusion rate and temperature, but due to differences in experience, as well as the pressure of decision-making in emergency situations, medical personnel may have difficulty making the best judgment. This has led to the occurrence of low blood pressure events, which has threatened the health of patients. Therefore, how to accurately control the infusion rate and temperature, prevent and control the occurrence of low blood pressure, has become a problem that needs to be solved in the field of dialysis technology. SUMMARY

[0005] In order to be able to calculate and adjust the parameters of the dialysis device and the liquid supplement device according to the real-time situation of the user, the present application provides a low blood pressure early monitoring and prevention system in hemodialysis.

[0006] The low blood pressure early monitoring and prevention system in hemodialysis provided by the present application adopts the following technical solution:

[0007] A low blood pressure early monitoring and prevention system in hemodialysis, comprising:

[0008] a dialysis device for performing hemodialysis on a user and collecting blood-related information of the user during dialysis;

[0009] a liquid supplement device for adding liquid supplement for the user to the dialysis device and capable of adjusting the temperature and output flow of the liquid supplement;

[0010] The low blood pressure early monitoring and prevention system in hemodialysis provided by the present application has the advantages that:A camera is configured to acquire an image of the user;

[0011] The memory is configured to store an image analysis and control program and a fluid supplement adjustment program, and the processor is configured to load the image analysis and control program to detect whether the user has a tremor symptom, and load the fluid supplement adjustment program to generate a temperature and an output flow rate of the fluid supplement, and adjust the temperature and the output flow rate of the fluid supplement in real time based on the detected tremor symptom of the user.

[0012] By using the above technical solutions, the system can early detect and prevent the occurrence of hypotension during dialysis by monitoring and analyzing the blood-related information and behavior of the user (such as tremor) in real time. The processor loads the fluid supplement adjustment program to generate the temperature and the output flow rate of the fluid supplement, thereby reducing the probability of hypotension of the user during dialysis. Since the user is prone to have adverse reactions such as tremor when the temperature of the fluid supplement is too low, the camera can detect the tremor symptom of the user in real time, and the processor can analyze the tremor symptom of the user in real time and adjust the temperature and the output flow rate of the fluid supplement accordingly. The system can more accurately and quickly respond, and is more effective and accurate than relying on manual judgment and manual adjustment. In addition, the system can automatically monitor and adjust, thereby reducing the work burden of medical staff and enabling them to better focus on other medical tasks that require manual operation and care.

[0013] Optionally, the fluid supplement adjustment program is configured to perform the following steps:

[0014] Obtaining a body characteristic parameter of the user;

[0015] Loading a blood pressure prediction model and importing the body characteristic parameter of the user to determine a fluid supplement amount, a fluid supplement speed, and a fluid supplement temperature;

[0016] Adjusting the temperature and the output flow rate of the fluid supplement in real time based on the detected tremor symptom of the user.

[0017] By using the above technical solutions, the fluid supplement amount, the fluid supplement speed, and the fluid supplement temperature can be determined according to the body characteristic parameter of the user through the blood pressure prediction model, and the dialysis parameters can be finely adjusted to make the fluid supplement process more in line with individual needs, thereby improving the effect of dialysis treatment. The parameters of the fluid supplement can be adjusted in real time through the blood pressure prediction model and according to the tremor symptom of the user, so that the dialysis process can dynamically adapt to the changes of the user. Through fine adjustment and real-time response, the occurrence of complications such as hypotension can be effectively prevented or reduced, the overall effect of dialysis treatment is improved, and the life quality of the patient is improved.

[0018] Optionally, the system further comprises a screen, and the step of obtaining the body characteristic parameter of the user comprises:

[0019] Outputting the body characteristic basic parameter information to the screen for display, wherein the body characteristic basic parameter is pre-recorded.

[0020] The blood real-time information is acquired based on the data acquisition device, stored and updated in the blood information database.

[0021] The user body feature parameters are acquired based on the historical information in the latest time period in the body feature information and the blood information database.

[0022] By outputting the body feature basic parameter information to the screen for display, the user can clearly understand the body state and the real-time condition of hemodialysis, increase the transparency of treatment, and help to improve the user's trust and safety. Based on the data acquisition device to acquire the blood real-time information and store it in the database, and based on the historical information in the latest time period in the body feature information and the blood information database, the user body feature parameters can be dynamically adjusted, which will make the blood pressure early monitoring and prevention system in hemodialysis have better real-time and accuracy, help to more accurately adjust the dialysis parameters, and further improve the effect and safety of hemodialysis. Based on the historical information in the latest time period in the body feature information and the blood information database, the user body feature parameters are acquired, which can make the system better understand and predict the user's body condition change, so as to more finely adjust the dialysis parameters, and help to improve the user's dialysis treatment effect.

[0023] Optionally, the step of loading the blood pressure prediction model and importing the user body feature parameters to determine the infusion volume, infusion speed and infusion temperature comprises:

[0024] The user body feature parameters, the standard infusion volume, the standard infusion speed and the standard infusion temperature are input into the blood pressure prediction model to predict the future blood pressure change;

[0025] The future blood pressure change predicted by the blood pressure prediction model is input into a preset reward function for operation and to obtain a reward value, wherein the output of the preset reward function is related to the difference between the future blood pressure change predicted by the blood pressure prediction model and the actually measured blood pressure change, the greater the difference, the lower the reward value, and the smaller the difference, the smaller the reward value;

[0026] Based on the reinforcement learning algorithm and the change of the reward value, the infusion volume, the infusion speed and the infusion temperature are adjusted to improve the reward value, so as to obtain new infusion volume, infusion speed and infusion temperature, which are substituted into the previous step and executed until the infusion volume, infusion speed and infusion temperature converge in a stable range.

[0027] By adopting the technical scheme, the user's body feature parameters and standard infusion parameters are input into the blood pressure prediction model to predict future blood pressure changes. Based on these information and actual measured blood pressure changes, the infusion volume, infusion speed and infusion temperature are adjusted by reinforcement learning algorithm and reward value changes to achieve the optimal infusion scheme. This individualized adjustment strategy can more accurately meet the special needs of users during hemodialysis, improving treatment effect and comfort. Based on the preset reward function and reinforcement learning algorithm, the system can self-adjust and optimize the infusion parameters, so that the infusion volume, infusion speed and infusion temperature can converge to a stable range, avoiding excessive infusion or insufficient infusion, and improving the safety of hemodialysis. By measuring the blood pressure changes in real time and comparing them with the output of the prediction model, the system can dynamically feedback and adjust the prediction model and infusion strategy to improve the accuracy of prediction and the rationality of infusion, thereby better preventing the occurrence of hypotension. The system continuously adjusts the infusion parameters by reinforcement learning algorithm to improve the reward value. With the increase of use time, the prediction accuracy and infusion strategy of the system will be more and more optimized, and the user experience will be better and better.

[0028] Optionally, the step of adjusting the temperature and output flow of the infusion in real time based on the detected user's tremor symptoms comprises:

[0029] determining whether the user has a tremor symptom, and if so, evaluating the severity of the tremor;

[0030] based on the severity of the tremor, the current infusion temperature and output flow, calculating the infusion temperature and output flow that need to be adjusted;

[0031] adjusting the temperature and output flow of the infusion according to the adjustment value calculated in the previous step, and continuously detecting whether the user's tremor state is improved, if so, further adjusting the temperature and output flow of the infusion until the user's tremor severity is lower than the standard degree.

[0032] By adopting the technical scheme, the system detects the user's tremor symptoms in real time and adjusts the temperature and output flow of the infusion, ensuring the continuity and stability of the hemodialysis process. This real-time feedback mechanism allows the system to discover and correct potential problems in a timely manner, reducing the risk of hypotension. The system not only detects the user's tremor symptoms, but also evaluates the severity of the tremor and adjusts the temperature and output flow of the infusion accordingly. This mechanism can avoid overreaction or neglect of the tremor symptoms, maintaining the stability of hemodialysis.

[0033] Optionally, the camera device comprises a camera and a posture adjustment device, the posture adjustment device is used to fix the camera on the frame body and has two rotational degrees of freedom; the image analysis and control program is used to perform the following steps:

[0034] detecting a human body image, and determining an arm corresponding to the dialysis channel based on the detected human body image, and taking the arm as a target arm;

[0035] controlling the posture adjustment device to adjust the orientation of the camera so that the target arm approaches the center of the viewfinder;

[0036] adjusting the focal length of the camera to increase the proportion of the target arm in the viewfinder, and returning to the previous step until the proportion of the target arm in the viewfinder exceeds a preset threshold.

[0037] By adopting the above technical solutions, since the arm corresponding to the dialysis channel is usually stationary when the user is dialyzing, it is suitable for being used as a judgment part of tremor. The system can determine the arm corresponding to the dialysis channel through human body image detection and take it as a target arm. This precise target positioning function enables the system to monitor and control specifically. The system can automatically adjust the orientation and focal length of the camera so that the target arm approaches the center of the viewfinder and the proportion of the target arm in the viewfinder reaches a preset threshold. This automatic adjustment function reduces the need for manual adjustment, reduces the operation difficulty, and can quickly and accurately adjust to the best angle in a short time. The intelligent adjustment of the camera can flexibly adapt to individual differences such as the body shape and posture of different users, providing a more personalized use experience.

[0038] Optionally, the image analysis and control program is further used to perform the following steps:

[0039] sampling images from the obtained image stream of the target arm according to a preset sampling interval;

[0040] judging the position change of the target arm in a plurality of groups of adjacent frame images in succession, and if repeated position oscillation occurs, it is judged as tremor, and if it is continuous change, it is filtered as noise;

[0041] reducing the preset sampling interval for several times and repeating the previous step.

[0042] By adopting the technical scheme, the position change of the target arm in continuous several groups of adjacent frame images is analyzed, if repeated position oscillation occurs, it is judged as tremor, if it is continuous change, it is filtered as noise. This judgment method can accurately distinguish the actual tremor and the arm position change caused by other reasons (such as non- abnormal activity of the user), thereby avoiding misjudgment and improving monitoring accuracy. By increasing the preset sampling interval and repeatedly judging, the tremor caused by the synchronization of the sampling interval and the tremor frequency can be avoided.

[0043] Optionally, the step of judging the position change of the target arm in continuous several groups of adjacent frames, if repeated position oscillation occurs, it is judged as tremor, if it is continuous change, it is filtered as noise comprises:

[0044] The position of the target arm in the image is obtained based on the instance segmentation technology, and a reference point is defined for the target arm image;

[0045] A coordinate system is established based on the current camera view area, and the position of the reference point at time point t is (x_t, y_t), the position of the reference point at time point t+1 is (x_{t+1}, y_{t+1}), and sqrt((x_{t+1}-x_t)^2+(y_{t+1}-y_t)^2) is defined as the displacement of the reference point in the adjacent two frame images;

[0046] Based on the current time point, continuous several frame images are selected forward, the displacement of the reference point of the first frame image and the last frame image is calculated, and it is judged whether the displacement is less than a first preset threshold, if not, it is judged that no tremor occurs, and the current step is repeated; if yes, it is judged whether the displacement of the reference point in the adjacent two frame images is greater than a preset threshold, and the ratio of the total displacement to the displacement of the reference point of the first frame image and the last frame image is greater than a second preset threshold, if yes, it is judged as tremor, if not, the current step is repeated.

[0047] By adopting the technical scheme, by judging the displacement of the reference point in the continuous frame images, if the displacement is less than the preset threshold, it is judged that no tremor occurs, this method helps to filter out the slight change of the arm position caused by noise and other non-tremor factors. By setting multiple preset thresholds, such as the displacement threshold and the threshold of the ratio of the total displacement to the displacement of the reference point in the first and last frame images, the fineness of tremor judgment can be improved. Through the above technical means, the system can realize early detection of user tremor, which helps to early warning and prevention of hypotension, and thus protects the life safety of the user and improves the effect and safety of hemodialysis. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 The program block diagram of the steps performed by the image analysis and control program in an embodiment of the application is shown.

[0049] Figure 2 A program flow chart illustrating steps performed by a fluid replacement adjustment procedure in an embodiment of the application. DETAILED DESCRIPTION

[0050] The application will be further described below in conjunction with the drawings. It should be understood that the specific embodiments described herein are intended to be illustrative only and are not intended to limit the scope of the application.

[0051] Most end-stage renal disease is treated by kidney replacement therapy, among which the number of maintenance hemodialysis (MHD) patients is increasing year by year. Intradialytic hypotension (IDH) is one of the serious complications of dialysis. The imbalance between fluid removal and plasma replenishment during hemodialysis leads to a decrease in effective arterial blood volume, which in turn reduces cardiac filling and cardiac output, ultimately leading to the occurrence of IDH. Patients with IDH will experience discomfort such as nausea, vomiting, dizziness, fatigue, and profuse sweating, and in severe cases, may experience muscle spasms, difficulty breathing, and transient syncope. IDH not only causes patients to interrupt or prematurely end dialysis, resulting in inadequate dialysis, decreased residual renal function, and internal fistula blockage, but also causes cardiovascular and cerebrovascular diseases such as arrhythmia and brain damage, and even death.

[0052] Continuous renal replacement therapy (CRRT) is a common treatment method, especially suitable for critically ill patients with unstable hemodynamics. CRRT is continuously performed within 24 hours, causing less interference to the patient's circulatory system, better controlling fluid balance and electrolyte homeostasis, improving the patient's metabolic state, and helping to stabilize critically ill patients.

[0053] CRRT mainly removes waste and excess fluid from the patient's body in the following ways:

[0054] Filtration: blood is made to pass through a semi-permeable membrane using a pressure difference, solutes and water flow through the semi-permeable membrane with the blood to form a filtrate, which contains waste and excess water.

[0055] Dialysis: solutes diffuse from the side with high concentration to the side with low concentration through the semi-permeable membrane.

[0056] Adsorption: a portion of toxic substances and inflammatory mediators are adsorbed by the filter membrane to achieve the purpose of removal.

[0057] CRRT can adjust the intensity of treatment by changing the flow rate of blood and dialysate or replacement fluid, and adjusting the filtration pressure and the degree of anticoagulation. Therefore, CRRT can be personalized according to the specific circumstances of the patient.

[0058] When using CRRT for treatment, special attention needs to be paid to some potential complications, such as hypotension, coagulation, filter clogging, etc. At the same time, since CRRT needs to be performed continuously for 24 hours, it needs to be monitored and managed by a dedicated nursing staff. In recent years, clinical research has increasingly recognized the definition of IDH in The European Best Practice Guidelines (EBPG), i.e. a decrease in systolic blood pressure ≥20 mmHg (1 mmHg=0.133 kPa), a decrease in mean arterial pressure ≥10 mmHg, accompanied by clinical events, and the need for intervention measures.

[0059] The incidence of IDH ranges from 20% to 50% in recent reports on IDH. Due to the inconsistency of diagnostic criteria, there are significant differences in the incidence of IDH among patients under different diagnostic criteria. In a prospective study of 124 patients with 3818 hemodialysis, the incidence of IDH was found to be 6.7%. Another researcher analyzed 43 MHD patients and found that the incidence of IDH was 30.7%. In another prospective study, it was found that the incidence of IDH was 39.9%.

[0060] For IDH, the temperature of the dialysate is too high, which can increase the core body temperature, and the skin and muscle blood vessels dilate to dissipate heat, resulting in a decrease in peripheral vascular resistance, and insufficient effective circulating blood volume leading to IDH 24. Lower plasma albumin can cause a decrease in intravascular colloid osmotic pressure, and fluid seeps into the interstitial space, reducing the effective circulating blood volume in the blood vessels, leading to the occurrence of IDH. Blood filtration through the dialyzer can activate leukocytes, platelets, complement, etc. reactions, leading to an increase in vascular permeability, causing albumin to leak out, and expanding blood vessels, thereby causing IDH.

[0061] In addition, when the circulating blood volume decreases during the period, the arteriolar side of the capillary bed contracts, the peripheral vascular resistance increases, and the blood pressure is maintained. However, when the blood vessels contract to a certain time and limit, ischemia leads to further release of adenosine, causing the arteriolar end of the capillary to dilate, venous blood to accumulate, reducing blood return to the heart, thereby reducing cardiac output and blood pressure, and the DeJager-Krogh phenomenon occurs. Due to ventricular hypertrophy, diastolic dysfunction, and decreased vascular compliance, cardiac filling and cardiac output are reduced, leading to a decrease in effective arterial blood volume, and ultimately causing the occurrence of IDH.

[0062] Therefore, adjusting dialysis treatment to extend the dialysis time reduces ultrafiltration rate and can reduce the frequency of IDH in patients dialyzed for less than 4 h. Increasing the number of dialysis sessions is another strategy to improve IDH. In a small retrospective cohort study of 12 MHD patients, daily short hemodialysis was found to reduce the onset of IDH. But it will lead to an increase in treatment costs, and is not suitable as the preferred option.

[0063] And low temperature dialysis can cause vasoconstriction, retain central blood volume, and maintain mean arterial pressure by increasing total peripheral resistance. In a meta-analysis of 26 randomized controlled trials involving 484 patients, the researchers found that reducing dialysate temperature can significantly reduce the incidence of IDH by 70%. But at the same time, it was found that the symptoms of shivering and other discomfort caused by temperature reduction were 2.95 times that of standard temperature dialysis. However, in a randomized crossover trial of 11 MHD patients, it was found that the systolic blood pressure during personalized low temperature dialysis was higher than that during 37°C dialysis, which can improve the symptoms caused by low temperature. Therefore, personalized low temperature dialysis is a safer and more effective intervention measure.

[0064] In this application, a blood dialysis low blood pressure early monitoring and prevention system is proposed, which is used for monitoring the state of the patient, generating the infusion parameters in the dialysis process and adjusting, which helps to reduce the probability of IDH occurrence.

[0065] The blood dialysis low blood pressure early monitoring and prevention system includes a dialysis device, an infusion device, a camera device, a memory and a processor. The dialysis device is used for blood dialysis of the user and collects blood related information of the user during dialysis; the infusion device is used to add infusion to the dialysis device for the user, and can adjust the temperature and output flow of the infusion; the camera device is used to acquire the image of the user; the memory is used to store the image analysis and control program and the infusion adjustment program, and the processor is used to load the image analysis and control program to detect whether the user has a shivering symptom, load the infusion adjustment program to generate the temperature and output flow of the infusion, and adjust the temperature and output flow of the infusion in real time based on the detected shivering symptom of the user.

[0066] Specifically, the dialysis device mainly consists of the following parts:

[0067] Dialysis machine: This is the main body of the blood dialysis device, which includes a liquid pump (used to push blood and dialysate), a dialyzer (also called artificial kidney), and a temperature control device (used to control the temperature of blood and dialysate).

[0068] Dialyzer: The dialyzer is the most important part of the blood dialysis device. Its interior contains a large number of small tubes, which are surrounded by a semi-permeable membrane. Blood flows through the tubes, and dialysate flows through the chamber outside the tubes.

[0069] Blood circulation system: This system includes blood in and blood out tubing, which is used to draw blood from the user, pass it through the dialyzer, and return it to the user.

[0070] Dialysate system: This system is used to prepare dialysate and pump it into the dialyzer.

[0071] Blood is drawn from the patient through the blood circulation system, into the interior of the thin tubes within the dialyzer. Outside the tubes is the dialysate, which contains appropriate amounts of electrolytes and glucose. The blood and dialysate are separated by a semipermeable membrane. Due to concentration differences, waste products and excess electrolytes in the blood diffuse through the semipermeable membrane into the dialysate. During dialysis, in order to remove excess water from the blood, the dialysis machine creates a pressure difference between the blood side and the dialysate side, forcing water from the blood to pass through the semipermeable membrane into the dialysate. After the above process, the cleaned blood is returned to the patient through the blood circulation system. The entire process is automatically controlled by the dialysis machine, including the flow rates of blood and dialysate, the composition and temperature of the dialysate, and the rate of ultrafiltration, to ensure the effectiveness of dialysis and the comfort of the patient.

[0072] Fluid replacement device mainly consists of the following parts:

[0073] Liquid storage part: Mainly used to store the liquid to be delivered, such as medicine, saline or specific dialysate.

[0074] Infusion pump: The infusion pump is the main component that controls the delivery of the liquid, which can accurately control the speed and pressure of the infusion.

[0075] Liquid delivery speed adjustment valve: It is an adjustable valve used to adjust the delivery speed of the liquid.

[0076] Temperature detection device: Used to detect the temperature of the liquid in real time to ensure that the temperature of the liquid is within an appropriate range.

[0077] Temperature adjustment device: When the temperature of the liquid is detected to be out of the preset range, the temperature adjustment device can adjust the temperature of the liquid to prevent discomfort to the user caused by too high or too low temperature.

[0078] The liquid storage part in the fluid replacement device stores the liquid to be delivered. When infusion is needed, the infusion pump will draw the liquid from the storage part and deliver it to the user through the infusion pipeline. During the infusion process, the liquid delivery speed adjustment valve can adjust the delivery speed of the liquid as needed. If it is necessary to speed up or slow down the infusion speed, medical personnel or the control system of the device itself can adjust the opening of the liquid delivery speed adjustment valve to change the delivery speed of the liquid.

[0079] The display screen is typically a liquid crystal display (LCD) or an organic light-emitting diode display (OLED), both of which can provide clear visual effects. On some higher-end devices, high-definition or ultra-high-definition displays can be used to provide better visual effects. During hemodialysis, the screen is often used to display various important information, such as the current amount of fluid replacement, the speed of fluid replacement, the temperature of fluid replacement, and the user's physiological parameters, etc. These information can be displayed in the form of numbers, words or charts on the screen. When user input is required, such as inputting the user's basic physical parameters, the graphical interface provides one or more input methods. Common input methods include virtual keyboards, drop-down menus, and sliders, etc. Users can select or input their choices by touching the screen. When the user makes a touch operation on the screen, the touch layer detects the operation and sends the operation information to the processor. The processor changes the displayed content or performs corresponding calculations according to the user's operation.

[0080] The basic physical parameters mainly refer to those physiological or biochemical parameters related to the individual's health condition, as well as some parameters closely related to the treatment effect. In the context of hemodialysis, the following are some optional basic physical parameters:

[0081] Age: Age has a great influence on the effect of hemodialysis and the required treatment measures. Older people may need more careful adjustment and care.

[0082] Gender: Men and women may have different physiological responses and needs in the process of hemodialysis.

[0083] Weight: Body weight affects the demand for fluid replacement and the dosage of drugs, etc.

[0084] Blood pressure: Blood pressure is an important parameter in the process of hemodialysis, which needs to be continuously monitored and adjusted.

[0085] Heart rate: Changes in heart rate can be an early sign of complications during dialysis, such as hypotension.

[0086] Blood biochemical parameters: including hemoglobin, blood urea nitrogen, blood creatinine, electrolytes such as potassium, sodium, calcium, phosphorus, etc. These parameters have an important influence on the evaluation of the effect of hemodialysis and the adjustment of the treatment plan.

[0087] Dialysis frequency: The frequency and history of hemodialysis are also important considerations.

[0088] These parameters may be entered before the device starts running and dynamically adjusted during the treatment process according to the monitoring results.

[0089] The camera is a device used to capture images, which can be a digital camera or a webcam, as long as it can capture high-definition images and transmit these image data to the processor for analysis. The camera should have the ability to adjust the focal length, so as to change the size or clarity of the image when necessary.

[0090] The posture adjustment device is a mechanical device that can change the position and angle of the camera, usually including a set of motors and gear systems, with two degrees of rotational freedom, which means it can rotate in two directions (such as up and down and left and right). This design allows the camera to flexibly adjust its viewing angle to best capture the image of the target arm.

[0091] In particular, with reference to Figure 1 , the image analysis and control program is used to perform the following steps:

[0092] S101. Perform human image detection and determine the arm corresponding to the dialysis channel based on the detected human image, and take the arm as the target arm.

[0093] In some embodiments, S101 can be implemented through the following sub-steps:

[0094] S1011. The camera captures images in the current environment.

[0095] S1012. Image data is sent to the processor for analysis.

[0096] S1013. Human detection is performed using computer vision techniques (such as deep learning models) to identify human parts in the image.

[0097] S1014. Identify and locate the arm corresponding to the dialysis channel.

[0098] S1015. Set the arm where the dialysis channel is located as the target arm.

[0099] For example, a camera, which is a 720p camera, is activated and starts capturing a video stream of the environment. It captures images at a rate of 100 frames per second, and the data from the camera is sent to a connected processor. The processor runs an image recognition algorithm, such as a convolutional neural network (CNN), to identify human shapes in the images. This algorithm has been trained to accurately recognize and locate human bodies. Through image recognition, the algorithm further analyzes the human body images, specifically looking for arm parts, especially the arm part that is performing dialysis. This can involve some specific image recognition techniques, such as shape recognition or edge detection, to determine the precise location of the arm and the dialysis channel. Once the location of the arm and the dialysis channel is determined, this location is set as the target location. The algorithm will track this location and ensure that the focus of the camera is always directed here. For example, if the arm is detected in the lower right corner of the image, then the location of this lower right corner is set as the target, and the focus of the camera is adjusted to track this target location.

[0100] S102. Control the pose adjustment device to adjust the orientation of the camera so that the target arm approaches the center of the view.

[0101] In some embodiments, S102 can be implemented by the following sub-steps:

[0102] S1021. Calculate the distance between the location of the target arm in the image and the center of the view.

[0103] S1022. Send instructions to the pose adjustment device to adjust the orientation of the camera.

[0104] S1023. The pose adjustment device operates the camera to rotate and adjust its orientation.

[0105] S1024. Recapture the image and determine whether the target arm is now in the center of the view.

[0106] Once the arm is targeted, the processor begins sending instructions to the connected pose adjustment device. For example, the instructions can include having the electric motor on the device move the camera a certain angle in the direction of the targeted arm. Upon receiving the instructions from the processor, the pose adjustment device responds to the instructions by driving the electric motor to change the orientation of the camera. For example, if the camera needs to move to the lower right corner, the pose adjustment device drives the electric motor to move the camera to the right and down. The processor will continue to analyze the image data from the camera to check the position of the targeted arm in the frame. For example, it can use real-time image analysis techniques to process each frame of the image to see if the targeted arm has moved to the center of the frame. If the processor detects that the targeted arm is not approaching the center of the frame, it will continue to send instructions to the pose adjustment device to instruct it to further adjust the orientation of the camera. For example, if the targeted arm is still in the lower right corner of the frame, the processor will send more instructions to the pose adjustment device to move the camera further to the lower right corner.

[0107] S103. Adjust the focal length of the camera to increase the proportion of the targeted arm in the view and return to the previous step until the proportion of the targeted arm in the view exceeds the preset threshold.

[0108] In some embodiments, S103 can be implemented by the following sub-steps:

[0109] S1031. Calculate the proportion of the targeted arm in the image and compare it with the preset threshold.

[0110] S1032. If the proportion is lower than the preset threshold, send instructions to the camera to adjust the focal length to enlarge the image.

[0111] S1033. Recapture the image and recalculate the proportion of the targeted arm in the image.

[0112] S1034. Repeat steps S1031 to S1034 until the proportion of the targeted arm in the image exceeds the preset threshold.

[0113] Once the position of the target arm is determined, the processor sends instructions to the camera to adjust the focal length. For example, the instructions can be to shorten or lengthen the focal length of the camera to increase the proportion of the target arm in the field of view. Upon receiving the instructions from the processor, the camera responds to the instructions by adjusting its lens or optical system to change its focal length. For example, if the proportion of the target arm in the field of view needs to be increased, the camera can shorten its focal length to "zoom in" on the target arm. The processor will continue to analyze the image data from the camera to check the proportion of the target arm in the field of view. For example, using image analysis techniques, each frame of image is processed to check whether the proportion of the target arm in the field of view has exceeded the preset threshold. If the processor detects that the proportion of the target arm in the field of view has not exceeded the preset threshold, it will continue to send instructions to the camera to further adjust the focal length. For example, if the proportion of the target arm in the field of view has not reached the preset threshold, the processor will issue more instructions to the camera to further shorten its focal length to increase the proportion of the target arm in the field of view.

[0114] S104. Sampling images from the acquired image stream of the target arm according to a preset sampling interval.

[0115] First, a preset sampling interval is determined, for example, every 0.01 seconds to capture an image. Then, according to the preset sampling interval, images are acquired from the image stream of the target arm. For example, if the sampling interval is 0.01 seconds, every 0.01 seconds an image is taken from the image stream. Finally, the captured images are stored for subsequent analysis.

[0116] S105. Judging the position change of the target arm in a plurality of consecutive groups of adjacent frames of images, and if repeated position oscillation occurs, it is judged as tremor, and if it is continuous change, it is filtered as noise.

[0117] In some embodiments, S105 can be implemented by the following sub-steps:

[0118] S1051. Obtaining the position of the target arm in the image based on instance segmentation technology, and defining a reference point for the target arm image.

[0119] S1052. Establishing a coordinate system based on the current camera view area, and defining the position of the reference point at time point t as (x_t, y_t), the position at time point t+1 as (x_{t+1}, y_{t+1}), and the displacement of the reference point in the adjacent two frames of images as sqrt((x_{t+1}-x_t)^2+(y_{t+1}-y_t)^2).

[0120] S1053. Based on the current time point, select a number of consecutive frames of images, calculate the displacement of the reference point of the first frame of image and the last frame of image, and determine whether the displacement is less than a first preset threshold, if not, it is determined that no tremor has occurred, and the current step is repeated; if yes, determine whether the displacement of the reference point in the adjacent two frames of images is greater than a preset threshold, and the ratio of the total displacement to the displacement of the reference point of the first frame of image and the last frame of image is greater than a second preset threshold, if yes, it is determined that tremor has occurred, if not, the current step is repeated.

[0121] Suppose there is a photo of a patient's arm during dialysis, the instance segmentation technology (such as Mask R-CNN) can be used to locate and identify the position of the arm, and mark the position. Then, the reference point can be defined as the position of the elbow.

[0122] A two-dimensional coordinate system can be established in the current camera's view area, such as taking the upper left corner of the camera's view area as the origin, the right as the positive x-axis, and the downward as the positive y-axis. Then, the position of the reference point at time point t can be measured, such as (200, 300), and the position at time point t+1 can be measured, such as (202, 302). The displacement of the reference point is sqrt((202-200)^2+(302-300)^2) = sqrt(8) ≈ 2.83.

[0123] Suppose 5 consecutive frames of images are selected, at time points t, t+1, t+2, t+3, t+4. Then, the displacement of the reference point of the first frame of image and the last frame of image is calculated as sqrt((x_{t+4}-x_t)^2+(y_{t+4}-y_t)^2), if this displacement is less than a preset threshold, such as 5 pixel units, it is determined that no tremor has occurred; if the displacement is greater than or equal to 5, but the displacement of the reference point in the adjacent two frames of images is less than another preset threshold, such as 2 pixel units, or the ratio of the total displacement to the displacement of the reference point of the first frame of image and the last frame of image is less than another preset threshold, such as 0.5, it is also determined that no tremor has occurred; otherwise, it is determined that tremor has occurred.

[0124] This determination method is derived from the fact that large-scale movement from the starting point to the ending point is generally user-controlled displacement, not tremor. For cases where the starting point and ending point positions are small, but the total movement distance is large, it is usually determined that tremor has occurred.

[0125] S106. Reduce the number of preset sampling intervals and repeat S105.

[0126] In some embodiments, S106 can be implemented through the following sub-steps:

[0127] S1061. Adjust the sampling interval, for example, reduce the sampling interval from 0.1 seconds to 0.08 seconds.

[0128] S1062. Resample the images using the new sampling interval and store these images.

[0129] S1063. Use the images with the new sampling interval to determine the change in arm position. If tremor is found, end early. If no tremor is found, return to step S1061 to further reduce the sampling interval until tremor is found or the minimum sampling interval is reached.

[0130] In actual operation, if the original sampling interval is to collect an image every 0.1 second, this interval can be adjusted to collect an image every 0.08 second. After the new sampling interval is set, the camera will collect images according to the new interval, for example, every 0.08 second, and then save these images to the memory. The processor will process the images stored in the memory according to the new sampling interval to determine the change in arm position. If tremor is found based on the new sampling interval, the processor will end the process early and send an alarm; if no tremor is found, the processor will further reduce the sampling interval, for example, from 0.08 seconds to 0.07 seconds, and then repeat the process until tremor is found or the maximum sampling interval, for example, 0.05 seconds, is reached.

[0131] Referring to Figure 2 , the liquid supplement adjustment program is used to perform the following steps:

[0132] S201. Obtain user body feature parameters.

[0133] In some embodiments, S201 can be implemented through the following sub-steps:

[0134] S2011. Output body feature basic parameter information to the screen for display, wherein the body feature basic parameters are pre-entered.

[0135] Assuming that the pre-entered body feature basic parameters include the user's age, gender, weight, height, etc. The screen will display these parameters, for example: "Age: 35 years old, Gender: Male, Weight: 70 kg, Height: 175 cm".

[0136] S2012. Obtain real-time blood information based on the data acquisition device and store and update it in the blood information database.

[0137] Data collection devices, such as blood analyzers, can collect blood information of users, including parameters such as blood glucose, blood pressure, hemoglobin, etc. These data will be updated in real time and saved to the blood information database, for example: "Time: 10:00, Blood glucose: 5.5 mmol / L, Blood pressure: 120 / 80 mmHg, Hemoglobin: 150 g / L".

[0138] S2013. Based on the body feature information and the latest historical information in the blood information database within the latest time period, the user's body feature parameters are obtained.

[0139] Based on the user's body feature information (such as age, gender, weight, height) and the latest historical information in the blood information database (such as recent blood glucose, blood pressure, hemoglobin, etc.), the system can calculate or predict other body feature parameters of the user. For example, the system may predict the user's insulin requirement, or estimate the user's oxygen demand based on their hemoglobin level, etc.

[0140] S202. Load the blood pressure prediction model and import the user's body feature parameters to determine the fluid replacement volume, fluid replacement speed, and fluid replacement temperature.

[0141] In some embodiments, S202 can be implemented through the following sub-steps:

[0142] S2021. Input the user's body feature parameters, standard fluid replacement volume, standard fluid replacement speed, and standard fluid replacement temperature into the blood pressure prediction model to predict future blood pressure changes.

[0143] The blood pressure prediction model is a complex machine learning model that can predict future blood pressure changes based on input human physical sign parameters (such as blood pressure, heart rate, dehydration rate, etc.). Determining dialysis parameters such as fluid replacement volume, fluid replacement speed, and fluid replacement temperature usually requires combining predicted blood pressure changes and specific conditions of the patient.

[0144] Fluid replacement volume: This refers to the total amount of fluid that needs to be supplemented to the patient during hemodialysis. This parameter is usually determined based on personal characteristics such as the patient's weight, height, age, and weight changes before and after hemodialysis. If the prediction model predicts that the patient's blood pressure may decrease, the fluid replacement volume may need to be increased to maintain stable blood pressure.

[0145] Fluid replacement speed: This refers to the speed at which fluid is supplemented during hemodialysis. If the prediction model predicts that the patient's blood pressure may rapidly decrease, the fluid replacement speed may need to be increased to prevent excessively low blood pressure. Conversely, if the prediction model predicts that the patient's blood pressure may rise, the fluid replacement speed may need to be reduced.

[0146] Fluid replacement temperature: The temperature of the fluid replacement can also affect blood pressure. Generally, a higher fluid replacement temperature can cause blood vessels to dilate, thereby lowering blood pressure, while a lower fluid replacement temperature can cause blood vessels to constrict, thereby increasing blood pressure. Therefore, if the prediction model predicts that the patient's blood pressure may rise, the fluid replacement temperature may need to be lowered; conversely, if the prediction model predicts that the patient's blood pressure may fall, the fluid replacement temperature may need to be increased.

[0147] In this step, assuming that the blood pressure prediction model is a machine learning model, the input parameters include: the user's physical feature parameters (such as age, gender, weight, height, etc.), the standard fluid replacement volume (such as 500 milliliters), the standard fluid replacement speed (such as 1 milliliter / minute), and the standard fluid replacement temperature (such as 37 degrees Celsius). The result of the model prediction may be the trend of blood pressure change in the next 30 minutes, for example: "Predicted in the next 30 minutes, blood pressure rises from 120 / 80 mmHg to 130 / 85 mmHg."

[0148] The blood pressure prediction model needs to be pre-trained, which can be trained through the following steps:

[0149] Step 1: Data collection. Collect all physiological data, medical record data, climate data, and dialysis treatment data for each dialysis session. These data can be obtained from the dialysis instrument or other medical devices, provided by the medical information system of the medical institution, or provided by external databases (such as the database of the weather center).

[0150] Specifically, the following data can be collected:

[0151] Physiological data: This includes various physiological indicators of the patient, such as heart rate, blood pressure, blood sugar, blood lipids, etc., which may directly affect the effectiveness of dialysis. For example, for a dialysis patient with diabetes, his blood sugar level before and after dialysis may need to be recorded.

[0152] Medical record data: This includes the patient's medical history, surgical records, medication use, etc. For example, if the patient has just undergone a surgery before dialysis, or he is using a drug that may affect blood thinning, these are important information that need to be recorded.

[0153] Climate data: Weather conditions can also affect the patient's blood pressure and physical condition, so it is necessary to collect climate data such as temperature, humidity, air pressure, etc. in the patient's location. For example, a dialysis patient may have different dialysis effects in summer and winter, or may experience dehydration when the weather is hot.

[0154] Dialysis treatment data: This includes the time of dialysis, the composition and concentration of the dialysis fluid, blood pressure changes during dialysis, etc. For example, a patient may feel more comfortable during dialysis in the evening than in the morning, or he may be allergic to the composition of a certain dialysis fluid, which are all information that need to be recorded.

[0155] It should be noted that data pre-screening is required, such as deleting data with a missing rate of more than 90%, deleting data with a single category rate of more than 90%, and deleting data with a coefficient of variation of less than 0.1. Since data missing is inevitable, this project will use four different methods to fill in the data, including no filling, simple filling, random forest filling, and improved random forest filling.

[0156] Step 2: Feature selection. According to the collected data, relevant features are selected for subsequent prediction models. This may include physiological data (such as blood pressure, heart rate), dialysis parameters (such as dialysis time, dialysis fluid composition), climate data (such as temperature, humidity), and other factors that may affect blood pressure.

[0157] In the data collection phase, a large amount of data is collected, including physiological data, medical record data, climate data, and dialysis treatment data. However, not all data is useful for the target. Therefore, it is necessary to find out which characteristic variables have a significant impact on hemodialysis effect. For example, the patient's age, gender, whether he has diabetes, dialysis time and dialysis fluid composition may be the most influential factors on hemodialysis effect.

[0158] Step 3: Create classification factors and patient clustering. Use the selected characteristics to generate classification factors, and use these factors to cluster patients into different groups. For example, clustering algorithms in machine learning (such as K-means, hierarchical clustering) can be used to complete this step.

[0159] After determining the characteristic variables, clustering algorithms (such as K-means or hierarchical clustering) can be used to cluster patients. The purpose of clustering is to group similar patients together so that a prediction model suitable for each group can be developed. For example, all patients over the age of 60, with diabetes, and who choose to dialyze in the morning are grouped into one group.

[0160] Clustering, also known as clustering, is an unsupervised machine learning method that aims to group similar samples into the same class, while different samples are grouped into different classes. In this case, patients will be grouped into different groups according to their characteristic variables, such as age, gender, whether they have diabetes, dialysis time and dialysis fluid composition.

[0161] The specific implementation process can be carried out through the following steps:

[0162] Feature selection and processing: Before clustering, the data needs to be pre-processed. This includes selecting features that have an impact on the target, and scaling or normalizing these features to ensure that different features have equal weight in the clustering.

[0163] Selecting a clustering algorithm: There are many different clustering algorithms to choose from, such as K-means, spectral clustering, DBSCAN, hierarchical clustering, etc. The choice of algorithm depends on the data and the goal. For example, if the dataset is spherical, K-means might be a good choice. If the dataset has a complex structure, spectral clustering or DBSCAN might be more appropriate.

[0164] Determining the number of clusters: Determining the number of clusters is a crucial step. In some algorithms, such as K-means, the number of clusters needs to be pre-specified. In some other algorithms, such as DBSCAN, the number of clusters is determined automatically by the algorithm. To determine the optimal number of clusters, some evaluation metrics can be used, such as the silhouette coefficient, Calinski-Harabasz index, or Davies-Bouldin index.

[0165] Performing clustering: After selecting a clustering algorithm and determining the number of clusters, the clustering can be performed. This step usually involves initializing cluster centers, computing the distance of each sample to the cluster centers, assigning samples to the nearest cluster, and then updating the cluster centers. This process is repeated until a certain stopping condition is reached, such as the change in cluster centers being less than a certain threshold, or reaching a pre-specified maximum number of iterations.

[0166] Evaluating the clustering result: Finally, the clustering result needs to be evaluated. This can be done using internal evaluation metrics, such as the silhouette coefficient, or external evaluation metrics, such as NMI (Normalized Mutual Information). If the clustering result is not satisfactory, the parameters of the clustering algorithm need to be adjusted, or other clustering algorithms need to be tried.

[0167] Step 4: Train blood pressure prediction model. For each patient group, select the patient data within the group as the training set, and train a machine learning model. The model can be linear regression, decision tree, neural network, etc. In this step, a prediction model for each patient group can be generated.

[0168] Selecting appropriate machine learning model: The appropriate machine learning model needs to be selected based on the nature of the problem and the characteristics of the data. For example, if the problem of interest is a regression problem (such as predicting blood pressure), then models such as linear regression, support vector regression (SVR), random forest regression, or neural networks can be chosen. If the data is non-linear, then neural networks or support vector regression may be a better choice.

[0169] Train the model: After selecting the model, it needs to be trained with the training data within the cohort. The training process usually includes initializing the model parameters, calculating the model's prediction results, calculating the error between the predicted results and the true results, and then updating the model parameters using optimization algorithms such as gradient descent to reduce the prediction error. This process is repeated until the model's prediction error is less than a certain threshold or reaches the preset maximum number of iterations.

[0170] For example, if the selected model is linear regression, the model's parameters (weights and biases) need to be initialized, then the model's prediction results on the training data are calculated, followed by the calculation of the squared error between the predicted results and the true results, and then the model's parameters are updated using gradient descent to reduce the squared error.

[0171] Evaluate the model: After training the model, it needs to be evaluated on the validation dataset to assess its performance. Evaluation metrics may be mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), etc. If the model's performance on the validation dataset is not satisfactory, it may be necessary to adjust the model's parameters or try other models.

[0172] Save the model: After evaluation is complete, if the model's performance meets the requirements, it can be saved for future use. The way the model is saved depends on the machine learning library used. For example, in Python's sklearn library, the joblib library can be used to save the model.

[0173] Through the above steps, a blood pressure prediction model can be trained for each patient cohort, which will be able to predict the patient's blood pressure based on their characteristic variables.

[0174] Step 5: Validate and adjust the model. Use test data to validate and adjust the model through methods such as hold-out method, cross-validation, etc. to improve the accuracy of the prediction.

[0175] Step 5.1. Hold-out method: In the hold-out method, all data is divided into two groups, one group is used to train the model (usually 70%-80% of the data), and the other group is used to test the model (20%-30% of the remaining data). For example, if there are 100 patient data, 70 patient data can be used to train the model, and the remaining 30 patient data can be used to test the model.

[0176] Suppose there is a set of hemodialysis data about 100 patients, the goal is to predict the blood pressure changes that may occur during hemodialysis. This problem can be considered as a regression problem, then the model needs to predict the patient's blood pressure changes based on the patient's physiological data, medical record data, climate data, etc.

[0177] Now, the leave-one-out method will be applied for training and testing the model. In this case, the data of the 100 patients will be first divided into two groups. One group includes the data of 70 patients, which will be used to train the model. The other group includes the data of 30 patients, which will be used to test the model.

[0178] In the training phase, the goal is to train the model based on the data of the 70 patients so that it can predict the possible blood pressure changes given the input data (e.g., physiological data, medical record data, etc.). Linear regression, neural networks, or any other suitable machine learning algorithm for this problem can be used.

[0179] Once the model is trained, it will be tested on the remaining 30 patients' data to evaluate its performance. The predicted blood pressure changes by the model are compared with the actual observed blood pressure changes to assess the model's predictive ability. Mean absolute error (MAE), mean squared error (MSE), or other appropriate metrics can be used to evaluate the model's prediction accuracy.

[0180] Step 5.2. Cross-validation: In cross-validation, all data is divided into K groups (usually K=5 or 10), and then K training and testing are performed, each time selecting one group as the test set and the rest as the training set. For example, if K=5 is chosen, the data of the 100 patients is divided into 5 groups, each with 20 patients. Then 5 training and testing are performed, each time selecting one group as the test set and the rest as the training set. This method can more accurately estimate the model's predictive ability, as each data is used for testing.

[0181] Suppose there is still hemodialysis data of 100 patients, which needs to be used to predict blood pressure changes during dialysis. The goal is to build a model that can accurately predict blood pressure changes.

[0182] Cross-validation can be used to train and validate the model. Suppose 5-fold cross-validation is chosen, which means that all data needs to be divided into 5 equal parts first, and then 5 training and validation processes are performed.

[0183] In each process, 4 parts of data (80 patients) are selected for model training, and the remaining 1 part of data (20 patients) is used for model validation. For example, in the first process, the 1st-4th parts of data are selected for training, and the 5th part of data is used for validation; in the second process, the 1st-3rd, 5th parts of data are selected for training, and the 4th part of data is used for validation, and so on.

[0184] In this way, each piece of data is used for one validation and four training. This maximizes the use of data and allows the model to perform on different datasets, improving its stability and generalization ability.

[0185] During these 5 processes, the validation results of each time, such as prediction error, are recorded. Finally, the average of the 5 validation results is taken as the final performance evaluation of the model. In this way, the performance of the model can be more accurately and comprehensively understood, avoiding the deviation of the performance evaluation of the model due to one-time accidental good or bad results.

[0186] Step 5.3. Adjust the model: After model validation, it may be found that the prediction accuracy of the model in some cases is not high enough. At this time, the model needs to be adjusted. For example, the number of hidden layer nodes of the neural network may be adjusted, the depth of the decision tree may be adjusted, or the regularization parameter of the linear regression may be adjusted, etc. The goal of model adjustment is to improve the prediction accuracy of the model on the test set.

[0187] In many machine learning algorithms, there are some parameters that need to be manually set, which are called hyperparameters. For example, in neural networks, the number of hidden layers and the number of neurons in each hidden layer are hyperparameters; in support vector machines (SVM), the penalty coefficient C and the parameters of the kernel function are hyperparameters. The setting of hyperparameters directly affects the performance of the model.

[0188] In general, it is not possible to know in advance which set of hyperparameters can make the model achieve the best performance, so model adjustment is needed to find the optimal hyperparameters. Common model adjustment methods include grid search (Grid Search) and random search (Random Search).

[0189] Taking a neural network as an example, suppose the hyperparameters to be adjusted are the number of hidden layers (which can be 1, 2, 3) and the number of neurons in each hidden layer (which can be 10, 20, 30). In grid search, all possible combinations of hyperparameters are listed, i.e. (1,10), (1,20), (1,30), (2,10), (2,20)…… and so on. Then the corresponding models are trained, and finally the hyperparameters corresponding to the best performing model are selected as the optimal hyperparameters.

[0190] During model adjustment, cross-validation is also used to evaluate the performance of the model under different hyperparameters, so model adjustment is usually combined with cross-validation.

[0191] Through model adjustment, the model that best fits the data collected before can be found, thereby improving the prediction accuracy of the model.

[0192] Step 5.4. Model evaluation: After the model is adjusted, it needs to be evaluated to confirm whether the prediction effect of the model meets the expectations. Different evaluation indicators may be used, such as prediction accuracy, recall rate, F1 score, area under the ROC curve, etc. The results of model evaluation will directly affect whether the model can be put into use. For example, if the prediction accuracy of the model reaches 90%, it is considered that the model is usable.

[0193] In the model optimization stage, the complexity, interpretability, and prediction accuracy of the model need to be balanced. For example, the depth of the decision tree is a key factor affecting the complexity of the model. If the depth of the decision tree is too large, the model may overfit; if the depth is too small, the model may underfit. Therefore, the depth of the decision tree needs to be adjusted to optimize the model.

[0194] Suppose the optimal model found in step 5.3 is a decision tree with a depth of 10, but if it is found that the interpretability of this model is not very good because the tree is too deep and it is difficult to explain the meaning of each decision node. In order to improve the interpretability of the model, the depth of the decision tree can be reduced to 5, and then cross-validation is performed again to see if the accuracy has decreased significantly.

[0195] If the decrease in accuracy is within an acceptable range, the decision tree with a depth of 5 can be selected as the final model because it has both certain prediction accuracy and good interpretability. This is an example of model optimization. Model optimization not only needs to consider the accuracy of the prediction, but also needs to consider the interpretability of the model, as well as the complexity of the model and other factors.

[0196] S2022. Input the future blood pressure change predicted by the blood pressure prediction model into the preset reward function to calculate and obtain a reward value, wherein the output of the preset reward function is related to the difference between the future blood pressure change predicted by the blood pressure prediction model and the actually measured blood pressure change, and the larger the difference, the lower the reward value, and the smaller the difference, the smaller the reward value.

[0197] In this step, suppose the actually measured blood pressure after 30 minutes is 125 / 82 mmHg, which is different from the model-predicted 130 / 85 mmHg. Therefore, the reward value calculated by the reward function may be -5 (this is just an example value, and the actual reward value will change according to the specific form of the preset reward function).

[0198] S2023. Adjust the infusion volume, infusion speed, and infusion temperature based on the reinforcement learning algorithm and the change of the reward value to improve the reward value, so as to obtain new infusion volume, infusion speed, and infusion temperature, which are substituted into the previous step and executed until the infusion volume, infusion speed, and infusion temperature converge within a stable range.

[0199] In this step, the reinforcement learning algorithm can decide to decrease the fluid volume (e.g., from 500 ml to 480 ml), increase the fluid rate (e.g., from 1 ml / min to 1.1 ml / min), and keep the fluid temperature unchanged. These new fluid parameters will be input into the blood pressure prediction model to predict the new blood pressure change. This process will be repeated until the adjustments of fluid volume, fluid rate, and fluid temperature no longer have a significant impact on the reward value, i.e., they converge to a stable range.

[0200] S203. Adjust the temperature and output flow rate of the fluid in real-time based on the detected tremor symptoms of the user.

[0201] In some embodiments, S2031 can be implemented through the following sub-steps:

[0202] S2031. Determine whether the user has tremor symptoms, and if so, assess the severity of the tremor.

[0203] S2032. Based on the severity of the tremor, the current fluid temperature and output flow rate, calculate the adjustments needed for the fluid temperature and output flow rate.

[0204] S2033. Adjust the fluid temperature and output flow rate according to the adjustments calculated in the previous step, and continue to detect whether the user's tremor state has improved. If so, further adjust the fluid temperature and output flow rate until the user's tremor severity is below a standard level.

[0205] Based on the severity of the tremor, and the current fluid temperature and output flow rate, calculate the adjustments needed for the fluid temperature and output flow rate. The specific calculation method may need to be determined according to the actual situation and pre-set rules. For example, the severity of the tremor can be assessed according to the weighted average of the tremor frequency and amplitude. After assessing the severity of the tremor, a certain mapping relationship or model can be used to determine how to adjust the fluid temperature and output flow rate.

[0206] Specific example: Suppose there is a simple rule that when the weighted average of the tremor frequency and amplitude exceeds a certain threshold (such as 10), it is considered that the tremor is severe, at which point the fluid temperature may need to be increased by 2 degrees Celsius and the fluid rate may need to be increased by 10%. If the weighted average of the tremor frequency and amplitude is below the threshold, but there is still tremor, such as a value of 5, it may only need to increase the fluid temperature by 1 degree Celsius and the fluid rate by 5%. This rule is a very simple example, and in actual applications, more complex rules or models may be needed to determine how to adjust the fluid temperature and output flow rate.

[0207] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above.

[0208] The above examples are only used to illustrate the technical solutions of the present application, but not to limit it; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can still be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A system for early monitoring and prevention of low blood pressure in hemodialysis, characterized in that, The application relates to a dialysis device, a liquid supplement device, a camera device, a memory and a processor. The dialysis device is used for hemodialysis of a user and collecting blood-related information of the user during the dialysis process. The liquid supplement device is used for adding liquid supplement for the user into the dialysis device and can adjust the temperature and output flow of the liquid supplement. The camera device is used for acquiring a user image. The memory is used for storing an image analysis and control program and a liquid supplement adjustment program. The processor is used for loading the image analysis and control program to detect whether the user has a tremor symptom, loading the liquid supplement adjustment program to generate the temperature and output flow of the liquid supplement, and adjusting the temperature and output flow of the liquid supplement in real time based on the detected tremor symptom of the user. The camera device comprises a camera and a posture adjustment device, the posture adjustment device is used for fixing the camera on a frame and has two rotation degrees of freedom. The image analysis and control program is used for performing the following steps: detecting a human body image and determining an arm corresponding to a dialysis channel based on the detected human body image, and taking the arm as a target arm; controlling the posture adjustment device to adjust the orientation of the camera so that the target arm approaches the center of the viewfinder; adjusting the focal length of the camera to increase the proportion of the target arm in the viewfinder, and returning to the previous step until the proportion of the target arm in the viewfinder exceeds a preset threshold; sampling images from the acquired image stream of the target arm at a preset sampling interval; judging the position change of the target arm in a plurality of groups of adjacent frame images, and judging the tremor if repeated position oscillation occurs, and filtering noise if continuous change occurs; reducing the preset sampling interval for several times and repeating the previous step; A coordinate system is established based on the current camera view area, and the position of the reference point at time point t is (x t , y t ), the position of the reference point at time point t+1 is (x t+1 , y t+1 ), and the displacement of the reference point between adjacent two frames of images is defined as ; the step of judging the position change of the target arm in a plurality of groups of adjacent frame images, and judging the tremor if repeated position oscillation occurs, and filtering noise if continuous change occurs comprises:

2. The system for early monitoring and prevention of hypotension in hemodialysis according to claim 1, characterized in that, obtaining the position of the target arm in the image based on an instance segmentation technology, and defining a reference point for the target arm image; based on the current time point, selecting a plurality of continuous frame images, calculating the displacement of the reference point of the first frame image and the last frame image, and judging whether the displacement is less than a first preset threshold, if not, judging that no tremor occurs, and repeating the current step; if yes, judging whether the displacement of the reference point of the adjacent two frame images is greater than a preset threshold, and whether the ratio of the total displacement to the displacement of the reference point of the first frame image and the last frame image is greater than a second preset threshold, if yes, judging that the tremor occurs, if not, repeating the current step. The liquid supplement adjustment program is used for performing the following steps: obtaining a user body feature parameter; 3. The system for early monitoring and prevention of hypotension in hemodialysis according to claim 2, characterized in that, loading a blood pressure prediction model and importing the user body feature parameter to determine the liquid supplement amount, the liquid supplement speed and the liquid supplement temperature; adjusting the temperature and output flow of the liquid supplement in real time based on the detected tremor symptom of the user. The step of obtaining the user body feature parameter comprises: outputting body feature basic parameter information to the screen for display, wherein the body feature basic parameter is pre-recorded; obtaining blood real-time information based on a data acquisition device, and storing and updating the blood information in a blood information database. Based on the body feature information and blood information database, the latest historical information in the time period is obtained to acquire user body feature parameters.

4. The system for early monitoring and prevention of hypotension in hemodialysis according to claim 3, characterized in that, The step of loading the blood pressure prediction model and importing the user body feature parameters to determine the fluid replacement amount, fluid replacement speed and fluid replacement temperature comprises: The user body feature parameters, standard fluid replacement amount, standard fluid replacement speed and standard fluid replacement temperature are input into the blood pressure prediction model to predict future blood pressure changes; The future blood pressure changes predicted by the blood pressure prediction model are input into a preset reward function to perform calculation and obtain a reward value, wherein the output of the preset reward function is related to the difference between the future blood pressure changes predicted by the blood pressure prediction model and the actually measured blood pressure changes, the greater the difference, the lower the reward value, and the smaller the difference, the smaller the reward value; Based on the reinforcement learning algorithm and the change of the reward value, the fluid replacement amount, fluid replacement speed and fluid replacement temperature are adjusted to improve the reward value, thereby obtaining new fluid replacement amount, fluid replacement speed and fluid replacement temperature, which are substituted into the previous step and executed until the fluid replacement amount, fluid replacement speed and fluid replacement temperature converge within a stable range.

5. The system for early monitoring and prevention of hypotension in hemodialysis according to claim 4, characterized in that, The step of adjusting the temperature and output flow of the fluid replacement in real time based on the detected user tremor symptoms comprises: Judging whether the user has a tremor symptom, and if so, evaluating the severity of the tremor; Based on the severity of the tremor, the current fluid replacement temperature and the output flow, the fluid replacement temperature and the output flow that need to be adjusted are calculated; according to the adjustment value calculated in the previous step, the temperature and output flow of the fluid replacement are adjusted, and it is continuously detected whether the user's tremor state is improved, if so, the fluid replacement temperature and output flow are further adjusted until the user's tremor severity is lower than the standard degree.

Images