Chronic kidney disease patient hemodialysis capacity load evaluation method

Abstract

The invention relates to the technical field of hemodialysis capacity load evaluation, in particular to a chronic kidney disease patient hemodialysis capacity load evaluation method, which comprises the following steps: acquiring staged ultrafiltration rate and body position angle time sequence data in dialysis in real time, generating ultrafiltration and body position coupling time period marks through a space-time matching algorithm, and determining the hemodialysis capacity load of a chronic kidney disease patient according to the ultrafiltration and body position coupling time period marks. During a coupling period, ambulatory blood pressure, jugular vein diameter and pulse wave conduction velocity data are collected at high frequency, an ultrafiltration rate fluctuation value, a body position angle change value and collected physiological index data are input into a pre-training interference calibration model, and blood vessel compliance false interference is decoupled through feature fusion and cross-modal correlation analysis. And finally, combining the weight difference before and after dialysis and biomarker detection data, calculating a dry weight deviation coefficient, and judging capacity overload, capacity insufficiency or a critical state, thereby providing an accurate basis for adjustment of a dialysis scheme, and improving the accuracy and safety of capacity load evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hemodialysis volume load evaluation, in particular to a method for evaluating the hemodialysis volume load of a chronic kidney disease patient. BACKGROUND

[0002] Hemodialysis volume load evaluation is an important technology, which is specifically applied to the precise evaluation of the volume load in the hemodialysis process of a chronic kidney disease patient. The core is to realize the scientific determination of the volume load and the dry body weight by eliminating the coupling interference and restoring the real physiological indicators, which meets the core demand of evaluation accuracy and safety in clinical dialysis.

[0003] During the hemodialysis of a chronic kidney disease patient, the dynamic fluctuation of the staged ultrafiltration rate may occur, and the patient's body position may also change. Due to these two factors, the coupling effect will affect the vascular compliance, resulting in false interference of the volume load related physiological indicators such as dynamic blood pressure, jugular vein diameter and pulse wave transmission velocity, and thus the original data collected cannot truly reflect the volume load state of the patient, affecting the scientificity of the dry body weight deviation evaluation and the volume load determination. In order to solve this technical problem, we provide a method for evaluating the hemodialysis volume load of a chronic kidney disease patient. SUMMARY

[0004] The present application aims to provide a method for evaluating the hemodialysis volume load of a chronic kidney disease patient to solve the problems raised in the background.

[0005] To achieve the above-mentioned purpose, one of the purposes of the present application is to provide a method for evaluating the hemodialysis volume load of a chronic kidney disease patient, comprising the following steps:

[0006] S1, real-time collection of staged ultrafiltration rate data and body position angle time series data in hemodialysis;

[0007] S2, based on the staged ultrafiltration rate data and the body position angle time series data, generating an ultrafiltration and body position coupling period mark through a spatiotemporal matching algorithm of the ultrafiltration rate fluctuation threshold and the body position change timestamp, obtaining the coupling period, and triggering high-frequency collection of dynamic blood pressure data, jugular vein diameter data and pulse wave transmission velocity data in the coupling period;

[0008] S3, inputting the ultrafiltration rate fluctuation value, the body position angle change value and the collected dynamic blood pressure data, jugular vein diameter data and pulse wave transmission velocity data into a pre-trained interference calibration model, and outputting the calibrated volume load related index value after eliminating the false interference of vascular compliance;

[0009] S4. Based on the calibrated volume load correlation index value, combined with the weight difference data before and after dialysis and the single biomarker detection data, generate the dry weight deviation assessment result and volume load status determination.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0011] This invention collects real-time time-series data on staged ultrafiltration rate and body position during hemodialysis. Using a spatiotemporal matching algorithm, it accurately identifies the coupling period between these two parameters, avoiding misjudgments due to unrelated fluctuations. This lays a precise data foundation for subsequent assessments. During the coupling period, it collects dynamic blood pressure, jugular vein diameter, and pulse wave velocity data at high frequencies, ensuring comprehensive capture of physiological indicators and precise correlation with the coupling event. Based on a pre-trained interference calibration model, it precisely decouples spurious interference caused by vascular compliance through feature fusion and cross-modal correlation analysis, restoring indicators that truly reflect volume overload. Furthermore, it calculates the dry weight deviation coefficient by combining the weight difference before and after dialysis, and uses biomarker detection data to determine the volume overload status, avoiding errors from single indicators and significantly improving the accuracy of volume overload assessment. This provides a reliable basis for doctors to adjust dry weight and dialysis protocols, effectively ensuring the safety and efficacy of hemodialysis for patients with chronic kidney disease. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating the overall workflow of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Please see Figure 1 As shown in the figure, this embodiment provides a method for assessing hemodialysis volume load in patients with chronic kidney disease, including the following steps:

[0015] S1. Real-time acquisition of staged ultrafiltration rate data and body position angle time-series data during hemodialysis;

[0016] S2. Based on the phased ultrafiltration rate data and body position angle time series data, the ultrafiltration and body position coupling period marker is generated by the spatiotemporal matching algorithm of ultrafiltration rate fluctuation threshold and body position change timestamp, the coupling period is obtained, and high-frequency acquisition of dynamic blood pressure data, jugular vein diameter data and pulse wave conduction velocity data is triggered within the coupling period.

[0017] S3. Input the ultrafiltration rate fluctuation value, body position angle change value, and the collected dynamic blood pressure data, jugular vein diameter data, and pulse wave conduction velocity data into the pre-trained interference calibration model, and output the calibrated volume load correlation index value after eliminating spurious interference from vascular compliance.

[0018] S4. Based on the calibrated volume load correlation index value, combined with the weight difference data before and after dialysis and the single biomarker detection data, generate the dry weight deviation assessment result and volume load status determination.

[0019] The spatiotemporal matching algorithm between the ultrafiltration rate fluctuation threshold and the body position change timestamp is as follows:

[0020] Based on the dialysis stage identifier in the phased ultrafiltration rate data, the preset ultrafiltration rate fluctuation threshold for the corresponding stage is invoked. When the real-time ultrafiltration rate data exceeds the current stage ultrafiltration rate fluctuation threshold N times consecutively, an ultrafiltration dynamic event start timestamp is generated. At the same time, the body position angle time series data is parsed. If the change in body position angle reaches the preset angle change threshold and is maintained for more than the preset duration, a body position change event start timestamp is generated. The timestamps of the ultrafiltration dynamic event and the body position change event are matched with a time window. If the time difference between the two is less than the preset tolerance window, it is determined that an ultrafiltration and body position coupling period marker is generated.

[0021] Ultrafiltration and positional coupling time markers, specifically including:

[0022] After the timestamps of the ultrafiltration rate fluctuation event and the body position change event are successfully matched, the peak time point of the ultrafiltration rate fluctuation and the stable time point of the body position angle change are extracted. The overlapping part of the two time intervals is taken as the core interval of the coupling period, and extended forward to the ultrafiltration rate fluctuation start time point and backward to the stable body position change time point to form a complete coupling period marker. The coupling period marker includes the quantified values ​​of the ultrafiltration rate fluctuation direction, the body position angle change direction, and the change amplitude of the two parameters.

[0023] During the coupling period, high-frequency acquisition of dynamic blood pressure data, jugular vein diameter data, and pulse wave velocity data is triggered, specifically including:

[0024] Based on the start time point marked by the coupling period, the ambulatory blood pressure monitor is activated to continuously collect systolic and diastolic blood pressure data at a preset high-frequency sampling interval. Simultaneously, the neck ultrasound probe is triggered to scan the cross-section of the jugular vein at a fixed short cycle within the coupling period and output the diameter measurement value. At the same time, the wearable pulse wave sensor is activated to calculate the pulse wave conduction velocity in real time through dual-point detection of the radial and femoral arteries. All collected data are bound to the coupling period mark number.

[0025] The pre-trained interference calibration model is constructed in the following way:

[0026] Data on the coupling periods marked in historical dialysis cases were collected, including ultrafiltration rate fluctuations, changes in body position angles, raw ambulatory blood pressure data, raw jugular vein diameter data, raw pulse wave velocity data, and their corresponding true volume load labels. The combination of ultrafiltration rate fluctuations and changes in body position angles was used as the input feature vector, and the deviation between the raw physiological index data and the true labels was used as the training objective. A neural network model was trained through supervised learning to enable the model to learn the mapping law of the coupling effect of ultrafiltration and body position on the pseudo-interference of vascular compliance.

[0027] The ultrafiltration rate fluctuation value, body position angle change value, and collected physiological index data are input into the pre-trained interference calibration model. The specific analysis logic is as follows:

[0028] The model first fuses the ultrafiltration rate fluctuation value with the body position angle change value to generate an interference feature vector representing the coupling strength. At the same time, it extracts the fluctuation amplitude feature from the dynamic blood pressure data according to the time window, the diameter change slope feature from the jugular vein diameter data, and the time delay change feature from the pulse wave conduction velocity data. Together, they form a physiological indicator feature group. The interference feature vector and the physiological indicator feature group are then subjected to cross-modal correlation analysis to identify the causal relationship chain between the coupling parameters and the changes in physiological indicators.

[0029] The working logic of cross-modal correlation analysis includes:

[0030] The interference calibration model has a built-in interference decoupling module. The interference decoupling module calculates the influence coefficient matrix of vascular compliance pseudo-interference through interference feature vectors. The influence coefficient matrix includes the conduction coefficient of blood pressure fluctuation, the contraction coefficient of jugular vein diameter, and the time delay coefficient of pulse wave conduction velocity. The influence coefficient matrix is ​​then multiplied by the physiological indicator feature group to deconstruct the contribution of coupling interference to each physiological indicator.

[0031] Output calibrated volume load correlation index values ​​after eliminating spurious interference from vascular compliance, specifically implemented as follows:

[0032] The interference calibration model subtracts the interference contribution component calculated by the interference decoupling module from the original physiological indicator feature group in reverse. For dynamic blood pressure data, it restores the true blood pressure fluctuation value by superimposing the inverse operation of the interference conduction coefficient. For jugular vein diameter data, it restores the actual diameter change by scaling the inverse of the interference contraction coefficient. For pulse wave conduction velocity data, it corrects the conduction delay using the compensation function of the time delay coefficient. Finally, it outputs the corrected dynamic blood pressure calibration value, jugular vein diameter calibration value, and pulse wave conduction velocity calibration value as volume load related index values.

[0033] The dry weight deviation assessment results are generated as follows:

[0034] The output calibrated volume load correlation index value is time-weighted averaged over the entire dialysis session to obtain the comprehensive calibration index value. The theoretical volume clearance is calculated by combining the weight difference data before and after dialysis. The ratio of the comprehensive calibration index value to the theoretical volume clearance is used as the dry weight deviation coefficient. If several weight deviation coefficients exceed the preset reasonable range, a dry weight assessment result including the deviation direction and adjustment amount is generated.

[0035] Capacity load status determination specifically includes:

[0036] Based on the dry weight deviation assessment results, if the deviation coefficient is within a reasonable range, the capacity load status is directly judged according to the trend of the comprehensive calibration index value. If the deviation coefficient exceeds the limit, the joint decision rule based on the comprehensive calibration index value and biomarker detection data is adopted first. When the comprehensive calibration index value indicates abnormal capacity load but the biomarker data does not support it, it is judged as a false interference residue, triggering model recalibration. When both support the abnormality, the results are divided into three categories: capacity overload, capacity underload, and critical state according to the preset threshold.

[0037] Further explanation is needed: After completing the real-time acquisition of staged ultrafiltration rate data and body position angle time-series data during hemodialysis for patients with chronic kidney disease, in order to accurately identify the correlation between ultrafiltration rate fluctuations and body position changes, and to avoid misjudging unrelated independent fluctuations as volume load-related signals, it is necessary to establish the temporal correlation between the two through a spatiotemporal matching algorithm. The specific implementation method is as follows:

[0038] The spatiotemporal matching algorithm for ultrafiltration rate fluctuation thresholds and body position change timestamps is based on the correlation between dialysis stage differences and event timestamps. First, according to the dialysis stage identifier in the staged ultrafiltration rate data, a preset ultrafiltration rate fluctuation threshold for the corresponding stage is called. The dialysis stage identifier is a stage label embedded in the ultrafiltration rate data, automatically generated by the dialysis machine according to the preset treatment procedure: 0-2 hours in the early stage, 2-4 hours in the middle stage, and 4-5 hours in the late stage. The ultrafiltration goals and patient tolerance differ in different stages, and the fluctuation threshold is also set differently. Clinical data statistics show that patients have relatively sufficient blood volume in the early stage of dialysis, so the fluctuation threshold is set accordingly. ±5ml / min is used to avoid oversensitivity; ±8ml / min is used in the mid-term for dynamic blood volume adjustment; and ±3ml / min is used in the later stage when blood volume approaches the target value and requires strict control. These thresholds are pre-stored in the parameter library of the assessment system. The system automatically retrieves and calls the corresponding threshold by parsing the stage identifier in the data. When the real-time ultrafiltration rate data continuously exceeds the current stage's ultrafiltration rate fluctuation threshold, an ultrafiltration dynamic event start timestamp is generated. Real-time ultrafiltration rate data is collected by the dialysis machine's flow sensor at a frequency of 1 time / minute. The criterion for continuous exceedance is that three consecutive sampling values ​​exceed the threshold. Misjudgments caused by transient interference, such as a mid-dialysis threshold of ±8 ml / min (three consecutive sampling values ​​of 15 ml / min, 16 ml / min, and 17 ml / min, all exceeding the upper limit of 8 ml / min), will result in the system recording the time of the first exceedance as the start timestamp of the over-filtering event, while simultaneously marking the direction of the fluctuation. Simultaneously, the system analyzes the body position angle time-series data, which is collected by an inertial measurement unit worn on the patient's chest and abdomen. This sensor outputs the angle between the patient's torso and the horizontal plane, forming a continuous angle time-series sequence. If the change in body position angle reaches a preset angle change threshold and... If the duration exceeds a preset time, a start timestamp for the position change event is generated. The preset angle change threshold is set to 15°, and the preset duration is set to 30 seconds to avoid misjudgment of instantaneous movements such as turning over. For example, if a patient changes from a supine position at 0° to a lateral position at 35°, the change is 35° ≥ 15°, and this angle is maintained for 32 seconds. The system records the time when the angle first reaches 35° as the start timestamp of the position change event. Subsequently, the timestamps of the ultrafiltration motion event and the position change event are matched using a time window. Time window matching calculates the absolute difference between the two timestamps. If the time difference is less than a preset tolerance window, a coupling period marker for ultrafiltration and position change is generated. The preset tolerance window is set to 60 seconds. For example, if the start timestamp of ultrafiltration motion is 10:05:30 and the start timestamp of position change is 10:06:10, the time difference of 40 seconds is less than 60 seconds, indicating a correlation between the two events. A coupling period marker is generated, containing the start timestamps of both events and the correlation determination result.

[0039] After determining the correlation between ultrafiltration and body position changes through a spatiotemporal matching algorithm, it is necessary to further generate precise markers for the coupling period between ultrafiltration and body position. These markers must fully cover the entire process of the associated events and quantify key parameters to provide clear data support for subsequent capacity load assessment. The specific implementation method is as follows:

[0040] The core of generating ultrafiltration and body position coupling time markers is defining the range of the coupling time period and extracting key quantitative information. First, after successfully matching the timestamps of the ultrafiltration motion event and the body position change event, the peak time point of the ultrafiltration rate fluctuation and the stable time point of the body position angle change are extracted. The peak time point of the ultrafiltration rate fluctuation refers to the time when the rate reaches its maximum value in the ultrafiltration motion event. For example, in the above fluctuation event, the rate reaches 20 ml / min (peak) at 10:08:20, and this time is the peak time point. The stable time point of the body position angle change refers to the starting time when the body position angle reaches the target value and the fluctuation of three consecutive sampling values ​​is less than or equal to 2°. For example, if the patient starts lying on their side at 10:06:10 and the angle stabilizes at 35°±1° after 10:06:40, then 10:06:40 is the stable time point. Then, the overlapping part of the two time intervals is taken as the core interval of the coupling time period, with the time intervals being the ultrafiltration motion interval and the body position interval, respectively. The change interval records the overlapping portion of the two. This interval is the core stage directly related to ultrafiltration rate and body position change. As the core interval of the coupling period, to ensure coverage of the complete process of the related events, the core interval needs to be extended forward to the start time stamp of ultrafiltration rate and backward to the stable time stamp of body position change, forming a complete coupling period marker. This coupling period marker contains three types of key quantitative values: first, the direction of ultrafiltration rate fluctuation, determined by the positive or negative difference between the peak rate and the initial rate; second, the direction of body position angle change, determined by comparing the angle values ​​before and after the body position change; and third, the quantitative value of the amplitude of the two parameter changes: the amplitude of ultrafiltration rate fluctuation is the difference between the peak rate and the initial rate, and the amplitude of body position angle change is the difference between the stable angle and the initial angle. These quantitative values, together with the time period range, constitute a complete marker, which is stored in the database of the evaluation system to provide accurate data support for subsequent analysis of the impact of body position change on volume load.

[0041] After generating ultrafiltration and positional coupling time markers, in order to accurately capture the dynamic changes of patients' vascular physiological indicators during this time period, it is necessary to simultaneously trigger multiple types of monitoring devices to collect key data in a high-frequency sampling mode, ensuring that each data point can be accurately correlated with the coupling event. The specific implementation method is as follows:

[0042] Based on the start time of the coupling period marker, the system automatically activates the ambulatory blood pressure monitor via a wireless communication module. This monitor is an upper-arm portable ambulatory blood pressure monitor that continuously collects systolic and diastolic blood pressure data at a preset high-frequency sampling interval. The preset high-frequency sampling interval is set to 30 seconds / sample. Considering that the coupling period is usually tens of seconds to 2 minutes, the 30-second interval can fully capture the fluctuations in blood pressure due to the ultrafiltration-position coupling effect without generating too much redundant data. After the monitor collects data, it automatically adds a collection timestamp to each set of systolic and diastolic blood pressure data, binds it to the current coupling period marker number, stores it in the local cache, and simultaneously uploads it to the evaluation system's database in real time. At the same time as activating the ambulatory blood pressure monitor, the system simultaneously triggers the neck ultrasound probe. This probe is a high-frequency linear array probe, mounted on an adjustable bracket. During the coupling period, it scans the cross-section of the jugular vein at a fixed short cycle, set to 10 seconds / sample. The response time of the jugular vein diameter affected by positional changes is approximately 5-8 seconds. The 10-second scanning cycle can accurately track diameter changes. The scanning area is selected as 2 cm above the right supraclavicular fossa of the patient. The probe automatically identifies the jugular vein lumen through the ultrasound host, calculates the maximum anteroposterior diameter of the lumen as the diameter measurement value, and adds a timestamp and coupling period marker number to the measurement result to achieve time synchronization with the blood pressure data. At the same time, the system activates the wearable pulse wave sensor, which contains two detection units, worn on the radial and femoral arteries on the same side of the patient, respectively. The sensor acquires the pulse wave signals of the two arteries in real time through photoplethysmography. The sensor records the pulse wave waveform at a sampling frequency of 50Hz, calculates the time difference of the pulse wave propagation from the femoral artery to the radial artery, and combines it with the preset distance between the two arteries to calculate the pulse wave conduction velocity in real time. The calculated PWV data is also timestamped and bound to the coupling period marker number to ensure that it forms a complete coupled event and multi-physiological index associated dataset with the dynamic blood pressure and jugular vein diameter data, providing complete input for subsequent interference calibration.

[0043] Because the raw physiological data collected during the coupling period are easily affected by the coupling effect of ultrafiltration rate fluctuations and body position changes, a pre-trained interference calibration model needs to be constructed to eliminate this spurious interference, so that the calibrated data can truly reflect the patient's volume load status. The specific construction process is as follows:

[0044] First, data from marked coupling time periods in historical dialysis cases were collected. Cases were derived from the dialysis records of chronic kidney disease patients at the hospital's dialysis center over the past three years, covering at least 300 patients. At least five valid coupling time periods were extracted from each patient. The collected data dimensions included: ultrafiltration rate fluctuation (the difference between the peak and initial ultrafiltration rate during the coupling time period); body position angle change (the difference between the stable and initial body position angle); raw ambulatory blood pressure data (systolic and diastolic blood pressure values ​​at all 30-second intervals during the coupling time period); raw jugular vein diameter data (diastolic vein diameter measurements at all 10-second intervals); raw pulse wave velocity data; and PWV values ​​calculated from all 50Hz samples. The corresponding true volume overload label, which is the clinically recognized gold standard for volume overload, is determined by physicians based on a comprehensive assessment of the patient's weight difference before and after dialysis, the real-time inferior vena cava diameter during dialysis, and the patient's clinical symptoms. It is classified into three levels: excessive, normal, and excessively low volume overload, or labeled as continuous numerical values, providing a supervisory signal for the model. Subsequently, the model's input feature vector and training target are constructed. The combination of ultrafiltration rate fluctuation and body position angle change is used as the input feature vector. First, the two parameters are normalized, scaling the ultrafiltration rate fluctuation to the 0-20 ml / min range and the body position angle change to the 0-90° range to eliminate dimensional differences. Then, the two normalized values ​​are... After normalization, the data is combined into two-dimensional vectors, and each coupling time period corresponds to an input feature vector. The deviation between the original physiological index data and the true volume load label is used as the training target. For example, if the average original systolic blood pressure in a certain coupling time period is 140 mmHg, while the corrected systolic blood pressure (i.e., the true value after eliminating interference) corresponding to the true volume load label is 132 mmHg, the deviation of 8 mmHg is the training target for the systolic blood pressure index in that time period. Similarly, the deviation values ​​of jugular vein diameter and pulse wave velocity are calculated to form a three-dimensional training target vector. Finally, a neural network model is trained through supervised learning. The model uses a shallow neural network structure to avoid overfitting and adapt to the limited medical data. The model consists of an input layer with two neurons, corresponding to the two dimensions of the input feature vector; a hidden layer with one layer containing 16 neurons, using the ReLU activation function to enhance nonlinear fitting ability; and an output layer with three neurons, corresponding to the deviation values ​​of three physiological indicators, using a linear activation function to adapt to continuous deviation targets. During training, the dataset is divided into training and validation sets in a 7:3 ratio. The Adam optimizer is used, with mean squared error as the loss function to measure the difference between the model's predicted deviation and the actual deviation. The model is trained iteratively for 50-100 rounds, and the loss is evaluated on the validation set after each round. Training stops when the loss on the validation set no longer decreases for 5 consecutive rounds, and the final model parameters are saved.This training process enables the model to learn the mapping rules of ultrafiltration and body position coupling on spurious interference with vascular compliance, resulting in a pre-trained interference calibration model. For example, the interference calibration model can learn that when the ultrafiltration rate fluctuates by +5 ml / min and the body position angle changes by +20°, the systolic blood pressure will produce a spurious increase deviation of +6 mmHg, providing an accurate basis for interference prediction for subsequent calibration of raw data.

[0045] After completing the pre-trained interference calibration model construction and collecting multi-dimensional data during the coupling period, the ultrafiltration and body position coupling parameters and physiological index data need to be input into the model for systematic analysis. The core is to accurately identify the false interference of coupling on physiological indicators through feature-level fusion and correlation, so as to avoid affecting the accuracy of volume load assessment. The specific analysis logic is as follows: The model first performs feature fusion on the input ultrafiltration rate fluctuation value and body position angle change value to generate an interference feature vector representing the coupling strength. Feature fusion is the process of forming a unified feature by normalizing and concatenating two independent parameters. First, the ultrafiltration rate fluctuation value is converted to 0.4 according to the normalization rule during model training, and the body position angle change value is mapped to 0 according to 0-90°. The rule of -1 is converted to 0.39, and then the two normalized values ​​are concatenated in the format of [ultrafiltration dynamic normalized value, postural change normalized value] to form a two-dimensional interference feature vector. The magnitude of this vector directly represents the strength of ultrafiltration-postural coupling. The closer the value is to 1, the stronger the potential interference of the coupling on physiological indicators. While generating the interference feature vector, the model extracts features from the three types of collected physiological indicator data to construct physiological indicator feature groups. For dynamic blood pressure data, the fluctuation amplitude feature is extracted according to the time window. The time window is set to three consecutive blood pressure sampling points within the coupling period, corresponding to 90 seconds, taking into account both data volume and timeliness. The difference between the maximum and minimum systolic blood pressure and the difference between the maximum and minimum diastolic blood pressure within each time window are calculated. The difference between the minimum and maximum values ​​is used as a feature of blood pressure fluctuation amplitude, reflecting the dynamic changes in blood pressure under coupling effects and avoiding the random errors of single-point data. For jugular vein diameter data, the slope feature of diameter change is extracted. Taking two adjacent jugular vein diameter measurements within the coupling period as a group, the difference between the later and previous measurements is calculated and then divided by the time interval to obtain the slope of diameter change every 10 seconds. This feature reflects the changing trend of jugular vein diameter due to position and ultrafiltration coupling effects. For pulse wave velocity (PWV) data, the time delay change feature is extracted, and the difference between the maximum and minimum PWV values ​​within the coupling period is calculated. This feature avoids the problem of the absolute value of PWV being affected by individual baseline arterial elasticity, focusing on coupling. The instantaneous changes in PWV caused by the interference are more in line with the requirements of interference calibration. Finally, the model performs cross-modal correlation analysis on the interference feature vector and the physiological indicator feature group to identify the causal relationship chain between the coupling parameters and the changes in physiological indicators. Cross-modal correlation refers to establishing the correspondence between the coupling strength feature and the dynamic features of physiological indicators. The model judges whether the coupling strength corresponding to the current interference feature vector matches the fluctuation trend in the physiological indicator feature group through the mapping rules learned during training. If the trends of the two are consistent, it is determined that there is a causal relationship between the change of the physiological indicator and the coupling interference, and it is marked as a feature to be calibrated. If the trends are not related, it is determined that the change of the indicator is caused by the actual capacity load and no calibration is required, providing a clear target direction for subsequent interference decoupling.

[0046] The core of cross-modal correlation analysis is to quantify and decompose the impact of coupling interference on various physiological indicators through the model's built-in interference decoupling module, ensuring that the calibrated physiological indicators can truly reflect the patient's volume overload status. Its working logic is as follows:

[0047] The key execution unit for cross-modal correlation analysis is the interference decoupling module built into the interference calibration model. This module pre-stores the coupling strength and interference influence mapping parameters learned during model training. Based on the input interference feature vector, it can accurately calculate the influence coefficient matrix of vascular compliance pseudo-interference. Vascular compliance pseudo-interference refers to the coupling effect between ultrafiltration rate fluctuations and body position changes, leading to temporary changes in vascular elasticity, which in turn causes pseudo-changes in physiological indicators that are unrelated to the true volume load. The influence coefficient matrix is ​​a 3x1 matrix containing three core coefficients: the conduction coefficient for blood pressure fluctuations, the contraction coefficient for jugular vein diameter, and the time delay coefficient for pulse wave conduction velocity. These coefficients are obtained by the model through historical data training and are dynamically adjusted according to the values ​​of the interference feature vector. After obtaining the influence coefficient matrix, the interference decoupling module performs a dot product operation with the physiological indicator feature group to deconstruct the contribution components of coupling interference in each physiological indicator. The dot product operation means multiplying each coefficient in the influence coefficient matrix by the physiological indicator feature group. Multiplying the corresponding eigenvalues ​​in the indicator feature group one by one yields the contribution component caused by coupling interference in that indicator feature. For example, if the blood pressure conduction coefficient in the influence coefficient matrix is ​​0.6, and the blood pressure fluctuation amplitude feature in the physiological indicator feature group is 15 mmHg, then the interference contribution component in blood pressure fluctuation = 0.6 × 15 mmHg = 9 mmHg. This indicates that 9 mmHg of the 15 mmHg blood pressure fluctuation is a false fluctuation caused by ultrafiltration-positional coupling interference. If the jugular vein diameter contraction coefficient... Given a value of -0.03 and a diameter change slope characteristic of 0.2 mm / s, the interference contribution component in the jugular vein diameter change is -0.03 × 0.2 mm / s = -0.006 mm / s. This indicates that 0.006 mm / s of the diameter increase trend is due to pseudo-inhibition caused by interference. Similarly, multiplying the PWV delay coefficient of 0.8 by the PWV delay change characteristic of 2 m / s yields an interference contribution component of 1.6 m / s, meaning that 1.6 m / s of the 2 m / s PWV change is caused by interference. Through this calculation, the model can accurately isolate pseudo-interference components from various physiological indicators, laying the foundation for subsequent assessment of volume load based on real physiological indicators.

[0048] After the interference decoupling module completes the decomposition of the coupling interference contribution components, the interference calibration model needs to perform targeted correction calculations to reversely eliminate the spurious interference of vascular compliance from the original physiological indicator feature set, restore the indicator values ​​that can truly reflect volume load, and ensure the accuracy of subsequent evaluation results. The specific implementation is as follows:

[0049] The interference calibration model first calls the interference decoupling module to calculate the various interference contribution components, and then subtracts them from the corresponding original physiological indicator feature groups. The core logic of the reverse subtraction is that the original feature value minus the interference contribution component equals the true feature value. However, it is necessary to perform targeted calculations based on the characteristics of different physiological indicators and the interference coefficient to avoid correction bias caused by simple subtraction. For ambulatory blood pressure data, the true blood pressure fluctuation value is restored by superimposing the inverse operation of the interference conduction coefficient. The interference conduction coefficient is the coefficient in the influence coefficient matrix mentioned above that quantifies the influence of interference on blood pressure fluctuation. Its inverse operation is to cancel the interference conduction effect through the inverse mapping of the coefficient. For example, if the original blood pressure fluctuation amplitude feature is 15m The interference contribution component is 9 mmHg (derived from 0.6 × 15 mmHg). First, calculate the inverse proportionality of the interference conduction coefficient (1 ÷ 0.6 ≈ 1.67). Then, multiply the result of the original fluctuation amplitude minus the interference contribution component by this inverse proportionality, i.e., (15-9) × 1.67 ≈ 10 mmHg. This 10 mmHg is the restored true blood pressure fluctuation value. The reason for this operation is that the interference conduction coefficient reflects the amplification or reduction effect of the interference on blood pressure. The inverse operation can accurately cancel this effect and avoid the incomplete correction caused by the omission of coefficients in direct subtraction. For the jugular vein diameter data, the actual diameter change is restored by scaling the inverse of the interference contraction coefficient.The interference contraction coefficient is a coefficient in the influence coefficient matrix that quantifies the impact of interference on the contraction / dilation of the jugular vein diameter. Its reciprocal is used to eliminate the distortion of the diameter change trend caused by interference through proportional scaling. For example, if the original jugular vein diameter change slope is 0.2 mm / s and the interference contribution component is -0.006 mm / s, first calculate the reciprocal of the absolute value of the interference contraction coefficient, then multiply the result of the original change slope minus the interference contribution component by this reciprocal, i.e., (0.2 - (-0.006)) × 33.33 ≈ 6.87 mm / min, which is approximately 0.114 mm / s. This value is the actual diameter change after recovery. This process can correct the deviation of the diameter change slope caused by interference, ensuring that the result closely matches the jugular vein diameter change pattern caused by the actual volume load. For pulse wave velocity (PWV) data, a compensation function of the time delay coefficient is used to correct the conduction delay. The time delay coefficient is a coefficient in the influence coefficient matrix that quantifies the impact of interference on the PWV time delay change, and the compensation function is... The corrected PWV time delay change is calculated as follows: (Original time delay change - Interference contribution component) × (1 + Time Delay Coefficient). For example, if the original PWV time delay change is 2 m / s and the interference contribution component is 1.6 m / s (derived from 0.8 × 2), substituting these values ​​into the compensation function yields (2 - 1.6) × (1 + 0.8) = 0.4 × 1.8 = 0.72 m / s. This value is the corrected PWV time delay change. The addition of the term "1 + Time Delay Coefficient" to the compensation function is because PWV is affected by both arterial elasticity and volume load. The time delay coefficient reflects the deviation of the interference in assessing arterial elasticity. Adding this term can simultaneously calibrate the elasticity assessment deviation, making the PWV data more accurately correlated with volume load. Finally, the model outputs the corrected dynamic blood pressure calibration value, jugular vein diameter calibration value, and pulse wave conduction velocity calibration value. These three types of values ​​serve as volume load correlation index values ​​and are stored in the patient dialysis records of the assessment system, providing core data support for subsequent dry weight deviation assessment.

[0050] After obtaining the calibrated volume load correlation index values, in order to further assess whether the patient's dry weight is suitable for the current dialysis protocol, dry weight is the target weight for the patient to reach the ideal volume load state after dialysis. Excessive deviation will lead to excessively high or low volume load. It is necessary to calculate the dry weight deviation coefficient by combining the data of the entire dialysis process, and finally generate the assessment results. The specific implementation is as follows:

[0051] First, the output calibrated volume load correlation index values ​​are time-weightedly averaged over the entire duration of a single dialysis session to obtain a comprehensive calibrated index value. Time-weighted averaging involves assigning different weights based on the clinical significance of the index value collection time period, rather than simply taking an arithmetic average. Since index values ​​within the coupled time period better reflect the dynamic changes in volume load influenced by ultrafiltration and body position, the weight is set to 0.7, while the weight of index values ​​in the non-coupled time period is set to 0.3, ensuring that the comprehensive value more closely reflects the volume load status during critical periods. During calculation, the three types of calibrated index values ​​are first extracted at a frequency of 1 minute / session throughout the dialysis session, and each time point is marked as belonging to a coupled time period. Then, each type of index value is calculated as (coupled time period index value × 0.7). The weighted average of the three indicators (0.7 + non-coupling period indicator value × 0.3) ÷ total time points is calculated logically. Finally, the weighted averages of the three indicators are summed according to the following proportions: blood pressure (40%), jugular vein diameter (30%), and PWV (30%), to obtain the comprehensive calibration indicator value. Then, the theoretical volume clearance is calculated by combining the pre- and post-dialysis weight difference data. This pre- and post-dialysis weight difference directly reflects the amount of fluid cleared during dialysis, i.e., the theoretical volume clearance. Since the cleared fluid is mainly water, 1 kg of weight difference is approximately equal to 1 L of fluid. Therefore, the theoretical volume clearance = pre- and post-dialysis weight difference × 1 L / kg (e.g., 2 kg × 1 L / kg). =2L). If the patient receives intravenous fluids during dialysis, the fluid volume must be deducted from the weight difference to ensure the result reflects the actual excess volume cleared. Then, the ratio of the comprehensive calibration index value to the theoretical volume clearance is used as the dry weight deviation coefficient. The calculation logic for the dry weight deviation coefficient is: Deviation coefficient = Comprehensive calibration index value ÷ Theoretical volume clearance. This coefficient reflects the degree of improvement in volume load corresponding to a unit volume clearance. If the coefficient is high, it indicates that the volume load is still too high after clearing a unit volume, possibly due to an underestimation of dry weight or a deviation in volume load assessment. If the coefficient is low, it indicates that the volume load is already too low after clearing a small amount of volume, possibly due to an overestimation of dry weight. Finally, a preset value for the dry weight deviation coefficient is set. The optimal range for dry weight setting is determined through clinical data statistics to be 2-4. If any weight deviation coefficient exceeds this range, a dry weight assessment result is generated, including the direction of deviation and the adjustment amount. If the coefficient is greater than 4, the dry weight setting is considered too low, and the adjustment amount is calculated as (deviation coefficient - 4) × 0.1 kg. If the coefficient is less than 2, the dry weight setting is considered too high, and the adjustment amount is calculated as (2 - deviation coefficient) × 0.1 kg (e.g., 2 - 1.8 = 0.2, adjustment amount ≈ 0.02 kg, i.e., a reduction of 0.02 kg is recommended). The assessment result is output in text form, along with the comprehensive calibration index value, theoretical volume clearance, and specific values ​​of the deviation coefficient, for doctors to refer to and adjust subsequent dialysis plans.

[0052] After completing the dry weight deviation assessment and obtaining the deviation coefficient, it is necessary to further combine the comprehensive calibration index values ​​and biomarker detection data to make the final determination of the patient's current volume load status. The core logic is:

[0053] When the dry weight deviation coefficient is within a reasonable range, it indicates that the dry weight setting is well-matched with the volume clearance rhythm. This can be directly judged by the dynamic trend of the comprehensive calibration index values. If the deviation coefficient exceeds the limit, cross-validation with biomarker data is required to avoid misjudgment due to error of a single index. The specific implementation method is as follows:

[0054] The determination of volume load status is first based on the assessment results of dry weight deviation, which divides the judgment path accordingly. If the deviation coefficient is within the preset reasonable range of 2-4, it indicates that the current dry weight setting matches the volume clearance effect, and no additional cross-validation is required. In this case, the volume load status is directly judged based on the trend of the comprehensive calibration index value. The trend of the comprehensive calibration index value refers to the direction and magnitude of the change of the comprehensive calibration index value over time throughout a single dialysis session. If the trend shows a continuous decline and eventually stabilizes within the normal range of 3-7, it is judged that the volume load clearance is effective and the current status is normal. If the trend is basically stable and always within the range of 3-7, it is judged that the initial volume load is normal and the clearance rhythm is appropriate. If the trend declines slowly and eventually remains above 7, it is judged that the volume load clearance is insufficient and the current status is too high. If the trend declines rapidly and falls below 3, it is judged as volume overload. Excessive body load clearance and a low current state necessitate excluding short-term fluctuations when assessing trends. Analysis should use a 15-minute time window to average the results, ensuring they reflect overall changes. If the deviation coefficient exceeds the reasonable range of 2-4, it indicates a potential bias in the dry weight setting or incomplete elimination of interference from the comprehensive calibration index. In this case, a joint decision-making rule based on comprehensive calibration index values ​​and biomarker detection data should be prioritized. Biomarker detection data refers to clinically recognized volume load-related biomarkers, specifically brain natriuretic peptide (BNP) or N-terminal pro-BNP (NT-proBNP). The levels of these two biomarkers increase significantly with increasing volume load. Detection timing is synchronized with the coupling period, and the detection method is immunochromatography. Reference ranges are stratified by patient age. The joint decision-making rule is implemented in two scenarios:

[0055] When the comprehensive calibration index indicates abnormal volume overload but the biomarker data does not support it, it is judged as residual false interference. This situation is usually due to the special interference pattern of ultrafiltration and body position coupling not being fully calibrated by the model. At this time, the model recalibration process needs to be triggered: the system automatically marks the coupling period data, raw physiological indicators, true values ​​of biomarkers, and judgment results of this dialysis as samples to be calibrated and adds them to the model's training dataset. When the cumulative number of samples to be calibrated reaches 50, supervised learning training is restarted to update the model parameters to improve the calibration ability for special interference patterns. When both the comprehensive calibration index value and the biomarker data support abnormality, the system classifies the volume overload, volume underload, and critical state according to preset thresholds. As a result, the preset thresholds were determined based on clinical data statistics. When the comprehensive calibration index value was greater than 7 and BNP was greater than 100 pg / mL (under 60 years old) / 150 pg / mL (over 60 years old), it was considered volume overload; when the comprehensive calibration index value was less than 3 and BNP was less than 50 pg / mL, it was considered volume insufficiency; when the comprehensive calibration index value was in the range of 6.5-7.5 or 2.5-3.5 and the biomarker data was close to the critical value, it was considered a critical state. In the critical state, a suggestion for close monitoring should be added to avoid volume fluctuations in subsequent dialysis. All judgment results should be accompanied by key data and output to the dialysis center's management system in tabular form for doctors to refer to and adjust the next dialysis plan.

[0056] This invention collects real-time time-series data on ultrafiltration rate and body position angle during dialysis at different stages. A spatiotemporal matching algorithm generates markers for the coupling of ultrafiltration and body position. During these coupling periods, dynamic blood pressure, jugular vein diameter, and pulse wave velocity data are collected at high frequency. The ultrafiltration rate fluctuations, body position angle changes, and collected physiological data are input into a pre-trained interference calibration model. Feature fusion and cross-modal correlation analysis decouple vascular compliance pseudo-interference, outputting calibrated volume load correlation index values. Finally, combining the weight difference before and after dialysis with biomarker detection data, the dry weight deviation coefficient is calculated to determine volume overload, underload, or critical state, providing a precise basis for adjusting dialysis protocols and improving the accuracy and safety of volume load assessment.

[0057] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for assessing the volume load of hemodialysis in a patient with chronic kidney disease, characterized in that: The method comprises the following steps: S1, real-time acquisition of phased ultrafiltration rate data and body position angle time series data in hemodialysis; S2, based on the phased ultrafiltration rate data and the body position angle time series data, an ultrafiltration and body position coupling time period marker is generated through a space-time matching algorithm of ultrafiltration rate fluctuation threshold and body position change timestamp, a coupling time period is obtained, and high-frequency acquisition of dynamic blood pressure data, jugular vein diameter data and pulse wave conduction velocity data is triggered within the coupling time period; S3, the ultrafiltration rate fluctuation value, the body position angle change value and the acquired dynamic blood pressure data, jugular vein diameter data and pulse wave conduction velocity data are input into a pre-trained interference calibration model, and a calibrated volume load correlation index value eliminating false interference of vascular compliance is output; S4, based on the calibrated volume load correlation index value, a dry weight deviation evaluation result and a volume load state judgment are generated in combination with pre-and post-dialysis body weight difference data and single biomarker detection data.

2. The method for evaluating the volume status of a chronic kidney disease patient according to claim 1, wherein: The space-time matching algorithm of the ultrafiltration rate fluctuation threshold and the body position change timestamp is specifically: According to the dialysis stage identifier in the phased ultrafiltration rate data, a preset corresponding stage ultrafiltration rate fluctuation threshold is called, when the real-time ultrafiltration rate data exceeds the current stage ultrafiltration rate fluctuation threshold for N times in succession, an ultrafiltration fluctuation event starting timestamp is generated, the body position angle time series data is analyzed, if the body position angle change amount reaches a preset angle change threshold and is maintained for more than a preset time length, a body position change event starting timestamp is generated, the timestamps of the ultrafiltration fluctuation event and the body position change event are matched in a time window, and if the time difference between the two is less than a preset tolerance window, it is determined that the ultrafiltration and body position coupling time period marker is generated.

3. The method for evaluating the volume status of a chronic kidney disease patient according to claim 2, wherein: The generation of the ultrafiltration and body position coupling time period marker specifically includes: After the ultrafiltration fluctuation event and the body position change event timestamp matching is successful, the ultrafiltration rate fluctuation peak time point and the body position angle change stable time point are extracted, the overlapping part of the time interval of the two is taken as the coupling time period core interval, and the coupling time period core interval is extended forward to the ultrafiltration fluctuation starting timestamp and backward to the body position change stable timestamp, to form a complete coupling time period marker, which contains the quantization values of the ultrafiltration rate fluctuation direction, the body position angle change direction and the double-parameter change amplitude.

4. The method for evaluating the volume status of a chronic kidney disease patient according to claim 1, wherein: The high-frequency acquisition of dynamic blood pressure data, jugular vein diameter data and pulse wave conduction velocity data within the coupling time period specifically includes: According to the starting time point of the coupling time period marker, a dynamic blood pressure monitor is activated to continuously collect systolic and diastolic pressure data at a preset high-frequency sampling interval, a neck ultrasonic probe is triggered to scan the cross section of the jugular vein at a fixed short period within the coupling time period and output diameter measurement values, and a wearable pulse wave sensor is started to calculate the pulse wave conduction velocity in real time through double-point detection of the radial artery and the femoral artery, and all collected data are bound to the coupling time period marker number.

5. The method for evaluating the volume status of a chronic kidney disease patient according to claim 1, wherein: The pre-trained interference calibration model is constructed by the following method: The labeled coupling period data in the historical dialysis cases is collected, including ultrafiltration rate fluctuation values, body position angle change values, original dynamic blood pressure data, original jugular vein diameter data, original pulse wave transmission velocity data and corresponding true volume load labels, the combination of the ultrafiltration rate fluctuation values and the body position angle change values is taken as an input feature vector, the deviation of the original physiological index data and the true label is taken as a training target, a neural network model is trained through supervised learning, the neural network model learns the mapping rule of the coupling of the ultrafiltration and the body position on the vascular compliance false interference, and a pre-trained interference calibration model is obtained.

6. The method for evaluating the volume status of a chronic kidney disease patient according to claim 1, wherein: The ultrafiltration rate fluctuation values, the body position angle change values and the collected physiological index data are input into the pre-trained interference calibration model, and the specific analysis logic is as follows: The interference calibration model first performs feature fusion on the ultrafiltration rate fluctuation values and the body position angle change values to generate an interference feature vector representing the coupling strength, extracts fluctuation amplitude features from the dynamic blood pressure data according to a time window, extracts diameter change slope features from the jugular vein diameter data, and extracts time delay change quantity features from the pulse wave transmission velocity data to jointly constitute a physiological index feature group. The interference feature vector and the physiological index feature group are analyzed in a cross-modal manner to identify the causal relationship between the coupling parameters and the physiological index changes.

7. The method for evaluating the volume status of a chronic kidney disease patient according to claim 6, wherein: The working logic of the cross-modal correlation analysis includes: The interference calibration model has an interference decoupling module built-in, which calculates an influence coefficient matrix of the vascular compliance false interference through the interference feature vector. The influence coefficient matrix includes a conduction coefficient of blood pressure fluctuation, a contraction coefficient of the jugular vein diameter and a time delay coefficient of the pulse wave transmission velocity. The influence coefficient matrix and the physiological index feature group are subjected to dot product operation to deconstruct the contribution components of the coupling interference in each physiological index.

8. The method for evaluating the volume status of a chronic kidney disease patient according to claim 1, wherein: The output calibrated volume load correlation index value that eliminates the vascular compliance false interference is specifically implemented as follows: The interference calibration model reversely deducts the interference contribution components calculated by the interference decoupling module from the original physiological index feature group. For the dynamic blood pressure data, the inverse operation of the interference conduction coefficient is used to restore the true blood pressure fluctuation value. For the jugular vein diameter data, the reciprocal of the interference contraction coefficient is used to restore the actual diameter change. For the pulse wave transmission velocity data, a compensation function of the time delay coefficient is used to correct the transmission time delay. Finally, the corrected dynamic blood pressure calibration value, the jugular vein diameter calibration value and the pulse wave transmission velocity calibration value are output as the volume load correlation index value.

9. The method for evaluating the volume status of a chronic kidney disease patient according to claim 1, wherein: The generation of the dry body weight deviation evaluation result is specifically implemented as follows: The calibrated volume load correlation index value is time-weighted averaged in the whole dialysis process to obtain a comprehensive calibration index value. The theoretical volume removal amount is calculated based on the weight difference before and after dialysis. The ratio of the comprehensive calibration index value to the theoretical volume removal amount is taken as a dry body weight deviation coefficient. If the dry body weight deviation coefficient exceeds a preset reasonable range, a dry body weight evaluation result including a deviation direction and an adjustment amount is generated.

10. The method for evaluating the volume status of a chronic kidney disease patient according to claim 1, wherein: The volume load state determination specifically includes: Based on the deviation evaluation result of dry body weight, if the deviation coefficient is within a reasonable range, the volume load status is directly determined according to the trend of the comprehensive calibration index value; if the deviation coefficient is out of limit, the joint decision rule based on the comprehensive calibration index value and the biomarker detection data is preferentially adopted; when the comprehensive calibration index value indicates abnormal volume load but the biomarker data does not support it, it is determined that false interference remains, and model recalibration is triggered; when both support abnormality, three types of results, i.e., volume overload, volume deficiency and critical state, are divided according to a preset threshold.

Images