Ai-based dialysis blood vessel replacement cycle prediction method
The AI-based method for predicting dialysis vessel replacement cycles addresses the limitations of subjective vascular management by using EMR, OCS, and OCR technologies to analyze time-series data, improving dialysis efficiency and patient safety through precise, personalized risk assessment.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- LIVIN AI INC
- Filing Date
- 2024-12-12
- Publication Date
- 2026-06-18
AI Technical Summary
Existing vascular management methods for dialysis patients fail to comprehensively analyze individual biometric data, relying on subjective judgments for determining the timing of vessel replacement, leading to missed early detection of stenosis and occlusion, and resulting in reduced dialysis efficiency and unnecessary medical resource waste.
An AI-based method utilizing EMR, OCS, and OCR technologies to collect and analyze time-series data, training a prediction model with LSTM and Transformer models to predict dialysis vessel replacement cycles, incorporating real-time data collection and multi-stage risk assessment.
Enhances dialysis safety and efficiency by accurately predicting vascular conditions, reducing unnecessary replacements, and optimizing medical resource use through personalized risk assessment.
Smart Images
Figure KR2024096800_18062026_PF_FP_ABST
Abstract
Description
AI-based method for predicting dialysis vessel replacement cycles
[0001] The present invention relates to a method for determining the timing of blood vessel replacement by predicting the vascular condition of a dialysis patient based on artificial intelligence (AI), and more specifically, to a method for accurately predicting the dialysis blood vessel replacement cycle through data collection utilizing EMR (Electronic Medical Record), OCS (Order Communication System), and OCR technologies, and training of an AI prediction model.
[0002] Dialysis refers to the removal of waste products from a patient's blood, and the types of dialysis commonly known are hemodialysis and peritoneal dialysis.
[0003] Among these, hemodialysis refers to a method of removing waste products from a patient's blood and maintaining electrolyte balance using a hemodialysis device. Hemodialysis treatment is typically performed two to four times a week, and there are various associated risks for patients undergoing this treatment. For example, problems such as vascular stenosis or occlusion may occur due to continuous vascular pressure, and cardiovascular diseases such as myocardial infarction may develop due to the burden on the heart caused by hemodialysis.
[0004] Furthermore, along with the global increase in the prevalence of chronic kidney disease (CKD), the need for hemodialysis as a treatment method for patients with end-stage renal disease (ESRD) is growing. This trend is attributed to the aging of the population and the rising prevalence of major risk factors for CKD, such as diabetes and hypertension. In particular, the number of ESRD patients in Korea has increased by 120% over the past nine years, leading to a rise in healthcare expenditures, with medical costs reaching approximately 2.1 trillion won in 2021. This trend is common globally and is driving the growth of the hemodialysis market.
[0005] Currently, hemodialysis is a process that purifies a patient's blood, making stable vascular access essential. However, blood vessels repeatedly used for dialysis are vulnerable to issues such as stenosis or occlusion, which can lead to reduced dialysis efficiency or, in severe cases, the loss of access routes. Furthermore, existing vascular management methods suffer from the problem of failing to comprehensively analyze individual patient biometric data and relying on the subjective judgment of medical staff to determine the timing of vessel replacement. This prevents the early detection of stenosis and occlusion, leading to missed opportunities for timely treatment and potentially resulting in the unnecessary waste of medical resources.
[0006] To address these issues, accurately predicting the vascular status of dialysis patients is crucial for early detection of vascular problems and the implementation of appropriate treatments, thereby maintaining dialysis efficiency and improving the patients' quality of life. Furthermore, there is a need for new predictive models that can improve treatment outcomes for dialysis patients and reduce medical costs by predicting and managing vascular stenosis or occlusion at an early stage.
[0007] The present invention has been devised to solve the aforementioned problems, and the purpose of the present invention is to provide an AI-based method for predicting the replacement cycle of dialysis patient blood vessels, which utilizes an artificial intelligence (AI)-based prediction model to precisely analyze the condition of a patient's dialysis blood vessels using time-series data, and to precisely predict the patient-specific blood vessel management and replacement cycle, thereby enhancing dialysis safety and maximizing treatment efficiency.
[0008] Furthermore, the present invention aims to provide an AI-based method for predicting the vascular replacement cycle of dialysis patients, which overcomes the limitations of existing manual observation methods and enables more accurate and rapid analysis of the patient's vascular condition by collecting real-time data through OCR technology.
[0009] To achieve the above objective, a method for predicting the timing of dialysis vessel replacement comprises: a data collection step (S10) in which electronic medical record (EMR) information of a hemodialysis patient and operation management system (OCS) information data including machine and setting information used during dialysis are collected and stored in an integrated DB, and key data (venous pressure (VP), blood flow velocity, systolic blood pressure (SBP), diastolic blood pressure (DBP), pulse, etc.) displayed on the dialysis machine screen are recognized in real time using Optical Character Recognition (OCR) technology and the corresponding data is stored in the integrated DB; a training data generation step (S20) in which prediction evaluation factor data necessary for predicting the dialysis vessel replacement cycle is extracted from the data collected in the data collection step (S10), and data is generated by normalizing it into a form suitable for training a prediction model; and a training step (S30) in which the data generated in the training data generation step (S20) is processed using a training GPU machine and a prediction model is trained. A data analysis step (S40) for analyzing data learned in the above learning step (S30) to identify patterns and statistical characteristics suitable for predicting the dialysis blood vessel replacement cycle; a risk evaluation step (S50) for setting threshold values for key data items (venous pressure, blood flow velocity, etc.) using an AI prediction model based on the results analyzed in the above data analysis step (S40), and sequentially applying risk evaluation models 1, 2, and 3 to evaluate the necessity of replacement stepwise; and a prediction step (S60) for predicting changes in the blood vessel condition using a learned prediction model based on the results of the above risk evaluation step (S50) and the patient's data from the last 5 times or more, and calculating the time when key data items reach a threshold level to predict the time when replacement is likely to be necessary.The present invention provides an AI-based method for predicting the replacement cycle of a dialysis vessel, characterized by comprising: an output step (S70) which classifies results including the patient's key data, risk grade, and recommended measures based on the results of the prediction step (S60), visualizes and provides the results to medical staff, transmits notification messages in conjunction with EMR and OCS systems if necessary, and integrates the results into a record.
[0010] The present invention has the effect of maximizing therapeutic efficacy by improving the quality of dialysis and enhancing patient safety by learning biological data based on artificial intelligence (AI) to predict the blood vessel replacement cycle of dialysis patients in advance and predicting the possibility of stenosis or occlusion of the patient at an early stage.
[0011] In addition, the present invention has the effect of maximizing medical resource efficiency and patient safety by precisely evaluating the need for replacement of each patient in a multi-stage manner through risk assessment models 1, 2, and 3, thereby enabling accurate risk assessment tailored to the individual condition of the patient and reducing unnecessary replacement of dialysis vessels.
[0012] In addition, the present invention significantly improves the flexibility and accessibility of the data collection method by directly recognizing and collecting data from the dialysis machine screen through OCR technology. In particular, by simply taking a picture of the dialysis machine screen using a mobile phone application, medical staff can automatically collect key data such as VP, blood flow velocity, blood pressure, and pulse rate and store them in a DB, thereby reducing errors that may occur in existing recording operations and enabling real-time data collection.
[0013] FIG. 1 is a diagram illustrating a method for predicting the replacement cycle of a dialysis vessel according to the present invention.
[0014] Figure 2 is a diagram showing the configuration of a dialysis blood vessel replacement cycle prediction system according to the present invention.
[0015] Specific details for implementing the present invention are described below with reference to Drawings 1 and 2.
[0016] The AI-based method for predicting the replacement cycle of a dialysis vessel according to an embodiment of the present invention comprises a data collection step (S10), a training data generation step (S20), a training step for training a model (S30), a data analysis step (S40), a risk assessment step (S50), a prediction step (S60), and an output step (S70).
[0017] The data collection step (S10) includes the process of collecting EMR (Electronic Medical Record) and OCS (Operation Management System) data to analyze and predict the condition and treatment process of a hemodialysis patient, storing them in an integrated DB (40), and collecting key data displayed on the dialysis machine screen in real time using OCR technology.
[0018] First, the EMR system provides basic patient information and medical record data, including fundamental details such as gender, age, medical history, dialysis history, vascular test results, and dialysis schedules. The EMR operates either in conjunction with the Order Communications System (OCS) or independently, recording and managing patient medical information in chronological order.
[0019] Next, OCS provides data including the dialysis schedule, information on the machine used during dialysis, and dialysis-related parameters.
[0020] In addition, key dialysis-related data can be collected in real time from the patient's dialysis machine screen by utilizing OCR technology. Specifically, after selecting a patient through a mobile phone application and photographing the dialysis machine screen, the application utilizes the OCR function to recognize and automatically extract data such as venous pressure (VP), blood flow velocity, systolic blood pressure (SBP), diastolic blood pressure (DBP), and pulse displayed on the screen. The extracted data is immediately stored in an integrated DB (40), and subsequently, in the training data generation step (S20), it undergoes data cleaning and normalization processes and is used to train a prediction model.
[0021] These data include data from at least five recent dialysis sessions to comprehensively evaluate the patient's past and present condition, and may include items such as the date of dialysis, venous pressure, arterial pressure, blood flow velocity, systolic blood pressure, diastolic blood pressure, heart rate, and age.
[0022] In order to collect data from EMR and OCS and store it in the integrated DB (40), data transmission technologies and protocols such as API, HL7 / FHIR standard, and ETL method are utilized through the network, and the collected data is stored in the integrated DB (40) while maintaining consistency.
[0023] The above-mentioned training data generation step (S20) is a step of preparing data in a form suitable for training a prediction model based on the data stored in the integrated DB (40) in the above-mentioned data collection step (S10), and includes the following tasks. First, data on the prediction evaluation factors of hemodialysis patients is filtered and extracted from the integrated DB (40). This includes basic information such as the patient's gender, age, medical history, dialysis history, vascular examination results, and dialysis schedule, as well as biological data such as systolic blood pressure (SBP), diastolic blood pressure (DBP), pulse, blood flow, venous pressure (VP), and arterial pressure (AP).
[0024] The extracted data is sorted chronologically and arranged continuously from the initial dialysis data to the most recent data, configured to track changes in biological data at each dialysis point along the time axis. Data sorted in chronological order plays a crucial role in enabling the predictive model to accurately learn patterns related to dialysis vessel turnover.
[0025] Aligned data is transformed into a form suitable for training a prediction model through a normalization process. Normalization converts data of different sizes and units into a fixed range (e.g., 0 to 1) so that the model can process and learn the data consistently.
[0026] Normalized data is converted into time-series data and configured so that the model can learn how the patient's vascular condition changes over time.
[0027] Finally, the generated data is stored back in the integrated DB (40) and is subsequently used to train the prediction model in the model training step (S30).
[0028] The learning step (S30) consists of a process of training various prediction models using a training GPU machine with the data generated in the learning data generation step (S20).
[0029] In the embodiments of the present invention, Long Short-Term Memory (LSTM) and Transformer models are used as representative examples, and data for predicting the timing of dialysis vessel replacement is learned.
[0030] The LSTM model processes data in chronological order and learns patterns such as systolic blood pressure (SBP), diastolic blood pressure (DBP), pulse, blood flow, and venous pressure (VP) based on the last five dialysis data. During the learning process, it calculates the error between predicted and actual values, adjusts model parameters to minimize this error, and improves prediction performance by repeating this process. The Transformer model utilizes an attention mechanism to learn relationships between data and effectively handles long-term dependencies.
[0031] During the training process, GPU machines for AI training are used to process large-scale data in parallel, allowing deep learning models to be trained much faster than CPUs. This enables the simultaneous processing of data from numerous patients and significantly improves the performance of prediction models by training with more data.
[0032] The model that has completed training is stored in the integrated DB (40), and is then used to predict the timing for dialysis vessel replacement by receiving new data in real time during the prediction step (S60).
[0033] Various techniques can be used to generate prediction models. While the embodiments of the present invention have focused on Long Short-Term Memory (LSTM) and Transformer models, various algorithms such as Recurrent Neural Networks (RNNs) or other time series prediction models may be used. Such algorithms can be selectively utilized depending on the performance and purpose of the prediction model.
[0034] The data analysis step (S40) consists of a process of analyzing the data learned in the learning step (S30) to derive patterns and statistical characteristics suitable for predicting the dialysis vessel replacement cycle.
[0035] First, the fluctuation patterns of key indicators such as venous pressure (VP), blood flow velocity, systolic blood pressure (SBP), and diastolic blood pressure (DBP) are analyzed, and basic statistical indicators such as the mean, variance, maximum, and minimum values of the data are calculated to identify the general distribution and characteristics of the data. In addition, the rate of change of the data over time is calculated to identify points where rapid changes occur or outliers.
[0036] It is verified whether the collected data has been recorded consistently over time, and patterns in the time-series data are analyzed to determine if the data is suitable for training a predictive model. During this process, the continuity and periodicity of the data are evaluated, and if specific patterns exist, they are utilized to predict replacement cycles.
[0037] Based on the analysis results, the most significant indicator for determining the dialysis vessel replacement cycle is selected in the risk assessment stage (S50), and weights are assigned according to the importance of each indicator. For example, if venous pressure (VP) is closely related to the replacement cycle, a high weight is assigned to it to improve the accuracy of the risk assessment and prediction model.
[0038] The risk assessment step (S50) consists of a process of evaluating the risk of dialysis vessel replacement stepwise based on patient data using an AI prediction model (LSTM model, Transformer model, or other time series prediction model) based on the results derived from the data analysis step (S40).
[0039] First, thresholds are set for key data items such as venous pressure (VP), blood flow velocity, systolic blood pressure (SBP), and diastolic blood pressure (DBP) to quantitatively analyze the patient's condition. These thresholds are set as criteria to determine whether the data has reached a dangerous range, enabling an objective assessment of the patient's safety and the need for blood vessel replacement.
[0040] The threshold is set by considering both general reference values and individual patient characteristics. For example, if the venous pressure is 180 mmHg or higher, or if the blood flow velocity decreases to 230 mL / min or lower, it is determined that the risk is high.
[0041] Next, the Attention mechanism of the Transformer model is utilized to learn the complex relationships between patient data, and the impact of changes in key data items on the need for vascular replacement is analyzed. In this process, the Transformer model rapidly processes large-scale data through parallel processing and performs accurate risk assessment.
[0042] Risk assessment is carried out step by step by sequentially applying risk assessment models 1, 2, and 3.
[0043] Risk assessment models 1, 2, and 3 each apply different threshold criteria to evaluate the patient's condition from various angles, and apply risk assessment models 1, 2, and 3 sequentially, while utilizing the F1 score (harmonic mean of precision and recall) to enhance the reliability of the evaluation results.
[0044] Risk assessment model 1 uses F1 scores to apply high-precision criteria to evaluate patients who absolutely need replacement, and selects based on data exceeding a threshold.
[0045] Risk assessment model 2 further evaluates patients with potential for replacement by analyzing data near the threshold based on the F1 score, while maintaining a balance of precision and recall.
[0046] Risk assessment model 3 broadly evaluates patients with even a slight possibility of replacement based on high recall in the F1 score and detects the need for replacement early.
[0047] The F1 score maintains a balance between the precision and recall of each evaluation model and is used to quantitatively verify evaluation results. This ensures the reliability of the predictive model and the validity of the evaluation results, allowing for a more accurate determination of the need for replacement.
[0048] Through this multi-stage evaluation process, the patient's condition is assessed quantitatively and reliably, and by providing customized management that reflects the importance of key data items, patient safety can be maximized and the need for replacement can be accurately determined.
[0049] The prediction step (S60) consists of a process of predicting changes in the vascular condition and calculating the time when replacement is required by utilizing a learned prediction model (LSTM model, Transformer model, or other time series prediction model) based on the risk level information derived from the risk assessment step (S50) and the patient's last five or more data (systolic blood pressure (SBP), diastolic blood pressure (DBP), pulse, blood flow, venous pressure (VP), etc.).
[0050] First, the Long Short-Term Memory (LSTM) model learns data patterns over time to predict long-term changes in vascular status. The Transformer model utilizes an attention mechanism to analyze complex relationships between data and processes large-scale data rapidly through parallel processing to accurately predict changes in vascular status. Depending on the characteristics of the patient's data, other time-series prediction models may be selectively applied to improve prediction accuracy.
[0051] The prediction model calculates when key data items reach a critical level; for example, if it is predicted that the venous pressure will be 190 mmHg or higher and the blood flow velocity will be 210 mL / min or lower on the next dialysis day, it determines that there is a possibility that a replacement will be necessary at that time. The reliability of the predicted results is reviewed by comparing them with data from similar patients or through feedback from medical staff, and the performance of the prediction model is improved through additional learning if necessary.
[0052] Finally, the prediction results are provided to medical staff, including information on the patient's vascular status and the timing of replacement needs; based on this, personalized management and replacement plans can be established to maximize patient safety and treatment efficiency.
[0053] The output step (S70) is a process of providing data on the necessity of replacing a patient's blood vessel to medical staff based on the results derived in the prediction step (S60). The prediction result may be provided in a simple form such as "Yes" (replacement needed) or "No" (replacement not needed), or subdivided into "low," "medium," or "high" depending on the risk level.
[0054] The output includes key patient data, risk levels, and recommended measures, and is visualized in the form of tables and graphs to help medical staff intuitively understand changes in the patient's condition.
[0055] In addition, results can be integrated into records or notification messages sent to medical staff by linking with EMR and OCS systems. This enables medical staff to monitor the patient's condition in real time and take timely action.
[0056] Medical staff can utilize the output prediction results to monitor changes in the patient's condition in real time, determine whether to replace the blood vessel, and take necessary measures in a timely manner.
[0057] Through the present invention, the possibility of stenosis or occlusion in a patient can be detected early to improve the quality of dialysis and enhance patient safety, and unnecessary replacements can be reduced to maximize the efficiency of medical resources and the therapeutic effect.
[0058] According to the present invention as described above, the AI-based method for predicting the blood vessel replacement cycle of a dialysis patient learns biological data based on artificial intelligence (AI) to predict the blood vessel replacement cycle of a dialysis patient in advance, thereby predicting the possibility of stenosis or occlusion in the patient at an early stage, which can improve the quality of dialysis and enhance patient safety, thereby maximizing treatment effectiveness.
[0059] In addition, by precisely evaluating the need for replacement for each patient in a multi-stage manner through risk assessment models 1, 2, and 3, it is possible to perform an accurate risk assessment tailored to the individual condition of the patient, thereby reducing unnecessary replacement of dialysis vessels and maximizing the efficiency of medical resources and patient safety.
[0060] Although the present invention has been described with reference to embodiments, this is merely illustrative, and those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible therefrom. Accordingly, the true technical scope of protection of the present invention should be determined by the technical spirit of the appended claims.
[0061] [Explanation of the symbol]
[0062] 10: Data Collection Unit 20: Data Processing Unit
[0063] 30: Control Unit 40: Integrated DB
[0064] 50: Learning model 60: Output section
[0065] S10: Data collection stage S20: Training data generation stage
[0066] S30: Learning phase S40: Data analysis phase
[0067] S50: Risk assessment stage S60: Prediction stage
[0068] S70: Output stage
[0069]
Claims
Regarding the method for predicting the timing of dialysis vessel replacement, A data collection step (S10) for collecting data including electronic medical record (EMR) information of a hemodialysis patient and operation management system (OCS) information used during dialysis, and for recognizing key data displayed on the dialysis machine screen in real time using OCR technology and storing it in an integrated DB; A training data generation step (S20) for extracting prediction evaluation factor data required for predicting the dialysis vessel replacement cycle from the data collected in the above data collection step (S10) and generating data by normalizing it into a form suitable for training a prediction model; A training step (S30) for processing the data generated in the above training data generation step (S20) using a training GPU machine and training a prediction model; A data analysis step (S40) for analyzing the data learned in the above learning step (S30) to identify patterns and statistical characteristics suitable for predicting the dialysis vessel replacement cycle; A risk assessment step (S50) that, based on the results analyzed in the above data analysis step (S40), sets threshold values for key data items (venous pressure, blood flow velocity, etc.) using an AI prediction model, and sequentially applies risk assessment models 1, 2, and 3 to calculate the necessity for replacement in stages; A prediction step (S60) that predicts changes in the vascular condition using a learned prediction model based on the results of the above risk assessment step (S50) and the patient's recent 5 or more data, calculates the time when key data items reach a threshold level, and predicts the time when replacement is likely to be necessary; An AI-based method for predicting the replacement cycle of a dialysis vessel, characterized by comprising: an output step (S70) that classifies results including the patient's key data, risk grade, and recommended measures based on the results of the prediction step (S60), visualizes and provides the results to medical staff, transmits notification messages in conjunction with EMR and OCS systems if necessary, and integrates the results into a record. In paragraph 1, An AI-based method for predicting the replacement cycle of a dialysis vessel, characterized in that the above-mentioned prediction evaluation factor data includes the gender, age, medical history, dialysis history, vascular examination results and dialysis schedule of a hemodialysis patient, systolic blood pressure (SBP), diastolic blood pressure (DBP), pulse, blood velocity, venous pressure (VP), and arterial pressure (AP). In paragraph 1, The above risk assessment step (S40) is characterized by multi-layered evaluation of patient data utilizing the Attention mechanism of a Transformer model and evaluating the risk through risk assessment models 1, 2, and 3, in an AI-based method for predicting the replacement cycle of a dialysis vessel. In paragraph 1, The above risk assessment step (S40) includes (a) risk assessment model 1 selecting patients who absolutely require replacement by applying high-precision criteria based on the F1 score, and (b) Risk assessment model 2 evaluates the possibility of additional replacement by considering the balance between precision and recall, and (c) AI-based dialysis vessel replacement cycle prediction method characterized by evaluating potential risks based on recall rate and evaluating them step by step to detect the need for replacement early. In paragraph 1, The above learning step (S30) is characterized by selecting one or more of an LSTM (Long Short-Term Memory) model, a Transformer model, and a Recurrent Neural Network (RNN) model and training them with prediction evaluation factor data, thereby providing an AI-based method for predicting the replacement cycle of a dialysis vessel. In paragraph 1, An AI-based method for predicting the replacement cycle of a dialysis vessel, characterized in that the above-mentioned AI training GPU machine rapidly processes a large amount of data through parallel processing and optimizes the training and prediction speed of the model. In paragraph 1, AI-based dialysis vessel replacement cycle prediction method characterized by including providing a user interface after the above prediction step (S60) that allows a medical professional to review the prediction results of the model and adjust them as needed.