Internal arteriovenous fistula blood vessel noninvasive intelligent monitor and control method thereof

Abstract

The invention discloses an internal arteriovenous fistula blood vessel noninvasive intelligent monitor, and belongs to the technical field of medical monitoring equipment. The sensor unit is composed of a photoelectric volume pulse wave sensor, a miniature ultrasonic sensor and a bioelectrical impedance sensor and used for collecting information such as heart rate, blood oxygen saturation, blood flow, angiography, tremor frequency, impedance parameters and vascular tension; the monitoring chip unit carries out amplification, filtering, analog-to-digital conversion and feature extraction on the information; the storage and preprocessing unit is used for realizing real-time storage and standardized processing of data; the communication unit uploads the data to a medical institution end or a cloud end through Bluetooth or Wi-Fi; the power management unit provides power and realizes low-power-consumption control; and the early warning unit performs risk assessment on the data based on an artificial intelligence model, and generates early warning information when a potential anomaly is detected. According to the invention, continuous, non-invasive and intelligent monitoring of internal arteriovenous fistula can be realized, and the early recognition and early warning capability of complications can be improved.

Description

TECHNICAL FIELD

[0001] The present specification relates to the field of blood vessel monitoring, and more particularly, the present application relates to an arteriovenous fistula blood vessel noninvasive intelligent monitor and a control method thereof. BACKGROUND

[0002] For maintenance hemodialysis patients, arteriovenous fistula is an extremely important blood vessel access, known as the "lifeline" of the patient. However, arteriovenous fistula is prone to complications such as thrombosis, vascular stenosis, and insufficient blood flow during use. If not discovered and handled in time, it will seriously affect the treatment effect and quality of life of the patient, and even endanger life.

[0003] Traditional monitoring methods mainly rely on physical examination such as visual, tactile, and auditory examination of medical personnel, as well as auxiliary examination methods such as B-ultrasound. These methods have many limitations, such as the need for patients to go to the hospital, the detection results being greatly affected by human factors, the inability to achieve continuous real-time monitoring, the lack of quantitative indicators and objectivity, and the like, which makes it difficult to meet the clinical needs of early identification, early warning, and intervention of arteriovenous fistula complications.

[0004] Therefore, it is necessary to propose an arteriovenous fistula blood vessel noninvasive intelligent monitor to at least solve some of the above problems. SUMMARY

[0005] A series of simplified concepts are introduced in the summary section, which will be further described in detail in the specific embodiments section. The summary section of the present application does not mean to attempt to limit the key features and essential technical features of the claimed technical solutions, nor does it mean to attempt to determine the protection scope of the claimed technical solutions.

[0006] In a first aspect, the present application proposes an arteriovenous fistula blood vessel noninvasive intelligent monitor, comprising:

[0007] A sensor unit comprising a photoplethysmographic pulse wave sensor, a miniature ultrasonic sensor, and a bioelectrical impedance sensor, for collecting monitoring information of a target patient, wherein the monitoring information comprises heart rate information, blood oxygen saturation information, blood flow information, vascular imaging information, tremor frequency information, impedance parameters, and vascular tension information;

[0008] A monitoring chip unit electrically connected to the sensor unit, for amplifying, filtering, analog-to-digital converting, and feature extracting the monitoring information to obtain a feature vector;

[0009] A storage and preprocessing unit electrically connected to the monitoring chip unit, for real-time storing and preprocessing the feature vector;

[0010] A communication unit is electrically connected with the storage and preprocessing unit, and is configured to transmit the preprocessed data to a medical institution terminal device or a cloud database through Bluetooth or Wi-Fi.

[0011] A power management unit is electrically connected with the sensor unit, the monitoring chip unit, the storage and preprocessing unit, and the communication unit, and is configured to supply power to each unit and realize low-power consumption control.

[0012] An early warning unit is electrically connected with the monitoring chip unit and the storage and preprocessing unit, and is configured to perform risk assessment on the preprocessed data based on an artificial intelligence model, and generate early warning information when potential abnormalities are detected.

[0013] In a feasible implementation, a wearable unit is further included, and the wearable unit includes:

[0014] A flexible electronic body is fixedly connected with the sensor unit, and is configured to support and adhere to the skin of a patient.

[0015] An alert unit is electrically connected with the early warning unit, and is configured to alert the patient through vibration, sound, or light when the early warning is triggered.

[0016] In a feasible implementation, a pressure adjusting unit is further included,

[0017] The pressure adjusting unit is mechanically connected with the sensor unit and electrically connected with the monitoring chip unit, and is configured to monitor the pressure value of the device on the blood vessel of the patient in real time, and automatically adjust the tightness of the watchband through an intelligent control system to ensure that the sensor stably contacts without compressing the blood vessel.

[0018] In a feasible implementation, a cloud data unit is further included,

[0019] The cloud data unit is wirelessly connected with the communication unit, and is configured to realize distributed storage of monitoring data, data security protection, statistical analysis, and visual display, and provide a remote access interface to medical staff and scientific researchers.

[0020] In a second aspect, the present application further provides a control method for the arteriovenous fistula blood vessel non-invasive intelligent monitor in any one of the first aspect.

[0021] The monitoring information of a target patient is collected based on the sensor unit, and the monitoring information includes heart rate information, blood oxygen saturation information, blood flow information, blood vessel imaging information, tremor frequency information, impedance parameters, and blood vessel tension information.

[0022] The monitoring chip unit is configured to amplify, filter, analog-to-digital convert, and feature extract the monitoring information to obtain a feature vector.

[0023] The feature vector is stored and pre-processed in real time based on the storage and preprocessing unit to obtain pre-processed data;

[0024] The pre-processed data is risk assessed by the early warning unit based on the artificial intelligence model, and early warning information is generated when potential abnormalities are detected.

[0025] In a feasible implementation, the artificial intelligence model comprises:

[0026] A multi-modal time sequence fusion encoding unit is configured to perform heart cycle alignment and time sequence feature extraction on the heart rate information, the blood oxygen saturation information, the blood flow information, the tremor frequency information, and the blood vessel tension information, and output a fusion feature vector;

[0027] A blood vessel atlas modeling unit is configured to receive the fusion feature vector, the blood vessel imaging information, and the impedance parameter, construct a blood vessel segment branch graph, extract blood vessel structure and hemodynamic features, and output a blood vessel atlas feature vector;

[0028] An individualized adaptive unit is configured to receive historical baseline data of a target patient, and perform individualized adjustment and cross-device, cross-scene distribution adaptation on the fusion feature vector and the blood vessel atlas feature vector to obtain an adjusted comprehensive feature representation;

[0029] An uncertainty evaluation and multi-task prediction unit is configured to receive the comprehensive feature representation to output multi-task prediction results of stenosis risk, thrombosis risk, and access failure risk, and generate a confidence interval for each prediction result.

[0030] In a feasible implementation, the multi-modal time sequence fusion encoding unit comprises:

[0031] A heart cycle alignment subunit is configured to perform cycle slicing and resampling on the heart rate information, the blood oxygen saturation information, the blood flow information, the tremor frequency information, and the blood vessel tension information based on feature point detection of the photoplethysmography signal;

[0032] A dual-flow time sequence encoding subunit is configured to respectively perform feature extraction on the resampled heart rate information, the resampled blood oxygen saturation information, and the resampled tremor frequency information using a time convolution network to obtain a first feature vector, and perform feature extraction on the resampled blood flow information and the resampled blood vessel tension information using a bidirectional long short-term memory network to obtain a second feature vector;

[0033] A cross-modal fusion subunit is configured to perform weighted fusion on the first feature vector and the second feature vector through a cross-attention mechanism to generate the fusion feature vector.

[0034] In an implementable embodiment, the blood vessel atlas modeling unit comprises:

[0035] a graph construction subunit configured to extract blood vessel centerlines based on the blood vessel imaging information and construct a blood vessel segment branch graph in combination with the impedance parameters, wherein the node features of the blood vessel segment branch graph comprise blood vessel diameter, wall thickness, pulsatility index and flow velocity, and the edge features comprise adjacent relationship between blood vessel segments and bifurcation angle;

[0036] a graph neural network subunit configured to update the node features by a graph attention mechanism and generate blood vessel segment level risk;

[0037] a physical consistency constraint subunit configured to introduce blood flow continuity and fluid mechanics formula constraints in the training process to improve the reliability of the blood vessel atlas feature vector.

[0038] In an implementable embodiment, the individualized adaptive unit comprises:

[0039] a baseline construction subunit configured to calculate resting heart rate, baseline blood oxygen, average blood vessel diameter and impedance baseline according to historical monitoring data of the target patient, and generate a patient baseline vector;

[0040] a feature modulation subunit configured to apply the baseline vector to the fusion feature vector and the blood vessel atlas feature vector by a layer-by-layer linear modulation mechanism to realize individualized parameter adjustment;

[0041] a de-domain subunit configured to weaken the influence of distribution differences in different devices and different scenarios by adversarial training, thereby improving the cross-scene generalization ability of the model, and obtaining the adjusted comprehensive feature representation.

[0042] In an implementable embodiment, the uncertainty assessment and multi-task prediction unit comprises:

[0043] a multi-task prediction subunit configured to simultaneously output multi-task prediction results of stenosis risk, thrombosis risk and pathway failure;

[0044] an uncertainty estimation subunit configured to generate a confidence interval for the multi-task prediction results based on a Monte Carlo Dropout method;

[0045] a double-threshold early warning subunit configured to trigger early warning when the comprehensive risk exceeds a preset risk threshold and the lower confidence limit exceeds the confidence interval.

[0046] In summary, the present application overcomes the subjectivity, inability to continuously monitor, and difficulty in timely warning caused by the physical examination and intermittent B-ultrasound in the background art. Firstly, in the information dimension, the sensor unit 11 introduces PPG (heart rate / SpO2), micro-ultrasound (vessel imaging / blood flow), and bioelectrical impedance (impedance spectrum / phase, etc.) in coordination, and combines tremor frequency and vascular tension information, to realize the synchronous acquisition of morphology, hemodynamics, electricity, and mechanical compliance, significantly improve the perceptibility and distinguishability of abnormal states such as stenosis, thrombus, and insufficient blood flow, and more comprehensive and sensitive than single B-ultrasound or physical assessment. Secondly, in terms of signal quality and objective quantification, the monitoring chip unit 12 performs amplification, filtering, and analog-to-digital conversion on the original monitoring information, and completes feature extraction on the chip side, cooperates with the denoising and standardization processing of the storage and preprocessing unit 13, effectively suppresses environmental light interference, motion artifacts, and electromagnetic noise, realizes high signal-to-noise ratio and reproducible quantitative index output, reduces the dependence on operator experience, and improves the consistency and comparability across time periods and personnel. Thirdly, in terms of timeliness and continuity, the embodiment uploads the preprocessed data to the medical institution end or cloud database in a Bluetooth / Wi-Fi manner through the communication unit 14, and the power management unit 15 guarantees low power consumption and long endurance, so that long-term and continuous data acquisition and trend tracking can be realized in and out of the hospital, breaking through the time blind area of traditional "hospitalization-intermittent examination", discovering risk signs accumulated from subclinical changes in advance, and shortening the time window from abnormality to identification. Further, in terms of decision support and active protection, the warning unit 16 evaluates the multi-source features based on an artificial intelligence model, triggers warning information and links the wearable end or clinical terminal when potential risks are detected, changes the monitoring from "passive discovery" to "active warning", helps medical staff to take intervention or review strategies at an early stage, reduces the probability of power failure events, and improves the timing selection of dialysis access maintenance. Finally, in terms of system expansion and whole-process management, the modular architecture of the embodiment facilitates the upgrade and iteration of subsequent sensor categories and algorithm versions, and the cloud side can realize visual analysis and audit traces for longitudinal follow-up and horizontal comparison, providing an objective data basis for clinical pathway optimization and scientific research verification; at the same time, the end-cloud collaborative structure enables remote follow-up and hierarchical management without increasing the patient's burden, improving the efficiency of medical resources. The method takes multi-modal non-invasive sensing, chip-level robust processing, and AI active warning as the core, and considers continuity, objectivity, and scalability, and effectively solves the key pain points of the background technology in early identification, real-time warning, and remote monitoring.

[0047] Other advantages, objects, and features of the present application will be apparent to those skilled in the art from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0048] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments with reference made to the accompanying drawings. The drawings are for purposes of illustration only and are not intended to be limiting in

[0049] Figure 1 A structural schematic diagram of an arteriovenous fistula blood vessel noninvasive intelligent monitor provided by an embodiment of the present application.

[0050] Figure 2 A flow schematic diagram of a control method provided by an embodiment of the present application. DETAILED DESCRIPTION

[0051] The terms "first", "second", "third", "fourth" and the like in the description and in the claims of the present application, if any, of the aforementioned drawings, are used for distinguishing between similar objects and not necessarily for describing a specific sequential or chronological order. It is to be understood that the use of the terms so construed herein is merely for convenience and only to aid in understanding the application and is in no way a limitation on its broader scope. It is also to be understood that the use of relational terms such as first, second and third, and the like, if any, are used solely to distinguish one from another but not to denote a required or a prioritized order, unless stated otherwise. The terms "comprises", "comprising", "includes", "including", "contains", "containing" or any other comparable terms are intended to cover a non-exclusive inclusion, such that any process, method, article, or apparatus that comprises, includes, contains or contains any of the stated steps or units is not thereby limited to the steps or units that are clearly described as such. The following will describe the technical solutions in the embodiments of the present application with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all embodiments of the present application.

[0052] Please refer to Figure 1 , Figure 1 A structural schematic diagram of an arteriovenous fistula blood vessel noninvasive intelligent monitor 10 provided by an embodiment of the present application, specifically can include:

[0053] A sensor unit 11, including a photoplethysmographic pulse wave sensor, a miniature ultrasonic sensor and a bioelectric impedance sensor, for collecting monitoring information of a target patient, the monitoring information including heart rate information, blood oxygen saturation information, blood flow information, vascular imaging information, tremor frequency information, impedance parameters and vascular tension information;

[0054] A monitoring chip unit 12, electrically connected with the above-mentioned sensor unit, for amplifying, filtering, analog-to-digital converting and feature extracting the monitoring information to obtain a feature vector;

[0055] The storage and preprocessing unit 13 is electrically connected with the monitoring chip unit, and is used for real-time storage and preprocessing of the feature vector.

[0056] The communication unit 14 is electrically connected with the storage and preprocessing unit, and is used for transmitting the preprocessed data to a medical institution terminal device or a cloud database through Bluetooth or Wi-Fi.

[0057] The power management unit 15 is electrically connected with the sensor unit, the monitoring chip unit, the storage and preprocessing unit, and the communication unit respectively, and is used for power supply and low-power consumption control of each unit.

[0058] The early warning unit 16 is electrically connected with the monitoring chip unit and the storage and preprocessing unit, and is used for risk assessment of the preprocessed data based on an artificial intelligence model, and generates early warning information when potential abnormalities are detected.

[0059] Exemplarily, the embodiment provides an arteriovenous fistula blood vessel noninvasive intelligent monitor, which is suitable for a maintenance hemodialysis patient, and the overall structure includes a sensor unit 11, a monitoring chip unit 12, a storage and preprocessing unit 13, a communication unit 14, a power management unit 15, and an early warning unit 16, and each unit is cooperatively worked through electrical connection and logical interaction.

[0060] The sensor unit 11 is provided with a photoelectric plethysmogram sensor, a miniature ultrasonic sensor, and a bioelectrical impedance sensor, which are used for real-time acquisition of arteriovenous fistula related multi-modal monitoring information when the patient wears the device. The monitoring information not only includes heart rate information, blood oxygen saturation information, blood flow information, vascular imaging information, and tremor frequency information, but also further covers impedance parameters and vascular tension information, so as to comprehensively reflect the functional state and dynamic changes of the patient's fistula blood vessel.

[0061] The monitoring chip unit 12 is electrically connected with the sensor unit 11, and is used for receiving the multi-source physiological signals, and performing amplification processing, filtering processing, and analog-to-digital conversion, and simultaneously using a built-in feature extraction algorithm to convert the original signals into feature vectors with clinical significance. Subsequently, the storage and preprocessing unit 13 is connected with the monitoring chip unit 12, and is used for real-time storage of the feature vectors, and performs preprocessing operations such as denoising, standardization, and normalization, so as to prepare for subsequent data analysis and remote transmission.

[0062] The communication unit 14 is electrically connected with the storage and preprocessing unit 13, and has a Bluetooth and Wi-Fi communication module integrated therein. According to different application scenarios, short-distance or long-distance communication mode can be selected to transmit the preprocessed data to a monitoring device at a medical institution end or a cloud database, so as to realize integrated data management and remote access in and out of the hospital. The power management unit 15 is electrically connected with the sensor unit 11, the monitoring chip unit 12, the storage and preprocessing unit 13 and the communication unit 14 respectively, and is used for providing stable power and low-power control of each unit, so as to ensure the endurance performance of the device and the convenience of clinical application.

[0063] The early warning unit 16 is electrically connected with the monitoring chip unit 12 and the storage and preprocessing unit 13, and is used for receiving the feature vector and the preprocessed data, and performing risk assessment on the internal arteriovenous fistula state of the patient based on an artificial intelligence model. When the model determines that there is a potential abnormality such as insufficient blood flow, thrombosis or vascular stenosis, the early warning unit 16 will generate early warning information in time, and can push the early warning information to the patient and medical staff through a wearable terminal or a remote terminal, so as to realize early identification and timely intervention.

[0064] In summary, the present application overcomes the subjectivity, inability to continuously monitor and difficulty in timely warning caused by the physical examination and intermittent B-ultrasound in the background art. Firstly, in the information dimension, the sensor unit 11 introduces PPG (heart rate / SpO2), micro-ultrasound (vessel imaging / blood flow) and bioelectrical impedance (impedance spectrum / phase, etc.) in coordination, and combines tremor frequency and vascular tension information, to realize the synchronous acquisition of morphology, hemodynamics, electricity and mechanical compliance, significantly improve the perceptibility and distinguishability of abnormal states such as stenosis, thrombus and blood flow deficiency, and the present application is more comprehensive and sensitive than single B-ultrasound or physical assessment. Secondly, in terms of signal quality and objectivity, the monitoring chip unit 12 amplifies, filters and digitizes the original monitoring information, and completes feature extraction on the chip side, cooperates with the denoising and standardization processing of the storage and preprocessing unit 13, effectively suppresses environmental light interference, motion artifacts and electromagnetic noise, realizes high signal-to-noise ratio and reproducible quantitative index output, reduces the dependence on operator experience, and improves the consistency and comparability across time periods and personnel. Thirdly, in terms of timeliness and continuity, the present embodiment uploads the preprocessed data to the medical institution end or cloud database in a Bluetooth / Wi-Fi manner through the communication unit 14, and the power management unit 15 guarantees low power consumption and long endurance, so that long-term and continuous data acquisition and trend tracking can be realized in and out of the hospital, breaking through the time blind area of traditional "hospitalization-intermittent examination", discovering risk signs accumulated by subclinical changes in advance, and shortening the time window from abnormality to identification. Further, in terms of decision support and active protection, the warning unit 16 evaluates the multi-source features based on an artificial intelligence model, triggers warning information and links the wearable end or clinical terminal when potential risks are detected, so that the monitoring is changed from "passive discovery" to "active warning", which helps medical staff to take intervention or review strategies at an early stage, reduces the probability of power failure events, and improves the timing selection of dialysis access maintenance. Finally, in terms of system expansion and whole-process management, the modular architecture of the present embodiment facilitates the upgrade and iteration of subsequent sensor categories and algorithm versions, and the cloud side can realize visual analysis and audit traces for longitudinal follow-up and horizontal comparison, providing an objective data basis for clinical pathway optimization and scientific research verification; at the same time, the end-cloud collaborative structure enables remote follow-up and hierarchical management without increasing the patient's burden, improving the efficiency of medical resources. The present method takes multi-modal non-invasive sensing, chip-level robust processing and AI active warning as the core, and takes into account continuity, objectivity and scalability, and effectively solves the key pain points of the background technology in early identification, real-time warning and remote monitoring.

[0065] In a feasible implementation manner, the wearable unit further comprises:

[0066] The flexible electronic body is fixedly connected with the sensor unit and is used for supporting and adhering to the patient's skin.

[0067] The prompting unit is electrically connected with the pre-warning unit, and is configured to prompt the patient through vibration, sound or light when the pre-warning is triggered.

[0068] Exemplarily, the present monitoring device further comprises a wearable unit. The wearable unit adopts a flexible electronic body and is fixedly connected with the sensor unit, so that the device can closely adhere to the patient's skin, thereby ensuring stable contact of the sensor when collecting physiological signals such as heart rate, blood flow and blood oxygen saturation, and improving monitoring accuracy and continuity. The wearable unit is further integrated with a prompting unit inside and is electrically connected with the pre-warning unit. When the artificial intelligence model identifies potential risks such as stenosis, thrombus or insufficient blood flow in the blood vessel, the prompting unit will remind the patient in the first time through vibration, sound or light, so that the patient can immediately perceive the abnormality in daily life and seek medical treatment as soon as possible. Thus, the wearable unit not only ensures the comfort and stability of the monitoring process, but also forms an intuitive risk prompting mechanism at the patient end, thereby improving the self-health management ability.

[0069] In a feasible implementation, a pressure adjusting unit is further included,

[0070] The pressure adjusting unit is mechanically connected with the sensor unit and is electrically connected with the monitoring chip unit, and is configured to monitor the pressure value of the device on the patient's blood vessel in real time, and automatically adjust the tightness of the watchband through the intelligent control system, so as to ensure stable contact of the sensor without pressing the blood vessel.

[0071] Exemplarily, the present monitoring device further comprises a pressure adjusting unit. The pressure adjusting unit is mechanically connected with the sensor unit and is electrically connected with the monitoring chip unit. The pressure value of the device acting on the patient's skin and blood vessel is collected in real time by the pressure sensor and is fed back to the monitoring chip unit, and the intelligent control system is used for calculation and adjustment. When it is detected that the device is too tight and may press the blood vessel, the system automatically controls the watchband to loosen; when it is detected that the watchband is too loose and may affect the stability of the signal, the system automatically tightens the watchband, so as to maintain in the appropriate pressure range. Through this dynamic adjustment mode, the sensor can be ensured to stably contact the skin and blood vessel for a long time, and discomfort or blood vessel function interference caused by improper pressure can be avoided, thereby significantly improving the reliability and safety of monitoring.

[0072] In a feasible implementation, a cloud data unit is further included,

[0073] The cloud data unit is wirelessly connected with the communication unit, and is configured to realize distributed storage of monitoring data, data security protection, statistical analysis and visual display, and provide a remote access interface to medical staff and scientific researchers.

[0074] Exemplarily, the monitor further comprises a cloud data unit. The cloud data unit is wirelessly connected with the communication unit, and is configured to upload the locally pre-processed monitoring data to a cloud database in real time. The cloud data unit adopts a distributed storage architecture, can realize efficient management of massive data, and has a strict data security protection mechanism. Through an intelligent data analysis engine, the cloud can statistically analyze and visually display long-term monitoring data of a patient, automatically generate trend curves, risk reports and other results, and facilitate medical staff to quickly assess the vascular condition of the patient. Meanwhile, researchers can also access anonymized data based on an authorized interface, for epidemiological research and development of new diagnostic methods.

[0075] In a second aspect, as shown in Figure 2 The present application also proposes a control method for the arteriovenous fistula blood vessel non-invasive intelligent monitor according to any one of the first aspect.

[0076] In S210, monitoring information of the target patient is collected based on the sensor unit. The monitoring information includes heart rate information, blood oxygen saturation information, blood flow information, vascular imaging information, tremor frequency information, impedance parameters and vascular tension information.

[0077] In S220, the monitoring chip unit amplifies, filters, digitizes and extracts features of the monitoring information to obtain a feature vector.

[0078] In S230, the storage and preprocessing unit stores and pre-processes the feature vector in real time to obtain pre-processed data.

[0079] In S240, the pre-warning unit performs risk assessment on the pre-processed data based on an artificial intelligence model, and generates pre-warning information when potential abnormalities are detected.

[0080] Exemplarily, in step S210, the sensor unit detects the target patient in real time and collects multi-modal monitoring information, which covers heart rate, blood oxygen saturation, blood flow, vascular imaging, tremor frequency, impedance parameters and vascular tension. By uniformly obtaining the above different types of physiological and vascular characteristic signals, the structure and function state of the arteriovenous fistula can be comprehensively reflected.

[0081] In step S220, the monitoring chip unit is electrically connected with the sensor unit, and is configured to receive the multi-source signals and amplify, filter and digitize them to eliminate noise interference, stabilize signal amplitude, and convert analog signals into a form available for digital processing. At the same time, the feature extraction algorithm built in the chip will perform pattern recognition and parameter calculation on various signals, so as to obtain a feature vector representing the vascular function characteristics of the patient.

[0082] In step S230, the feature vector is input to a storage and preprocessing unit, which not only stores the data in real time, but also performs preprocessing operations such as denoising, standardization, normalization, etc., to improve the comparability and analysis quality of the data. This process ensures that the subsequent artificial intelligence model can perform calculations and judgments on a unified data scale, avoiding biases caused by differences in raw signals.

[0083] In step S240, the pre-warning unit further analyzes the pre-processed data. Specifically, the artificial intelligence model embedded in the pre-warning unit performs fusion and risk assessment on the feature vector. Based on the model's ability to identify potential abnormalities such as vascular stenosis, thrombus formation, and insufficient blood flow, it outputs a risk level in real time. When the model determines that there is a potential risk, the system immediately generates a pre-warning message and can notify the patient and medical staff through the wearable terminal, mobile terminal, or medical institution interface, thereby achieving early identification and rapid intervention.

[0084] In summary, the embodiment constructs a complete data flow link through four consecutive steps of sensor acquisition, chip processing, storage preprocessing, and artificial intelligence analysis, achieving dynamic, non-invasive, and intelligent monitoring of the arteriovenous fistula blood vessels. This method not only improves the accuracy and stability of signal processing, but also significantly shortens the time window from abnormality occurrence to pre-warning response, addressing the shortcomings of traditional detection methods such as reliance on manual intervention, intermittency, and lack of quantitative indicators.

[0085] In a feasible implementation, the artificial intelligence model includes:

[0086] A multi-modal time series fusion encoding unit is used to perform heart cycle alignment and time series feature extraction on the heart rate information, blood oxygen saturation information, blood flow information, tremor frequency information, and vascular tension information, and output a fusion feature vector;

[0087] A vascular atlas modeling unit is used to receive the fusion feature vector, vascular imaging information, and impedance parameters, construct a vascular segment branch graph, extract vascular structure and hemodynamic features, and output a vascular atlas feature vector;

[0088] An individualized adaptive unit is used to receive historical baseline data of a target patient, and perform individualized adjustment and cross-device, cross-scene distribution adaptation on the fusion feature vector and the vascular atlas feature vector, to obtain an adjusted comprehensive feature representation;

[0089] An uncertainty assessment and multi-task prediction unit is used to receive the comprehensive feature representation, output stenosis risk, thrombus risk, and access failure risk multi-task prediction results, and generate confidence intervals for each prediction result.

[0090] Exemplarily, the artificial intelligence model of the present application is composed of a multi-modal temporal fusion encoding unit, a vascular atlas modeling unit, an individualized adaptive unit, and an uncertainty evaluation and multi-task prediction unit, which are sequentially connected and work cooperatively.

[0091] Firstly, the multi-modal temporal fusion encoding unit is used to process the heart rate information, blood oxygen saturation information, blood flow information, tremor frequency information and vascular tension information collected by the sensor. This unit uses the heart cycle alignment method to slice and resample the dynamic signals of different modalities according to the heart cycle, eliminating the differences in time scale of different signals. Then, the temporal feature extraction network is used to analyze the features of each type of signal, and the cross-modal fusion mechanism is used to generate a fusion feature vector that can comprehensively reflect the hemodynamic and vascular function state, providing a unified high-dimensional representation for subsequent analysis.

[0092] Next, the vascular atlas modeling unit receives the above-mentioned fusion feature vector, and at the same time, combines the vascular imaging information provided by the ultrasound sensor and the impedance parameters provided by the bioelectrical impedance sensor to construct a vascular segment branch graph. This graph not only describes the structural morphology of the blood vessels, including diameter, wall thickness, bifurcation angle, etc., but also introduces hemodynamic characteristics such as impedance and flow rate. Through the modeling and updating of the graph neural network, the deep features of the blood vessel structure and blood flow state can be extracted, and a vascular atlas feature vector can be generated, thereby providing structured and topological evidence for abnormal identification.

[0093] On this basis, the individualized adaptive unit further introduces the patient's historical baseline data, such as resting heart rate, baseline blood oxygen level, average vascular diameter, and impedance baseline. This unit uses the layer-by-layer feature modulation mechanism to apply baseline information to the fusion feature vector and the vascular atlas feature vector, so that the model can adaptively adjust according to the individual differences of different patients. At the same time, through adversarial training and distribution correction, this unit can also weaken the signal differences between different devices and different scenes, improve the generalization ability of the model in cross-hospital and cross-device applications, and thus obtain a comprehensive feature representation that is more consistent with the actual situation of the target patient.

[0094] Finally, the uncertainty evaluation and multi-task prediction unit receives the above-mentioned comprehensive feature representation and simultaneously outputs multiple prediction results such as narrow risk, thrombus risk and pathway failure risk through a multi-task learning architecture. During the prediction process, this unit also generates a confidence interval for each prediction result through an uncertainty estimation method, and triggers an early warning when the comprehensive risk exceeds the set threshold and its lower confidence interval is still higher than the confidence threshold, thereby reducing the risk of false positives and false negatives and ensuring the timeliness and reliability of clinical intervention.

[0095] The artificial intelligence model proposed in this embodiment not only realizes the time sequence fusion of multi-source physiological information and the atlas modeling of blood vessel structure, but also improves the accuracy and robustness of prediction through individualized adaptation and uncertainty evaluation. Compared with the traditional risk assessment method which relies on a single parameter or artificial experience, the model can provide continuous, comprehensive and confidence-annotated risk prediction results, which helps to identify internal arteriovenous fistula complications early and guide personalized intervention.

[0096] In a feasible implementation, the multi-modal time sequence fusion encoding unit comprises:

[0097] A cardiac cycle alignment subunit is configured to perform cycle slicing and resampling on the heart rate information, the blood oxygen saturation information, the blood flow information, the tremor frequency information and the blood vessel tension information based on feature point detection of the photoplethysmogram signal;

[0098] A dual-flow time sequence encoding subunit is configured to perform feature extraction on the resampled heart rate information, blood oxygen saturation information and tremor frequency information respectively using a time convolution network to obtain a first feature vector, and perform feature extraction on the resampled blood flow information and blood vessel tension information using a bidirectional long short-term memory network to obtain a second feature vector;

[0099] A cross-modal fusion subunit is configured to perform weighted fusion on the first feature vector and the second feature vector through a cross-attention mechanism to generate the fusion feature vector.

[0100] In a feasible implementation, the blood vessel atlas modeling unit comprises:

[0101] A graph construction subunit is configured to extract a blood vessel centerline based on the blood vessel imaging information and construct a blood vessel segment branch graph in combination with the impedance parameter, wherein the node features of the blood vessel segment branch graph include blood vessel diameter, wall thickness, pulsatility index and flow rate, and the edge features include adjacent relationship between blood vessel segments and bifurcation angle;

[0102] A graph neural network subunit is configured to update the node features through a graph attention mechanism and generate a blood vessel segment level risk;

[0103] A physical consistency constraint subunit is configured to introduce blood flow continuity and fluid mechanics formula constraints during the training process to improve the reliability of the blood vessel atlas feature vector.

[0104] In a feasible implementation, the individualized adaptation unit comprises:

[0105] A baseline construction subunit is configured to calculate a resting heart rate, a basic blood oxygen, an average blood vessel diameter and an impedance baseline according to historical monitoring data of a target patient to generate a patient baseline vector.

[0106] characteristic modulation subunit, configured to apply the baseline vector to the fusion feature vector and the vessel atlas feature vector through a layer-by-layer linear modulation mechanism to realize individualized parameter adjustment;

[0107] de-domain subunit, configured to weaken the influence of distribution differences in different devices and different scenes through adversarial training, thereby improving the cross-scene generalization capability of the model, and obtaining the adjusted comprehensive feature representation.

[0108] In a feasible implementation, the uncertainty evaluation and multi-task prediction unit comprises:

[0109] multi-task prediction subunit, configured to simultaneously output multi-task prediction results of the stenosis risk, the thrombosis risk and the pathway failure;

[0110] uncertainty estimation subunit, configured to generate a confidence interval based on the multi-task prediction results through a Monte Carlo Dropout method;

[0111] double-threshold early warning subunit, configured to trigger early warning when the comprehensive risk exceeds a preset risk threshold and the lower confidence limit exceeds the confidence interval.

[0112] For example, the artificial intelligence model proposed in the present application is composed of a multi-modal time series fusion encoding unit, a vessel atlas modeling unit, an individualized adaptive unit and an uncertainty evaluation and multi-task prediction unit, each of which is further subdivided into several subunits, and cooperates to realize the whole process intelligent analysis from data preprocessing to risk output.

[0113] The multi-modal time series fusion encoding unit is responsible for standardization processing and feature extraction of the time series information collected by the sensor. Specifically, the heart cycle alignment subunit aligns the multi-source information such as heart rate, blood oxygen saturation, blood flow, tremor frequency and vessel tension to the same heart cycle scale through feature point detection of the photoplethysmogram signal, and completes slicing and resampling operations to ensure the consistency of different modal data in the time dimension. Subsequently, the dual-flow time series encoding subunit inputs the heart rate, blood oxygen and tremor data into a time convolution network to capture local time series patterns, obtaining a first feature vector; at the same time, the blood flow and vessel tension data are input into a bidirectional long short-term memory network to model bidirectional dependency, obtaining a second feature vector. The cross-modal fusion subunit then performs weighted fusion on the two feature vectors based on a cross-attention mechanism, and finally outputs a fusion feature vector containing multi-dimensional time series information.

[0114] The vascular graph modeling unit is used to structure the morphological and hemodynamic features. The graph construction subunit extracts the vascular centerline based on the vascular imaging information and generates a vascular segment branch graph combined with impedance parameters, wherein the node features cover parameters such as vascular diameter, wall thickness, pulsatility index and flow velocity, and the edge features record the connection relationship and bifurcation angle of adjacent vascular segments. The graph neural network subunit introduces a graph attention mechanism on this basis to update the nodes and edges and generate vascular segment-level features that can characterize local stenosis risk. To ensure that the results conform to physiological laws, the physical consistency constraint subunit introduces blood flow continuity and fluid mechanics formulas as regularization conditions during the training stage to ensure that the vascular graph feature vector output by the model is not only data-driven but also complies with blood flow conservation and pipe wall mechanics constraints, thereby improving the credibility of the prediction.

[0115] The baseline construction subunit first calculates the resting heart rate, baseline blood oxygen, average vascular diameter and impedance baseline according to the historical monitoring data of the patient to generate a baseline vector reflecting individual differences. Then, the feature modulation subunit uses a layer-by-layer linear modulation mechanism to apply the baseline vector to the fusion feature vector and the vascular graph feature vector, thereby adjusting the internal parameters of the model to make it more consistent with the individual characteristics of the target patient. At the same time, the de-territorialization subunit weakens the distribution differences caused by different devices and different acquisition scenarios through an adversarial training method to improve the generalization ability of the model when applied across hospitals and devices, and finally outputs the individualized adjusted comprehensive feature representation.

[0116] The uncertainty evaluation and multi-task prediction unit is used to convert the above-mentioned comprehensive feature representation into a clinical risk prompt. The multi-task prediction subunit can simultaneously output three prediction results of arteriovenous fistula stenosis risk, thrombosis risk and access failure risk, realizing parallel evaluation of multiple complications. The uncertainty estimation subunit generates a corresponding confidence interval for each prediction result based on the Monte Carlo Dropout method to measure the reliability of the model prediction. The dual-threshold early warning subunit triggers the early warning mechanism only when the comprehensive risk value exceeds the preset threshold and the lower confidence limit of the prediction result still exceeds the confidence requirement, thereby effectively reducing the occurrence of false positives and false negatives and ensuring that the early warning result is more clinically operable.

[0117] In summary, the present application combines multi-modal time series feature extraction, vascular structured graph modeling, individualized adaptive adjustment and uncertainty-driven multi-task prediction to achieve comprehensive, accurate and robust risk assessment of arteriovenous fistula status. Compared with traditional analysis methods that rely on a single signal or static image, the present application can capture potential abnormalities at an early stage and provide intelligent early warning results with confidence annotations, significantly improving the safety and scientificity of vascular access management for hemodialysis patients.

[0118] The above examples are only used to illustrate the technical solutions of the present application, but not to limit the same; although the present application has been described in detail with reference to the foregoing examples, it should be understood by those of ordinary skill in the art that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An arteriovenous fistula blood vessel non-invasive intelligent monitor, characterized in that, The application relates to a wearable device for monitoring a patient's blood vessel, comprising: a sensor unit comprising a photoplethysmogram sensor, a micro-ultrasound sensor and a bioelectrical impedance sensor for collecting monitoring information of a target patient, the monitoring information comprising heart rate information, blood oxygen saturation information, blood flow information, vascular imaging information, tremor frequency information, impedance parameters and vascular tension information; a monitoring chip unit electrically connected to the sensor unit for amplifying, filtering, analog-to-digital converting and feature extracting the monitoring information to obtain a feature vector; a storage and preprocessing unit electrically connected to the monitoring chip unit for real-time storing and preprocessing the feature vector; a communication unit electrically connected to the storage and preprocessing unit for transmitting the preprocessed data to a medical institution terminal device or a cloud database through Bluetooth or Wi-Fi; a power management unit electrically connected to the sensor unit, the monitoring chip unit, the storage and preprocessing unit and the communication unit for powering each unit and realizing low-power consumption control; an early warning unit electrically connected to the monitoring chip unit and the storage and preprocessing unit for risk assessment of the preprocessed data based on an artificial intelligence model and generating early warning information when potential abnormalities are detected.

2. The arteriovenous fistula vascular non-invasive intelligent monitor according to claim 1, characterized in that, The application further comprises a wearable unit, which comprises: a flexible electronic main body fixedly connected to the sensor unit for supporting and adhering to the patient's skin; an alert unit electrically connected to the early warning unit for prompting the patient through vibration, sound or light when the early warning is triggered.

3. The arteriovenous fistula vascular non-invasive intelligent monitor according to claim 1, characterized in that, The application further comprises a pressure adjusting unit, which is mechanically connected to the sensor unit and electrically connected to the monitoring chip unit for real-time monitoring of the pressure value of the device on the patient's blood vessel and automatically adjusting the tightness of the watchband through an intelligent control system to ensure that the sensor stably contacts without pressing the blood vessel.

4. The arteriovenous fistula vascular non-invasive intelligent monitor according to claim 1, characterized in that, The application further comprises a cloud data unit, which is wirelessly connected to the communication unit for realizing distributed storage of monitoring data, data security protection, statistical analysis and visual display and providing a remote access interface for medical staff and scientific researchers.

5. A control method for the arteriovenous fistula blood vessel noninvasive intelligent monitor according to any one of claims 1 to 4, characterized in that, The application comprises: collecting monitoring information of a target patient based on the sensor unit, the monitoring information comprising heart rate information, blood oxygen saturation information, blood flow information, vascular imaging information, tremor frequency information, impedance parameters and vascular tension information; amplifying, filtering, analog-to-digital converting and feature extracting the monitoring information based on the monitoring chip unit to obtain a feature vector; real-time storing and preprocessing the feature vector based on the storage and preprocessing unit to obtain preprocessed data; risk assessment of the preprocessed data based on an artificial intelligence model through the early warning unit and generating early warning information when potential abnormalities are detected.

6. The control method according to claim 5, characterized by The artificial intelligence model comprises: a multi-modal time sequence fusion encoding unit for heart cycle alignment and time sequence feature extraction of the heart rate information, the blood oxygen saturation information, the blood flow information, the tremor frequency information and the vascular tension information and outputting a fusion feature vector; a blood vessel atlas modeling unit configured to receive the fusion feature vector, the blood vessel imaging information, and the impedance parameter, construct a blood vessel segment branch graph, extract blood vessel structure and hemodynamic features, and output a blood vessel atlas feature vector; an individualized adaptive unit configured to receive historical baseline data of a target patient, and perform individualized adjustment and cross-device and cross-scene distribution adaptation on the fusion feature vector and the blood vessel atlas feature vector to obtain an adjusted comprehensive feature representation; an uncertainty evaluation and multi-task prediction unit configured to receive the comprehensive feature representation to output multi-task prediction results of stenosis risk, thrombosis risk, and access failure risk, and generate a confidence interval for each prediction result.

7. The control method according to claim 6, characterized by The multi-modal time series fusion encoding unit comprises: a cardiac cycle alignment sub-unit configured to perform cycle slicing and resampling on the heart rate information, the blood oxygen saturation information, the blood flow information, the tremor frequency information, and the blood vessel tension information based on feature point detection of a photoplethysmogram signal; a dual-flow time series encoding sub-unit configured to perform feature extraction on the resampled heart rate information, the blood oxygen saturation information, and the tremor frequency information respectively using a time convolution network to obtain a first feature vector, and perform feature extraction on the resampled blood flow information and the blood vessel tension information using a bidirectional long short-term memory network to obtain a second feature vector; a cross-modal fusion sub-unit configured to perform weighted fusion on the first feature vector and the second feature vector through a cross-attention mechanism to generate the fusion feature vector.

8. The control method according to claim 6, characterized by, The blood vessel atlas modeling unit comprises: a graph construction sub-unit configured to extract a blood vessel centerline based on the blood vessel imaging information and construct a blood vessel segment branch graph in combination with the impedance parameter, wherein node features of the blood vessel segment branch graph include blood vessel diameter, wall thickness, pulsatility index, and flow velocity, and edge features include adjacent relationship between blood vessel segments and bifurcation angle; a graph neural network sub-unit configured to update the node features and generate blood vessel segment level risk through a graph attention mechanism; a physical consistency constraint sub-unit configured to introduce blood flow continuity and fluid mechanics formula constraints in a training process to improve credibility of the blood vessel atlas feature vector.

9. The control method according to claim 6, characterized by, The individualized adaptive unit comprises: a baseline construction sub-unit configured to calculate resting heart rate, basic blood oxygen, average blood vessel diameter, and impedance baseline from historical monitoring data of a target patient to generate a patient baseline vector; a feature modulation sub-unit configured to apply the baseline vector to the fusion feature vector and the blood vessel atlas feature vector through a layer-by-layer linear modulation mechanism to achieve individualized parameter adjustment; a de-domaining sub-unit configured to weaken the influence of distribution differences in different devices and different scenes through adversarial training to improve cross-scene generalization capability of the model, and obtain the adjusted comprehensive feature representation.

10. The control method according to claim 6, characterized by, The uncertainty evaluation and multi-task prediction unit comprises: a multi-task prediction sub-unit configured to simultaneously output multi-task prediction results of stenosis risk, thrombosis risk, and access failure. An uncertainty estimation subunit is configured to generate a confidence interval for the multi-task prediction result based on a Monte Carlo Dropout method. A double-threshold early warning subunit is configured to trigger an early warning when the comprehensive risk exceeds a preset risk threshold and the lower confidence limit thereof exceeds the confidence interval.

Images