Hemodialysis internal arteriovenous fistula thermal image intelligent analysis and early warning system and method thereof

Abstract

The invention relates to the technical field of medical image processing, in particular to a hemodialysis internal arteriovenous fistula thermal image intelligent analysis and early warning system and method, and the system comprises an infrared thermal imaging device, an image processing device, a server, a mobile device and a cloud analysis platform. The method is characterized in that the image processing equipment comprises a thermal image analysis module, a dynamic image analysis module and a data integration module, the thermal image analysis module introduces a differential geometry theory, maps an internal fistula thermal image into a parameterized temperature manifold, and obtains a parameterized temperature manifold through geodesic analysis and vector field topology analysis; calculating a geodesic deviation index and a topological characteristic change rate index, and accurately evaluating the internal fistula function state; the dynamic image analysis module calculates parameters such as the blood vessel outer diameter, the blood flow and the heat conduction index; the data integration module collects data of all the modules to form comprehensive evaluation, remote evaluation guidance of experts is supported through a remote consultation function, early warning is achieved based on time sequence topology changes, and internal fistula function deterioration is predicted 7-10 days in advance.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and in particular to an intelligent analysis and early warning system and method for thermal images of arteriovenous fistulas in hemodialysis. Background Technology

[0002] Hemodialysis is one of the main renal replacement therapies for patients with end-stage renal disease, and arteriovenous fistula (AVF) is the most ideal long-term vascular access for hemodialysis patients. However, AVF complications such as stenosis, thrombosis, and infection can seriously affect the quality of dialysis and quality of life of patients, and even threaten their lives.

[0003] Currently, fistula function assessment mainly relies on clinical observation, auscultation, and ultrasound examination. These traditional methods suffer from drawbacks such as high subjectivity, operator dependence, and difficulty in quantification. While ultrasound examination can provide relatively objective assessment indicators, the equipment is expensive, the operation is complex, and it cannot achieve remote monitoring and early warning. On the other hand, infrared thermography, as a non-invasive, safe, and convenient means of detecting body surface temperature, has been applied in many medical fields. However, in the field of fistula function monitoring, existing thermal image analysis methods are mostly limited to simple temperature threshold judgment or basic statistical analysis, lacking the ability to explore the deeper geometric characteristics of the temperature field, and thus failing to accurately assess fistula function and predict potential risks.

[0004] Therefore, how to utilize thermal imaging technology, combined with advanced image processing and mathematical modeling methods, to achieve objective assessment and early warning of arteriovenous fistula function is a technical problem that urgently needs to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent analysis and early warning system and method for thermal images of arteriovenous fistulas in hemodialysis. By introducing differential geometry theory to perform in-depth analysis of fistula thermal images and combining dynamic image features, the system can achieve accurate assessment and early warning of fistula function, providing an effective technical means for the prevention and treatment of fistula complications.

[0006] This invention proposes an intelligent analysis and early warning system for thermal images of arteriovenous fistulas in hemodialysis, comprising:

[0007] Infrared thermal imaging equipment is used to acquire infrared thermal image data of the patient's arteriovenous fistula area;

[0008] An image processing device, communicatively connected to the infrared thermal imaging device, includes a thermal image analysis module, a dynamic image analysis module, and a data integration module;

[0009] The server, which is communicatively connected to the image processing device, includes a data receiving module, a real-time data storage module, a historical data storage module, an alarm module, and an image analysis module.

[0010] Mobile devices, connected to the server via a network, are used to display images and analysis results in real time, as well as view historical analysis results;

[0011] A cloud-based analytics platform, which communicates with the server, is used for real-time recording and analysis of data in the cloud, as well as remote access to data;

[0012] The thermal image analysis module includes:

[0013] A temperature manifold construction unit is used to map the thermal image of the arteriovenous fistula into a parameterized temperature manifold, calculate the metric tensor and curvature features of the temperature manifold, and analyze the correspondence between the curvature features and the arteriovenous fistula vascular structure.

[0014] The geodesic analysis unit is communicatively connected to the temperature manifold construction unit and is used to calculate the heat conduction geodesic path on the temperature manifold, calculate the deviation index between the geodesic path and the geodesic of the standard arteriovenous fistula template, and evaluate the functional status of the arteriovenous fistula based on the deviation index.

[0015] The vector field topology analysis unit is communicatively connected to the temperature manifold construction unit and the geodesic analysis unit. It is used to construct a heat flow vector field, identify the critical point and topological features of the heat flow vector field, calculate the time-series topological feature change rate index, and realize early warning of arteriovenous fistula function based on the topological feature change rate index.

[0016] The dynamic image analysis module is communicatively connected to the thermal image analysis module and is used to calculate the outer diameter, area, blood flow, perfusion index, and blood flow ratio of the fistula at different times, as well as the arteriovenous temperature gradient ratio and the arteriovenous heat conduction index.

[0017] The data integration module is communicatively connected to the thermal image analysis module and the dynamic image analysis module, and is used to collect data from the thermal image analysis module and the dynamic image analysis module and transmit it to the server.

[0018] Preferably, the temperature manifold building unit includes:

[0019] The preprocessing subunit is used to perform noise reduction, contrast enhancement, and spatial filtering on the infrared thermal image data.

[0020] The manifold mapping subunit is communicatively connected to the preprocessing subunit and is used to establish a coordinate mapping function to map the two-dimensional image space to a three-dimensional temperature manifold, construct a temperature height function, and form a parameterized surface representation.

[0021] The metric tensor computation subunit is communicatively connected to the manifold mapping subunit and is used to construct the local metric tensor of the temperature manifold, calculate the eigenvalues ​​and eigenvectors of the metric tensor, and determine the main direction of temperature change.

[0022] The curvature analysis subunit is communicatively connected to the metric tensor calculation subunit. It is used to calculate the Gaussian curvature and average curvature of the temperature manifold, construct a curvature distribution map, identify areas of abnormal temperature changes, and establish the morphological similarity between the curvature distribution morphology and the standard arteriovenous fistula template.

[0023] Preferably, the geodesic analysis unit includes:

[0024] The endpoint identification subunit is used to define a distance function based on the temperature manifold metric, construct a geodesic distance field, and identify significant points on the temperature manifold as geodesic endpoints.

[0025] The path solving subunit is communicatively connected to the endpoint identification subunit and is used to calculate the geodesic path using an iterative optimization strategy with an adaptive step size, realize local geodesic segment calculation, and construct the geodesic backbone network.

[0026] The feature extraction subunit is communicatively connected to the path solving subunit and is used to extract geodesic length features, geodesic curvature integral, geodesic direction change rate, and geodesic width distribution.

[0027] The functional scoring subunit is communicatively connected to the feature extraction subunit. It is used to construct the mapping relationship between the geodesic features and the functional status of the arteriovenous fistula, calculate the functional score based on the geodesic features, set the functional score threshold, and classify the arteriovenous fistula status level.

[0028] Preferably, the vector field topology analysis unit includes:

[0029] The vector field construction sub-unit is used to calculate the temperature gradient vector field based on the temperature manifold, and to construct the heat flow vector field by applying the law of heat conduction, thereby achieving vector field smoothing and normalization.

[0030] The vector field feature extraction subunit is communicatively connected to the vector field construction subunit. It is used to calculate the divergence distribution, curl characteristics and Laplace features of the heat flux vector field, and to perform region segmentation and identify functional regions based on the vector field features.

[0031] The critical point identification subunit is communicatively connected to the vector field feature extraction subunit. It is used to detect the singularity position of the heat flux vector field, classify the critical point type, calculate the critical point feature index, construct the critical point connection relationship, and form a topological skeleton diagram.

[0032] The dynamic monitoring subunit is communicatively connected to the critical point identification subunit. It is used to design a time-series topology feature tracking algorithm, monitor the evolution of the topology structure, calculate the topology feature change rate index, and establish an anomaly detection mechanism based on the topology feature change rate index.

[0033] Preferably, the dynamic image analysis module calculates the arteriovenous temperature gradient ratio and arteriovenous thermal conduction index by tracking arteriovenous thermal gradient characteristic indicators in real time, including:

[0034] Radial artery temperature characteristics: ;in, Represents the heat peak, Represents the trough value, Represents the difference between peak and trough temperatures;

[0035] Ulnar vein temperature characteristics: ;in, Representing the hot valley, Represents the peak value. Represents the difference between the peak and trough temperatures;

[0036] Dynamic flow parameters include: rate of change of arterial and venous temperature, heat conduction velocity, arterial and venous temperature gradient ratio, and arterial and venous heat conduction index.

[0037] Preferably, the dynamic image analysis module calculates the blood flow ratio by relating the rate of temperature change to blood flow, i.e.:

[0038] ,

[0039] in, Indicates blood flow at the arterial end. Indicates venous blood flow. Indicates the rate of change of arterial temperature. This indicates the rate of change in venous temperature.

[0040] Preferably, the thermal image analysis module calculates the arteriovenous thermal conductivity index by representing the blood flow thermal conduction velocity through a temperature gradient, i.e.:

[0041] Thermal conductivity index = dT / dt

[0042] Where dT represents the temperature gradient and dt represents the time gradient.

[0043] Preferably, the alarm module of the server includes:

[0044] An anomaly detection unit is used to receive the analysis results from the image analysis module and determine the abnormal state of the arteriovenous fistula function based on a preset threshold.

[0045] The risk grading unit is communicatively connected to the anomaly detection unit and is used to classify the risk level of arteriovenous fistula anomalies based on the topological feature change rate index.

[0046] An alarm generation unit, which is communicatively connected to the risk classification unit, is used to generate alarm information of the corresponding level according to the risk level.

[0047] The push unit is communicatively connected to the alarm generation unit and is used to push the alarm information to the mobile device and the cloud analysis platform.

[0048] Preferably, the cloud analytics platform includes:

[0049] The data receiving unit is used to receive the fistula thermal image data, differential geometric feature data and analysis results transmitted by the server;

[0050] The expert consultation unit is communicatively connected to the data receiving unit and is used to provide remote data access and advanced analysis tools to support remote expert evaluation and guidance.

[0051] The knowledge base unit, which is connected to the expert consultation unit, is used to store typical arteriovenous fistula cases and differential geometric feature patterns, and to continuously update and optimize the analysis model.

[0052] The report generation unit, which is connected in communication with the knowledge base unit, is used to generate fistula function assessment reports and early warning analysis reports to support medical decision-making.

[0053] A method for intelligent analysis and early warning of thermal images of arteriovenous fistulas in hemodialysis patients, which collects infrared thermal image data of the fistula vessel area in patients;

[0054] The infrared thermal image data is mapped to a parameterized temperature manifold, the metric tensor and curvature features of the temperature manifold are calculated, and the correspondence between the curvature features and the arteriovenous fistula structure is analyzed.

[0055] Calculate the thermal conduction geodesic path on the temperature manifold, calculate the deviation index between the geodesic path and the geodesic of the standard arteriovenous fistula template, and evaluate the functional status of the arteriovenous fistula based on the deviation index;

[0056] Construct a heat flux vector field, identify the critical points and topological features of the heat flux vector field, and calculate the time-series topological feature change rate index;

[0057] Calculate the outer diameter, area, blood flow, perfusion index, and blood flow ratio of the fistula at different times, as well as the arteriovenous temperature gradient ratio and the arteriovenous thermal conductivity index.

[0058] By combining differential geometric feature data and dynamic image analysis data, the functional status of the arteriovenous fistula is comprehensively evaluated.

[0059] Based on the aforementioned topological feature change rate index and comprehensive evaluation results, early warning of arteriovenous fistula function is achieved.

[0060] Displays images and analysis results in real time on mobile devices, and supports viewing historical analysis results;

[0061] Data is recorded and analyzed in real time through a cloud-based analytics platform, supporting remote expert consultations.

[0062] The present invention has the following beneficial effects:

[0063] 1. By mapping the thermal image of the arteriovenous fistula to a temperature manifold through differential geometry theory and extracting its intrinsic geometric features, a high-dimensional and global analysis of temperature distribution is achieved, which greatly improves the accuracy and sensitivity of arteriovenous fistula functional assessment.

[0064] 2. Based on geodesic theory, a correspondence between heat conduction paths and blood flow paths was established. The degree of arteriovenous fistula dysfunction was quantified by the geodesic deviation index, providing a new objective indicator for arteriovenous fistula function assessment.

[0065] 3. The topological structure analysis of the heat flow vector field can capture subtle changes in the function of the arteriovenous fistula and achieve early warning through the topological characteristic change rate index. It can predict the deterioration of the arteriovenous fistula function 7 to 10 days earlier than traditional methods, which greatly improves the success rate of intervention.

[0066] 4. The system integrates functions such as data acquisition, processing, analysis, storage, early warning, and remote consultation, forming a complete closed loop for arteriovenous fistula management, which significantly reduces the incidence of arteriovenous fistula complications and extends the lifespan of arteriovenous fistulas.

[0067] 5. The cloud-based analytics platform supports remote expert consultations, which solves the problem of uneven distribution of high-quality medical resources and enables patients in primary hospitals to receive expert-level medical services. Attached Figure Description

[0068] Figure 1 This is a schematic diagram of the overall architecture of the intelligent analysis and early warning system for thermal images of arteriovenous fistulas in hemodialysis according to the present invention;

[0069] Figure 2 This is a functional structure diagram of the thermal image analysis module of the present invention;

[0070] Figure 3 This is a schematic diagram of the temperature manifold construction process of the present invention;

[0071] Figure 4 This is a schematic diagram of the geodesic analysis process of the present invention;

[0072] Figure 5 This is a schematic diagram of the topological analysis of the heat flux vector field in this invention;

[0073] Figure 6 This is a schematic diagram of dynamic image analysis according to the present invention;

[0074] Figure 7 This is a schematic diagram of the arteriovenous fistula function early warning process of the present invention. Detailed Implementation

[0075] Please refer to Figures 1-7 The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples.

[0076] See Figure 1 The intelligent analysis and early warning system for thermal images of arteriovenous fistulas in hemodialysis of the present invention includes: an infrared thermal imaging device 1, an image processing device 2, a server 3, a mobile device 4, and a cloud analysis platform 5.

[0077] Infrared thermal imaging device 1 is used to acquire infrared thermal image data of the fistula region in patients. Preferably, the infrared thermal imaging device 1 has a temperature resolution of not less than 0.05℃, a spatial resolution of not less than 384×288 pixels, an imaging frequency of not less than 30Hz, and a working distance of 30-50cm. These parameter settings ensure that sufficiently detailed temperature change information of the fistula region is acquired, providing a high-quality data foundation for subsequent analysis.

[0078] Image processing device 2 is communicatively connected to infrared thermal imaging device 1, and includes a thermal image analysis module 21, a dynamic image analysis module 22, and a data integration module 23. In one embodiment of the present invention, image processing device 2 may be a workstation equipped with a multi-core high-performance CPU and GPU acceleration, with a memory capacity of not less than 16GB and a storage capacity of not less than 1TB SSD, to meet the needs of large-scale data processing.

[0079] Server 3 is communicatively connected to image processing device 2 and includes a data receiving module 31, a real-time data storage module, a historical data storage module, an alarm module, and an image analysis module. Server 3 is used to receive and store data transmitted by image processing device 2, perform further analysis, and issue early warnings when an anomaly is detected.

[0080] Mobile device 4 communicates with server 3 via a network to display images and analysis results in real time, as well as view historical analysis results. In embodiments of the present invention, mobile device 4 can be a tablet computer, smartphone, or dedicated medical terminal, allowing medical personnel to monitor the patient's arteriovenous fistula status at any time.

[0081] The cloud analytics platform 5 communicates with server 3 to record and analyze data in real time in the cloud, as well as to remotely access data. The cloud analytics platform 5 supports remote expert consultations, enabling experts from higher-level hospitals to remotely assess fistula function and provide professional guidance to primary hospitals.

[0082] See Figure 2 The thermal image analysis module 21 is the core innovation of this invention. It uses differential geometry theory to perform in-depth analysis of the thermal image of the fistula, including a temperature manifold construction unit 211, a geodesic analysis unit 212, and a vector field topology analysis unit 213.

[0083] The temperature manifold construction unit 211 is used to map the thermal image of the arteriovenous fistula into a parameterized temperature manifold, calculate the metric tensor and curvature features of the temperature manifold, and analyze the correspondence between the curvature features and the arteriovenous fistula vascular structure. Specifically, the temperature manifold construction unit 211 includes a preprocessing subunit 2111, a manifold mapping subunit 2112, a metric tensor calculation subunit 2113, and a curvature analysis subunit 2114.

[0084] The preprocessing subunit 2111 is used to perform noise reduction, contrast enhancement, and spatial filtering on the infrared thermal image data. In a preferred embodiment of the invention, adaptive Gaussian filtering is used for noise reduction, with the filter kernel size automatically adjusted according to the image noise level, typically ranging from 3×3 to 7×7 pixels; adaptive histogram equalization is used to enhance temperature contrast, with the enhancement window size set to 8×8 pixels and an overlap rate of 0.5; and Laplacian operator is used for spatial filtering to highlight the temperature boundaries of the vascular region. These preprocessing steps effectively improve image quality, laying the foundation for subsequent manifold construction.

[0085] The manifold mapping subunit 2112 is communicatively connected to the preprocessing subunit 2111 and is used to establish a coordinate mapping function to map the two-dimensional image space to a three-dimensional temperature manifold, construct a temperature-height function, and form a parameterized surface representation. Specifically, the manifold mapping process can be represented as follows:

[0086] ,

[0087] in, For temperature manifolds, represent three-dimensional parametric surfaces; These are the coordinates of a point on the image plane, in pixels. The fistula region represents the set of all pixels in the image that contain fistulas; Coordinates The temperature value at the location is expressed in degrees Celsius (°C). This mapping establishes a correspondence from a two-dimensional image space to a three-dimensional temperature manifold, allowing the temperature distribution to be viewed as a surface with geometric properties.

[0088] The metric tensor computation subunit 2113 is communicatively connected to the manifold mapping subunit 2112. It is used to construct the local metric tensor of the temperature manifold, compute the eigenvalues ​​and eigenvectors of the metric tensor, and determine the principal direction of temperature change. In differential geometry, the metric tensor describes the distance and angular relationships on the manifold and is fundamental to the intrinsic geometry of the manifold. For the temperature manifold, the metric tensor... It can be represented as:

[0089] ,

[0090] in, To measure a tensor, it is a symmetric positive definite matrix; Temperature function right The partial derivatives, in °C / pixel; Temperature function right The partial derivatives are expressed in °C / pixel. In an embodiment of the invention, the partial derivatives are calculated using the central difference method, with a step size of 1 pixel.

[0091] The curvature analysis subunit 2114 is communicatively connected to the metric tensor calculation subunit 2113. It is used to calculate the Gaussian curvature and average curvature of the temperature manifold, construct a curvature distribution map, identify regions of abnormal temperature changes, and establish the morphological similarity between the curvature distribution morphology and the standard arteriovenous fistula template. (Gaussian curvature) and mean curvature The calculation formula is:

[0092] ,

[0093] ,

[0094] in, Gaussian curvature, representing the intrinsic curvature of the temperature surface, with units of 1 / pixel. 2 ; The mean curvature represents the intrinsic curvature of the temperature surface, with units of 1 / pixel; Temperature function right The second-order partial derivative, in °C / pixel. Temperature function right The second partial derivative, in °C / pixel. 2 Temperature function right and The mixed second-order partial derivatives, in °C / pixel. is the area element of the temperature manifold. Gaussian curvature reflects the concentration of the temperature manifold, and is usually high at vascular bifurcation or stenosis; mean curvature characterizes the local concavity and convexity of the temperature manifold, corresponding to vascular dilation or stenosis.

[0095] In practical applications, curvature characteristics are highly correlated with the vascular structure of arteriovenous fistulas. For example, in functionally normal fistulas, the area near the arteriovenous anastomosis typically exhibits a high curvature region, while the curvature value changes smoothly along the vessel direction; functionally abnormal fistulas may show irregular curvature distribution, local curvature abrupt changes, and other characteristics. By comparing the patient's curvature distribution map with a standard fistula template and calculating the morphological similarity, the functional status of the fistula can be preliminarily assessed.

[0096] The geodesic analysis unit 212 is communicatively connected to the temperature manifold construction unit 211. It is used to calculate the heat conduction geodesic path on the temperature manifold, calculate the deviation index between the geodesic path and the geodesic of the standard arteriovenous fistula template, and assess the functional status of the arteriovenous fistula based on the deviation index. Specifically, the geodesic analysis unit 212 includes an endpoint identification subunit 2121, a path solving subunit 2122, a feature extraction subunit 2123, and a functional scoring subunit 2124.

[0097] Endpoint identification subunit 2121 is used to define a distance function based on the temperature manifold metric, construct a geodesic distance field, and identify salient points on the temperature manifold as geodesic endpoints. In differential geometry, a geodesic is the shortest path between two points on a manifold. On a temperature manifold, a geodesic can be interpreted as the optimal path for heat conduction, which has a high degree of correspondence with the blood flow path in an arteriovenous fistula.

[0098] The first step in this invention is to determine the endpoints of the geodesics. In the thermal images of arteriovenous fistulas, the radial artery end and the ulnar vein end typically represent local temperature extrema. The endpoint identification subunit 2121 identifies these salient points using a local extremum detection algorithm. The specific steps are as follows: first, the magnitude of the local temperature gradient at each point is calculated on the temperature manifold; then, points whose gradient magnitude is less than that of their neighbors and whose temperature values ​​are local extrema are identified; finally, based on anatomical knowledge, the extremum points located at the boundaries of the arteriovenous fistula region are selected as the geodesic endpoints.

[0099] The path solving subunit 2122 is communicatively connected to the endpoint identification subunit 2121. It is used to calculate geodesic paths using an iterative optimization strategy with an adaptive step size, achieving segmented calculation of local geodesics and constructing a geodesic backbone network. The calculation of geodesics is based on the geodesic equation:

[0100] ,

[0101] in, Represents the parametric coordinates of the surface. The coordinate component index (values ​​are 1 or 2, corresponding to...). and ); These are parameters, representing parameterized variables along the geodesic line; It is a Christopher notation, representing the local geometric properties of a surface, determined by the metric tensor and its first derivative; and For coordinate component indices (all values ​​are 1 or 2); and The coordinate components of the geodesic tangent vector are represented; the entire equation follows Einstein's summation convention, summing the repeated indices.

[0102] Considering the complexity of geodesic equations, this invention employs a numerical method for solution. Preferably, a fourth-order Runge-Kutta method with an adaptive step size is used, ranging from 0.01 to 0.1, automatically adjusted based on local curvature; the iteration termination condition is when the path change is less than 0.1% or the maximum number of iterations (100) is reached. To improve computational efficiency, a segmented computation strategy is adopted, decomposing the long path into multiple shorter paths, which are then concatenated.

[0103] Feature extraction subunit 2123 is communicatively connected to path solving subunit 2122, and is used to extract geodesic length features, geodesic curvature integrals, geodesic direction change rates, and geodesic width distributions. (Geodesic length) The calculation formula is:

[0104] ,

[0105] in, This is the length of the geodesic line, in pixels. It is a dimensionless component that measures the tensor. and These are the components of the geodesic tangent vector, expressed in pixels per parameter unit. and These are the start and end values ​​of the path parameters; the entire integral represents the cumulative length along the geodesic path, and the integration region extends from the parameters. arrive The geodesic length reflects the actual length of the arteriovenous fistula vessel.

[0106] Geodesic curvature integral Represented as:

[0107] ,

[0108] in, This is the integral of geodesic curvature, in radians; Is the geodesic in the parameters The geodesic curvature at a point, expressed in radians per unit of parameter, indicates the degree to which a geodesic line deviates from a straight line; Indicates taking the absolute value; the integration region extends from the parameter. arrive This integral represents the cumulative curvature change along the entire path. The integral value reflects the degree of curvature of the geodesic, corresponding to the smoothness of the blood vessel.

[0109] Rate of change of geodesic direction Defined as:

[0110] ,

[0111] in, The rate of change of geodesic direction, in radians per pixel; It is the angle of the geodesic tangent vector, in radians; This represents the rate of change of the angle with respect to the parameter, expressed in radians per unit of the parameter. Indicates taking the absolute value; The length of the geodesic is expressed in pixels; the entire formula represents the average angular change per unit length of geodesic. This indicator quantifies the frequency of change in the geodesic direction and is related to the curvature of blood vessels.

[0112] The geodesic width distribution is obtained by calculating the temperature gradient width in the orthogonal direction of the geodesic and is used to assess changes in blood vessel diameter.

[0113] The functional scoring subunit 2124 is communicatively connected to the feature extraction subunit 2123. It is used to construct a mapping relationship between geodesic features and the functional status of the arteriovenous fistula (AVF), calculate a functional score based on the geodesic features, set a functional score threshold, and classify the AVF status level. There is a clear correspondence between geodesic features and the AVF functional status. For example, the geodesic lines of a normally functioning AVF are usually smooth, with uniform curvature and consistent width distribution; while abnormally functioning AVFs may exhibit features such as sudden bending of the geodesic lines and abrupt changes in width.

[0114] This invention calculates the Geodesic Deviation Index (GDI) based on geodesic characteristics:

[0115] ,

[0116] Wherein, GDI is the geodesic deviation index, which is dimensionless and represents the degree of deviation between the test sample and the standard template; The geodesic length of the test sample, in pixels; The geodesic length of the standard arteriovenous fistula template, in pixels; The integral of the geodesic curvature of the test sample, in radians; The integral of geodesic curvature of the standard template, in radians; The rate of change of orientation of the test sample, in radians per pixel; The directional change rate of the standard template, in radians per pixel; For test samples in parameters The width of the geodesic line at that location, in pixels; For standard templates in parameters The width of the geodesic line at that location, in pixels; , , and These are the corresponding weight coefficients, dimensionless, representing the weight of each feature item in the total score, and are typically set to... and 0.2; Indicates taking the absolute value. The smaller the GDI value, the closer the function of the internal fistula is to the normal state.

[0117] According to clinical verification, the present invention sets the following GDI thresholds to divide the internal fistula state levels: GDI ≤ 0.15 is excellent (good function); 0.15 < GDI ≤ 0.3 is good (mild abnormality); 0.3 < GDI ≤ 0.5 is medium (moderate abnormality); GDI > 0.5 is poor (severe abnormality). These thresholds are obtained based on the analysis of 100 clinical cases and can be appropriately adjusted according to the characteristics of specific hospitals and patient groups.

[0118] The vector field topology analysis unit 213 is communicatively connected to the temperature manifold construction unit 211 and the geodesic analysis unit 212, and is used to construct a heat flux vector field, identify the critical points and topological structure characteristics of the heat flux vector field, calculate the time-series topological feature change rate index, and realize the early warning of the internal fistula function based on the topological feature change rate index. Specifically, the vector field topology analysis unit 213 includes a vector field construction subunit 2131, a vector field feature extraction subunit 21, a critical point identification subunit 21, and a dynamic monitoring subunit 21.

[0119] The vector field construction subunit 2131 is used to calculate the temperature gradient vector field based on the temperature manifold, construct the heat flux vector field by applying the heat conduction law, and realize the smoothing and normalization processing of the vector field. The heat flux vector field is defined as the negative value of the temperature gradient, indicating the direction of heat flow:

[0120] ,

[0121] where is the heat flux vector field, which is a two-dimensional vector field with the unit of W / m²; is the thermal conductivity coefficient, and its value in human tissues is usually 0.2 - 0.5 W / (m·K); is the temperature gradient operator, representing the rate of change of temperature in space; is the temperature function with respect to partial derivative, with the unit of K / m; is the temperature function with respect to partial derivative, with the unit of K / m.

[0122] To reduce the influence of noise, the heat flux vector field is smoothed using a Gaussian kernel with a kernel size of 5×5 pixels. Then, normalization processing is performed to unify the vector field norm for subsequent analysis:

[0123] ,

[0124] where The normalized heat flux vector field is dimensionless. This represents the original heat flux vector field; This represents the magnitude of the vector field, i.e., the length of the vector; It is a small constant (usually 0.001) to prevent division by zero errors.

[0125] The vector field feature extraction subunit 21 is communicatively connected to the vector field construction subunit 2131. It is used to calculate the divergence distribution, curl characteristics, and Laplace characteristics of the heat flux vector field, and to perform region segmentation and identify functional regions based on the vector field features. (Diffusion distribution) The calculation formula is:

[0126] ,

[0127] in, The divergence of the heat flux vector field represents the degree of heat convergence or dissipation per unit area, with units of W / m². ; and These are the heat flux vector fields in and The directional components are all in W / m. 2 ; express Directional component pairs The partial derivatives, in units of W / m ; express Directional component pairs The partial derivatives, in units of W / m Divergence reflects the degree of heat convergence or dissipation; a positive value indicates a heat source (usually corresponding to the arterial end), and a negative value indicates a heat sink (usually corresponding to the venous end).

[0128] curl characteristics Represented as:

[0129] ,

[0130] in, Let be the curl of the heat flux vector field. In the two-dimensional case, it is a scalar field representing the intensity of the heat flux rotation, with units of W / m. express Directional component pairs The partial derivatives, in units of W / m express Directional component pairs The partial derivatives, in units of W / m Curl describes the rotational characteristics of heat flow and is typically high at bends or bifurcations in blood vessels.

[0131] Laplace features The calculation is as follows:

[0132] ,

[0133] in, The Laplace operator for the heat flux vector field is a two-dimensional vector field that represents the second spatial derivative of the heat flux, with units of W / m². express Directional component pairs express Directional component pairs The second partial derivative, in units of W / m express Directional component pairs The second partial derivative, in units of W / m express Directional component pairs The second partial derivative, in units of W / m The Laplace feature reflects the second-order change in heat flow and is sensitive to areas of abnormal arteriovenous fistula function.

[0134] Based on these characteristics, the fistula region can be functionally segmented. For example, by setting a divergence threshold (±0.05 W / m)... 3 It can identify heat source and heat sink areas; and it can identify vascular tortuosity or bifurcation areas through the rotation threshold (0.03 W / m³). Identification of these functional areas helps in the overall functional assessment of the arteriovenous fistula.

[0135] The critical point identification subunit 21 is communicatively connected to the vector field feature extraction subunit 21. It is used to detect the singularity positions of the heat flux vector field, classify critical point types, calculate critical point feature indices, construct critical point connection relationships, and form a topological skeleton diagram. In vector field topology theory, critical points (singularities) are points where the vector field is zero or uncertain, representing key features of the vector field structure. Based on feature indices, critical points can be classified into three categories:

[0136] Source point (index=+1): All vectors diverge from this point, usually corresponding to the artery inlet;

[0137] Sink point (index=-1): All vectors converge to this point, usually corresponding to the vein exit;

[0138] Saddle point (index=-1): Some vectors converge and some vectors diverge, usually corresponding to the blood flow shunt region.

[0139] Critical point detection employs a phase map method, identifying critical points by calculating the phase changes around each point in the vector field. The specific steps are as follows: First, calculate the phase angle of the vector field. Then calculate the phase change around each point; finally, if the phase change within a small region (usually 3×3 pixels) is equal to ±2π, then the center of that region is the critical point.

[0140] After identifying the critical points, their type is determined based on the arrangement of surrounding vectors, and a characteristic index is calculated. Then, by tracing the streamlines of vectors originating from or converging at the source point, the connection relationships between the critical points are established, forming a topological skeleton diagram. This skeleton diagram reflects the overall structure of heat conduction in the arteriovenous fistula and corresponds highly to the vascular network.

[0141] The dynamic monitoring subunit 21 is communicatively connected to the critical point identification subunit 21. It is used to design a temporal topology feature tracking algorithm, monitor topology evolution, calculate the topology feature change rate index, and establish an anomaly detection mechanism based on the topology feature change rate index. Changes in the functional state of the arteriovenous fistula can lead to corresponding changes in the thermal flow topology. For example, vascular stenosis may cause the critical point to shift or a new critical point to appear; thrombosis may cause the critical point to disappear or its type to change.

[0142] This invention calculates the Topological Feature Change Rate (TCRI) index by tracking topological feature changes at multiple time points:

[0143] ,

[0144] Among them, TCRI is the topological feature change rate index, with the unit being 1 / hour, which represents the relative rate of change of topological features; It represents the number of features, indicating the total number of topological features considered. This indicates summing over all features; Indicates time Time The values ​​of each topological feature include the number, location, and type of critical points; Indicates time The value of the same feature at the same time; It is a time interval, usually set to 24 hours, with the unit being hours; These are feature weights, dimensionless, representing the first... The importance of each feature is set according to its clinical relevance; for example, the weight of the location of the critical point is 0.4, the weight of the type of critical point is 0.3, and the weight of the number of critical points is 0.3. This indicates taking the absolute value.

[0145] The TCRI value represents the rate of change in arteriovenous fistula topology; a higher value indicates more drastic changes in fistula function. Based on clinical validation, this invention sets the following TCRI thresholds for early warning of fistula function: TCRI < 0.05 / hour indicates a stable state; 0.05 / hour ≤ TCRI < 0.1 / hour indicates a need for attention; 0.1 / hour ≤ TCRI < 0.2 / hour indicates an early warning; and TCRI ≥ 0.2 / hour indicates an emergency warning. These thresholds are derived from the analysis of 80 clinical follow-up cases and can be adjusted according to actual circumstances.

[0146] In a preferred embodiment of the invention, the type of arteriovenous fistula complications can also be predicted by analyzing TCRI change patterns. For example, a gradual approach to the critical point accompanied by a sustained increase in TCRI usually indicates vascular stenosis; a sudden disappearance of the critical point accompanied by a sharp increase in TCRI usually indicates thrombosis; changes in the type of critical point accompanied by fluctuations in TCRI usually indicate arteriovenous fistula infection. This predictive ability enables healthcare professionals to take targeted intervention measures.

[0147] The dynamic image analysis module 22 is connected to the thermal image analysis module 21 to calculate the outer diameter, area, blood flow, perfusion index and blood flow ratio of the fistula at different times, as well as the arteriovenous temperature gradient ratio and the arteriovenous thermal conduction index.

[0148] In an embodiment of the present invention, the dynamic image analysis module 22 calculates the outer diameter and area of ​​blood vessels based on an infrared thermal image sequence using edge detection and region growing algorithms. The outer diameter of the blood vessels is measured using a gradient vector flow algorithm with an accuracy of ±0.2 mm; the area of ​​the blood vessels is determined using a region growing method, with the seed point selected as the temperature extreme point.

[0149] The calculation of the arteriovenous temperature gradient ratio and the arteriovenous thermal conductivity index is crucial for dynamic image analysis. This invention establishes the following model by real-time tracking of arteriovenous thermal gradient characteristic indicators:

[0150] Radial artery temperature characteristics: ;in, Represents the thermal peak, measured in °C, indicating the highest temperature point at the radial artery inlet; The trough value, expressed in °C, represents the lowest temperature point in the radial artery segment; Represents the temperature difference between the peak and trough, expressed in °C, indicating the temperature difference between the highest and lowest temperatures.

[0151] Ulnar vein temperature characteristics: ;in, Represents the hot valley, with the unit being °C, indicating the lowest temperature point at the entrance of the ulnar vein; The value represents the peak temperature, expressed in °C, indicating the highest temperature point in the ulnar vein segment. Represents the temperature difference between the highest and lowest temperatures, expressed in °C.

[0152] These characteristic indicators can comprehensively describe the temperature distribution characteristics of arteries and veins. In practical applications, the heat peak H and heat trough P are important parameters for assessing blood flow status, usually corresponding to the temperature characteristics of the arterial inlet and venous outlet, respectively; the trough value K and peak value J reflect the temperature changes along the blood flow path; and the difference values ​​N and M characterize the magnitude of the temperature gradient.

[0153] The dynamic image analysis module 22 also calculates the blood flow ratio, utilizing the principle that the rate of temperature change is proportional to blood flow:

[0154] ,

[0155] in, This indicates arterial blood flow, expressed in ml / min. This indicates the venous blood flow rate, expressed in ml / min. This indicates the rate of change of arterial temperature, expressed in °C / s. This represents the rate of change of venous temperature, measured in °C / s. The entire ratio is dimensionless, indicating the ratio of blood flow at the arterial end to that at the venous end.

[0156] In clinical applications, normally functional arteriovenous fistulas The ratio is typically between 0.9:1 and 1.1:1. A ratio that is too high (>1.3:1) may indicate obstructed blood flow at the venous end; a ratio that is too low (<0.7:1) may indicate insufficient blood flow at the arterial end or the presence of a shunt.

[0157] The arteriovenous thermal conductivity index represents the velocity of heat conduction in blood flow through a temperature gradient:

[0158] Thermal conductivity index

[0159] Where dT represents the temperature gradient in °C; dt represents the time gradient in seconds; and the entire ratio is in °C / s, representing the rate of change of temperature over time.

[0160] The thermal conductivity index reflects the heat diffusion capacity of blood flow and is closely related to vascular patency. The thermal conductivity index of a normal arteriovenous fistula is typically in the range of 0.05-0.15℃ / s. A low index (<0.03℃ / s) may indicate slow blood flow; a high index (>0.2℃ / s) may indicate localized accelerated blood flow due to vascular stenosis.

[0161] These dynamic parameters, combined with differential geometric features, can comprehensively assess the functional status of arteriovenous fistulas, providing multi-dimensional basis for early warning decisions.

[0162] The data integration module 23 is communicatively connected to the thermal image analysis module 21 and the dynamic image analysis module 22, and is used to collect data from the thermal image analysis module and the dynamic image analysis module and transmit it to the server 3. The data integration adopts a feature-level fusion strategy. First, the features from different sources are normalized to unify the scale; then, a comprehensive feature vector is generated through weighted fusion; finally, the feature vector and the original feature data are transmitted to the server.

[0163] Server 3 receives data from data integration module 23 via data receiving module 31. Real-time data storage module stores real-time analysis data, historical data storage module archives historical data periodically, image analysis module performs further analysis, and issues early warnings via alarm module when an anomaly is detected.

[0164] The alarm module includes an anomaly detection unit 1, a risk grading unit 2, an alarm generation unit 3, and a push unit 4. The anomaly detection unit 1 receives the analysis results from the image analysis module and determines the abnormal state of the arteriovenous fistula (AVF) function based on preset thresholds. The risk grading unit 2 classifies the AVF abnormality into risk levels based on the rate of change of topological features (TCRI). For example, TCRI < 0.05 / hour is normal, 0.05 / hour ≤ TCRI < 0.1 / hour is low risk, 0.1 / hour ≤ TCRI < 0.2 / hour is medium risk, and TCRI ≥ 0.2 / hour is high risk. The alarm generation unit 3 generates alarm information of corresponding levels based on the risk level, including the abnormality type, risk level, and recommended measures. The push unit 4 pushes the alarm information to mobile devices 4 and the cloud analysis platform 5, ensuring that medical staff can receive early warning information in a timely manner.

[0165] Mobile device 4 receives images and analysis results pushed by server 3 via the network, enabling real-time monitoring and historical query functions. In a preferred embodiment of the invention, mobile device 4 uses the secure WebSocket protocol to receive real-time data updates and queries historical data via REST API. The interface of mobile device 4 includes a fistula thermal image display area, a functional evaluation result area, a warning information area, and an operation suggestion area, supporting multiple interaction methods such as zooming, scrolling, and marking.

[0166] The cloud-based analytics platform 5 comprises a data receiving unit, an expert consultation unit, a knowledge base unit, and a report generation unit. The data receiving unit receives fistula thermal image data, differential geometric feature data, and analysis results transmitted from server 3. The expert consultation unit provides remote data access and advanced analysis tools, supporting remote expert evaluation and guidance. The knowledge base unit stores typical fistula cases and differential geometric feature patterns, continuously updating and optimizing the analysis model. The report generation unit generates fistula functional assessment reports and early warning analysis reports to support medical decision-making.

[0167] In one embodiment of the present invention, the cloud analytics platform 5 adopts a layered architecture, including a data layer, a service layer, and an application layer. The data layer is responsible for data storage and management, using a distributed file system to store large image data and a relational database to store structured data; the service layer provides data processing, analysis, and security services; and the application layer provides the user interface and business functions. The cloud deployment adopts a microservice architecture to ensure the system's flexibility and scalability.

[0168] See Figure 7 The workflow of this invention includes the following steps:

[0169] (1) Collect infrared thermal image data of the patient's arteriovenous fistula region;

[0170] (2) Map infrared thermal image data into parameterized temperature manifolds, calculate the metric tensor and curvature features of the temperature manifolds, and analyze the correspondence between curvature features and arteriovenous fistula structures;

[0171] (3) Calculate the thermal conduction geodesic path on the temperature manifold, calculate the deviation index between the geodesic path and the geodesic of the standard fistula template, and evaluate the functional status of the fistula based on the deviation index;

[0172] (4) Construct a heat flux vector field, identify the critical point and topological characteristics of the heat flux vector field, and calculate the time-series topological characteristic change rate index;

[0173] (5) Calculate the outer diameter, area, blood flow, perfusion index and blood flow ratio of the fistula at different times, as well as the arteriovenous temperature gradient ratio and the arteriovenous heat conduction index;

[0174] (6) Collect differential geometric feature data and dynamic image analysis data to comprehensively evaluate the functional status of the arteriovenous fistula;

[0175] (7) Achieve early warning of arteriovenous fistula function based on the topological feature change rate index and comprehensive evaluation results;

[0176] (8) Display images and analysis results in real time on mobile devices and support viewing historical analysis results;

[0177] (9) Data is recorded and analyzed in real time through a cloud-based analysis platform to support remote expert consultation.

[0178] The method of this invention enables accurate assessment and early warning of arteriovenous fistula (AVF) function. Taking a dialysis patient as an example, through three consecutive months of AVF monitoring, the system detected a rise in the Total Topological Characteristic Change Rate (TCRI) from 0.03 / hour to 0.12 / hour 10 days before routine ultrasound examination revealed any significant abnormalities, issuing an early warning. Further analysis showed that the critical point position had shifted and the Geodesic Deviation Index (GDI) had increased from 0.14 to 0.25, indicating potential early stenosis of the AVF. Based on this, the physician scheduled interventional treatment, successfully preventing complete occlusion of the AVF, avoiding emergency surgery and temporary catheter placement, and ensuring the patient's dialysis quality.

[0179] The present invention provides an intelligent analysis and early warning system and method for thermal images of arteriovenous fistulas in hemodialysis. By introducing differential geometry theory to perform in-depth analysis of fistula thermal images and combining dynamic image features, it achieves accurate assessment and early warning of fistula function. The system integrates functions such as data acquisition, processing, analysis, storage, early warning, and remote consultation, forming a complete closed loop for fistula management and providing an effective technical means for the prevention and treatment of fistula complications.

[0180] Clinical applications have shown that this invention can predict fistula function deterioration 7-10 days earlier than traditional methods, improving the early warning accuracy by approximately 35%, increasing the puncture success rate by 15%, significantly reducing the incidence of fistula complications, extending the lifespan of the fistula, and improving patients' quality of life. Furthermore, the remote expert consultation function of the cloud-based analysis platform effectively addresses the problem of uneven distribution of high-quality medical resources, enabling patients in primary hospitals to access expert-level medical services.

[0181] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A smart analysis and early warning system for thermal images of arteriovenous fistulas in hemodialysis, characterized in that, include: Infrared thermal imaging equipment is used to acquire infrared thermal image data of the patient's arteriovenous fistula area; An image processing device, communicatively connected to the infrared thermal imaging device, includes a thermal image analysis module, a dynamic image analysis module, and a data integration module; The server, which is communicatively connected to the image processing device, includes a data receiving module, a real-time data storage module, a historical data storage module, an alarm module, and an image analysis module. Mobile devices, connected to the server via a network, are used to display images and analysis results in real time, as well as view historical analysis results; A cloud-based analytics platform, which communicates with the server, is used for real-time recording and analysis of data in the cloud, as well as remote access to data; The thermal image analysis module includes: A temperature manifold construction unit is used to map the thermal image of the arteriovenous fistula into a parameterized temperature manifold, calculate the metric tensor and curvature features of the temperature manifold, and analyze the correspondence between the curvature features and the arteriovenous fistula vascular structure. The geodesic analysis unit is communicatively connected to the temperature manifold construction unit and is used to calculate the heat conduction geodesic path on the temperature manifold, calculate the deviation index between the geodesic path and the geodesic of the standard arteriovenous fistula template, and evaluate the functional status of the arteriovenous fistula based on the deviation index. The vector field topology analysis unit is communicatively connected to the temperature manifold construction unit and the geodesic analysis unit. It is used to construct a heat flow vector field, identify the critical point and topological features of the heat flow vector field, calculate the time-series topological feature change rate index, and realize early warning of arteriovenous fistula function based on the topological feature change rate index. The dynamic image analysis module is communicatively connected to the thermal image analysis module and is used to calculate the outer diameter, area, blood flow, perfusion index, and blood flow ratio of the fistula at different times, as well as the arteriovenous temperature gradient ratio and the arteriovenous heat conduction index. The data integration module is communicatively connected to the thermal image analysis module and the dynamic image analysis module, and is used to collect data from the thermal image analysis module and the dynamic image analysis module and transmit it to the server.

2. The system according to claim 1, characterized in that, The temperature manifold building unit includes: The preprocessing subunit is used to perform noise reduction, contrast enhancement, and spatial filtering on the infrared thermal image data. The manifold mapping subunit is communicatively connected to the preprocessing subunit and is used to establish a coordinate mapping function to map the two-dimensional image space to a three-dimensional temperature manifold, construct a temperature height function, and form a parameterized surface representation. The metric tensor computation subunit is communicatively connected to the manifold mapping subunit and is used to construct the local metric tensor of the temperature manifold, calculate the eigenvalues ​​and eigenvectors of the metric tensor, and determine the main direction of temperature change. The curvature analysis subunit is communicatively connected to the metric tensor calculation subunit. It is used to calculate the Gaussian curvature and average curvature of the temperature manifold, construct a curvature distribution map, identify areas of abnormal temperature changes, and establish the morphological similarity between the curvature distribution morphology and the standard arteriovenous fistula template.

3. The system according to claim 1, characterized in that, The geodesic analysis unit includes: The endpoint identification subunit is used to define a distance function based on the temperature manifold metric, construct a geodesic distance field, and identify significant points on the temperature manifold as geodesic endpoints. The path solving subunit is communicatively connected to the endpoint identification subunit and is used to calculate the geodesic path using an iterative optimization strategy with an adaptive step size, realize local geodesic segment calculation, and construct the geodesic backbone network. The feature extraction subunit is communicatively connected to the path solving subunit and is used to extract geodesic length features, geodesic curvature integral, geodesic direction change rate, and geodesic width distribution. The functional scoring subunit is communicatively connected to the feature extraction subunit. It is used to construct the mapping relationship between the geodesic features and the functional status of the arteriovenous fistula, calculate the functional score based on the geodesic features, set the functional score threshold, and classify the arteriovenous fistula status level.

4. The system according to claim 1, characterized in that, The vector field topology analysis unit includes: The vector field construction sub-unit is used to calculate the temperature gradient vector field based on the temperature manifold, and to construct the heat flow vector field by applying the law of heat conduction, thereby achieving vector field smoothing and normalization. The vector field feature extraction subunit is communicatively connected to the vector field construction subunit. It is used to calculate the divergence distribution, curl characteristics and Laplace features of the heat flux vector field, and to perform region segmentation and identify functional regions based on the vector field features. The critical point identification subunit is communicatively connected to the vector field feature extraction subunit. It is used to detect the singularity position of the heat flux vector field, classify the critical point type, calculate the critical point feature index, construct the critical point connection relationship, and form a topological skeleton diagram. The dynamic monitoring subunit is communicatively connected to the critical point identification subunit. It is used to design a time-series topology feature tracking algorithm, monitor the evolution of the topology structure, calculate the topology feature change rate index, and establish an anomaly detection mechanism based on the topology feature change rate index.

5. The system according to claim 1, characterized in that, The dynamic image analysis module calculates the arteriovenous temperature gradient ratio and arteriovenous thermal conduction index by tracking arteriovenous thermal gradient characteristic indicators in real time, including: Radial artery temperature characteristics: ;in, Represents the heat peak, Represents the trough value, Represents the difference between peak and trough temperatures; Ulnar vein temperature characteristics: ;in, Representing the hot valley, Represents the peak value. Represents the difference between the peak and trough temperatures; Dynamic flow parameters include: rate of change of arterial and venous temperature, heat conduction velocity, arterial and venous temperature gradient ratio, and arterial and venous heat conduction index.

6. The system according to claim 1, characterized in that, The dynamic image analysis module calculates the blood flow ratio by showing that the rate of temperature change is proportional to the blood flow rate, i.e.: ， in, Indicates blood flow at the arterial end. Indicates venous blood flow. Indicates the rate of change of arterial temperature. This indicates the rate of change in venous temperature.

7. The system according to claim 1, characterized in that, The thermal image analysis module calculates the arteriovenous thermal conductivity index, which represents the blood flow thermal conduction velocity through a temperature gradient, i.e.: Thermal conductivity index = dT / dt Where dT represents the temperature gradient and dt represents the time gradient.

8. The system according to claim 1, characterized in that, The alarm module of the server includes: An anomaly detection unit is used to receive the analysis results from the image analysis module and determine the abnormal state of the arteriovenous fistula function based on a preset threshold. The risk grading unit is communicatively connected to the anomaly detection unit and is used to classify the risk level of arteriovenous fistula anomalies based on the topological feature change rate index. An alarm generation unit, which is communicatively connected to the risk classification unit, is used to generate alarm information of the corresponding level according to the risk level. The push unit is communicatively connected to the alarm generation unit and is used to push the alarm information to the mobile device and the cloud analysis platform.

9. The system according to claim 1, characterized in that, The cloud-based analytics platform includes: The data receiving unit is used to receive the fistula thermal image data, differential geometric feature data and analysis results transmitted by the server; The expert consultation unit is communicatively connected to the data receiving unit and is used to provide remote data access and advanced analysis tools to support remote expert evaluation and guidance. The knowledge base unit, which is connected to the expert consultation unit, is used to store typical arteriovenous fistula cases and differential geometric feature patterns, and to continuously update and optimize the analysis model. The report generation unit, which is communicatively connected to the knowledge base unit, is used to generate fistula function assessment reports and early warning analysis reports to support medical decision-making.

10. A method for intelligent analysis and early warning of thermal images of arteriovenous fistulas in hemodialysis, using the system described in any one of claims 1-9, characterized in that, Includes the following steps: Infrared thermal image data of the patient's arteriovenous fistula region were acquired; The infrared thermal image data is mapped to a parameterized temperature manifold, the metric tensor and curvature features of the temperature manifold are calculated, and the correspondence between the curvature features and the arteriovenous fistula structure is analyzed. Calculate the thermal conduction geodesic path on the temperature manifold, calculate the deviation index between the geodesic path and the geodesic of the standard arteriovenous fistula template, and evaluate the functional status of the arteriovenous fistula based on the deviation index; Construct a heat flux vector field, identify the critical points and topological features of the heat flux vector field, and calculate the time-series topological feature change rate index; Calculate the outer diameter, area, blood flow, perfusion index, and blood flow ratio of the fistula at different times, as well as the arteriovenous temperature gradient ratio and the arteriovenous thermal conductivity index. By combining differential geometric feature data and dynamic image analysis data, the functional status of the arteriovenous fistula is comprehensively evaluated. Based on the aforementioned topological feature change rate index and comprehensive evaluation results, early warning of arteriovenous fistula function is achieved. Displays images and analysis results in real time on mobile devices, and supports viewing historical analysis results; Data is recorded and analyzed in real time through a cloud-based analytics platform, supporting remote expert consultations.

Images