An auxiliary evaluation system for risk of deep vein thrombosis of severe patients in plateau region

Abstract

The present application belongs to the technical field of intelligent medical treatment, and discloses a highland region severe patient deep vein thrombosis risk auxiliary evaluation system; comprising: a data acquisition module for acquiring patient basic data, numerical characteristic data and image characteristic data; an index evaluation module for grouping and dividing the numerical characteristic data and evaluating real-time indexes; an index detection module for acquiring historical indexes, identifying abnormal indexes in the real-time indexes based on the historical indexes and the real-time indexes; an image analysis module for analyzing the image characteristic data, extracting blood vessel characteristic data and quantifying a morphological abnormality index; a risk evaluation module for fusing the abnormal indexes and the morphological abnormality index, evaluating a thrombosis risk coefficient and acquiring a risk degree; the present application can effectively improve the evaluation accuracy of the deep vein thrombosis risk, timely identify high-risk patients and reduce the incidence of deep vein thrombosis and related complications.

Description

Technical Field

[0001] This invention relates to the field of smart medical technology, and more specifically, to an auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas. Background Technology

[0002] The unique low-oxygen and low-temperature environment of high-altitude areas significantly impacts the physiological functions of critically ill patients, particularly increasing the risk of deep vein thrombosis (DVT). DVT refers to the formation of thrombi in deep veins, commonly found in the lower extremities, which can cause blood flow obstruction, pain, and swelling, and in severe cases, lead to fatal complications such as pulmonary embolism. Studies have shown that high-altitude exposure is a significant risk factor for thrombotic diseases; the thrombotic tendency in people at altitudes above 2000 meters is more than twice that of people at lower altitudes, while the risk can increase to 30 times in people at altitudes above 3000 meters. The pathological mechanism of this increased risk is related to... High-altitude environments induce coagulation abnormalities: hypoxia and hypothermia activate hypoxia-inducible factors, upregulate transferrin expression, and thus enhance the activity of thrombin and coagulation factors, leading to a hypercoagulable state. In addition, critically ill patients in high-altitude areas often experience further exacerbations of blood flow stagnation and vascular endothelial damage due to factors such as trauma, surgery, infection, or prolonged bed rest, forming a cumulative effect of the Virchow triad (blood stasis, hypercoagulable state, and vascular injury), which significantly increases the risk of deep vein thrombosis. Therefore, developing an intelligent deep vein thrombosis risk assessment system suitable for critically ill patients in high-altitude areas has become an urgent problem to be solved.

[0003] Patent application CN116264116A discloses a modeling method for predicting the risk of deep vein thrombosis (DVT) in patients after lower extremity fracture surgery. The method includes: S1, establishing a database and collecting data; S2, processing the collected clinical characteristic data to enable its use in a patterned manner; S3, establishing a decision model based on the processed data, enabling the model to predict the incidence of DVT in patients after lower extremity fracture surgery; S4, optimizing the model and parameters; S5, selecting preferred options from multiple models and parameters; and S6, integrating multiple models and using deep neural network and recurrent neural network algorithms to finally complete the model establishment. This invention can significantly improve assessment accuracy and reduce bias errors.

[0004] However, while the aforementioned technologies can assess the risk of deep vein thrombosis, they primarily target patients after lower extremity fracture surgery. They do not consider the unique impact of environmental factors such as hypoxia and low air pressure in high-altitude areas on thrombosis, nor the physiological differences between critically ill and ordinary patients in high-altitude areas. Therefore, these technologies cannot comprehensively identify the specific factors contributing to thrombosis risk in high-altitude areas, leading to decreased accuracy in risk assessment results. Furthermore, they make it difficult to identify high-risk patients in a timely manner, causing them to miss the optimal time for prevention and treatment, thus increasing the incidence of deep vein thrombosis and related complications.

[0005] In view of this, the present invention proposes an auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas to solve the above problems. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas, comprising:

[0007] The data acquisition module is used to collect patient characteristic data in real time, which includes basic patient data, numerical characteristic data, and image characteristic data.

[0008] The indicator evaluation module is used to divide the numerical feature data into groups using a clustering algorithm, evaluate the corresponding physiological indicators based on the data within each group, and mark them as real-time indicators.

[0009] The indicator detection module is used to obtain historical indicators based on the patient's basic data, calculate the coefficient of variation for each real-time indicator based on the historical and real-time indicators, analyze the coefficient of variation, and identify abnormal indicators in the real-time indicators.

[0010] The image analysis module is used to analyze image feature data, extract vascular feature data, and quantify the morphological abnormality index based on the vascular feature data.

[0011] The risk assessment module is used to integrate abnormal indicators and morphological abnormality indices to assess the thrombosis risk coefficient and obtain the risk level corresponding to the thrombosis risk coefficient based on pre-constructed risk classification criteria.

[0012] Furthermore, the steps for grouping numerical feature data include:

[0013] Step S101: Using a pre-trained word embedding model, each data point in the numerical feature data is converted into a corresponding vector and labeled as a physiological vector;

[0014] Step S102: Each physiological vector is used as a point to be divided, and each point to be divided corresponds one-to-one with a physiological vector;

[0015] Step S103: Preset the number of groups a, randomly select a points to be divided as center points, and mark each center point as α. b b∈[1,a]; mark the sample points that are not used as center points as partition points, and label each partition point as β. c c∈[1,Aa], where A is the number of physiological vectors;

[0016] Step S104: Establish a corresponding a groups based on a center points, and calculate the cosine similarity between each dividing point and each center point in turn, and mark it as similarity;

[0017] Step S105: Divide the point β c Compare all corresponding similarities, mark the center point corresponding to the highest similarity as the optimal point, and set the split point β. c Divide into groups corresponding to the optimal points;

[0018] Step S106: Let c = c + 1;

[0019] Step S107: Repeat steps S105 to S106 until c = Aa, then proceed to step S108.

[0020] Step S108: Recalculate the center point corresponding to each group;

[0021] Step S109: Repeat steps S104 to S108 until the center point of each group recalculated in step S108 is consistent with the corresponding center point calculated in the previous cycle. The cycle ends when a groups and their corresponding division points are obtained. The groups correspond one-to-one with the physiological indicators.

[0022] In step S108, the method for recalculating the center point corresponding to each group includes:

[0023] Count the number of points to be divided in each group and mark them as the number of punctuation marks; add up the center points corresponding to each group in turn, and then divide by the corresponding number of punctuation marks to obtain the new center point of each group.

[0024] Furthermore, methods for assessing physiological indicators include:

[0025] Each group's data is treated as a dataset, with each dataset corresponding to a group. Each dataset is then input into its corresponding evaluation model to assess the corresponding physiological indicator. The evaluation model includes 'a' evaluation models, each corresponding to a physiological indicator. All 'a' evaluation models are deep neural network models, and the training process for all 'a' evaluation models is identical.

[0026] The training process for evaluating the model includes:

[0027] R sets of data are pre-collected, each corresponding to a physiological indicator. A corresponding physiological indicator is set for each of the r sets of data, where r is an integer greater than 1. The data sets and their corresponding physiological indicators are converted into a set of feature vectors. Each set of feature vectors is used as input to an evaluation model. The evaluation model outputs a set of predicted physiological indicators corresponding to each set of data, and uses the actual physiological indicators corresponding to each set of data as the prediction target. The actual physiological indicators are the pre-set physiological indicators corresponding to the data sets. The training objective is to minimize the sum of prediction errors for all data sets. The evaluation model is trained until the sum of prediction errors converges, at which point training stops.

[0028] Furthermore, the historical indicators are physiological indicators of the same patient under normal physiological conditions assessed at a historical time; the same patient refers to a patient whose basic data is the same as the patient's basic data collected in real time.

[0029] The steps for calculating the discrete coefficients for each real-time indicator include:

[0030] Step S201: Set numerical labels for each physiological indicator in ascending order and mark them as indicator labels. The range of indicator labels is [1, a].

[0031] Step S202: Select the physiological indicator with the indicator label d and mark it as an element, d∈[1,a];

[0032] Step S203: Select physiological indicators corresponding to the elements from historical indicators and mark them as detection elements. Mark the real-time indicators corresponding to the elements as real-time elements. Use both detection elements and real-time elements as analysis elements.

[0033] Step S204: Calculate the element difference and baseline distance for each analytical element;

[0034] Step S205: Determine the baseline element corresponding to each analysis element;

[0035] Step S206: Calculate the adjacent distance between each analytical element and its corresponding reference element based on the element difference and the reference distance;

[0036] Step S207: Calculate the local density corresponding to each analysis element based on the adjacent distance;

[0037] Step S208: Calculate the discrete coefficients corresponding to the real-time elements based on the local density;

[0038] Step S209: Let d = d + 1;

[0039] Step S210: Loop through steps S202 to S209 until all physiological indicators are marked as elements, and then the loop ends;

[0040] The method for identifying abnormal indicators in real-time indicators includes:

[0041] Preset an identification coefficient h, where 0 < h < 1; subtract 1 from the coefficient of variation of each real-time element, then take the absolute value to obtain the judgment coefficient of each real-time element; compare the judgment coefficient of each real-time element with the identification coefficient; if the judgment coefficient is greater than or equal to the identification coefficient, mark the real-time indicator corresponding to the corresponding real-time element as an abnormal indicator; if the judgment coefficient is less than the identification coefficient, do not mark the real-time indicator corresponding to the corresponding real-time element.

[0042] Furthermore, in the step S204, the method for calculating the element difference corresponding to each analysis element is as follows: Obtain the adjacent elements corresponding to each analysis element, where the adjacent elements are the remaining f analysis elements, and f is the number of detection elements; subtract each analysis element from its corresponding adjacent element to obtain the element difference corresponding to each analysis element;

[0043] The method for calculating the reference distance corresponding to each analysis element is as follows:

[0044] Take the element difference corresponding to each analysis element as an element set, and the element set corresponds to the analysis element one by one; sort the element differences in each element set from small to large to generate an element sorting table corresponding to each element set; preset a screening quantity g, and take the element difference ranked at the gth position in each element sorting table as the reference distance of the analysis element corresponding to the corresponding element set;

[0045] In the step S205, the method for determining the reference element corresponding to each analysis element is as follows: Take the adjacent elements corresponding to the first g element differences in each element sorting table as the reference element of the analysis element corresponding to the corresponding element set;

[0046] In the step S206, the expression for the adjacent distance is: xl(p,q) = max(jl(p), yc(p,q)); where xl(p,q) is the adjacent distance between the pth analysis element and the corresponding qth reference element, max is the maximum value function, jl(p) is the reference distance of the pth analysis element, yc(p,q) is the element difference between the pth analysis element and the corresponding qth reference element, p ∈ [1, f + 1], q ∈ [1, g], and f > g;

[0047] In the step S207, the method for calculating the local density corresponding to each analysis element is as follows: successively add the adjacent distances corresponding to each analysis element to obtain the comprehensive distance corresponding to each analysis element; take the reciprocal of the comprehensive distance corresponding to each analysis element as the local density corresponding to each analysis element.

[0048] In the step S208, the method for calculating the dispersion coefficient corresponding to the real-time element is as follows: obtain the reference element corresponding to the real-time element and mark it as the evaluation element; divide the local density corresponding to each evaluation element by the local density corresponding to the real-time element respectively to obtain the relative density corresponding to each evaluation element; successively add the relative densities corresponding to each evaluation element and then divide by g to obtain the dispersion coefficient corresponding to the real-time element.

[0049] Furthermore, the vascular feature data includes a curvature set and a diameter set.

[0050] The method for extracting the curvature set includes:

[0051] Preset a gray threshold, obtain the gray value corresponding to each pixel point in the image feature data, and compare the gray value of each pixel point with the gray threshold respectively; if the gray value is greater than the gray threshold, mark the corresponding pixel point as a vascular point; if the gray value is less than or equal to the gray threshold, do not mark the corresponding pixel point; based on all vascular points, use a skeletonization algorithm to extract the vascular centerline; randomly select k vascular points from all the vascular points corresponding to the vascular centerline and mark them as midline points, where 1 < k < l and l is the number of vascular points on the vascular centerline.

[0052] According to the image coordinate system built in the image feature data, obtain the coordinates corresponding to the k midline points and mark them as midline coordinates; according to the midline coordinates of the k midline points, use a curve fitting algorithm to perform curve fitting on the centerline to obtain the mathematical expression corresponding to the centerline and mark it as the curve expression; perform a first-order derivative on the curve expression to obtain the first-order expression; perform a second-order derivative on the curve expression to obtain the second-order expression; substitute each midline coordinate into the first-order expression to obtain the first-order value corresponding to each midline point; substitute each midline coordinate into the second-order expression to obtain the second-order value corresponding to each midline point; square each first-order value and then add 1 to obtain the first value; take the square root of the cube of each first value to obtain the second value corresponding to each midline point; divide the absolute value of the second-order value corresponding to each midline point by the corresponding second value to obtain the vascular curvature corresponding to each midline point; take the vascular curvature corresponding to each midline point as the curvature set.

[0053] Furthermore, the method for extracting the diameter set includes:

[0054] The negative reciprocal of the first-order value corresponding to each midline point is used as the slope of the normal line corresponding to each midline point. Based on the midline coordinates of each midline point and the slope of the corresponding normal line, the normal line equation corresponding to each midline point is expressed in point-slope form. The expression for the normal line equation is: y = z(x - x0) + y0; where y is the ordinate of any point on the normal line, z is the slope, x is the abscissa of any point on the normal line, x0 is the abscissa of the midline point, and y0 is the ordinate of the midline point. The adjacent points corresponding to each vessel point are analyzed; adjacent points are the pixels adjacent to the vessel point. If there are pixels among the adjacent points that are not marked as vessel points, then the corresponding vessel point is marked as an edge. Edge points; if adjacent points are all blood vessel points, then the corresponding blood vessel points are not marked; substitute each edge point into each normal equation, if the normal equation holds, then the corresponding edge point is taken as the point satisfying the corresponding normal equation; obtain the coordinates corresponding to each satisfying point and mark them as satisfying coordinates; calculate the Euclidean distance between the two satisfying points corresponding to each normal equation based on the satisfying coordinates, and take it as the satisfying length; take the satisfying length corresponding to each normal equation as the image diameter of the corresponding midline point; obtain the scaling factor, multiply each image diameter by the scaling factor to obtain the blood vessel diameter corresponding to each midline point; take the blood vessel diameter corresponding to each midline point as the diameter set.

[0055] Furthermore, methods for quantifying morphological anomaly indices include:

[0056] Sum the curvatures of each vessel sequentially, then divide by k to obtain the mean curvature. Subtract the mean curvature from each vessel's curvature and square the result to obtain the squared curvature difference. Sum the squared curvature differences of each vessel sequentially, then divide by k and take the square root to obtain the standard deviation of curvature. Divide the standard deviation of curvature by the mean curvature to obtain the curvature abnormality index. Sum the diameters of each vessel sequentially, then divide by k to obtain the mean diameter. Subtract the mean diameter from each vessel's diameter and square the result to obtain the squared diameter difference. Sum the squared diameter differences of each vessel sequentially, then divide by k and take the square root to obtain the standard deviation of diameter. Divide the standard deviation of diameter by the mean diameter to obtain the diameter abnormality coefficient.

[0057] A preset threshold coefficient is used, which includes a curvature coefficient and a diameter coefficient. The curvature of each blood vessel is compared with the curvature coefficient. If the curvature is greater than or equal to the curvature coefficient, the corresponding blood vessel point is marked as a curvature abnormality point; if the curvature is less than the curvature coefficient, the corresponding blood vessel point is not marked. The diameter of each blood vessel is compared with the diameter coefficient. If the diameter is greater than or equal to the diameter coefficient, the corresponding blood vessel point is marked as a diameter abnormality point; if the diameter is less than the diameter coefficient, the corresponding blood vessel point is not marked. Both curvature abnormalities and diameter abnormalities are considered local abnormalities. The number of local abnormalities is counted and marked as the local abnormality number. The local abnormality coefficient is obtained by dividing the local abnormality number by 2k.

[0058] A preset weight set is used, which includes the weight coefficients corresponding to the curvature anomaly index, diameter anomaly index, and local anomaly coefficient. The curvature anomaly index, diameter anomaly index, and local anomaly coefficient are multiplied by their respective weight coefficients and then added together to obtain the morphological anomaly index.

[0059] Furthermore, the method for assessing the thrombosis risk factor includes:

[0060] Environmental factors, including ambient temperature, atmospheric pressure, and oxygen concentration, are acquired. These environmental factors, abnormal indicators, and morphological abnormality indices are used as assessment data. The assessment data is then input into a trained risk assessment model to evaluate the patient's thrombosis risk coefficient. The training process of the risk assessment model is consistent with that of the assessment model, and both are deep neural network models.

[0061] Furthermore, the method for obtaining the risk level corresponding to the thrombosis risk coefficient includes:

[0062] Based on environmental factors, corresponding severity assessment standards are selected from the pre-constructed risk classification criteria and marked as real-time assessment standards. Among them, the risk classification criteria include u different severity assessment standards, where u is an integer greater than 1. Each severity assessment standard includes a coefficient range corresponding to different risk levels. The thrombosis risk coefficient is compared with each coefficient range in the real-time assessment standards to obtain the risk level corresponding to the coefficient range in which the thrombosis risk coefficient is located.

[0063] The steps for selecting real-time evaluation criteria include:

[0064] Step S301: Construct a corresponding fuzzy set for each data point in the environmental factors, with each fuzzy set including multiple fuzzy categories;

[0065] Step S302: Map each data point in the environmental factors to the membership degree of the corresponding fuzzy category using fuzzification technology;

[0066] Step S303: Define fuzzy rules;

[0067] Step S304: Match the fuzzified environmental factors with fuzzy rules, perform fuzzy inference, and obtain the fuzzy inference results. The fuzzy inference results are the membership degrees corresponding to each level of evaluation criteria.

[0068] Step S305: Compare each membership degree in the fuzzy inference results and select the evaluation standard with the highest membership degree as the real-time evaluation standard.

[0069] The technical effects and advantages of the present invention: A system for assisting in the risk assessment of deep vein thrombosis in critically ill patients in high-altitude areas.

[0070] By collecting various characteristic data of patients in real time, key information on patients' health status can be comprehensively obtained. Clustering algorithms and deep learning technologies are used to assess physiological indicators in different groups, and anomaly detection is performed on these indicators to identify abnormalities and promptly identify any abnormalities. Image analysis is used to quantify and extract the curvature and diameter features of blood vessels, calculate the morphological abnormality index, and combine it with abnormal indicators and environmental factors to form multi-dimensional information, accurately assessing patients' thrombosis risk coefficients and obtaining specific risk levels based on pre-constructed risk classification criteria. The system fully considers the impact of environmental factors such as hypoxia and low air pressure in high-altitude areas, comprehensively identifying special factors contributing to thrombosis risk in these regions. This improves the accuracy of assessing the risk of deep vein thrombosis in critically ill patients at high altitudes, enabling timely identification of high-risk patients and providing effective early warning and intervention measures for clinical practice, thereby reducing the incidence of deep vein thrombosis and related complications. Attached Figure Description

[0071] Figure 1 This is a schematic diagram of an auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas, according to Embodiment 1 of the present invention.

[0072] Figure 2 This is a flowchart illustrating the functional relationships between the modules in Embodiment 1 of the present invention. Detailed Implementation

[0073] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] Example 1

[0075] Please see Figure 1 As shown in this embodiment, a risk assessment system for deep vein thrombosis in critically ill patients in high-altitude areas includes a data acquisition module, an indicator assessment module, an indicator detection module, an image analysis module, and a risk assessment module. These modules are connected via wired and / or wireless means to achieve data transmission between them. The functional relationships between the modules are as follows: Figure 2 As shown.

[0076] The data acquisition module is used to collect patient characteristic data in real time, which includes basic patient data, numerical characteristic data, and image characteristic data.

[0077] Patient basic data includes personal information and medical history. Personal information includes gender, age, height, and weight. Medical history includes past medical history and family medical history. Past medical history refers to diseases the patient had before seeking medical treatment, such as heart disease and hypertension, which helps to assess the patient's health status and potential risks. Family medical history refers to diseases that the patient's family members (such as parents, siblings, etc.) have had or currently have, including hereditary diseases, chronic diseases, or other health problems, which helps to identify genetic risks, understand the health problems the patient may face, and their susceptibility to diseases. Patient basic data can affect the normal range of physiological indicators, so it is necessary to obtain personalized normal physiological indicators based on the patient's basic data. This helps to provide basic information for subsequent indicator testing and ensure the accuracy of abnormal indicator identification.

[0078] Numerical characteristic data refers to the patient's physiological data, which are objective, physiologically related data that can be quantified numerically, such as blood oxygen saturation, hematocrit, blood viscosity, prothrombin time, and fibrinogen level. Numerical characteristic data helps to assess the patient's physiological indicators, detect the patient's physiological state, detect abnormalities in a timely manner, and realize real-time physiological monitoring and early warning of thrombosis.

[0079] The image feature data is magnetic resonance imaging, which is used to assess the patient's vascular status and provide important visual information for subsequent deep vein thrombosis risk assessment;

[0080] Patient basic data is obtained through the hospital's electronic medical record system.

[0081] The indicator evaluation module is used to divide the numerical feature data into groups using a clustering algorithm, evaluate the corresponding physiological indicators based on the data within each group, and mark them as real-time indicators.

[0082] The steps for grouping numerical feature data include:

[0083] Step S101: Using a pre-trained word embedding model (such as Word2Vec, GloVe, BERT, etc.), each data in the numerical feature data is converted into a corresponding vector and labeled as a physiological vector. It should be understood that when performing vector conversion on the numerical feature data, the name of each data in the numerical feature data is converted, not the specific value corresponding to these data. The word embedding model is an existing technology, and the specific training process will not be described in detail here.

[0084] Step S102: Each physiological vector is used as a point to be divided, and each point to be divided corresponds one-to-one with a physiological vector;

[0085] Step S103: Preset the number of groups a, randomly select a points to be divided as center points, and mark each center point as α. bb∈[1,a]; mark the sample points that are not used as center points as partition points, and label each partition point as β. c c∈[1,Aa], where A is the number of physiological vectors; the number of groups a is preset by those skilled in the art based on the actual number of physiological indicators;

[0086] Step S104: Establish a corresponding a groups based on a center points, and calculate the cosine similarity between each division point and each center point in turn, and mark it as similarity. The calculation method of cosine similarity is existing technology, and will not be elaborated on here.

[0087] Step S105: Divide the point β c Compare all corresponding similarities, mark the center point corresponding to the highest similarity as the optimal point, and set the split point β. c Divide into groups corresponding to the optimal points;

[0088] Step S106: Let c = c + 1;

[0089] Step S107: Repeat steps S105 to S106 until c = Aa, then proceed to step S108.

[0090] Step S108: Recalculate the center point corresponding to each group;

[0091] Step S109: Repeat steps S104 to S108 until the center point of each group recalculated in step S108 is consistent with the corresponding center point calculated in the previous cycle. The cycle ends when the recalculation ends and a groups and their corresponding division points are obtained. The groups correspond one-to-one with physiological indicators. Physiological indicators include hypoxia tolerance, coagulation status, and hemodynamic status. Hypoxia tolerance corresponds to blood oxygen saturation and hematocrit in the numerical feature data. Coagulation status corresponds to prothrombin time and fibrinogen level in the numerical feature data. Hemodynamic status corresponds to blood viscosity in the numerical feature data.

[0092] In step S108 above, the method for recalculating the center point corresponding to each group includes:

[0093] Count the number of points to be divided in each group and mark them as the number of punctuation marks; add up the center points corresponding to each group in turn, and then divide by the corresponding number of punctuation marks to obtain the new center point of each group.

[0094] Methods for assessing physiological indicators include:

[0095] Each group's data is treated as a dataset, with each dataset corresponding to a group. Each dataset is then input into a corresponding indicator evaluation model to evaluate the corresponding physiological indicator. The indicator evaluation model includes 'a' evaluation models, each corresponding to a physiological indicator, meaning one evaluation model is used to evaluate one physiological indicator. All 'a' evaluation models are deep neural network models, and the training process for all 'a' evaluation models is consistent.

[0096] The training process for evaluating the model includes:

[0097] R sets of data are collected in advance, each corresponding to a physiological indicator. A corresponding physiological indicator is set for each of the r sets of data, where r is an integer greater than 1. The data sets and their corresponding physiological indicators are converted into a set of feature vectors. The physiological indicators corresponding to the data sets are collected by those skilled in the art during the historical evaluation of physiological indicators. Each set of data is analyzed in turn based on the actual situation, and the physiological indicators corresponding to each set of data are evaluated. The corresponding physiological indicators are set for each of the r sets of data in turn.

[0098] Each set of feature vectors is used as input to the evaluation model, which outputs a set of predicted physiological indicators corresponding to each dataset and uses the actual physiological indicators corresponding to each dataset as the prediction target. The actual physiological indicators are the pre-set physiological indicators corresponding to the dataset. The training objective is to minimize the sum of prediction errors for all datasets. The prediction error is calculated using the formula η. w =(θ w -ε w ) 2 , where η w Let w be the prediction error, w be the group number of the feature vector corresponding to the dataset, and θ be the prediction error. w Let ε be the predicted physiological index corresponding to the w-th data set. w Let w be the actual physiological indicators corresponding to the w-th data set; train the evaluation model until the sum of prediction errors converges and then stop training.

[0099] The indicator detection module is used to obtain historical indicators based on the patient's basic data, calculate the coefficient of variation for each real-time indicator based on the historical and real-time indicators, analyze the coefficient of variation, and identify abnormal indicators in the real-time indicators.

[0100] Historical indicators are physiological indicators of the same patient under normal physiological conditions at a historical time; the same patient refers to a patient whose basic data is the same as the patient's basic data collected in real time; historical indicators are obtained through the electronic medical record system within the hospital system.

[0101] The steps for calculating the coefficient of variation for each real-time metric include:

[0102] Step S201: Set numerical labels for each physiological indicator in ascending order and mark them as indicator labels. The range of indicator labels is [1, a].

[0103] Step S202: Select the physiological indicator with the indicator label d and mark it as an element, d∈[1,a];

[0104] Step S203: Select physiological indicators corresponding to the elements from historical indicators and mark them as detection elements. Mark the real-time indicators corresponding to the elements as real-time elements. Use both detection elements and real-time elements as analysis elements.

[0105] Step S204: Calculate the element difference and baseline distance for each analytical element;

[0106] Step S205: Determine the baseline element corresponding to each analysis element;

[0107] Step S206: Calculate the adjacent distance between each analytical element and its corresponding reference element based on the element difference and the reference distance;

[0108] Step S207: Calculate the local density corresponding to each analysis element based on the adjacent distance;

[0109] Step S208: Calculate the discrete coefficients corresponding to the real-time elements based on the local density;

[0110] Step S209: Let d = d + 1;

[0111] Step S210: Repeat steps S202 to S209 until all physiological indicators are marked as elements, then the cycle ends.

[0112] In step S204 above, the method for calculating the element difference corresponding to each analysis element is as follows: obtain the neighboring elements corresponding to each analysis element, where the neighboring elements are the other f analysis elements, and f is the number of detected elements; subtract the corresponding neighboring elements from each analysis element to obtain the element difference corresponding to each analysis element.

[0113] The method for calculating the baseline distance for each analysis element is as follows:

[0114] The element difference corresponding to each analytical element is taken as an element set, and the element set corresponds one-to-one with the analytical element; the element differences in each element set are sorted from smallest to largest to generate an element sorting table corresponding to each element set; the screening quantity g is preset, which is set by those skilled in the art according to the actual situation; the element difference ranked at the g position in each element sorting table is taken as the reference distance of the analytical element corresponding to the element set.

[0115] In the above step S205, the method for determining the reference element corresponding to each analysis element is as follows: The adjacent elements corresponding to the differences of the first g elements in each element sorting table are used as the reference elements of the analysis elements corresponding to the corresponding element sets.

[0116] In the above step S206, the expression for the adjacent distance is: xl(p,q) = max(jl(p), yc(p,q)); where xl(p,q) is the adjacent distance between the p-th analysis element and the corresponding q-th reference element, max is the maximum value function, jl(p) is the reference distance of the p-th analysis element, yc(p,q) is the element difference between the p-th analysis element and the corresponding q-th reference element, p ∈ [1, f + 1], q ∈ [1, g], and f > g.

[0117] In the above step S207, the method for calculating the local density corresponding to each analysis element is as follows: The adjacent distances corresponding to each analysis element are added in sequence to obtain the comprehensive distance corresponding to each analysis element; the reciprocal of the comprehensive distance corresponding to each analysis element is used as the local density corresponding to each analysis element.

[0118] In the above step S208, the method for calculating the dispersion coefficient corresponding to the real-time element is as follows: Obtain the reference element corresponding to the real-time element and mark it as the evaluation element; divide the local density corresponding to each evaluation element by the local density corresponding to the real-time element to obtain the relative density corresponding to each evaluation element; add the relative densities corresponding to each evaluation element in sequence and then divide by g to obtain the dispersion coefficient corresponding to the real-time element.

[0119] The method for identifying abnormal indicators in real-time indicators includes:

[0120] Preset an identification coefficient h, 0 < h < 1, and the identification coefficient h is preset by those skilled in the art according to the actual situation; subtract 1 from the dispersion coefficient of each real-time element, and then take the absolute value to obtain the judgment coefficient of each real-time element; compare the judgment coefficient of each real-time element with the identification coefficient; if the judgment coefficient is greater than or equal to the identification coefficient, mark the real-time indicator corresponding to the corresponding real-time element as an abnormal indicator; if the judgment coefficient is less than the identification coefficient, do not mark the real-time indicator corresponding to the corresponding real-time element.

[0121] An image analysis module, which is used to analyze image feature data, extract vascular feature data, and quantify the morphological abnormality index based on the vascular feature data.

[0122] The vascular feature data includes a curvature set and a diameter set;

[0123] The method for extracting the curvature set includes:

[0124] A preset gray threshold value, which is preset by those skilled in the art according to the actual situation; obtain the gray value corresponding to each pixel point in the image feature data, and compare the gray value of each pixel point with the gray threshold value respectively; if the gray value is greater than the gray threshold value, mark the corresponding pixel point as a blood vessel point; if the gray value is less than or equal to the gray threshold value, do not mark the corresponding pixel point; based on all blood vessel points, use a skeletonization algorithm (such as Zhang-Suen thinning algorithm, Guo-Hall thinning algorithm, etc.) to extract the blood vessel center line; randomly select k blood vessel points from all the blood vessel points corresponding to the blood vessel center line and mark them as midline points, where 1 < k < l and l is the number of blood vessel points on the blood vessel center line;

[0125] According to the image coordinate system built in the image feature data, obtain the coordinates corresponding to the k midline points and mark them as midline coordinates; according to the midline coordinates of the k midline points, use a curve fitting algorithm (such as the least squares method, B-spline curve algorithm, etc.) to perform curve fitting on the center line, obtain the mathematical expression corresponding to the center line and mark it as the curve expression; perform a first-order derivative on the curve expression to obtain the first-order expression; perform a second-order derivative on the curve expression to obtain the second-order expression; substitute each midline coordinate into the first-order expression to obtain the first-order value corresponding to each midline point; substitute each midline coordinate into the second-order expression to obtain the second-order value corresponding to each midline point; square each first-order value and then add one to obtain the first value; take the square root of the cube of each first value to obtain the second value corresponding to each midline point; divide the absolute value of the second-order value corresponding to each midline point by the corresponding second value to obtain the blood vessel curvature corresponding to each midline point; use the blood vessel curvature corresponding to each midline point as the curvature set.

[0126] The method for extracting the diameter set includes:

[0127] The negative reciprocal of the first-order value corresponding to each midline point is used as the slope of the normal line corresponding to each midline point. Based on the midline coordinates of each midline point and the slope of the corresponding normal line, the normal line equation corresponding to each midline point is expressed in point-slope form. The expression for the normal line equation is: y = z(x - x0) + y0; where y is the ordinate of any point on the normal line, z is the slope, x is the abscissa of any point on the normal line, x0 is the abscissa of the midline point, and y0 is the ordinate of the midline point. The adjacent points corresponding to each vessel point are analyzed; adjacent points are pixels adjacent to the vessel point. If there are unmarked pixels among the adjacent points, the corresponding vessel point is marked as an edge point; if all adjacent points are vessel points, the corresponding vessel point is not marked. Each edge point... Substitute each edge point into each normal equation. If the normal equation holds, then the corresponding edge point is taken as the point that satisfies the normal equation. Obtain the coordinates corresponding to each satisfying point and mark them as satisfying coordinates. Calculate the Euclidean distance between the two satisfying points corresponding to each normal equation based on the satisfying coordinates, and use this distance as the satisfying length. The method for calculating the Euclidean distance is existing technology and will not be elaborated upon here. Use the satisfying length corresponding to each normal equation as the image diameter of the corresponding midline point. Obtain the scaling factor, multiply each image diameter by the scaling factor to obtain the blood vessel diameter corresponding to each midline point. Use the blood vessel diameter corresponding to each midline point as the diameter set. The scaling factor is obtained based on the image resolution of the image feature data, and the image resolution is obtained through the metadata of the image feature data.

[0128] Methods for quantifying morphological anomaly indices include:

[0129] Sum the curvatures of each vessel sequentially, then divide by k to obtain the mean curvature. Subtract the mean curvature from each vessel's curvature and square the result to obtain the squared curvature difference. Sum the squared curvature differences of each vessel sequentially, then divide by k and take the square root to obtain the standard deviation of curvature. Divide the standard deviation of curvature by the mean curvature to obtain the curvature abnormality index. Sum the diameters of each vessel sequentially, then divide by k to obtain the mean diameter. Subtract the mean diameter from each vessel's diameter and square the result to obtain the squared diameter difference. Sum the squared diameter differences of each vessel sequentially, then divide by k and take the square root to obtain the standard deviation of diameter. Divide the standard deviation of diameter by the mean diameter to obtain the diameter abnormality coefficient.

[0130] A preset threshold coefficient is used, which includes a curvature coefficient and a diameter coefficient. The threshold coefficients are preset by those skilled in the art based on the actual situation. Each vessel curvature is compared with the curvature coefficient. If the vessel curvature is greater than or equal to the curvature coefficient, the corresponding vessel point is marked as a curvature abnormality point. If the vessel curvature is less than the curvature coefficient, the corresponding vessel point is not marked. Each vessel diameter is compared with the diameter coefficient. If the vessel diameter is greater than or equal to the diameter coefficient, the corresponding vessel point is marked as a diameter abnormality point. If the vessel diameter is less than the diameter coefficient, the corresponding vessel point is not marked. Both curvature abnormality points and diameter abnormality points are regarded as local abnormality points. The number of local abnormality points is counted and marked as the number of local abnormality points. The number of local abnormality points is divided by 2k to obtain the local abnormality coefficient.

[0131] A preset weight set is provided, which includes weight coefficients corresponding to the curvature anomaly index, diameter anomaly index, and local anomaly coefficient. The weight set is preset by those skilled in the art based on the actual situation. The curvature anomaly index, diameter anomaly index, and local anomaly coefficient are multiplied by their corresponding weight coefficients and then added together to obtain the morphological anomaly index.

[0132] The risk assessment module is used to integrate abnormal indicators and morphological abnormality indices to assess the thrombosis risk coefficient and obtain the risk level corresponding to the thrombosis risk coefficient based on pre-constructed risk classification criteria.

[0133] Methods for assessing the risk of thrombosis include:

[0134] Environmental factors, including ambient temperature, air pressure, and oxygen concentration, were collected. Ambient temperature was obtained using temperature sensors installed in the intensive care unit (ICU), as were air pressure and oxygen concentration sensors. The reason for collecting these environmental factors is that the lower air pressure at high altitudes leads to thinner air and reduced oxygen partial pressure. To adapt to this hypoxic environment, patients may experience increased blood viscosity and impaired venous return, thus increasing the risk of deep vein thrombosis (DVT). Furthermore, the thin air at high altitudes reduces blood oxygen concentration, which affects blood circulation, including vasoconstriction and increased clotting factors, further increasing the risk of DVT. Additionally, the typically lower temperatures at high altitudes cause vasoconstriction, affecting blood circulation, particularly slowing blood flow in the extremities, exacerbating the risk of DVT. Therefore, collecting these environmental factors allows for a comprehensive consideration of the unique impact of hypoxia and low air pressure at high altitudes on thrombosis, contributing to a more accurate assessment of the risk of DVT in critically ill patients at high altitudes.

[0135] Environmental factors, abnormal indicators, and morphological abnormality indices are used as assessment data. The assessment data is input into a trained risk assessment model to evaluate the patient's thrombosis risk coefficient. The training process of the risk assessment model is consistent with that of the assessment model, and both are deep neural network models.

[0136] Methods for obtaining the risk level corresponding to the thrombosis risk coefficient include:

[0137] Based on environmental factors, corresponding severity assessment standards are selected from pre-constructed risk classification criteria and marked as real-time assessment standards. The risk classification criteria include u different severity assessment standards, where u is an integer greater than 1. Each severity assessment standard includes a coefficient range corresponding to different risk levels, such as the coefficient range for high risk, medium risk, and low risk. The risk classification criteria are pre-constructed by those skilled in the art based on actual conditions. The thrombosis risk coefficient is compared with each coefficient range in the real-time assessment standards to obtain the risk level corresponding to the coefficient range in which the thrombosis risk coefficient falls.

[0138] The steps for selecting real-time evaluation criteria include:

[0139] Step S301: Construct a corresponding fuzzy set for each data point in the environmental factors. Each fuzzy set includes multiple fuzzy categories. For example, the fuzzy categories corresponding to the ambient temperature are low temperature, medium temperature, and high temperature; the fuzzy categories corresponding to the ambient air pressure are low air pressure, medium air pressure, and high air pressure; and the fuzzy categories corresponding to the oxygen concentration are low concentration, medium concentration, and high concentration.

[0140] Step S302: Map each data point in the environmental factors to the membership degree of the corresponding fuzzy category using fuzzification techniques. Fuzzification is the process of mapping precise numerical values ​​to the membership degree of a fuzzy category. Fuzzification techniques include triangular membership functions, trapezoidal membership functions, etc. For example, if the environmental air pressure value is low, it is inferred that the membership degree of low air pressure is 0.9, the membership degree of medium air pressure is 0.1, and the membership degree of high air pressure is 0.

[0141] Step S303: Define fuzzy rules. Fuzzy rules are defined based on expert knowledge or relevant literature. For example, if the ambient temperature, ambient air pressure, and oxygen concentration are low, then the probability of using inference degree assessment standard A as a real-time assessment standard is high; if the ambient temperature, ambient air pressure, and oxygen concentration are medium, then the probability of using inference degree assessment standard B as a real-time assessment standard is high. Among these, in degree assessment standard A, the coefficient range corresponding to low risk is relatively narrow, while the coefficient range corresponding to medium and high risk is relatively wide, making it suitable for plateau areas; in degree assessment standard B, the coefficient ranges corresponding to low risk, medium risk, and high risk are evenly distributed, making it suitable for conventional environments.

[0142] Step S304: Match the fuzzified environmental factors with fuzzy rules, perform fuzzy inference, and obtain the fuzzy inference results. The fuzzy inference results are the membership degrees corresponding to each degree evaluation standard. The fuzzy inference method is, for example, the Mamdani or Sugeno fuzzy inference method. For example, the membership degree of degree evaluation standard C is 0.1, the membership degree of degree evaluation standard B is 0.3, and the membership degree of degree evaluation standard A is 0.6.

[0143] Step S305: Compare each membership degree in the fuzzy inference results and select the evaluation standard with the highest membership degree as the real-time evaluation standard.

[0144] This embodiment comprehensively acquires key information about patients' health status by collecting various characteristic data of patients in real time; it uses clustering algorithms and deep learning technology to evaluate physiological indicators of different groups and detects abnormalities in these indicators, enabling timely identification of abnormalities in patients' physiological indicators; through image analysis, it quantitatively extracts the curvature and diameter features of blood vessels, calculates the morphological abnormality index, and forms multi-dimensional information with abnormal indicators and environmental factors to accurately assess the patient's thrombosis risk coefficient, and obtains the specific risk level based on a pre-constructed risk classification standard; it fully considers the impact of environmental factors such as hypoxia and low air pressure in high-altitude areas, comprehensively identifies special factors of thrombosis risk in high-altitude areas, improves the accuracy of assessing the risk of deep vein thrombosis in critically ill patients at high altitudes, and timely identifies high-risk patients, thereby providing effective early warning and intervention measures for clinical practice and reducing the incidence of deep vein thrombosis and related complications.

[0145] Example 2

[0146] This application also provides an electronic device. The electronic device may include one or more processors and one or more memories. The memories store computer-readable code that, when executed by the one or more processors, can perform the aforementioned auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas.

[0147] The methods or systems according to the embodiments of this application can also be implemented using the architecture of the electronic device shown in this application. The electronic device may include a bus, one or more CPUs, ROM, RAM, a communication port connected to a network, input / output, a hard disk, etc. The storage device in the electronic device, such as a ROM or hard disk, may store the auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas provided in this application. Furthermore, the electronic device may also include a user interface. Of course, the architecture shown in this application is merely exemplary; when implementing different devices, one or more components in the electronic device shown in this application may be omitted according to actual needs.

[0148] Example 3

[0149] One embodiment of this application discloses a computer-readable storage medium. The computer-readable storage medium stores computer-readable instructions. When the computer-readable instructions are executed by a processor, they can perform an auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas, as described in the above-described figures, according to an embodiment of this application. The storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0150] Furthermore, according to embodiments of this application, the processes described in the above-referenced flowcharts can be implemented as computer software programs. For example, this application provides a non-transitory machine-readable storage medium storing machine-readable instructions that can be executed by a processor to perform instructions corresponding to the method steps provided in this application, such as a system for assisting in the assessment of the risk of deep vein thrombosis in critically ill patients in high-altitude areas. When this computer program is executed by a central processing unit (CPU), it performs the functions defined in the method of this application.

[0151] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0152] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A system for assisting in the risk assessment of deep vein thrombosis in critically ill patients in high-altitude areas, characterized in that, include: The data acquisition module is used to collect patient characteristic data in real time, which includes basic patient data, numerical characteristic data, and image characteristic data. The indicator evaluation module uses a clustering algorithm to group numerical feature data, evaluates the corresponding physiological indicators based on the data within each group, and labels them as real-time indicators. The method for grouping the numerical feature data is as follows: a pre-trained word embedding model is used to convert each type of data in the numerical feature data into a corresponding physiological vector; clustering is then performed on all physiological vectors to obtain... Each group is divided into groups, and each group corresponds to a physiological indicator. The indicator detection module is used to obtain historical indicators based on the patient's basic data, calculate the coefficient of variation for each real-time indicator based on the historical and real-time indicators, analyze the coefficient of variation, and identify abnormal indicators in the real-time indicators. The steps for calculating the coefficient of variation for each real-time metric include: Step S201: For each physiological indicator, assign a numerical label sequentially and increment it, then mark it as an indicator label. The range of the indicator labels is... ; Step S202: Select indicator label as Physiological indicators, and labeled as elements. ; Step S203: Select physiological indicators corresponding to the elements from historical indicators and mark them as detection elements. Mark the real-time indicators corresponding to the elements as real-time elements. Use both detection elements and real-time elements as analysis elements. Step S204: Calculate the element difference and baseline distance for each analytical element; Step S205: Determine the baseline element corresponding to each analysis element; Step S206: Based on the element difference and the baseline distance, calculate the adjacent distance between each analyzed element and its corresponding baseline element; the expression for the adjacent distance is: In the formula, For the first The analysis element and its corresponding first element Adjacent distances between reference elements It is a function with maximum value. For the first The baseline distance of each analysis element, For the first The analysis element and its corresponding first element The elemental difference between each reference element , , ;in, To detect the number of elements, The preset screening quantity; Step S207: Calculate the local density corresponding to each analysis element based on the adjacent distance; Step S208: Calculate the discrete coefficients corresponding to the real-time elements based on the local density; Step S209: Let ; Step S210: Repeat steps S202 to S209 until all physiological indicators are marked as elements, then the loop ends; The image analysis module is used to analyze image feature data, extract vascular feature data, and quantify the morphological abnormality index based on the vascular feature data. The risk assessment module is used to integrate abnormal indicators and morphological abnormality indices to assess the thrombosis risk coefficient and obtain the risk level corresponding to the thrombosis risk coefficient based on the pre-constructed risk classification criteria. Methods for obtaining the risk level corresponding to the thrombosis risk coefficient include: Based on environmental factors, fuzzy reasoning is used to select corresponding severity assessment criteria from pre-constructed risk classification standards, and these are marked as real-time assessment standards. The risk classification standards include... Different risk assessment standards are used, each of which includes a coefficient range corresponding to different risk levels. The thrombosis risk coefficient is compared with each coefficient range in the real-time assessment standard to obtain the risk level corresponding to the coefficient range in which the thrombosis risk coefficient is located.

2. The auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas according to claim 1, characterized in that, The steps for grouping numerical feature data include: Step S101: Using a pre-trained word embedding model, each data point in the numerical feature data is converted into a corresponding vector and labeled as a physiological vector; Step S102: Each physiological vector is used as a point to be divided, and each point to be divided corresponds one-to-one with a physiological vector; Step S103: Preset the number of groups Random selection Each of the points to be divided is taken as a center point and marked as... , ; Mark the sample points that are not used as center points as partition points, and mark each partition point as , , The number of physiological vectors; Step S104: According to Establish corresponding centers The data is divided into groups, and the cosine similarity between each dividing point and each center point is calculated and marked as similarity. Step S105: Divide the points Compare all corresponding similarities, mark the center point corresponding to the highest similarity as the optimal point, and divide the points. Divide into groups corresponding to the optimal points; Step S106: Let ; Step S107: Repeat steps S105 to S106 until... The time loop ends, and proceed to step S108; Step S108: Recalculate the center point corresponding to each group; Step S109: Repeat steps S104 to S108 until the center point of each group recalculated in step S108 is consistent with the corresponding center point calculated in the previous iteration. Then the loop ends, and the data is obtained. Each group and its corresponding dividing point, with each group corresponding to a physiological indicator; In step S108, the method for recalculating the center point corresponding to each group includes: Count the number of points to be divided in each group and mark them as the number of punctuation marks; add up the center points corresponding to each group in turn, and then divide by the corresponding number of punctuation marks to obtain the new center point of each group.

3. The auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas according to claim 2, characterized in that, Methods for assessing physiological indicators include: Each group's data is treated as a dataset, with each dataset corresponding to a specific group. Each dataset is then input into its corresponding evaluation model to assess the relevant physiological indicators. The evaluation model includes... Each assessment model corresponds one-to-one with a physiological indicator. All evaluation models are deep neural network models, and The training process for each evaluation model is consistent; The training process for evaluating the model includes: Pre-collection Group data set, Each dataset corresponds to a physiological indicator, for Each dataset has corresponding physiological indicators. For integers greater than 1, the dataset and corresponding physiological indicators are converted into a set of feature vectors. Each set of feature vectors is used as input to the evaluation model, which outputs a set of predicted physiological indicators corresponding to each dataset and uses the actual physiological indicators corresponding to each dataset as the prediction target. The actual physiological indicators are the pre-set physiological indicators corresponding to the dataset. The training objective is to minimize the sum of prediction errors of all datasets. The evaluation model is trained until the sum of prediction errors converges, at which point training stops.

4. The auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas according to claim 3, characterized in that, The historical indicators are physiological indicators of the same patient under normal physiological conditions at a historical time. The same patient is a patient whose basic patient data is the same as the patient basic data collected in real time. Methods for identifying outliers in real-time metrics include: Preset recognition coefficient , Subtract 1 from the discrete coefficient of each real-time element, then take the absolute value to obtain the judgment coefficient of each real-time element; compare the judgment coefficient of each real-time element with the recognition coefficient. If the judgment coefficient is greater than or equal to the recognition coefficient, the real-time indicator corresponding to the real-time element will be marked as an abnormal indicator; if the judgment coefficient is less than the recognition coefficient, the real-time indicator corresponding to the real-time element will not be marked.

5. The auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas according to claim 4, characterized in that, In step S204, the method for calculating the element difference corresponding to each analysis element is as follows: obtain the neighboring elements corresponding to each analysis element, where the neighboring elements are the remaining elements. Each analysis element is analyzed; each analysis element is subtracted from its corresponding neighboring element to obtain the element difference for each analysis element. The method for calculating the baseline distance for each analysis element is as follows: Each element difference corresponding to an analysis element is treated as an element set, and each element set corresponds one-to-one with the analysis element; the element differences in each element set are sorted in ascending order to generate an element sorting table corresponding to each element set. Rank each element in the sorted list as number 1 The element difference of the corresponding element set is used as the reference distance for the analysis element corresponding to the corresponding element set; In step S205, the method for determining the reference element corresponding to each analysis element is as follows: sort the first element in the sorting table of each element. The adjacent elements corresponding to the bit difference are used as the reference elements of the analysis elements corresponding to the element set. In step S207, the method for calculating the local density corresponding to each analysis element is as follows: the adjacent distances corresponding to each analysis element are added sequentially to obtain the comprehensive distance corresponding to each analysis element; the reciprocal of the comprehensive distance corresponding to each analysis element is used as the local density corresponding to each analysis element. In step S208, the method for calculating the discrete coefficients corresponding to the real-time elements is as follows: obtain the reference elements corresponding to the real-time elements and mark them as evaluation elements; divide the local density corresponding to each evaluation element by the local density corresponding to the real-time elements to obtain the relative density corresponding to each evaluation element; add the relative densities of each evaluation element sequentially, and then divide by the reference density. Obtain the discrete coefficients corresponding to the real-time elements.

6. The auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas according to claim 5, characterized in that, The vascular feature data includes a curvature set and a diameter set; Methods for extracting curvature sets include: A preset grayscale threshold is used to obtain the grayscale value of each pixel in the image feature data. The grayscale value of each pixel is compared with the grayscale threshold. If the grayscale value is greater than the threshold, the corresponding pixel is marked as a blood vessel point; if the grayscale value is less than or equal to the threshold, the corresponding pixel is not marked. Based on all blood vessel points, a skeletonization algorithm is used to extract the blood vessel centerline. From all blood vessel points corresponding to the blood vessel centerline, random selections are made... One blood vessel point, marked as the midline point. , The number of vascular points along the central line of the blood vessel; Based on the image coordinate system built into the image feature data, obtain The coordinates corresponding to each midline point are marked as midline coordinates; according to The centerline coordinates of each centerline point are used to fit a curve to the centerline using a curve fitting algorithm, obtaining the corresponding mathematical expression for the centerline, which is then marked as a curve expression. The first derivative of the curve expression is then taken to obtain a first-order expression. The second derivative of the curve expression is then taken to obtain a second-order expression. Each centerline coordinate is substituted into the first-order expression to obtain a first-order value for each centerline point. Each centerline coordinate is then substituted into the second-order expression to obtain a second-order value for each centerline point. Each first-order value is squared and then incremented by one to obtain a first value. The cube root of each first value is then taken to obtain a second value for each centerline point. The absolute value of the second-order value for each centerline point is divided by the corresponding second value to obtain the vascular curvature for each centerline point. The vascular curvatures corresponding to each centerline point are then used as a curvature set.

7. The auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas according to claim 6, characterized in that, Methods for extracting the diameter set include: The negative reciprocal of the first-order value corresponding to each midline point is used as the slope of the normal line corresponding to that midline point. Based on the midline coordinates of each midline point and the slope of the corresponding normal line, the equation of the normal line corresponding to each midline point is expressed in point-slope form. The expression for the normal line equation is: In the formula, Let be the ordinate of any point on the normal. The slope Let x be the x-coordinate of any point on the normal. The x-coordinate of the midline point is... The ordinate of the midline point is used. For each blood vessel point, its adjacent points are analyzed; adjacent points are pixels adjacent to the blood vessel point. If there are unmarked pixels among the adjacent points, the corresponding blood vessel point is marked as an edge point; if all adjacent points are blood vessels, the corresponding blood vessel point is not marked. Each edge point is substituted into each normal equation; if the normal equation holds, the corresponding edge point is taken as the point satisfying the normal equation. The coordinates of each satisfying point are obtained and marked as satisfying coordinates. Based on the satisfying coordinates, the Euclidean distance between two satisfying points corresponding to each normal equation is calculated and used as the satisfying length. The satisfying length corresponding to each normal equation is used as the image diameter of the corresponding midline point. A scaling factor is obtained, and each image diameter is multiplied by the scaling factor to obtain the blood vessel diameter corresponding to each midline point. The blood vessel diameter corresponding to each midline point is used as the diameter set.

8. The auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas according to claim 7, characterized in that, Methods for quantifying morphological anomaly indices include: Add up the curvature of each blood vessel in sequence, then divide by . Obtain the mean curvature; subtract the mean curvature from the curvature of each vessel and square the result to obtain the squared curvature difference; sum the squared curvature differences of each vessel and then divide by the mean curvature difference. The standard deviation of curvature is obtained by taking the square root of the standard deviation of curvature and dividing it by the mean curvature to obtain the curvature abnormality index. The diameters of each vessel are summed sequentially and then divided by the mean. Obtain the mean diameter; subtract the mean diameter from the diameter of each vessel and square the result to obtain the squared diameter difference; sum the squared diameter differences and divide by the mean. Then, take the square root to obtain the standard deviation of the diameter; divide the standard deviation of the diameter by the mean of the diameter to obtain the diameter anomaly coefficient; A preset threshold coefficient is used, including a curvature coefficient and a diameter coefficient. The curvature of each blood vessel is compared with the curvature coefficient. If the curvature is greater than or equal to the curvature coefficient, the corresponding blood vessel point is marked as a curvature abnormality point; if the curvature is less than the curvature coefficient, the corresponding blood vessel point is not marked. The diameter of each blood vessel is compared with the diameter coefficient. If the diameter is greater than or equal to the diameter coefficient, the corresponding blood vessel point is marked as a diameter abnormality point; if the diameter is less than the diameter coefficient, the corresponding blood vessel point is not marked. Both curvature abnormalities and diameter abnormalities are considered local abnormalities. The number of local abnormalities is counted and marked as the local abnormality count. The local abnormality count is divided by... Obtain local anomaly coefficients; A preset weight set is used, which includes the weight coefficients corresponding to the curvature anomaly index, diameter anomaly index, and local anomaly coefficient. The curvature anomaly index, diameter anomaly index, and local anomaly coefficient are multiplied by their respective weight coefficients and then added together to obtain the morphological anomaly index.

9. The auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas according to claim 8, characterized in that, The methods for assessing the risk factor for thrombosis include: Environmental factors, including ambient temperature, atmospheric pressure, and oxygen concentration, are acquired. These environmental factors, abnormal indicators, and morphological abnormality indices are used as assessment data. The assessment data is then input into a trained risk assessment model to evaluate the patient's thrombosis risk coefficient. The training process of the risk assessment model is consistent with that of the assessment model, and both are deep neural network models.

10. The auxiliary assessment system for the risk of deep vein thrombosis in critically ill patients in high-altitude areas according to claim 9, characterized in that, The steps for selecting real-time evaluation criteria include: Step S301: Construct a corresponding fuzzy set for each data point in the environmental factors, with each fuzzy set including multiple fuzzy categories; Step S302: Map each data point in the environmental factors to the membership degree of the corresponding fuzzy category using fuzzification technology; Step S303: Define fuzzy rules; Step S304: Match the fuzzified environmental factors with fuzzy rules, perform fuzzy inference, and obtain the fuzzy inference results. The fuzzy inference results are the membership degrees corresponding to each level of evaluation criteria. Step S305: Compare each membership degree in the fuzzy inference results and select the evaluation standard with the highest membership degree as the real-time evaluation standard.

Images