Intelligent temperature control system and method for whole body thermal therapy instrument

Abstract

The invention discloses an intelligent temperature control system and method for a whole body thermal therapy instrument, and relates to the technical field of temperature control monitoring, and the system comprises a monitoring data acquisition module, a space-time alignment module, a characteristic index analysis module, a graph structure construction module, a probability model analysis module, a thermal damage risk score calculation module, and a real-time monitoring response module. The monitoring data acquisition module is used for acquiring monitoring data stored when the whole body thermal therapy instrument executes a temperature control regulation event; the space-time alignment module is used for aligning monitoring data of different sampling frequencies by adopting dynamic time warping and generating a data array at the same moment, and the feature index analysis module is used for extracting feature indexes for evaluating user states based on the monitoring data; the graph structure construction module is used for constructing the monitoring data in the data array in a graph structure form; and the probability model analysis module is used for analyzing and outputting associated abnormal nodes of all the hot area nodes and corresponding abnormal temperature control probability models.

Description

Technical Field

[0001] The present invention relates to the technical field of temperature control and monitoring, and in particular to an intelligent temperature control system and method for a whole-body hyperthermia therapy apparatus. Background Art

[0002] Whole-body hyperthermia devices, as medical devices that achieve therapeutic goals by raising the body's overall temperature, play an important role in clinical treatment. Their operating principles are primarily based on infrared and graphene heating. For example, some devices generate heat by focusing infrared light. Leveraging the unique penetrating power of infrared light, they penetrate the epidermis and reach the subcutaneous tissue and subcutaneous capillary network, primarily heating the blood within this network. This heat is then transferred from the inside out through the bloodstream, gradually raising the patient's overall body temperature. Other devices use graphene, leveraging its excellent thermal conductivity to generate heat and provide a heat source for hyperthermia. However, existing whole-body hyperthermia devices face numerous challenges in practical application. Current monitoring methods for hyperthermia devices are extremely limited. Most are equipped with simple temperature sensors that can only capture surface or localized temperature information. During prolonged hyperthermia treatment, important physiological parameters such as heart rate and respiratory rate can significantly change due to thermal stimulation, but existing devices struggle to simultaneously monitor these changes. For example, when a patient's heart rate suddenly accelerates during heat therapy, traditional heat therapy devices cannot detect it in time, which makes it difficult for medical staff to accurately grasp the potential abnormal reactions in the patient's body, greatly increasing the risk of heat therapy.

[0003] At the same time, different parts of the human body differ significantly in their absorption and tolerance of heat. For example, the limbs and torso have different temperature sensitivities and requirements during heat therapy. However, the heat field generated by traditional heat therapy devices fails to fully account for these differences, making it difficult to achieve precise heat field control for different parts of the body. Furthermore, the heat field data is independent of the patient's physiological indicators, without effective correlation analysis. This results in the inability to dynamically adjust the heat field parameters based on the patient's real-time physiological state during the heat therapy process, seriously affecting the accuracy and safety of heat therapy. Summary of the Invention

[0004] The object of the present invention is to provide an intelligent temperature control system and method for a whole body hyperthermia apparatus to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent temperature control method for a whole-body hyperthermia device, the method comprising:

[0006] Step S100: Acquire monitoring data stored by the whole-body thermal therapy device when executing temperature control events, recording data types and monitoring values; data types include ECG, blood oxygen concentration SpO2, heart rate HR, core temperature Tcore, and surface temperature of each region; use dynamic time warping to align monitoring data of different sampling frequencies to generate a data array at the same time, and extract characteristic indicators for evaluating user status based on the monitoring data;

[0007] Step S200: The monitoring data in the data array is constructed in the form of a graph structure, generating an association structure containing nodes, edges, and recording node features; the nodes include seven hyperthermia zones and three physiological nodes; based on the constructed graph structure, the monitoring values ​​at the same time are input to form a data layer that records the spatial edge weights and node feature values ​​of each node at different times;

[0008] Step S300: searching for monitoring data marked as abnormal temperature control events by clinical samples as target data, and outputting the associated abnormal nodes of each hot zone node and the corresponding abnormal temperature control probability model based on the response analysis of the target data;

[0009] Step S400: Based on step S300, the spatiotemporal correlation features of the abnormal temperature control and adjustment event records are captured, and the spatiotemporal correlation features are associated with the analysis data of the corresponding abnormal temperature control and adjustment event records to generate response feature groups, which are stored in the temperature control monitoring database, and a thermal damage risk score is assigned to each response feature group;

[0010] Step S500: Acquire real-time monitoring data of the whole-body thermal therapy device and match it with the temperature control monitoring database. When the match is successful, output the corresponding spatiotemporal correlation characteristics and the predicted abnormal temperature control adjustment event response probability; when the match is unsuccessful, return to step S300 and substitute the abnormal temperature control probability model for a new prediction and respond whether to issue an early warning.

[0011] Furthermore, step S100 extracts characteristic indicators for evaluating user status based on monitoring data, including the following specific steps:

[0012] Obtain the time interval between two adjacent heartbeats from ECG monitoring, that is, the time difference RRj-RRj+1 from the R wave peak of the jth heartbeat to the R wave peak of the j+1th heartbeat, and the total number of RR intervals N; use the formula:

[0013] RMSSD={[1 / (N-1)]*∑(RR j -RR j+1 ) 2} 1 / 2 ;

[0014] Calculate RMSSD, a measurement indicator of heart rate variability (HRV) used to evaluate the user's status;

[0015] Get the monitoring value corresponding to the core temperature Tcore using the formula:

[0016]

[0017] Calculate the thermal meter accumulation CEM43(t) of the user to evaluate the user status;

[0018] Thermal measurement accumulation, heart rate variability, temperature change rate, momentary temperature, heart rate and blood oxygen concentration SpO2 are used as characteristic indicators for evaluating user status; temperature change rate refers to the temperature change value based on a certain monitoring node per unit time, and momentary temperature refers to the temperature measured based on a certain monitoring node.

[0019] Furthermore, step S200 includes the following specific steps:

[0020] Based on the connection relationship of the physiological structure, each node is connected in sequence through edges, the actual area distance d of adjacent nodes after connection is obtained, and the edge weight k is calculated. d , k d =exp(-d / c), c=10cm;

[0021] The characteristic indicators of the moment temperature T(t), temperature change rate ΔT(t), heart rate HR(t), blood oxygen concentration SpO2(t) and thermal measurement accumulation CEM43(t) are used to form the node feature f(t) of each node in the corresponding graph structure at time t, f(t) = [T(t), ΔT(t), HR(t), SpO2(t), CEM43(t)];

[0022] Then a data layer is generated in which nodes record node features and edges between nodes record spatial edge weights.

[0023] Furthermore, step S300 includes the following specific steps:

[0024] The node that initially records the temperature anomaly in each abnormal temperature control adjustment event is obtained as the target node, and all adjacent nodes of the target node in the graph structure are searched as reference nodes using the target node as the initial search node;

[0025] Based on the node features of the target node and the control node, the weight matrix W is mapped to the eight-dimensional space to obtain W·f 目标 (t) and W·f 对照 (t); where each weight w in the weight matrix ij Set by model training; w ij Indicates the weight value corresponding to the i-th row and j-th column;

[0026] Calculate the attention coefficient e of the target node and each control node 目标-对照 , e 目标-对照=LeakyReLU(a T [Wf 目标 (t)||Wf 对照 (t)]), where a T Represents the attention vector, “||” represents feature concatenation; the attention coefficient e formed by the target node and the i-th control node 目标-对照i Normalize to generate attention weight ɑ 目标-对照i ,ɑ 目标-对照 i=e 目标-对照i / ∑(e 目标-对照 );∑(e 目标-对照 ) represents the sum of the attention coefficients of the same target node and all control nodes; calculates the GAT output feature z of the target node 目标 (t), z 目标 (t)=∑[ɑ 目标-对照 *Wf 对照 (t)];

[0027] Calculate the LSTM layer: Get the GAT output feature z of all target nodes 目标 (t) is concatenated into an 80-dimensional vector and the hidden state h is obtained. t Update, pass the LSTM hidden state h through the decoder t Restored to input feature f'(t), f'(t) = σ(W d ·h t +b d ), where σ represents the activation function, W d represents the decoder weight matrix, b d Represents the error term; obtain the original input feature f(t), and use the formula: v=||f(t)-f'(t)||2 to calculate the reconstruction error v;

[0028] Set the attention anomaly judgment condition: when the target node area temperature is greater than g1 degrees Celsius and the control node area temperature is less than g2 degrees Celsius, g1>g2, the output feature value is 1; otherwise the output is 0; obtain the control node that satisfies the output feature value of 1 and has the largest edge weight as the associated abnormal node, and extract the attention weight ɑ of the associated abnormal node record 目标-对照 1. Calculate the attention abnormality u, u = ɑ 目标-对 According to 1*1; and calculate the total abnormality score S(t), S(t) = 0.7*v+0.3*u;

[0029] Based on the total abnormality score, the abnormal temperature control probability model P(control) is constructed, P(control) = 1 / [1+e -S(t)*3 ].

[0030] Furthermore, step S300 includes the following specific steps:

[0031] Obtain monitoring data of the target node and associated abnormal nodes during the monitoring process, calculate the dynamic changes of attention weights through GAT, and record the dynamic change ratio of attention weights; capture temporal feature content through LSTM;

[0032] The dynamic change ratio and time feature content are used as the response features of the target node and the associated abnormal node, and the output value of the abnormal temperature control probability model constitutes a response feature group and is stored in the temperature control monitoring database;

[0033] Extract the monitoring data to determine the dynamic change ratio moment, using the formula:

[0034] HDRS=0.6·CEM43+0.3·MAX(T 目标 -40)+0.1·(100-SpO2)

[0035] The heat injury risk score HDRS corresponding to the response feature group was calculated.

[0036] The model architecture formed by GAT and LSTM in this application can effectively and quickly extract data features from spatial and temporal perspectives, capture characteristic conditions that are directly effective for monitoring and judgment, thereby realizing intelligent monitoring in the face of large amounts of data and improving the safety of hyperthermia. It is no longer a simple single judgment monitoring based on temperature thresholds.

[0037] Furthermore, step S500 includes the following specific processes:

[0038] A successful match means that the response feature group corresponding to the thermal damage risk score calculated based on the real-time monitoring data is the same as the target node and the associated abnormal node reflected by the real-time monitoring data; the spatiotemporal correlation features recorded by the corresponding response feature group and the response probability of the abnormal temperature control adjustment event are output;

[0039] When the matching is unsuccessful, the thermal damage risk score of each node is first calculated, and the thermal damage risk score threshold is set. When the thermal damage risk score is greater than or equal to the thermal damage risk score threshold, the target node is determined and the associated abnormal nodes are further located. The response probability of the abnormal temperature control adjustment event is calculated and the corresponding spatiotemporal correlation characteristics are output in response.

[0040] An intelligent temperature control system for a whole-body hyperthermia device, comprising a monitoring data acquisition module, a spatiotemporal alignment module, a characteristic indicator analysis module, a graph structure construction module, a probability model analysis module, a thermal injury risk score calculation module, and a real-time monitoring response module;

[0041] The monitoring data acquisition module is used to obtain the monitoring data stored by the whole body hyperthermia device when performing temperature control adjustment events;

[0042] The spatiotemporal alignment module is used to align monitoring data with different sampling frequencies using dynamic time warping to generate data arrays at the same time.

[0043] The characteristic indicator analysis module is used to extract characteristic indicators for evaluating user status based on monitoring data;

[0044] The graph structure construction module is used to construct the monitoring data in the data array in the form of a graph structure, generating an association structure containing nodes, edges and recording node features;

[0045] The probability model analysis module is used to analyze and output the associated abnormal nodes of each hot zone node and the corresponding abnormal temperature control probability model;

[0046] The heat injury risk score calculation module is used to calculate the heat injury risk score for each response feature group;

[0047] The real-time monitoring response module is used to output the corresponding spatiotemporal correlation characteristics and the predicted abnormal temperature control adjustment event response probability when the match is successful; when the match is unsuccessful, it returns to step S300 to substitute the abnormal temperature control probability model for a new prediction and respond whether to issue an early warning.

[0048] Furthermore, the probability model analysis module includes a node type determination unit, an attention weight calculation unit, a reconstruction error calculation unit, a total anomaly score calculation unit, and an abnormal temperature control probability model construction unit;

[0049] The node type determination unit is used to determine the target node and the control node;

[0050] The attention weight calculation unit is used to calculate the attention weight of each target node and all corresponding control nodes;

[0051] The reconstruction error calculation unit is used for the reconstruction error of each target node;

[0052] The total anomaly score calculation unit is used to set the attention anomaly item judgment conditions and calculate the total anomaly score based on the satisfied nodes;

[0053] The abnormal temperature control probability model building unit is used to build an abnormal temperature control probability model based on the total abnormality score.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] 1. The present invention uses a dynamic graph neural network to deeply integrate a variety of physiological indicator data during the patient's thermal therapy, such as heart rate, respiratory rate, body temperature, etc., with the complex thermal field data inside the thermal therapy device, including the temperature and heat flow distribution of different areas formed by infrared and graphene heating. By exploring the complex relationship between the two in the time and space dimensions, it is possible to comprehensively and accurately assess the risk of thermal damage faced by patients during thermal therapy. During the thermal therapy process, the system can simultaneously analyze the changing trend of the patient's heart rate and the temperature fluctuations of different parts of the body due to the thermal field exposed to infrared or graphene heating. Compared with the traditional single data monitoring and analysis method, it greatly improves the accuracy of the patient's thermal therapy safety risk assessment and provides a strong guarantee for the safety of the thermal therapy process.

[0056] 2. Ability to track the changing characteristics of thermal field data and physiological index data in time and space dimensions during hyperthermia treatment in real time;

[0057] 3. This system, combined with a specially constructed thermal injury risk scoring system, provides a quantitative early warning of hyperthermia risks. By comprehensively analyzing and calculating the fused physiological and thermal field data, it outputs a specific thermal injury risk score, clearly and unambiguously informing medical staff of the severity of the patient's current risk. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 The figure is a structural schematic diagram of an intelligent temperature control method for a whole body hyperthermia therapy apparatus according to the present invention. DETAILED DESCRIPTION

[0059] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0060] Example: Figure 1 As shown, the present invention provides an intelligent temperature control method for a whole body hyperthermia device, the method comprising:

[0061] Step S100: Acquire monitoring data stored by the whole-body thermal therapy device when executing temperature control events, recording data types and monitoring values; data types include ECG, blood oxygen concentration SpO2, heart rate HR, core temperature Tcore, and surface temperature of each region; use dynamic time warping to align monitoring data of different sampling frequencies to generate a data array at the same time, and extract characteristic indicators for evaluating user status based on the monitoring data;

[0062] Step S200: The monitoring data in the data array is constructed in the form of a graph structure, generating an association structure containing nodes, edges, and recording node features; the nodes include seven hyperthermia zones and three physiological nodes; based on the constructed graph structure, the monitoring values ​​at the same time are input to form a data layer that records the spatial edge weights and node feature values ​​of each node at different times;

[0063] Step S300: searching for monitoring data marked as abnormal temperature control events by clinical samples as target data, and outputting the associated abnormal nodes of each hot zone node and the corresponding abnormal temperature control probability model based on the response analysis of the target data;

[0064] Step S400: Based on step S300, the spatiotemporal correlation features of the abnormal temperature control and adjustment event records are captured, and the spatiotemporal correlation features are associated with the analysis data of the corresponding abnormal temperature control and adjustment event records to generate response feature groups, which are stored in the temperature control monitoring database, and a thermal damage risk score is assigned to each response feature group;

[0065] Step S500: Acquire real-time monitoring data of the whole-body thermal therapy device and match it with the temperature control monitoring database. When the match is successful, output the corresponding spatiotemporal correlation characteristics and the predicted abnormal temperature control adjustment event response probability; when the match is unsuccessful, return to step S300 and substitute the abnormal temperature control probability model for a new prediction and respond whether to issue an early warning.

[0066] Extracting characteristic indicators for evaluating user status based on monitoring data in step S100 includes the following specific steps:

[0067] Obtain the time interval between two adjacent heartbeats from ECG monitoring, that is, the time difference RRj-RRj+1 from the R wave peak of the jth heartbeat to the R wave peak of the j+1th heartbeat, and the total number of RR intervals N. For example, if N = 70 heartbeats, then N-1 = 69 RR interval differences are obtained within one minute. Use the formula:

[0068] RMSSD={[1 / (N-1)]*∑(RR j -RR j+1 ) 2} 1 / 2 ;

[0069] Calculate RMSSD, a measurement indicator of heart rate variability (HRV) used to evaluate the user's status;

[0070] Get the monitoring value corresponding to the core temperature Tcore using the formula:

[0071]

[0072] Calculate the thermal meter accumulation CEM43(t) of the user to evaluate the user status;

[0073] Thermal accumulation, heart rate variability, temperature change rate, momentary temperature, heart rate, and blood oxygen concentration (SpO2) are used as characteristic indicators to assess user status. The temperature change rate refers to the temperature change per unit time at a monitoring node, and the momentary temperature refers to the temperature measured at a monitoring node. The monitoring node can be any of the seven hyperthermia zones and three physiological nodes.

[0074] Step S200 includes the following specific steps:

[0075] Based on the connection relationship of the physiological structure, each node is connected in sequence through edges, the actual area distance d of adjacent nodes after connection is obtained, and the edge weight k is calculated. d , k d =exp(-d / c), c=10cm;

[0076] The characteristic indicators of the moment temperature T(t), temperature change rate ΔT(t), heart rate HR(t), blood oxygen concentration SpO2(t) and thermal measurement accumulation CEM43(t) are used to form the node feature f(t) of each node in the corresponding graph structure at time t, f(t) = [T(t), ΔT(t), HR(t), SpO2(t), CEM43(t)];

[0077] Then a data layer is generated in which nodes record node features and edges between nodes record spatial edge weights.

[0078] Step S300 includes the following specific steps:

[0079] The node that initially records the temperature anomaly in each abnormal temperature control adjustment event is obtained as the target node, and all adjacent nodes of the target node in the graph structure are searched as reference nodes using the target node as the initial search node;

[0080] Based on the node features of the target node and the control node, the weight matrix W is mapped to the eight-dimensional space to obtain W·f 目标 (t) and W·f 对照 (t); where each weight w in the weight matrix ij Set by model training; w ij Indicates the weight value corresponding to the i-th row and j-th column;

[0081] Calculate the attention coefficient e of the target node and each control node 目标-对照 , e 目标-对照 =LeakyReLU(a T [Wf 目标 (t)||Wf 对照 (t)]), where a Trepresents the attention vector, “||” represents feature concatenation; a is obtained by training the attention allocation strategy based on the node feature data through gradient descent optimization; the attention coefficient e for the target node and the i-th control node is 目标-对照i Normalize to generate attention weight ɑ 目标-对照 i,ɑ 目标-对照 i=e 目标-对照i / ∑(e 目标-对照 );∑(e 目标 - control) represents the sum of the attention coefficients of the same target node and all control nodes; calculates the GAT output feature z of the target node 目标 (t), z 目标 (t)=∑[ɑ 目标-对照 *Wf 对照 (t)];

[0082] Calculate the LSTM layer: Get the GAT output feature z of all target nodes 目标 (t) is concatenated into an 80-dimensional vector and the hidden state h is obtained. t Update, pass the LSTM hidden state h through the decoder t Restored to input feature f'(t), f'(t) = σ(W d ·h t +b d ), where σ represents the activation function, W d represents the decoder weight matrix, b d Represents the error term; obtain the original input feature f(t), and use the formula: v=||f(t)-f'(t)||2 to calculate the reconstruction error v;

[0083] In this application, the LSTM layer calculation for each abnormal temperature control event is carried out around the target node. For example, when the target node is the chest area, and through data input, the LSTM hidden state ht contains the time feature of "chest area temperature rises by 0.3 degrees Celsius per minute, and HR increases by 15 beats per minute". The calculated reconstruction error is also based on the node feature data obtained from this time feature.

[0084] Set the attention anomaly judgment condition: when the target node area temperature is greater than g1 degrees Celsius and the control node area temperature is less than g2 degrees Celsius, g1>g2, the output feature value is 1; otherwise the output is 0; obtain the control node that satisfies the output feature value of 1 and has the largest edge weight as the associated abnormal node, and extract the attention weight ɑ of the associated abnormal node record 目标-对照 1. Calculate the attention abnormality u, u = ɑ 目标-对 According to 1*1; and calculate the total abnormality score S(t), S(t) = 0.7*v+0.3*u;

[0085] Based on the total abnormality score, the abnormal temperature control probability model P(control) is constructed, P(control) = 1 / [1+e -S(t)*3 ].

[0086] As shown in the examples:

[0087] The chest region node (tumor region) is used as the target node, and the control nodes are the heart, left arm, and abdomen. The node features of the chest region are obtained as follows:

[0088] f(t)=[T(t),ΔT(t),HR(t),SpO2(t),CEM43(t)]=[41,0.3,118,18,28];

[0089] The node features of the heart are [37.8, 0.1, 118, 18, 28];

[0090] Calculate the edge weights k chest area - heart = 0.223, k chest area - skin = 0.607;

[0091] The attention weight is then calculated based on the GAT layer. Since the calculated edge weights indicate that the distance from the chest area to the skin and heart is relatively close, the parameters of these two types of nodes can be prioritized for analysis in subsequent calculations to determine attention anomalies. The control feature that satisfies the attention anomaly judgment adjustment can be locked as skin. The reconstruction error is then calculated through the LSTM layer to comprehensively obtain the total score anomaly. Based on the skin feature, the specific probability of erythema on the skin can be further estimated.

[0092] In this application, the feature values ​​in the attention anomaly determination process can be dynamically amplified based on the influence of the LSTM hidden state.

[0093] Step S300 includes the following specific steps:

[0094] Obtain monitoring data of the target node and associated abnormal nodes during the monitoring process, calculate the dynamic changes of attention weights through GAT, and record the dynamic change ratio of attention weights; capture temporal feature content through LSTM;

[0095] The dynamic change ratio and time feature content are used as the response features of the target node and the associated abnormal node, and the output value of the abnormal temperature control probability model constitutes a response feature group and is stored in the temperature control monitoring database;

[0096] Extract the monitoring data to determine the dynamic change ratio moment, using the formula:

[0097] HDRS=0.6·CEM43+0.3·MAX(T 目标 -40)+0.1·(100-SpO2)

[0098] The heat injury risk score HDRS corresponding to the response feature group was calculated.

[0099] The model architecture formed by GAT and LSTM in this application can effectively and quickly extract data features from spatial and temporal perspectives, capture characteristic conditions that are directly effective for monitoring and judgment, thereby realizing intelligent monitoring in the face of large amounts of data and improving the safety of hyperthermia. It is no longer a simple single judgment monitoring based on temperature thresholds.

[0100] Step S500 includes the following specific processes:

[0101] A successful match means that the response feature group corresponding to the thermal damage risk score calculated based on the real-time monitoring data is the same as the target node and the associated abnormal node reflected by the real-time monitoring data; the spatiotemporal correlation features recorded by the corresponding response feature group and the response probability of the abnormal temperature control adjustment event are output;

[0102] When the matching is unsuccessful, the thermal damage risk score of each node is first calculated, and the thermal damage risk score threshold is set. When the thermal damage risk score is greater than or equal to the thermal damage risk score threshold, the target node is determined and the associated abnormal nodes are further located. The response probability of the abnormal temperature control adjustment event is calculated and the corresponding spatiotemporal correlation characteristics are output in response.

[0103] In this application, a successful match indicates that there is an anomaly in the real-time monitoring data, because the temperature control monitoring database records abnormal temperature control events;

[0104] As shown in the embodiment; if the matching is not successful, the user's input data at a certain moment is obtained.

[0105] Chest temperature 41.2 degrees Celsius for three minutes, HR: 118 bpm (baseline 82 bpm); HRV: 18 ms (baseline 45 ms); CEM 43: 28 minutes;

[0106] Can be calculated

[0107] HDRS=0.6·CEM43+0.3·MAX(T 目标 -40)+0.1·(100-SpO2)=17.6>8; belongs to

[0108] within the scope of early warning;

[0109] Further analysis to determine the change in attention weights shows that the attention weight of the node chest area to the node skin changes the most, so the associated abnormal node is located as the skin. Then, GAT and LSTM are used to obtain the spatiotemporal correlation features and the output value of the abnormal temperature control probability model to respond.

[0110] An intelligent temperature control system for a whole-body hyperthermia device, comprising a monitoring data acquisition module, a spatiotemporal alignment module, a characteristic indicator analysis module, a graph structure construction module, a probability model analysis module, a thermal injury risk score calculation module, and a real-time monitoring response module;

[0111] The monitoring data acquisition module is used to obtain the monitoring data stored by the whole body hyperthermia device when performing temperature control adjustment events;

[0112] The spatiotemporal alignment module is used to align monitoring data with different sampling frequencies using dynamic time warping to generate data arrays at the same time.

[0113] The characteristic indicator analysis module is used to extract characteristic indicators for evaluating user status based on monitoring data;

[0114] The graph structure construction module is used to construct the monitoring data in the data array in the form of a graph structure, generating an association structure containing nodes, edges and recording node features;

[0115] The probability model analysis module is used to analyze and output the associated abnormal nodes of each hot zone node and the corresponding abnormal temperature control probability model;

[0116] The heat injury risk score calculation module is used to calculate the heat injury risk score for each response feature group;

[0117] The real-time monitoring response module is used to output the corresponding spatiotemporal correlation characteristics and the predicted abnormal temperature control adjustment event response probability when the match is successful; when the match is unsuccessful, it returns to step S300 to substitute the abnormal temperature control probability model for a new prediction and respond whether to issue an early warning.

[0118] The probability model analysis module includes a node type determination unit, an attention weight calculation unit, a reconstruction error calculation unit, a total anomaly score calculation unit, and an abnormal temperature control probability model construction unit;

[0119] The node type determination unit is used to determine the target node and the control node;

[0120] The attention weight calculation unit is used to calculate the attention weight of each target node and all corresponding control nodes;

[0121] The reconstruction error calculation unit is used for the reconstruction error of each target node;

[0122] The total anomaly score calculation unit is used to set the attention anomaly item judgment conditions and calculate the total anomaly score based on the satisfied nodes;

[0123] The abnormal temperature control probability model building unit is used to build an abnormal temperature control probability model based on the total abnormality score.

[0124] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. An intelligent temperature control method for a whole-body hyperthermia device, characterized by: The method comprises: Step S100: Acquire monitoring data stored by the whole-body thermal therapy device when executing a temperature control adjustment event, wherein the monitoring data records data types and monitoring values; the data types include ECG, blood oxygen concentration SpO2, heart rate HR, core temperature Tcore, and surface temperature of each region; use dynamic time warping to align monitoring data of different sampling frequencies to generate a data array at the same time, and extract characteristic indicators for evaluating the user's status based on the monitoring data; Step S200: Constructing the monitoring data in the data array in the form of a graph structure to generate an association structure containing nodes, edges, and recording node features; the nodes include seven hyperthermia zones and three physiological nodes; based on the constructed graph structure, input monitoring values ​​at the same time to form a data layer that records spatial edge weights and node feature values ​​for each node at different times; Step S300: searching for monitoring data marked as abnormal temperature control events by clinical samples as target data, and outputting the associated abnormal nodes of each hot zone node and the corresponding abnormal temperature control probability model based on the response analysis of the target data; Step S400: Based on step S300, the spatiotemporal correlation features of the abnormal temperature control and adjustment event records are captured, and the spatiotemporal correlation features are associated with the analysis data of the corresponding abnormal temperature control and adjustment event records to generate response feature groups, which are stored in the temperature control monitoring database, and a thermal damage risk score is assigned to each response feature group; Step S500: Acquire real-time monitoring data of the whole-body thermal therapy device and match it with the temperature control monitoring database. When the match is successful, output the corresponding spatiotemporal correlation characteristics and the predicted abnormal temperature control adjustment event response probability; when the match is unsuccessful, return to step S300 and substitute the abnormal temperature control probability model for a new prediction and respond whether to issue an early warning.

2. The intelligent temperature control method for a whole body hyperthermia device according to claim 1, characterized in that: Extracting characteristic indicators for evaluating user status based on monitoring data in step S100 includes the following specific steps: Obtain the time interval between two adjacent heartbeats from ECG monitoring, that is, the time difference RRj-RRj+1 from the R wave peak of the jth heartbeat to the R wave peak of the j+1th heartbeat, and the total number of RR intervals N; use the formula: RMSSD={[1 / (N-1)]*∑(RR j -RR j+1 ) 2 } 1 / 2 ; Calculate RMSSD, a measurement indicator of heart rate variability (HRV) used to evaluate the user's status; Get the monitoring value corresponding to the core temperature Tcore using the formula: Calculate the thermal meter accumulation CEM43(t) of the user to evaluate the user status; Thermal measurement accumulation, heart rate variability, temperature change rate, momentary temperature, heart rate and blood oxygen concentration SpO2 are used as characteristic indicators for evaluating user status; the temperature change rate refers to the temperature change value based on a certain monitoring node per unit time, and the momentary temperature refers to the temperature measured based on a certain monitoring node.

3. The intelligent temperature control method for a whole body hyperthermia device according to claim 2, characterized in that: The step S200 includes the following specific steps: Based on the connection relationship of the physiological structure, each node is connected in sequence through edges, the actual area distance d of adjacent nodes after connection is obtained, and the edge weight k is calculated. d , k d =exp(-d / c), c=10cm; The characteristic indicators of the moment temperature T(t), temperature change rate ΔT(t), heart rate HR(t), blood oxygen concentration SpO2(t) and thermal measurement accumulation CEM43(t) are used to form the node feature f(t) of each node in the corresponding graph structure at time t, f(t) = [T(t), ΔT(t), HR(t), SpO2(t), CEM43(t)]; Then a data layer is generated in which nodes record node features and edges between nodes record spatial edge weights.

4. The intelligent temperature control method for a whole body hyperthermia device according to claim 3, characterized in that: The step S300 includes the following specific steps: The node that initially records the temperature anomaly in each abnormal temperature control adjustment event is obtained as the target node, and all adjacent nodes of the target node in the graph structure are searched as reference nodes using the target node as the initial search node; Based on the node features of the target node and the control node, the weight matrix W is mapped to the eight-dimensional space to obtain W·f 目标 (t) and W·f 对照 (t); where each weight w in the weight matrix ij Set by model training; w ij Indicates the weight value corresponding to the i-th row and j-th column; Calculate the attention coefficient e of the target node and each control node 目标-对照 , e 目标-对照 =LeakyReLU(a T [Wf 目标 (t)||Wf 对照 (t)]), where a T represents the attention vector, "||" represents feature concatenation; The attention coefficient e of the target node and the i-th control node 目标-对照i Normalize to generate attention weight ɑ 目标-对照 i,ɑ 目标-对照 i=e 目标-对照i / ∑(e 目标-对照 );∑(e 目标-对照 ) represents the sum of the attention coefficients of the same target node and all control nodes; calculates the GAT output feature z of the target node 目标 (t), z 目标 (t)=∑[ɑ 目标-对照 *Wf 对照 (t)]; Calculate the LSTM layer: Get the GAT output feature z of all target nodes 目标 (t) is concatenated into an 80-dimensional vector and the hidden state h is obtained. t Update, pass the LSTM hidden state h through the decoder t Restored to input feature f'(t), f'(t) = σ(W d ·h t +b d ), where σ represents the activation function, W d represents the decoder weight matrix, b d Represents the error term; obtain the original input feature f(t), and use the formula: v=||f(t)-f'(t)||2 to calculate the reconstruction error v; Set the attention anomaly judgment condition: when the target node area temperature is greater than g1 degrees Celsius and the control node area temperature is less than g2 degrees Celsius, g1>g2, the output feature value is 1; otherwise the output is 0; obtain the control node that satisfies the output feature value of 1 and has the largest edge weight as the associated abnormal node, and extract the attention weight ɑ of the associated abnormal node record 目标-对照 1. Calculate the attention abnormality u, u = ɑ 目标-对 According to 1*1; and calculate the total abnormality score S(t), S(t) = 0.7*v+0.3*u; Based on the total abnormality score, the abnormal temperature control probability model P(control) is constructed, P(control) = 1 / [1+e -S(t)*3 ].

5. The intelligent temperature control method for a whole body hyperthermia device according to claim 4, characterized in that: The step S300 includes the following specific steps: Obtain monitoring data of the target node and associated abnormal nodes during the monitoring process, calculate the dynamic changes of attention weights through GAT, and record the dynamic change ratio of attention weights; capture temporal feature content through LSTM; The dynamic change ratio and the time characteristic content are used as the response characteristics of the target node and the associated abnormal node, and the output value of the abnormal temperature control probability model constitutes a response feature group and is stored in the temperature control monitoring database; Extract the monitoring data to determine the dynamic change ratio moment, using the formula: HDRS=0.6·CEM43+0.3·MAX(T 目标 -40)+0.1·(100-SpO2) The heat injury risk score HDRS corresponding to the response feature group was calculated.

6. The intelligent temperature control method for a whole body hyperthermia device according to claim 4, characterized in that: The step S500 includes the following specific processes: The successful matching means that the response feature group corresponding to the thermal damage risk score calculated based on the real-time monitoring data is the same as the target node and the associated abnormal node reflected by the real-time monitoring data; the spatiotemporal correlation features recorded in the corresponding response feature group and the abnormal temperature control adjustment event response probability are output; When the matching is unsuccessful, the thermal damage risk score of each node is first calculated, and the thermal damage risk score threshold is set. When the thermal damage risk score is greater than or equal to the thermal damage risk score threshold, the target node is determined and the associated abnormal nodes are further located. The response probability of the abnormal temperature control adjustment event is calculated and the corresponding spatiotemporal correlation characteristics are output in response.

7. An intelligent temperature control system for a whole body hyperthermia apparatus, comprising: The system includes a monitoring data acquisition module, a spatiotemporal alignment module, a characteristic index analysis module, a graph structure construction module, a probability model analysis module, a heat damage risk score calculation module and a real-time monitoring response module; The monitoring data acquisition module is used to acquire the monitoring data stored by the whole body hyperthermia device when performing a temperature control adjustment event; The spatiotemporal alignment module is used to align monitoring data with different sampling frequencies using dynamic time warping to generate data arrays at the same time. The characteristic index analysis module is used to extract characteristic indicators for evaluating user status based on monitoring data; The graph structure building module is used to build the monitoring data in the data array in the form of a graph structure, generating an association structure including nodes, edges and recording node features; The probability model analysis module is used to analyze and output the associated abnormal nodes of each hot zone node and the corresponding abnormal temperature control probability model; The thermal injury risk score calculation module is used to perform a thermal injury risk score on each response feature group; The real-time monitoring response module is used to output the corresponding spatiotemporal correlation characteristics and the predicted abnormal temperature control adjustment event response probability when the match is successful; when the match is unsuccessful, return to step S300 to substitute the abnormal temperature control probability model for a new prediction and respond whether to issue an early warning.

8. The intelligent temperature control system for whole body hyperthermia therapy apparatus according to claim 7, characterized in that: The probability model analysis module includes a node type determination unit, an attention weight calculation unit, a reconstruction error calculation unit, a total anomaly score calculation unit and an abnormal temperature control probability model construction unit; The node type determination unit is used to determine the target node and the reference node; The attention weight calculation unit is used to calculate the attention weight of each target node and all corresponding control nodes; The reconstruction error calculation unit is used for the reconstruction error of each target node; The total anomaly score calculation unit is used to set the attention anomaly item determination conditions and calculate the total anomaly score based on the satisfied nodes; The abnormal temperature control probability model construction unit is used to construct an abnormal temperature control probability model based on the total abnormality score.

Images