Method and system for predicting body temperature trend during anesthesia process
By combining vascular dynamics and metabolic environment characteristics with a phase-aware time fusion transformer, the focus on features is dynamically adjusted, which solves the problem of the two-stage time-varying characteristics of perioperative hypothermia during anesthesia, and achieves accurate prediction of body temperature trends, avoiding the shortcomings of traditional models.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING QICHENG MEDICAL TECHNOLOGY CO LTD
- Filing Date
- 2025-12-29
- Publication Date
- 2026-04-21
AI Technical Summary
Existing deep learning prediction models cannot accurately fit the two-stage time-varying characteristics of perioperative hypothermia during anesthesia, resulting in an underestimation of the cooling magnitude during induction and an overestimation of the cooling rate during maintenance, and they cannot capture the heat redistribution initiation point in real time.
A phase-aware time fusion transformer is used to identify the anesthesia induction time, process the vascular dynamics and metabolic environment features in the physiological monitoring data in groups, dynamically adjust the focus of feature attention, and achieve accurate prediction of body temperature trends.
It enables accurate capture of the onset of heat redistribution without the need for anesthetic drug injection recording, avoiding the underestimation and false alarms of traditional models, providing stable prediction of body temperature trends, and has high interpretability and low deployment cost.
Smart Images
Figure CN121890962A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for predicting intraoperative body temperature trends during anesthesia. Background Technology
[0002] The mechanism of perioperative hypothermia (IPH) exhibits significant two-stage time-varying characteristics, differing markedly from the fluctuation patterns of routine physiological indicators such as blood pressure and heart rate. Specifically, it is divided into a heat redistribution phase and a linear heat dissipation phase. The heat redistribution phase typically occurs 0-60 minutes after anesthesia induction, with core body temperature rapidly decreasing by 1.0-1.5°C due to peripheral vasodilation caused by anesthetic drugs. Vascular tone parameters are the core influencing factors during this phase. The linear heat dissipation phase usually begins around 60 minutes after induction, with body temperature decreasing slowly and linearly. This is mainly due to insufficient metabolic heat production to offset heat loss from the environment, with metabolic rate and the temperature difference between the body and the environment being the core influencing factors.
[0003] Existing deep learning prediction models (such as standard LSTM and TFT) use global static weights to process features throughout the entire time process, paying the same level of attention to features at different stages. This results in a "phase agnostic" defect, leading to underestimation of the precipitous temperature drop during the induction phase and overestimation of the temperature drop rate during the maintenance phase, causing false alarms. At the same time, the lag in clinical recording makes it impossible for the model to obtain the anesthetic drug injection time in real time, further increasing the difficulty of capturing the heat redistribution initiation point and making it difficult to accurately fit the two-stage body temperature curve.
[0004] Therefore, there is an urgent need for a method to predict intraoperative body temperature trends during anesthesia that can sense the phase of the surgical process and automatically switch the focus of attention. Summary of the Invention
[0005] In view of this, the present invention proposes a method and system for predicting the intraoperative body temperature trend during the anesthesia process, which can accurately predict the intraoperative body temperature trend during the anesthesia process.
[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for predicting intraoperative body temperature trends during anesthesia, comprising: Collect physiological monitoring data during clinical anesthesia; The anesthesia induction time was identified based on the physiological monitoring data; Features are extracted from the physiological monitoring data, and the extracted features are grouped to obtain vascular dynamics feature group and metabolic environment feature group; The anesthesia induction time, vasomotor features, and metabolic environment features are input into a pre-trained phase-aware time fusion transformer to obtain body temperature prediction results for multiple future time steps.
[0007] Based on the above technical solution, the present invention can be further improved as follows: Optionally, identifying the anesthesia induction time based on the physiological monitoring data includes: The hemodynamic parameters of physiological monitoring data are monitored in real time. When the instantaneous change of the hemodynamic parameters exceeds the preset mutation threshold and the duration of the change reaches the preset duration threshold, the moment when the hemodynamic parameters undergo a significant mutation and the consciousness index changes synchronously is marked as the logical anesthesia induction moment.
[0008] Optionally, the vascular dynamics feature set includes mean arterial pressure, heart rate, pulse pressure variability, and perfusion index; The metabolic environment characteristics group includes age, body mass index, patient preoperative health status classification standards, and anesthesia method.
[0009] Optionally, the phase-aware time fusion transformer includes a phase encoding module, a feature weighting module, and a timing processing module; The phase encoding module is used to convert the anesthesia induction time into a phase vector; The feature weighting module is used to concatenate the vascular dynamics features, metabolic environment features, and phase vector to obtain concatenated features, generate feature weights for the concatenated features, and adjust the information throughput of the vascular dynamics feature group and the metabolic environment feature group based on the feature weights to obtain a weighted feature sequence. The time-series processing module is used to extract the time-series features of the weighted feature sequence, output the body temperature prediction results for multiple future time steps based on the time-series features, and extract the extreme values of body temperature within different time windows from the body temperature prediction results.
[0010] Optionally, converting the anesthesia induction time into a phase vector via a phase encoding module includes: Calculate the relative time between the current moment and the time since the anesthesia induction. Discretization embedding based on preset rules, sinusoidal position encoding, or learnable embedding layers can map relative duration into a high-dimensional phase vector.
[0011] Optionally, the feature weights for generating the spliced features include: When the phase vector indicates the heat redistribution period, the weights of the hemodynamic feature groups are automatically amplified by the gating unit; When the phase vector indicates a linear heat dissipation period, the weights of the metabolic environment feature group are automatically amplified by the gating unit.
[0012] Optionally, during the thermal redistribution period, the phase-aware time fusion converter is configured in a high dynamic response mode. During the heat dissipation period, the phase-aware time fusion converter is configured to a smooth hold mode.
[0013] A system for predicting intraoperative body temperature trends during anesthesia includes: The data acquisition module is used to collect physiological monitoring data during clinical anesthesia. The time recognition module is used to identify the anesthesia induction time based on the physiological monitoring data; The feature grouping module is used to extract features from the physiological monitoring data and group the extracted features to obtain vascular dynamics feature group and metabolic environment feature group. The prediction module is used to input the anesthesia induction time, vasodynamic features, and metabolic environment features into a pre-trained phase-aware time fusion transformer to obtain body temperature prediction results for multiple future time steps.
[0014] An electronic device includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the method described herein.
[0015] A non-transitory computer-readable storage medium having a computer program stored thereon, the computer program implementing the steps of the method when executed by a processor.
[0016] The present invention has the following advantages: The method for predicting intraoperative body temperature trends during anesthesia in this invention utilizes a phase-sensing time fusion transformer that combines anesthesia induction time, vasodynamic features, and metabolic environment features. This allows for precise adaptation to the two-stage time-varying characteristics of perioperative hypothermia, enabling accurate prediction of future body temperature trends across multiple time steps. It eliminates the need for anesthetic drug injection records, capturing the onset of heat redistribution through physiological monitoring data, effectively avoiding the shortcomings of traditional models that underestimate the degree of temperature drop during induction and issue false warnings during maintenance. Attached Figure Description
[0017] For illustrative and not limiting purposes, the present invention will now be described in conjunction with embodiments and accompanying drawings, wherein: Figure 1 This is a flowchart illustrating the method for predicting intraoperative body temperature trends during anesthesia in an embodiment of the present invention. Figure 2 A schematic diagram of the main components of the system for predicting intraoperative body temperature trends during anesthesia in an embodiment of the present invention; Figure 3 This is a schematic diagram of the physical structure of the electronic device provided by the present invention. Detailed Implementation
[0018] To enable those skilled in the art to better understand the present invention, 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 a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0019] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be used interchangeably where appropriate for the embodiments of the invention described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0020] It should be noted that, where there is no conflict, the embodiments and features of the present invention can be combined with each other. The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0021] Figure 1 This is a flowchart illustrating the method for predicting intraoperative body temperature trends during anesthesia according to an embodiment of the present invention. Figure 1 As shown, the method for predicting intraoperative body temperature trends during anesthesia provided in this embodiment of the invention includes the following steps S101 to S104.
[0022] S101, collects physiological monitoring data during clinical anesthesia.
[0023] Physiological monitoring data during clinical anesthesia is collected using a monitor. This physiological monitoring data includes, but is not limited to, hemodynamic data (mean arterial pressure, heart rate, pulse pressure, etc.), consciousness index data, and basic physiological information data. Redundant and interfering data, such as records of human events, are simultaneously removed during the collection process to ensure data integrity and accuracy.
[0024] S102, identifies the anesthesia induction time based on physiological monitoring data.
[0025] Real-time monitoring of hemodynamic parameters (such as mean arterial pressure or heart rate) of physiological monitoring data. When the instantaneous change of the hemodynamic parameters exceeds the preset mutation threshold and the duration of the change reaches the preset duration threshold, the moment when the hemodynamic parameters undergo a significant mutation and the consciousness index changes synchronously is marked as the logical anesthesia induction time T0.
[0026] S103 extracts features from physiological monitoring data and groups the extracted features to obtain vascular dynamics feature group and metabolic environment feature group.
[0027] The vascular dynamics feature set includes mean arterial pressure (MAP), heart rate (HR), pulse pressure variability (PPV), and perfusion index (PI). The metabolic environment characteristics group includes age (Age), body mass index (BMI), patient preoperative health status classification standard (ASA), and anesthesia method (ANES_TYPE).
[0028] S104 inputs the anesthesia induction time, vasomotor features, and metabolic environment features into a pre-trained phase-aware time fusion transformer to obtain body temperature prediction results for multiple future time steps.
[0029] The phase-aware time fusion transformer includes a phase encoding module, a feature weighting module, and a timing processing module; The anesthesia induction time is converted into a phase vector using a phase encoding module; The phase vector provides explicit context information (Context) for the phase-aware time fusion transformer, indicating which stage of the surgery is currently underway.
[0030] Calculate the relative time between the current moment and the time since the anesthesia induction. Discretization embedding based on preset rules, sinusoidal position encoding, or learnable embedding layers can map relative duration into a high-dimensional phase vector.
[0031] The vascular dynamics feature group, metabolic environment feature group and phase vector are concatenated by the feature weighting module to obtain concatenated features. Feature weights of the concatenated features are generated. Based on the feature weights, the information throughput of the vascular dynamics feature group and the metabolic environment feature group is adjusted to obtain a weighted feature sequence. The Phase-Aware Temporal Fusion Transformer (PA-TFT) improves the Variable Selection Network (VSN) in the TFT architecture by introducing a phase-gated unit.
[0032] The model dynamically weights the input features, and the weights are generated not only based on the features themselves but also conditionally on the phase vector. .
[0033] The feature weights for generating the spliced features include: When the phase vector indicates the heat redistribution period, the weights of hemodynamic feature groups (such as pulse pressure variability rate, perfusion index, and heart rate HR) are automatically amplified by the gating unit; because at this time, the increase in PPV means vasodilation, which directly drives the cooling caused by heat redistribution.
[0034] When the phase vector indicates a linear heat dissipation period, the weights of metabolic environmental characteristics (such as BMI, age, ASA classification, and anesthesia method) are automatically amplified by the gating unit. This is because heat loss at this time is mainly controlled by metabolic rate and ambient temperature difference.
[0035] In the formula, For feature weights, For splicing features, The information throughput of the vascular dynamics feature group and the metabolic environment feature group; The time series processing module extracts the time series features of the weighted feature sequence, outputs the body temperature prediction results for multiple future time steps based on the time series features, and extracts the extreme values of body temperature within different time windows from the body temperature prediction results.
[0036] The model's output layer contains prediction heads for short, medium, and long time windows. During training, a phase-adaptive trend regularization mechanism is introduced. Multidimensional time series output: The model synchronously outputs the predicted body temperature values for multiple discrete time steps (such as 3, 5, and 10 minutes) to form a complete future body temperature trajectory.
[0037] During the thermal redistribution phase, the phase-aware time fusion transformer is configured in a high dynamic response mode, allowing the predicted trajectory to have a large gradient to keenly capture the rapid cooling caused by vasodilation. During the heat dissipation period, the phase-aware time fusion transformer is configured in smooth hold mode. The weights of high-frequency fluctuation features are suppressed through the attention mechanism, or the penalty for the second derivative of the trajectory is added to the loss function, thereby ensuring that the prediction curve exhibits stable linear decay characteristics.
[0038] Through the above mechanism, the predicted curve output by the model is forced to strictly approximate the "rapid at first and then slow" physiological manifold in its overall shape, thus avoiding the problem of false fluctuations and false alarms in the maintenance period of traditional models.
[0039] To accurately predict “heat redistribution” without drug administration records, this invention constructs vasomotor surrogate features.
[0040] Pulse pressure and shock index are calculated using high-frequency arterial pressure waveforms.
[0041] Logical reasoning: Although the model doesn't know when propofol was administered, the vasodilation caused by propofol is reflected in real time through a combination of "widened pulse pressure" or "decreased blood pressure + insufficient heart rate compensation." Combined with the aforementioned "phase gating," the model can highly sensitively capture these changes in vascular tension in the early stages of surgery, thereby accurately predicting the subsequent sudden drop in body temperature.
[0042] Training the phase-aware time fusion transformer; Stratified sampling: Ensure that the training data contains a balanced mix of samples from the "induction period" and the "maintenance period" to prevent the model from being biased towards learning the patterns of a particular stage.
[0043] Loss function: Quantile loss is used, with a focus on optimizing the prediction of the lower bound of body temperature.
[0044] This invention utilizes a phase-sensing mechanism, allowing the model to "know" that it is currently in the heat redistribution phase. Therefore, it intelligently uses vasodilation indicators such as PPV (percutaneous pulse volume) to estimate the magnitude of body temperature decrease. This solves the problem of delayed early body temperature prediction caused by traditional models neglecting vascular factors.
[0045] The dynamic attention weights of this invention are highly consistent with the pathophysiological mechanisms (looking at blood vessels in the early stages and metabolism in the later stages), which makes the model's prediction logic highly interpretable and easily accepted by clinicians.
[0046] This invention does not require inputting specific anesthetic types and dosages. It can accurately infer the thermal effects of drugs by relying solely on the monitored "vascular tension characteristics" combined with "time phase," thus adapting to real surgical environments where information flow is incomplete.
[0047] This invention's single model solves the prediction problem of both the rapid cooling period and the stable heat dissipation period simultaneously through a smooth phase transition, eliminating the need to train multiple sub-models for different stages and resulting in low deployment and maintenance costs.
[0048] Figure 2This is a schematic diagram of the main components of the system for predicting intraoperative body temperature trends during anesthesia in an embodiment of the present invention. Figure 2 As shown, the system 1 for predicting intraoperative body temperature trends during anesthesia provided in this embodiment of the invention includes a data acquisition module 10, a time identification module 20, a feature grouping module 30, and a prediction module 40.
[0049] Data acquisition module 10 is used to collect physiological monitoring data during clinical anesthesia. Time recognition module 20 is used to identify the anesthesia induction time based on the physiological monitoring data; The feature grouping module 30 is used to extract features from the physiological monitoring data and group the extracted features to obtain a vascular dynamics feature group and a metabolic environment feature group. The prediction module 40 is used to input the anesthesia induction time, vasodynamic features and metabolic environment features into a pre-trained phase-aware time fusion transformer to obtain the body temperature prediction results for multiple future time steps.
[0050] Figure 3 This is a schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention, such as... Figure 3 As shown, the electronic device 50 includes: a processor 501, a memory 502, and a bus 503; The processor 501 and the memory 502 communicate with each other via the bus 503. The processor 501 is used to call program instructions in the memory 502 to execute the methods provided in the above-described method embodiments, and to execute the methods provided in the embodiments of the present invention.
[0051] This embodiment provides a non-transitory computer-readable storage medium that stores computer instructions, which cause a computer to execute the method provided in this embodiment of the invention.
[0052] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various storage media capable of storing program code, such as ROM, RAM, magnetic disk, or optical disk.
[0053] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.
Claims
1. A method for predicting intraoperative body temperature trends during anesthesia, characterized in that, include: Collect physiological monitoring data during clinical anesthesia; The anesthesia induction time was identified based on the physiological monitoring data; Features are extracted from the physiological monitoring data, and the extracted features are grouped to obtain vascular dynamics feature group and metabolic environment feature group; The anesthesia induction time, vasomotor features, and metabolic environment features are input into a pre-trained phase-aware time fusion transformer to obtain body temperature prediction results for multiple future time steps.
2. The method for predicting intraoperative body temperature trends during anesthesia according to claim 1, characterized in that, The identification of the anesthesia induction time based on the physiological monitoring data includes: The hemodynamic parameters of physiological monitoring data are monitored in real time. When the instantaneous change of the hemodynamic parameters exceeds the preset mutation threshold and the duration of the change reaches the preset duration threshold, the moment when the hemodynamic parameters undergo a significant mutation and the consciousness index changes synchronously is marked as the logical anesthesia induction moment.
3. The method for predicting intraoperative body temperature trends during anesthesia according to claim 1, characterized in that, The vascular dynamics feature set includes mean arterial pressure, heart rate, pulse pressure variability, and perfusion index; The metabolic environment characteristics group includes age, body mass index, patient preoperative health status classification standards, and anesthesia method.
4. The method for predicting intraoperative body temperature trends during anesthesia according to claim 1, characterized in that, The phase-aware time fusion transformer includes a phase encoding module, a feature weighting module, and a timing processing module; The phase encoding module is used to convert the anesthesia induction time into a phase vector; The feature weighting module is used to concatenate the vascular dynamics features, metabolic environment features, and phase vector to obtain concatenated features, generate feature weights for the concatenated features, and adjust the information throughput of the vascular dynamics feature group and the metabolic environment feature group based on the feature weights to obtain a weighted feature sequence. The time-series processing module is used to extract the time-series features of the weighted feature sequence, output the body temperature prediction results for multiple future time steps based on the time-series features, and extract the extreme values of body temperature within different time windows from the body temperature prediction results.
5. The method for predicting intraoperative body temperature trends during anesthesia according to claim 4, characterized in that, The step of converting the anesthesia induction time into a phase vector through a phase encoding module includes: Calculate the relative time between the current moment and the time since the anesthesia induction. Discretization embedding based on preset rules, sinusoidal position encoding, or learnable embedding layers can map relative duration into a high-dimensional phase vector.
6. The method for predicting intraoperative body temperature trends during anesthesia according to claim 4, characterized in that, The feature weights for generating the spliced features include: When the phase vector indicates the heat redistribution period, the weights of the hemodynamic feature groups are automatically amplified by the gating unit; When the phase vector indicates a linear heat dissipation period, the weights of the metabolic environment feature group are automatically amplified by the gating unit.
7. The method for predicting intraoperative body temperature trends during anesthesia according to claim 1, characterized in that, During the thermal redistribution period, the phase-aware time fusion converter is configured in a high dynamic response mode. During the heat dissipation period, the phase-aware time fusion converter is configured to a smooth hold mode.
8. A system for predicting intraoperative body temperature trends during anesthesia, characterized in that, include: The data acquisition module is used to collect physiological monitoring data during clinical anesthesia. The time recognition module is used to identify the anesthesia induction time based on the physiological monitoring data; The feature grouping module is used to extract features from the physiological monitoring data and group the extracted features to obtain vascular dynamics feature group and metabolic environment feature group. The prediction module is used to input the anesthesia induction time, vasodynamic features, and metabolic environment features into a pre-trained phase-aware time fusion transformer to obtain body temperature prediction results for multiple future time steps.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.
10. A non-transitory computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.