Perioperative period hypothermia real-time prediction model based on multi-modal data and deep learning

Abstract

The invention relates to the technical field of real-time body temperature prediction, in particular to a perioperative low-body-temperature real-time prediction model based on multi-modal data and deep learning, which is characterized in that a large amount of performed operation data is collected in advance by establishing an intelligent auxiliary system, and the data comprises preoperative information, intraoperative information and postoperative data; the health condition of the patient is comprehensively analyzed; after collected preoperative health data of a patient, historical operation records and ASA classification standards are combined through the system, the probability of occurrence of hypothermia of the patient in the operation process can be accurately predicted, a risk score or probability of occurrence of hypothermia can be provided, doctors can be helped to consider the factor when making an operation plan, and the probability of occurrence of hypothermia can be accurately predicted. By combining the environment data with the real-time data of the physical signs of the patient, the occurrence risk of the hypothermia in the operation of the patient can be better evaluated, a doctor can be assisted in adjusting the warm keeping measures in the operation in time, and a safer and more comfortable operation environment is provided for the patient.

Description

Technical Field

[0001] The present invention relates to the technical field of real-time body temperature prediction, and in particular to a real-time perioperative hypothermia prediction model based on multimodal data and deep learning. Background Art

[0002] With the development of social economy, the development of medical resources has become more in-depth and extensive, such as clinical practice guidelines and recommendations for human body temperature, temperature management goals and operating procedures. Currently, if patients experience perioperative hypothermia (perioperative hypothermia is defined as a core temperature less than 36°C), it will cause a higher rate of complications and bring many adverse consequences.

[0003] After searching, Chinese patent CN111248871A discloses a perioperative hypothermia prediction method. The method constructs a hypothermia prediction model based on surgery type parameters, intravenous input volume parameters, anesthesia time parameters, warming method parameters, BMI parameters, surgical environment temperature parameters, and preoperative core body temperature parameters.

[0004] The perioperative hypothermia prediction method is to construct a hypothermia prediction model based on the surgery type parameters, surgical environment temperature parameters and preoperative core body temperature parameters. The traditional analysis system does not record the physical constitution of different patients in advance before surgery, resulting in the prediction model being unable to take this factor into account when predicting surgery, thereby affecting the prediction results. To this end, the present invention proposes a real-time perioperative hypothermia prediction model based on multimodal data and deep learning. By collecting comprehensive patient information and surgical monitoring parameters before surgery, a hypothermia prediction model is constructed to predict the probability of hypothermia occurring in patients during surgery. Summary of the Invention

[0005] Technical problem solved: The physical constitution of different patients was not recorded in advance before surgery, resulting in the prediction model being unable to take this factor into account when predicting surgery, thus affecting the prediction results.

[0006] In response to the shortcomings of the existing technology, the present invention provides a real-time prediction model for perioperative hypothermia based on multimodal data and deep learning, thereby solving the technical problems mentioned in the background technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0008] The following steps are involved:

[0009] S1. Patient information and multimodal data acquisition module:

[0010] Establish an intelligent assistance system to collect a large amount of surgical data in advance, including preoperative, intraoperative and postoperative information, and comprehensively analyze the patient's health status;

[0011] The system uses the ASA (American Society of Anesthesiologists) classification criteria to categorize patients based on their health information and the type of surgery they are undergoing. By combining the collected preoperative health data and historical surgical records with the ASA classification criteria, the system can more accurately predict the probability of hypothermia during surgery, helping doctors consider this factor when formulating surgical plans.

[0012] Before surgery, temperature sensors are placed on the patient's body at various locations besides the surgical site. These sensors monitor the patient's body temperature in real time, providing a comprehensive understanding of temperature fluctuations. In combination with a multi-parameter monitor, the patient's key physiological indicators are monitored during surgery. This data provides a scientific basis for timely decision-making during surgery, particularly for early warning of hypothermia.

[0013] Temperature and humidity sensors are installed in the operating room to monitor environmental conditions in real time, especially changes in temperature and humidity. This environmental data, combined with real-time data on patients' vital signs, can better assess the risk of intraoperative hypothermia, assisting doctors in timely adjusting intraoperative warming measures and providing patients with a safer and more comfortable surgical environment.

[0014] S2, data preprocessing module:

[0015] Process detected missing values ​​and outliers, and filter noisy data to improve data accuracy and quality; ensure that data from different sensors can be compared and analyzed under the same standards, and standardize or normalize these data to improve the effectiveness of data analysis or model training;

[0016] Time-align the multimodal data from multiple sensors when merging and processing them to ensure temporal consistency, avoid data distortion or prediction errors caused by time lag or asynchrony, and improve the subsequent prediction accuracy and real-time response capabilities of the model system.

[0017] S3. Selection and construction of deep learning models:

[0018] Using recurrent neural networks (RNNs) to process time-series multimodal data can effectively establish the complex dependencies between physiological signals such as body temperature, ECG, and blood oxygen levels over time. The long short-term memory (LSTM) network in deep learning can address the vanishing gradient problem of traditional RNNs when capturing long-span temporal information, thereby accurately modeling the temporal dynamics of real-time intraoperative data.

[0019] By using ensemble learning methods to fuse the prediction results of multiple models, we can improve prediction accuracy, explore the potential correlations between different data sources, and provide more comprehensive and accurate information for subsequent decision support systems, thereby providing a solid theoretical foundation and practical support for clinical prediction and decision-making.

[0020] S4. Model training, evaluation and deployment:

[0021] In the application scenario of perioperative hypothermia prediction, clinicians can annotate the data based on historical patient temperature data before, during, and after surgery, combined with clinical observation results. When the patient's temperature drops below a preset threshold (such as 36°C) during surgery, it can be determined that a hypothermia event has occurred.

[0022] Deploy the trained model into the hospital's real-time surgical environment to ensure timely response to hypothermia. The deployed model should be able to receive and process the patient's current temperature data in real time and continuously update hypothermia predictions based on this data. When temperature data changes, the model can immediately predict the occurrence of hypothermia and alert the doctor whether appropriate intraoperative warming measures are needed.

[0023] S5. Model optimization:

[0024] After perioperative hypothermia prediction is completed, the model's performance is evaluated by comparing the model's prediction results with actual hypothermia events, and the model is further optimized based on the evaluation results; through an iterative feedback mechanism, the model continuously learns and adjusts to improve the accuracy and reliability of predictions in subsequent applications; through continuous optimization and adjustment, more accurate and effective hypothermia predictions can ultimately be achieved, improving the application effect of the model in actual surgical scenarios.

[0025] In one possible implementation, the ASA classification system divides patients into different risk levels based on their health status. ASA Class I represents healthy patients, while ASA Class IV and V represent patients in poor health. Based on these classifications, doctors can determine the different risks patients may face during surgery, especially the probability of hypothermia. The core goal of the ASA classification is to help doctors assess the patient's preoperative condition so that they can make more reasonable anesthesia and surgical plans.

[0026] In one possible implementation, the missing values ​​should be filled or deleted through reasonable methods, outliers need to be detected and removed through statistical methods, and noise data should be filtered through appropriate filtering techniques, such as low-pass filtering or median filtering, to remove unnecessary interference and ensure that the data is purer; standardization or normalization can make data from different sources, scales, and units in the same range, eliminating inconsistencies caused by differences in sensor output.

[0027] In one possible implementation, the continuity of the time dimension and the temporal relationship of the data are very important during the model training process. Cross-validation is used during the training process. Especially in the scenario of time series data, cross-validation helps ensure that the training set and the validation set have consistent distribution characteristics in the time dimension, avoiding model overfitting or data leakage problems caused by inconsistent time spans or improper data division.

[0028] Beneficial effects compared with existing technologies:

[0029] 1. In this solution, by collecting a large amount of data on existing surgeries and patients' health status in advance, and combining the collected preoperative health data and historical surgical records with the ASA classification standards through the system, the probability of hypothermia occurring in patients during surgery can be accurately predicted, providing a risk score or probability of hypothermia, which can help doctors take this factor into account when formulating surgical plans, thereby achieving a preliminary prediction effect of perioperative hypothermia in patients. By wearing temperature sensors and multi-parameter monitors on multiple parts of the patient and using them in conjunction with temperature and humidity sensors installed in the operating room, and by combining environmental data with real-time data of patients' vital signs, the risk of hypothermia occurring during surgery can be better assessed, assisting doctors in timely adjusting intraoperative warming measures, and providing patients with a safer and more comfortable surgical environment.

[0030] 2. In this solution, a recurrent neural network (RNN) is used to process the monitored multimodal data. Using ensemble learning methods, the prediction results of multiple models are integrated to effectively improve prediction accuracy, providing a solid theoretical foundation and practical support for clinical prediction and decision-making. Clinicians use preoperative, intraoperative, and postoperative temperature data from patients, combined with clinical observations, to label the data. When a patient's temperature drops below a preset threshold during surgery, a hypothermia event can be identified.

[0031] 3. In this solution, the trained model is deployed in the hospital's real-time surgical environment. The deployed model should be able to receive and process the patient's current temperature data in real time, and continuously update the hypothermia prediction results based on this data. When the temperature data changes, the model can immediately predict the occurrence of hypothermia and remind the doctor whether to take appropriate intraoperative warming measures, thereby effectively preventing further aggravation of hypothermia and ensuring the safety of the patient. It can provide doctors with timely decision support, auxiliary reminders, and optimized intraoperative patient management. Through an iterative feedback mechanism, the model can continuously learn and adjust to improve the accuracy and reliability of the prediction in subsequent applications, ultimately achieving more accurate and effective hypothermia prediction and improving the application effect of the model in actual surgical scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention with reference to the accompanying drawings.

[0033] Figure 1 It is a schematic diagram of the process structure of the present invention; DETAILED DESCRIPTION

[0034] In the description of the present invention, it should be understood that the terms "center", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "axial", "radial", "circumferential", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention.

[0035] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature identified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0036] In the present invention, unless otherwise clearly specified and limited, terms such as "install", "connect", "connect", and "fix" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integrated connection; it can be a mechanical connection, an electrical connection, or a communication connection; it can be a direct connection or an indirect connection through an intermediate medium, and it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0037] In the present invention, unless otherwise clearly specified and limited, a first feature "above" or "below" a second feature may be such that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. In the description of this specification, the descriptions with reference to the terms "one scheme", "some schemes", "examples", "specific examples", or "some examples" etc. mean that the specific features, structures, materials or characteristics described in conjunction with the scheme or example are included in at least one scheme or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same scheme or example. Moreover, the specific features, structures, materials or characteristics described may be combined in an appropriate manner in any one or more schemes or examples;

[0038] In order to more clearly and completely illustrate the technical solution of the present invention, the present invention will be further described below with reference to the accompanying drawings:

[0039] Please refer to Figure 1 As shown, this embodiment introduces a real-time prediction model for perioperative hypothermia based on multimodal data and deep learning, including the following steps:

[0040] S1. Patient information and multimodal data acquisition module:

[0041] Establish an intelligent assistance system to collect a large amount of surgical data in advance. This data includes preoperative, intraoperative, and postoperative information, and comprehensively analyze the patient's health status. Combined with the type of surgery and the patient's specific physical condition, the risk of hypothermia can be systematically assessed. First, before the patient undergoes surgery, detailed preoperative information should be collected, including but not limited to the patient's basic information, as well as more in-depth health status data. In addition, physical data such as body mass index (BMI), metabolic rate, and blood test results should also be included in the collection scope as key data to fully understand the patient's physical condition.

[0042] The ASA (American Society of Anesthesiologists) classification system is used to classify patients based on their health information and the type of surgery. This classification system divides patients into different risk levels based on their health status. ASA Class I represents healthy patients, while ASA Class IV and Class V represent patients in poor health. Based on these classifications, doctors can determine the different risks patients may face during surgery, especially the probability of hypothermia. The core goal of the ASA classification is to help doctors assess the patient's preoperative condition in order to make more reasonable anesthesia and surgical plans. By combining the collected patient's preoperative health data and historical surgical records with the ASA classification standards, the system can more accurately predict the probability of hypothermia in patients during surgery. By providing a risk score or probability of hypothermia, doctors can take this factor into account when formulating surgical plans.

[0043] Before surgery, to ensure effective temperature monitoring, temperature sensors can be worn on multiple locations on the patient's body, in addition to the surgical site. These sensors can monitor the temperature of key body parts in real time, providing a comprehensive understanding of temperature fluctuations. At the same time, a multi-parameter monitor can be used to monitor key intraoperative physiological indicators, including electrocardiogram (ECG), respiratory rate, heart rate, and blood oxygen saturation (SpO2), ensuring a comprehensive understanding of the patient's vital signs. These monitoring data will provide a scientific basis for timely decision-making during surgery, especially early warning of hypothermia.

[0044] Temperature and humidity sensors are installed in the operating room to monitor environmental conditions in real time, especially changes in temperature and humidity. This environmental data, combined with real-time data on patients' vital signs, can better assess the risk of intraoperative hypothermia and assist doctors in timely adjusting intraoperative warming measures, providing patients with a safer and more comfortable surgical environment.

[0045] S2, data preprocessing module:

[0046] Process the detected missing values ​​and outliers, and filter the noisy data to improve the accuracy and quality of the data; missing values ​​should be filled or deleted through reasonable methods, outliers need to be detected and removed through statistical methods, and noisy data should be filtered through appropriate filtering techniques, such as low-pass filtering or median filtering, to remove unnecessary interference and ensure that the data is purer; to ensure that data from different sensors can be compared and analyzed under the same standards, these data must be standardized or normalized; standardization or normalization can make data from different sources, scales, and units in the same range, eliminating the inconsistency caused by differences in sensor output, thereby improving the effectiveness of data analysis or model training;

[0047] When merging and processing multimodal data from multiple sensors, time alignment must be performed. Data collected by all sensors should be guaranteed to have the same timestamp. Especially when processing dynamic data streams, each piece of data must maintain consistent time synchronization with other data to ensure data consistency in time and avoid data distortion or prediction errors caused by time lag or asynchrony, thereby improving the subsequent prediction accuracy and real-time responsiveness of the model system.

[0048] S3. Selection and construction of deep learning models:

[0049] Using recurrent neural networks (RNNs) to process time-series multimodal data can effectively establish the complex dependencies between physiological signals such as body temperature, ECG, and blood oxygen levels over time. The long short-term memory (LSTM) network in deep learning can address the vanishing gradient problem of traditional RNNs when capturing long-span temporal information, thereby accurately modeling the temporal dynamics of real-time intraoperative data.

[0050] By using ensemble learning methods to fuse the prediction results of multiple models, we can not only improve prediction accuracy but also explore the potential correlations between different data sources, providing more comprehensive and accurate information for subsequent decision support systems, thereby providing a solid theoretical foundation and practical support for clinical prediction and decision-making.

[0051] S4. Model training, evaluation and deployment:

[0052] In the application scenario of perioperative hypothermia prediction, clinicians can annotate the data based on the patient's preoperative, intraoperative and postoperative temperature data, combined with clinical observation results; when the patient's temperature drops below a preset threshold (such as 36°C) during the operation, it can be determined that a hypothermia event has occurred; at this time, the annotated data will become the key input for training the model; since the dataset used is usually presented as time series data, containing the dynamic physiological changes of the patient during the operation, the continuity of the time dimension and the temporal relationship of the data are very important in the model training process; cross-validation can be used during training, especially in the scenario of time series data, cross-validation helps to ensure that the training set and the validation set have consistent distribution characteristics in the time dimension, thereby avoiding model overfitting or data leakage problems caused by inconsistent time spans or improper data division; a reasonable cross-validation strategy can effectively improve the performance of the model on unknown data, ensuring the robustness and accuracy of the model;

[0053] Whether a model can accurately capture the occurrence of hypothermia depends on its prediction accuracy and recognition ability. When the model predicts hypothermia, whether the prediction is consistent with the actual hypothermia situation is usually measured by sensitivity or recall, indicating the proportion of hypothermia cases correctly predicted by the model. Whether the model correctly predicts non-hypothermia cases focuses on the model's accurate prediction of non-hypothermia events, that is, whether the model can correctly judge non-hypothermia in the absence of hypothermia. This evaluation metric is usually measured by specificity or accuracy, indicating the model's ability to correctly identify non-hypothermia events. During the model evaluation process, in addition to focusing on the accurate prediction rate of hypothermia, the model must also focus on the model's correct recognition of non-hypothermia situations. The combination of these two aspects of evaluation can fully reflect the model's predictive ability in actual applications, ensuring that it can not only promptly detect hypothermia risks in actual clinical operations, but also avoid false alarms, ensuring the accuracy and reliability of predictions.

[0054] The purpose of deploying the trained model into the hospital's real-time surgical environment is to ensure timely response to hypothermia, especially considering that hypothermia may occur rapidly and worsen in a short period of time during surgery. The deployed model should be able to receive and process the patient's current temperature data in real time and continuously update the hypothermia prediction results based on this data. When the temperature data changes, the model can immediately predict the occurrence of hypothermia and remind the doctor whether to take appropriate intraoperative warming measures. Through this real-time prediction and feedback mechanism, the doctor can respond in time and adjust the warming measures or take other necessary intervention measures according to the prompts of the model, thereby effectively preventing the further aggravation of hypothermia, ensuring the safety of the patient, and achieving the effect of providing timely decision support, auxiliary reminders and optimized intraoperative patient management for the doctor.

[0055] S5. Model optimization:

[0056] After perioperative hypothermia prediction is completed, the model's performance can be evaluated by comparing the model's prediction results with actual hypothermia events, and the model can be further optimized based on the evaluation results. Specifically, by comparing the predicted results with the actual results, it not only helps to identify deviations and errors in the model's predictions, but also provides feedback information for the model, guiding the model to improve in parameter adjustment, feature selection, and model architecture.

[0057] Through this iterative feedback mechanism, the model can continuously learn and adjust to improve the accuracy and reliability of predictions in subsequent applications. This process usually includes specific optimization based on the situation of erroneous predictions, such as adjusting model hyperparameters, introducing more features, optimizing data preprocessing, etc., so that the model is more adapted to the needs of hypothermia prediction in actual clinical environments. Through continuous optimization and adjustment, more accurate and effective hypothermia predictions can be achieved, improving the application effect of the model in actual surgical scenarios.

[0058] Finally, it should be noted that the above embodiments are merely examples for the purpose of illustrating the present invention and are not intended to limit the embodiments. Those skilled in the art will readily appreciate that other variations or modifications based on the above description are possible. It is not necessary and impossible to provide an exhaustive list of all embodiments. However, obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. A real-time prediction model for perioperative hypothermia based on multimodal data and deep learning, characterized by: The following steps are involved: S1. Patient information and multimodal data acquisition module: Establish an intelligent assistance system to collect a large amount of surgical data in advance, including preoperative, intraoperative and postoperative information, and comprehensively analyze the patient's health status; The system uses the ASA (American Society of Anesthesiologists) classification criteria to categorize patients based on their health information and the type of surgery they are undergoing. By combining the collected preoperative health data and historical surgical records with the ASA classification criteria, the system can more accurately predict the probability of hypothermia during surgery, helping doctors consider this factor when formulating surgical plans. Before surgery, temperature sensors are placed on the patient's body at various locations besides the surgical site. These sensors monitor the patient's body temperature in real time, providing a comprehensive understanding of temperature fluctuations. In combination with a multi-parameter monitor, the patient's key physiological indicators are monitored during surgery. This data provides a scientific basis for timely decision-making during surgery, particularly for early warning of hypothermia. Temperature and humidity sensors are installed in the operating room to monitor environmental conditions in real time, especially changes in temperature and humidity. This environmental data, combined with real-time data on patients' vital signs, can better assess the risk of intraoperative hypothermia, assisting doctors in timely adjusting intraoperative warming measures and providing patients with a safer and more comfortable surgical environment. S2, data preprocessing module: Process detected missing values ​​and outliers, and filter noisy data to improve data accuracy and quality; ensure that data from different sensors can be compared and analyzed under the same standards, and standardize or normalize these data to improve the effectiveness of data analysis or model training; Time-align the multimodal data from multiple sensors when merging and processing them to ensure temporal consistency, avoid data distortion or prediction errors caused by time lag or asynchrony, and improve the subsequent prediction accuracy and real-time response capabilities of the model system. S3. Selection and construction of deep learning models: Using recurrent neural networks (RNNs) to process time-series multimodal data can effectively establish the complex dependencies between physiological signals such as body temperature, ECG, and blood oxygen levels over time. The long short-term memory (LSTM) network in deep learning can address the vanishing gradient problem of traditional RNNs when capturing long-span temporal information, thereby accurately modeling the temporal dynamics of real-time intraoperative data. By using ensemble learning methods to fuse the prediction results of multiple models, we can improve prediction accuracy, explore the potential correlations between different data sources, and provide more comprehensive and accurate information for subsequent decision support systems, thereby providing a solid theoretical foundation and practical support for clinical prediction and decision-making. S4. Model training, evaluation and deployment: In the application scenario of perioperative hypothermia prediction, clinicians can annotate the data based on historical patient temperature data before, during, and after surgery, combined with clinical observation results. When the patient's temperature drops below a preset threshold (such as 36°C) during surgery, it can be determined that a hypothermia event has occurred. Deploy the trained model into the hospital's real-time surgical environment to ensure timely response to hypothermia. The deployed model should be able to receive and process the patient's current temperature data in real time and continuously update hypothermia predictions based on this data. When temperature data changes, the model can immediately predict the occurrence of hypothermia and alert the doctor whether appropriate intraoperative warming measures are needed. S5. Model optimization: After perioperative hypothermia prediction is completed, the model's performance is evaluated by comparing the model's prediction results with actual hypothermia events, and the model is further optimized based on the evaluation results; through an iterative feedback mechanism, the model continuously learns and adjusts to improve the accuracy and reliability of predictions in subsequent applications; through continuous optimization and adjustment, more accurate and effective hypothermia predictions can ultimately be achieved, improving the application effect of the model in actual surgical scenarios.

2. The real-time prediction model for perioperative hypothermia based on multimodal data and deep learning according to claim 1, characterized in that: The ASA classification system divides patients into different risk levels based on their health status. ASA Class I represents healthy patients, while ASA Class IV and V represent patients in poor health. Based on these classifications, doctors can determine the different risks patients may face during surgery, especially the probability of hypothermia. The core goal of the ASA classification is to help doctors assess the patient's preoperative condition in order to make more reasonable anesthesia and surgical plans.

3. The real-time prediction model for perioperative hypothermia based on multimodal data and deep learning according to claim 1, characterized in that: The missing values ​​should be filled or deleted through reasonable methods, outliers need to be detected and removed through statistical methods, and noise data should be filtered through appropriate filtering techniques, such as low-pass filtering or median filtering, to remove unnecessary interference and ensure that the data is purer; standardization or normalization can make data from different sources, scales, and units in the same range, eliminating inconsistencies caused by differences in sensor output.

4. The real-time prediction model for perioperative hypothermia based on multimodal data and deep learning according to claim 1, characterized in that: As mentioned above, during the model training process, the continuity of the time dimension and the temporal relationship of the data are very important. Cross-validation is used during the training process. Especially in the scenario of time series data, cross-validation helps ensure that the training set and the validation set have consistent distribution characteristics in the time dimension, avoiding model overfitting or data leakage problems caused by inconsistent time spans or improper data division.

Images