Deep learning-based operative patient body temperature early warning method and related equipment

Abstract

The invention discloses a deep learning-based operative patient body temperature early warning method and related equipment, and relates to the field of intelligent medical treatment, and the method comprises the steps: collecting multi-source body temperature data, environment and scene data and individual basic information of a patient, and constructing a time sequence feature data set through abnormal value elimination, missing value filling and standardization processing; and inputting the data into a feature extraction sub-network formed by a convolutional neural network and a bidirectional long-short-term memory network to generate a fusion feature vector, and inputting the fusion feature vector into a trend prediction sub-network formed by a gating circulation unit to obtain a future body temperature prediction sequence. And constructing an individual normal body temperature probability distribution model based on a kernel density estimation algorithm, and identifying an abnormal body temperature risk according to whether a predicted value exceeds a dynamic threshold value or whether the probability is lower than the threshold value. When early warning is triggered, an early warning object identifier, the current body temperature and a future prediction curve are output, and early risk identification of the body temperature of the surgical patient is achieved.

Description

Technical Field

[0001] This specification relates to the field of smart healthcare, and more specifically, this application relates to a method and related equipment for early warning of surgical patients' body temperature based on deep learning. Background Technology

[0002] Body temperature is an important physiological indicator of human health, and abnormal changes in it are often an early signal of disease (such as infectious diseases and inflammatory reactions). Traditional body temperature monitoring methods mainly rely on manual handheld thermometers (such as mercury thermometers and electronic thermometers) to measure each person individually. This is not only inefficient and unable to meet the needs of large-scale rapid monitoring in densely populated places, but also has measurement lag, making it difficult to achieve real-time early warning.

[0003] With the development of infrared thermometry technology, infrared thermal imagers and other equipment have been applied to rapid body temperature screening. However, the core of this method still relies on traditional threshold judgment methods (such as classifying a forehead temperature exceeding 37.3℃ as abnormal) for early warning. This method has significant drawbacks: firstly, it is easily affected by environmental factors (such as ambient temperature, humidity, and airflow), measurement distance, and individual differences (such as skin color, hair coverage, and basal body temperature), leading to a high false alarm rate; secondly, it cannot capture the dynamic trend of body temperature changes and can only identify cases where obvious abnormalities have already occurred, making it difficult to achieve the prevention and control goal of "early detection and early warning."

[0004] In recent years, machine learning technology has been applied to some extent in the field of medical data processing. However, existing body temperature analysis models based on traditional machine learning (such as support vector machines and random forests) have problems such as feature engineering relying on manual design, insufficient fitting ability for complex nonlinear body temperature data, and poor generalization performance. They are difficult to adapt to the diverse body temperature monitoring needs in different scenarios and cannot effectively improve the accuracy and foresight of body temperature early warning.

[0005] Therefore, there is an urgent need for a body temperature early warning technology that can overcome environmental interference, mine dynamic body temperature characteristics, and has high accuracy and foresight, in order to solve the pain points of low efficiency, high misjudgment rate and delayed early warning in existing technologies. Summary of the Invention

[0006] The summary section introduces a series of simplified concepts, which will be further explained in detail in the detailed description section. This summary section is not intended to limit the key and essential technical features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.

[0007] Firstly, this application proposes a deep learning-based method for early warning of surgical patient body temperature, including: Collect multi-source body temperature data, environmental and scene data, and individual basic information of surgical patients to form raw monitoring data; Data cleaning, missing value imputation and standardization were performed on the above raw monitoring data. The cleaned body temperature data, environmental data, scene parameters and individual basic information were then fused to construct a time-series feature dataset. The aforementioned temporal feature dataset is input into a feature extraction subnetwork composed of a convolutional neural network and a bidirectional long short-term memory network to extract local temperature change features and long temporal dependence features of the patient, and generate a fused feature vector. The aforementioned fused feature vectors are input into a trend prediction subnetwork composed of gated recurrent units to obtain the predicted body temperature sequence of the surgical patient within the future prediction step. An individual normal body temperature probability distribution model is constructed based on the kernel density estimation algorithm. The above body temperature prediction sequence is then substituted into the above probability distribution model. Based on whether the predicted value exceeds the dynamic threshold range or whether its probability value is lower than the set probability threshold, it is determined whether to trigger an abnormal body temperature warning. When an abnormal body temperature warning is triggered, the system outputs the warning target identifier, current body temperature, future body temperature prediction curve, and analysis of the cause of the abnormality, and feeds the warning results back to the medical institution's monitoring terminal.

[0008] In one feasible implementation, the above-mentioned raw monitoring data undergoes data cleaning, missing value imputation, and standardization processing. The cleaned body temperature data, environmental data, scene parameters, and individual basic information are then fused to construct a time-series feature dataset, including: The isolated forest algorithm is used to remove transient outliers caused by changes in body position or equipment malfunctions. When the missing duration is less than or equal to 15 minutes, linear interpolation is used for filling; when the missing duration is greater than 15 minutes, interpolation based on a weighted sliding window is used for filling. Z-Score standardization was performed on core body temperature, surface body temperature, and ambient temperature. The standardized body temperature features, environmental features, surgical scenario features, and individual basic information were then concatenated into a temporal feature vector set with a fixed time step and feature dimensions.

[0009] In one feasible implementation, the feature extraction subnetwork includes three one-dimensional convolutional layers, a batch normalization layer and a max pooling layer, as well as two bidirectional long short-term memory network layers. The aforementioned trend prediction subnetwork includes a gated recurrent unit network that outputs a body temperature prediction sequence for the next 30 minutes during the intraoperative phase and a body temperature prediction sequence for the next 2 hours before and after the operation.

[0010] In one feasible implementation, the aforementioned dynamic thresholds include an intraoperative core body temperature range of 36.0℃-37.5℃ and a postoperative body temperature range of 36.2℃-38.0℃. The above probability distribution model is constructed using the kernel density estimation method of Gaussian kernel, with a probability threshold of 0.03 during the intraoperative stage and a probability threshold of 0.05 during the postoperative stage. When the predicted body temperature exceeds the normal range for the corresponding stage or its probability in the probability distribution model is lower than the probability threshold, a low body temperature warning or an infectious fever warning is triggered.

[0011] In one feasible implementation, the method further includes a step of training models for the feature extraction subnetwork and the trend prediction subnetwork, including: Historical time-series feature data containing surgical patients’ body temperature characteristics, environmental characteristics, scene parameters and individual basic information are divided into training set, validation set and test set according to a preset ratio; The training set is input into the feature extraction subnetwork composed of the convolutional neural network and the bidirectional long short-term memory network, and the fused feature vector output by the feature extraction subnetwork is input into the trend prediction subnetwork composed of the gated recurrent unit to calculate the corresponding body temperature prediction value. A hybrid loss function is constructed based on the mean squared error loss used for body temperature prediction error and the cross-entropy loss used for anomaly prediction accuracy. An adaptive moment estimation optimizer is then used to iteratively optimize the model parameters of the feature extraction subnetwork and the trend prediction subnetwork. During training, a learning rate decay strategy, Dropout regularization, and L2 regularization are used to suppress overfitting, and an early stopping strategy is triggered to terminate training when the validation set loss does not decrease in several consecutive training rounds. After training, the body temperature prediction performance and abnormal warning accuracy of the feature extraction subnetwork and the trend prediction subnetwork are evaluated using the test set. The qualified model is used as the deployment model of the body temperature warning method.

[0012] In one feasible implementation, the above-mentioned model for constructing the probability distribution of normal body temperature for an individual based on the kernel density estimation algorithm includes: Historical normal body temperature data of the above-mentioned surgical patients at different stages before, during and after the operation, and establish the body temperature probability distribution for each stage. The density of the above historical normal body temperature data was estimated using a Gaussian kernel function to obtain a first probability distribution model for the intraoperative stage and a second probability distribution model for the postoperative stage. Based on the temperature prediction sequence output by the aforementioned trend prediction subnetwork, the corresponding probability values ​​are calculated and compared with the aforementioned dynamic threshold to determine whether the patient has entered the intraoperative hypothermia risk zone or the postoperative infectious fever risk zone.

[0013] In one feasible implementation, when an abnormal body temperature warning is triggered, the output of the warning object identifier, current body temperature, future body temperature prediction curve, and abnormal cause analysis includes: The warning levels are generated based on the changing trends of the above-mentioned body temperature prediction sequence, which include mild abnormality warning and severe abnormality warning. When the above-mentioned body temperature prediction sequence shows a continuous upward trend and exceeds the upper limit of the above-mentioned dynamic threshold, an infectious fever risk warning is generated; When the above-mentioned body temperature prediction sequence shows a rapid downward trend and falls below the lower limit of the above-mentioned dynamic threshold, an intraoperative hypothermia risk warning is generated.

[0014] Secondly, this invention also proposes a deep learning-based surgical patient body temperature early warning system, comprising: The data acquisition unit is used to collect multi-source body temperature data, environmental and scene data, and individual basic information of surgical patients to form raw monitoring data; The construction unit is used to perform data cleaning, missing value imputation and standardization on the above-mentioned raw monitoring data, and to fuse the cleaned body temperature data, environmental data, scene parameters and individual basic information to construct a time-series feature dataset. The fusion unit is used to input the above-mentioned temporal feature dataset into the feature extraction subnetwork composed of a convolutional neural network and a bidirectional long short-term memory network, so as to extract the local change features and long temporal dependence features of the patient's body temperature and generate a fused feature vector. The prediction unit is used to input the above-mentioned fused feature vector into the trend prediction subnetwork composed of gated recurrent units to obtain the predicted body temperature sequence of the surgical patient within the future prediction step. The judgment unit is used to construct an individual normal body temperature probability distribution model based on the kernel density estimation algorithm, and substitute the above body temperature prediction sequence into the above probability distribution model. Based on whether the predicted value exceeds the dynamic threshold range or whether its probability value is lower than the set probability threshold, it determines whether to trigger an abnormal body temperature warning. The early warning unit is used to output the early warning object identifier, current body temperature, future body temperature prediction curve and abnormal cause analysis when it is determined that an abnormal body temperature warning has been triggered, and to feed the early warning results back to the medical institution's monitoring terminal.

[0015] Thirdly, the present invention also proposes an electronic device comprising: a memory and a processor, characterized in that the processor is used to execute a computer program stored in the memory to implement the steps of the deep learning-based surgical patient body temperature early warning method described in any of the first aspects.

[0016] Fourthly, the present invention also proposes a computer-readable storage 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 deep learning-based surgical patient body temperature early warning method as described in any of the first aspects.

[0017] In summary, this invention, by collecting multi-source body temperature data before, during, and after surgery, and combining it with environmental parameters, scene parameters, and individual basic information, constructs a more comprehensive and refined data input system than traditional thermometers or infrared temperature measurement devices, effectively overcoming the interference caused by environmental temperature, measurement methods, and individual differences in body temperature judgment. By combining a feature extraction subnetwork of convolutional neural networks and bidirectional long short-term memory networks, this invention can simultaneously capture local fluctuations and long-term temporal dependencies in body temperature, overcoming the shortcomings of traditional machine learning models that rely on manual features and struggle to model nonlinear changes in body temperature. Simultaneously, a trend prediction subnetwork based on gated recurrent units enables continuous prediction of future body temperature, allowing the system to identify risk trends before significant temperature abnormalities occur, thus achieving an early warning capability 30 minutes to 2 hours earlier than traditional threshold methods. This invention constructs an individualized body temperature probability distribution model through kernel density estimation and combines it with dynamic thresholds during and after surgery, achieving truly individualized body temperature risk identification. This approach not only avoids misjudgments caused by fixed thresholds but also makes anomaly assessments more consistent with the physiological characteristics of surgical patients and their postoperative recovery patterns. It significantly reduces the misjudgment rate that is easily affected by factors such as distance, humidity, and skin color in traditional methods. Upon triggering an alert, this invention outputs the alert target identifier, current body temperature, a predicted future body temperature curve, and the cause of the anomaly. This helps medical staff to grasp the patient's condition immediately, enabling timely adjustments to surgical warming measures or early identification of postoperative infection risks, thus improving clinical decision-making efficiency and patient safety.

[0018] Other advantages, objectives and features of this application will be apparent in part from the description which follows, and in part from what those skilled in the art will understand through study and practice of this application. Attached Figure Description

[0019] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit this specification. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1A schematic flowchart of a deep learning-based surgical patient body temperature early warning method provided in this application embodiment; Figure 2 A structural schematic diagram of a deep learning-based surgical patient body temperature early warning system provided for an embodiment of this application; Figure 3 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0020] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application 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 interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or 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. The technical solutions of the embodiments of this application will now be clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0021] Please see Figure 1 This is a flowchart illustrating a deep learning-based method for early warning of surgical patient body temperature, provided in an embodiment of this application. Specifically, it may include: S110. Collect multi-source body temperature data, environmental and scene data, and individual basic information of surgical patients to form raw monitoring data; S120. Perform data cleaning, missing value imputation and standardization on the above raw monitoring data, and fuse the cleaned body temperature data, environmental data, scene parameters and individual basic information to construct a time-series feature dataset. S130. Input the above-mentioned time-series feature dataset into the feature extraction sub-network composed of a convolutional neural network and a bidirectional long short-term memory network to extract the local change features and long-term time-dependent features of the patient's body temperature and generate a fused feature vector. S140. Input the above-mentioned fused feature vector into the trend prediction sub-network composed of gated recurrent units to obtain the body temperature prediction sequence of the surgical patient within the future prediction step. S150. Construct an individual normal body temperature probability distribution model based on the kernel density estimation algorithm, and substitute the above body temperature prediction sequence into the above probability distribution model. Determine whether to trigger an abnormal body temperature warning based on whether the predicted value exceeds the dynamic threshold range or whether its probability value is lower than the set probability threshold. S160. When it is determined that an abnormal body temperature warning has been triggered, output the warning object identifier, current body temperature, future body temperature prediction curve and abnormal cause analysis, and feed the warning result back to the medical institution's monitoring terminal.

[0022] For example, in step S110, multi-dimensional information is collected from the surgical patient. Preoperative, intraoperative, and postoperative multi-source temperature data are obtained using a patch-type temperature sensor, an esophageal temperature probe from the anesthesia monitor, and an infrared thermometer. Simultaneously, environmental and scene parameters such as ambient temperature, humidity, and airflow speed are obtained from the operating room or ward environmental monitoring system, and individual basic information such as the patient's age, weight, underlying diseases, and preoperative basal body temperature are extracted from the hospital's electronic medical record system, thus forming complete raw monitoring data.

[0023] In step S120, data cleaning and missing value imputation operations are performed on the above-mentioned raw monitoring data to eliminate instantaneous outliers caused by changes in body position, equipment vibration, etc., and the missing values ​​are imputed using linear interpolation or weighted sliding window methods according to the duration of the missing values. To eliminate the dimensional differences between different data sources, key features such as core body temperature, body surface temperature, and ambient temperature are Z-score standardized, and the above-mentioned body temperature features, environmental features, scene parameters, and individual basic information are fused to construct a time-series feature dataset with a fixed time step and feature dimensions.

[0024] In step S130, the system inputs the aforementioned temporal feature dataset into a feature extraction subnetwork composed of a convolutional neural network and a bidirectional long short-term memory network. The convolutional neural network is used to extract local body temperature fluctuation patterns, while the bidirectional long short-term memory network is used to capture the long-term temporal dependencies of body temperature development before, during, and after surgery, thereby generating a fused feature vector that can truly reflect the patient's physiological state.

[0025] In step S140, the system inputs the aforementioned fused feature vector into the trend prediction subnetwork composed of gated recurrent units to predict the patient's body temperature change trend within the future prediction step. Leveraging the time-series modeling capabilities of the gated recurrent units, the system can output multi-time-point body temperature prediction sequences for early identification of potential abnormal risks.

[0026] In step S150, the system constructs an individual normal body temperature probability distribution model for the surgical patient based on a kernel density estimation algorithm. The aforementioned body temperature prediction sequence is substituted into this probability distribution model to calculate the probability that the patient's body temperature falls within the normal range. If a predicted value exceeds the aforementioned dynamic threshold range, or its corresponding probability is lower than a preset probability threshold, the system can determine that the patient has a risk of abnormal body temperature.

[0027] In step S160, when the system determines that an abnormal body temperature warning has been triggered, it automatically outputs the patient's warning target identifier, current body temperature information, future body temperature prediction curve, and possible abnormal cause analysis, and pushes the warning results to the monitoring terminal of the medical institution in real time to support medical staff to take timely intervention measures.

[0028] In summary, this invention, by collecting multi-source body temperature data before, during, and after surgery, and combining it with environmental parameters, scene parameters, and individual basic information, constructs a more comprehensive and refined data input system than traditional thermometers or infrared temperature measurement devices, effectively overcoming the interference caused by environmental temperature, measurement methods, and individual differences in body temperature judgment. By combining a feature extraction subnetwork of convolutional neural networks and bidirectional long short-term memory networks, this invention can simultaneously capture local fluctuations and long-term temporal dependencies in body temperature, overcoming the shortcomings of traditional machine learning models that rely on manual features and struggle to model nonlinear changes in body temperature. Simultaneously, a trend prediction subnetwork based on gated recurrent units enables continuous prediction of future body temperature, allowing the system to identify risk trends before significant temperature abnormalities occur, thus achieving an early warning capability 30 minutes to 2 hours earlier than traditional threshold methods. This invention constructs an individualized body temperature probability distribution model through kernel density estimation and combines it with dynamic thresholds during and after surgery, achieving truly individualized body temperature risk identification. This approach not only avoids misjudgments caused by fixed thresholds but also makes anomaly assessments more consistent with the physiological characteristics of surgical patients and their postoperative recovery patterns. It significantly reduces the misjudgment rate that is easily affected by factors such as distance, humidity, and skin color in traditional methods. Upon triggering an alert, this invention outputs the alert target identifier, current body temperature, a predicted future body temperature curve, and the cause of the anomaly. This helps medical staff to grasp the patient's condition immediately, enabling timely adjustments to surgical warming measures or early identification of postoperative infection risks, thus improving clinical decision-making efficiency and patient safety.

[0029] In one feasible implementation, the above-mentioned raw monitoring data undergoes data cleaning, missing value imputation, and standardization. The cleaned body temperature data, environmental data, scene parameters, and individual basic information are then fused to construct a time-series feature dataset, including: The isolated forest algorithm is used to remove transient outliers caused by changes in body position or equipment malfunctions. When the missing duration is less than or equal to 15 minutes, linear interpolation is used for filling; when the missing duration is greater than 15 minutes, interpolation based on a weighted sliding window is used for filling. Z-Score standardization was performed on core body temperature, surface body temperature, and ambient temperature. The standardized body temperature features, environmental features, surgical scenario features, and individual basic information were then concatenated into a temporal feature vector set with a fixed time step and feature dimensions.

[0030] For example, the system performs data cleaning, missing value imputation, and standardization on the collected raw monitoring data, and constructs a time-series feature dataset for deep learning model training and inference based on this. To ensure the reliability of body temperature data, the system uses the isolated forest algorithm to identify and remove instantaneous abnormal body temperature values ​​caused by changes in patient position, sensor pressure, temporary probe detachment, or equipment malfunction, thereby avoiding interference from outliers in subsequent trend analysis. When handling missing body temperature data, if the missing duration is less than or equal to 15 minutes, the system uses linear interpolation to smoothly imput the missing data based on normal body temperature data on both sides of the missing point. If the missing duration exceeds 15 minutes, a weighted sliding window-based interpolation method is used to estimate the missing interval using the local trend of the time-series data, making the imputation result more closely match the actual trajectory of the patient's body temperature change.

[0031] After completing the data cleaning process described above, the system performs Z-Score standardization on key temperature features, including core body temperature, surface body temperature, and ambient temperature, to eliminate numerical differences between different temperature units, equipment accuracies, and acquisition methods. The standardized temperature features are then concatenated with environmental features, surgical scenario features, and the patient's individual basic information to construct a temporal feature vector set with a fixed time step and feature dimensions. This vector set fully expresses the correlation between patient body temperature changes and environmental conditions within that time window. The resulting temporal feature dataset can be used as input to a deep learning model for predicting surgical patient body temperature trends and providing early warnings of abnormalities.

[0032] In one feasible implementation, the feature extraction subnetwork includes three one-dimensional convolutional layers, a batch normalization layer and a max pooling layer, as well as two bidirectional long short-term memory network layers. The aforementioned trend prediction subnetwork includes a gated recurrent unit network that outputs a body temperature prediction sequence for the next 30 minutes during the intraoperative phase and a body temperature prediction sequence for the next 2 hours before and after the operation.

[0033] For example, the deep learning structure of the aforementioned body temperature early warning model consists of a feature extraction subnetwork and a trend prediction subnetwork. The feature extraction subnetwork uses three one-dimensional convolutional layers as the initial feature extraction module. Each layer uses an appropriate kernel size and number to capture local fluctuation patterns of the body temperature signal over a short timescale. To further improve the stability of the feature distribution and accelerate the model's convergence process, a batch normalization layer is added after each convolutional layer, and a max-pooling layer is used to downsample the convolutional output, thereby effectively suppressing noise interference and reducing the model's computational complexity.

[0034] After extracting local dynamic features using the convolutional structure, the feature extraction subnetwork models the long-term dependencies of time-series body temperature data through a two-layer bidirectional long short-term memory network. Because the bidirectional structure can simultaneously consider both forward and backward information from the time series, it effectively preserves key trend features even when the patient's body temperature changes exhibit non-linear fluctuations, phased changes, or cross-period coupling. After the combined processing of the convolutional layers and the bidirectional long short-term memory network, the resulting fused feature vector comprehensively reflects both the short-term fluctuations and long-term trends of the patient's body temperature.

[0035] After feature extraction, this embodiment inputs the fused feature vector into a trend prediction subnetwork composed of gated recurrent units to output body temperature prediction sequences for different stages. Considering that surgical patients experience more rapid temperature changes during surgery, and that small fluctuations may have clinical significance, the trend prediction subnetwork outputs a body temperature prediction sequence for the next 30 minutes during the intraoperative stage to identify potential hypothermia risks in advance. In the preoperative and postoperative stages, body temperature is influenced by a combination of environmental and physiological factors, and the change cycle is usually longer. Therefore, the trend prediction subnetwork outputs a body temperature prediction sequence for the next 2 hours during these stages to better identify medium- to long-term risk changes such as infectious fever. Through the design of the above network structure, this embodiment enables the system to maintain high prediction accuracy and strong clinical adaptability in different surgical scenarios.

[0036] In one feasible implementation, the aforementioned dynamic thresholds include an intraoperative core body temperature range of 36.0℃-37.5℃ and a postoperative body temperature range of 36.2℃-38.0℃. The above probability distribution model is constructed using the kernel density estimation method of Gaussian kernel, with a probability threshold of 0.03 during the intraoperative stage and a probability threshold of 0.05 during the postoperative stage. When the predicted body temperature exceeds the normal range for the corresponding stage or its probability in the probability distribution model is lower than the probability threshold, a low body temperature warning or an infectious fever warning is triggered.

[0037] For example, to achieve accurate judgment of abnormal body temperature in surgical patients, the system adopts a combination of a phased dynamic threshold mechanism and a probability distribution model based on Gaussian kernel density estimation for anomaly identification. Specifically, considering that the patient's core body temperature is usually in a low and narrow range during surgery due to the influence of anesthesia, surgical position, and surgical environment temperature, this embodiment sets the normal range of intraoperative core body temperature to 36.0℃ to 37.5℃; while in the postoperative stage, due to the influence of the body's inflammatory response, metabolic changes, and postoperative nursing operations, the body temperature may show a certain degree of increase, so the normal range of postoperative body temperature is set to 36.2℃ to 38.0℃ to better reflect clinical reality.

[0038] After defining the normal range, this embodiment further constructs a probability distribution model of individual patient body temperature based on the Gaussian kernel density estimation method. By performing density estimation on normal body temperature data, the system can obtain the probability distribution characteristics of body temperature values ​​occurring under normal conditions. Based on this, the system sets a probability threshold of 0.03 during the intraoperative stage. That is, if the predicted probability of body temperature occurring in the probability distribution is lower than 0.03, it is considered that the body temperature value deviates significantly from the patient's normal physiological state during surgery and has a risk of hypothermia. In the postoperative stage, considering the greater uncertainty of body temperature fluctuations, the probability threshold is set to 0.05, making the abnormality detection more sensitive and helping to identify early signs of infectious fever in a timely manner.

[0039] When the predicted body temperature value output by the trend prediction subnetwork exceeds the normal range for the corresponding stage, or when the probability of the predicted value in the individual probability distribution model is lower than the set probability threshold, the system triggers a corresponding abnormal body temperature warning. Specifically, when the predicted value is low and the probability decreases significantly, the system generates a low body temperature warning, indicating that intraoperative temperature maintenance measures may be insufficient; when the predicted value is high and the probability decreases, an infectious fever warning is generated, providing medical staff with a basis for early intervention. This embodiment improves the accuracy and foresight of abnormal body temperature identification through the joint judgment of dynamic thresholds and probability models.

[0040] In one feasible implementation, the method further includes a step of training models for the feature extraction subnetwork and the trend prediction subnetwork, including: Historical time-series feature data containing surgical patients’ body temperature characteristics, environmental characteristics, scene parameters and individual basic information are divided into training set, validation set and test set according to a preset ratio; The training set is input into the feature extraction subnetwork composed of the convolutional neural network and the bidirectional long short-term memory network, and the fused feature vector output by the feature extraction subnetwork is input into the trend prediction subnetwork composed of the gated recurrent unit to calculate the corresponding body temperature prediction value. A hybrid loss function is constructed based on the mean squared error loss used for body temperature prediction error and the cross-entropy loss used for anomaly prediction accuracy. An adaptive moment estimation optimizer is then used to iteratively optimize the model parameters of the feature extraction subnetwork and the trend prediction subnetwork. During training, a learning rate decay strategy, Dropout regularization, and L2 regularization are used to suppress overfitting, and an early stopping strategy is triggered to terminate training when the validation set loss does not decrease in several consecutive training rounds. After training, the body temperature prediction performance and abnormal warning accuracy of the feature extraction subnetwork and the trend prediction subnetwork are evaluated using the test set. The qualified model is used as the deployment model of the body temperature warning method.

[0041] For example, to enable the feature extraction subnetwork and the trend prediction subnetwork to have high-precision modeling capabilities for changes in the body temperature of surgical patients, the system jointly trains the two subnetworks. First, the system collects a large amount of representative historical surgical patient data, including time-series samples of body temperature characteristics, environmental characteristics, scene parameters, and individual basic information, and divides them into training set, validation set, and test set according to a preset ratio to ensure data balance at different stages and good generalization ability of the model training.

[0042] During model training, the system inputs the temporal feature data corresponding to the training set into the feature extraction subnetwork, which consists of a convolutional neural network and a bidirectional long short-term memory network. This subnetwork extracts local fluctuation features through the convolutional structure, captures the dependencies between preceding and subsequent events through the bidirectional long short-term memory structure, and outputs a fused feature vector. Subsequently, the system inputs the fused feature vector into the trend prediction subnetwork, which consists of gated recurrent units, to calculate the corresponding predicted body temperature value and compare it with the actual body temperature data to form the prediction error.

[0043] To simultaneously improve the accuracy of body temperature trend prediction and anomaly early warning performance, this embodiment constructs a hybrid loss function based on the mean squared error loss of body temperature prediction error and the cross-entropy loss of anomaly classification accuracy. An adaptive moment estimation optimizer is then used to iteratively optimize the parameters of the two sub-networks. During training, to prevent overfitting, the system introduces a learning rate decay strategy, Dropout regularization, and L2 regularization to ensure the model remains robust even in complex surgical scenarios. Furthermore, if the validation set loss does not decrease over several consecutive training rounds, the system automatically triggers an early stopping strategy to prevent the model from degrading its generalization performance during overtraining.

[0044] After completing all stages of training, the system uses a test set to comprehensively evaluate the body temperature prediction capabilities and abnormal warning accuracy of the feature extraction subnetwork and the trend prediction subnetwork. Once the model reaches the preset performance standard, it is used as the final deployment model in the surgical patient body temperature warning process. Through this training process, this embodiment achieves high-precision prediction and highly reliable early warning of surgical patient body temperature dynamics, providing important support for intraoperative hypothermia management and postoperative infection monitoring.

[0045] In one feasible implementation, the above-mentioned model for constructing the probability distribution of normal body temperature for an individual based on the kernel density estimation algorithm includes: Historical normal body temperature data of the above-mentioned surgical patients at different stages before, during and after the operation, and establish the body temperature probability distribution for each stage. The density of the above historical normal body temperature data was estimated using a Gaussian kernel function to obtain a first probability distribution model for the intraoperative stage and a second probability distribution model for the postoperative stage. Based on the temperature prediction sequence output by the aforementioned trend prediction subnetwork, the corresponding probability values ​​are calculated and compared with the aforementioned dynamic threshold to determine whether the patient has entered the intraoperative hypothermia risk zone or the postoperative infectious fever risk zone.

[0046] For example, to improve the accuracy and individualized adaptability of abnormal body temperature identification, the system constructs an individual normal body temperature probability distribution model for surgical patients based on a kernel density estimation algorithm. First, historical normal body temperature data of the target patient at different stages before, during, and after surgery are acquired to reflect the patient's temperature fluctuation characteristics under different physiological states. Since the surgical process affects body temperature through different mechanisms, the system models the three types of staged data separately, thereby constructing the normal body temperature probability distribution corresponding to each stage.

[0047] Based on this, a Gaussian kernel function was used to estimate the density of the aforementioned historical normal body temperature data. By smoothing the body temperature data, the system obtained a probability distribution model that better reflects the actual physiological changes of patients. The probability distribution formed by the intraoperative normal body temperature data was used as the first probability distribution model, while the probability distribution formed by the postoperative normal body temperature data was used as the second probability distribution model, providing a probabilistic reference for subsequent risk assessment.

[0048] After acquiring the predicted body temperature sequence output by the trend prediction subnetwork, the system substitutes each predicted body temperature value into the probability distribution model of the corresponding stage to calculate its probability value within the normal body temperature distribution. Subsequently, the system jointly compares this probability value with a dynamic threshold mechanism to comprehensively determine health risks. When the probability of the predicted body temperature value in the first probability distribution model is lower than the corresponding threshold, the system identifies it as potentially entering the intraoperative hypothermia risk zone; when the probability of the predicted body temperature value in the second probability distribution model is lower than the set threshold, it is identified as potentially entering the postoperative infectious fever risk zone. This embodiment significantly enhances the sensitivity and accuracy of abnormal body temperature identification through the joint determination of the probability density model and the dynamic threshold.

[0049] In one feasible implementation, when an abnormal body temperature warning is triggered, the output of the warning object identifier, current body temperature, future body temperature prediction curve, and abnormal cause analysis includes: The warning levels are generated based on the changing trends of the above-mentioned body temperature prediction sequence, which include mild abnormality warning and severe abnormality warning. When the above-mentioned body temperature prediction sequence shows a continuous upward trend and exceeds the upper limit of the above-mentioned dynamic threshold, an infectious fever risk warning is generated; When the above-mentioned body temperature prediction sequence shows a rapid downward trend and falls below the lower limit of the above-mentioned dynamic threshold, an intraoperative hypothermia risk warning is generated.

[0050] For example, when the system determines that an abnormal body temperature warning has been triggered based on the predicted body temperature results, it will automatically execute the specific process of generating and outputting the warning information. Based on the body temperature prediction sequence output by the trend prediction sub-network, the system analyzes the overall trend of the prediction curve to generate the corresponding warning level. If the predicted body temperature deviates slightly or the trend is relatively slow, the system generates a mild abnormality warning. If the predicted body temperature shows a rapid shift, a large change, or crosses multiple thresholds within a short period, the system generates a severe abnormality warning to remind medical staff to pay closer attention to the patient's condition.

[0051] When further analyzing the predicted body temperature trend, the system incorporates a dynamic threshold mechanism to identify potential risk types. When the predicted body temperature sequence shows a continuous upward trend over a period of time and exceeds the upper limit of the dynamic threshold, the system will generate an infectious fever risk warning based on its predicted characteristics to alert medical staff that the patient may have an early infection or postoperative inflammation. Conversely, when the predicted sequence shows a significant and rapid downward trend and falls below the lower limit of the dynamic threshold, the system generates an intraoperative hypothermia risk warning, indicating that hypothermia may be caused by anesthesia effects, insufficient surgical environment temperature, or hemodynamic changes.

[0052] After generating the aforementioned warning results, the system simultaneously outputs the warning target identifier, current body temperature value, future body temperature prediction curve, and possible abnormal cause analysis. This warning information is then transmitted to the monitoring terminal of the medical institution via a graphical interface, audio-visual prompts, or push notifications. Through this warning output mechanism, this embodiment can identify potential abnormal body temperature states in advance and provide clear risk types, helping medical staff to take timely intervention measures and improve intraoperative safety and postoperative recovery quality.

[0053] Secondly, this invention also proposes a deep learning-based surgical patient body temperature early warning system, such as... Figure 2 As shown, it includes: The data acquisition unit 21 is used to collect multi-source body temperature data, environmental and scene data, and individual basic information of surgical patients to form raw monitoring data; Construction unit 22 is used to perform data cleaning, missing value imputation and standardization on the above-mentioned original monitoring data, and to fuse the cleaned body temperature data, environmental data, scene parameters and individual basic information to construct a time-series feature dataset; The fusion unit 23 is used to input the above-mentioned temporal feature dataset into the feature extraction sub-network composed of a convolutional neural network and a bidirectional long short-term memory network to extract the local change features and long temporal dependence features of the patient's body temperature and generate a fused feature vector. Prediction unit 24 is used to input the above-mentioned fused feature vector into the trend prediction subnetwork composed of gated recurrent units to obtain the predicted body temperature sequence of the surgical patient within the future prediction step. Judgment unit 25 is used to construct an individual normal body temperature probability distribution model based on kernel density estimation algorithm, and substitute the above body temperature prediction sequence into the above probability distribution model. Based on whether the predicted value exceeds the dynamic threshold range or whether its probability value is lower than the set probability threshold, it determines whether to trigger an abnormal body temperature warning. The early warning unit 26 is used to output the early warning object identifier, current body temperature, future body temperature prediction curve and abnormal cause analysis when it is determined that an abnormal body temperature warning has been triggered, and to feed the early warning results back to the medical institution's monitoring terminal.

[0054] In one feasible implementation, a deep learning-based surgical patient body temperature early warning system can also perform any step of the method proposed in the first aspect.

[0055] Thirdly, the present invention also proposes an electronic device 300, such as... Figure 3 As shown, it includes a memory 310, a processor 320, and a computer program 311 stored on the memory 310 and executable on the processor. When the processor 320 executes the computer program 311, it implements the steps of the deep learning-based surgical patient body temperature early warning method as described in any of the first aspects.

[0056] Fourthly, the present invention also proposes a computer-readable storage 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 deep learning-based surgical patient body temperature early warning method as described in any one of the first aspects.

[0057] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0058] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0059] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0060] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0061] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.

[0062] This application also provides a computer program product, which includes computer software instructions. When the computer software instructions are executed on a processing device, the processing device performs the voice-based identity recognition process in the corresponding embodiment. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0063] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0064] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0065] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0066] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0067] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0068] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A deep learning-based method for early warning of body temperature in surgical patients, characterized in that, include: Collect multi-source body temperature data, environmental and scene data, and individual basic information of surgical patients to form raw monitoring data; The original monitoring data is cleaned, missing value imputation and standardization are performed, and the cleaned body temperature data, environmental data, scene parameters and individual basic information are fused to construct a time-series feature dataset. The temporal feature dataset is input into a feature extraction subnetwork composed of a convolutional neural network and a bidirectional long short-term memory network to extract local temperature change features and long temporal dependence features of the patient, and generate a fused feature vector. The fused feature vector is input into a trend prediction subnetwork composed of gated recurrent units to obtain the predicted body temperature sequence of the surgical patient within the future prediction step. An individual normal body temperature probability distribution model is constructed based on the kernel density estimation algorithm. The predicted body temperature sequence is then substituted into the probability distribution model. Based on whether the predicted value exceeds the dynamic threshold range or whether its probability value is lower than the set probability threshold, it is determined whether to trigger an abnormal body temperature warning. When an abnormal body temperature warning is triggered, the system outputs the warning target identifier, current body temperature, future body temperature prediction curve, and analysis of the cause of the abnormality, and feeds the warning results back to the medical institution's monitoring terminal.

2. The deep learning-based surgical patient body temperature early warning method according to claim 1, characterized in that, The original monitoring data undergoes data cleaning, missing value imputation, and standardization. The cleaned body temperature data, environmental data, scene parameters, and individual basic information are then fused to construct a time-series feature dataset, including: The isolated forest algorithm is used to remove transient outliers caused by changes in body position or equipment malfunctions. When the missing duration is less than or equal to 15 minutes, linear interpolation is used for filling; when the missing duration is greater than 15 minutes, interpolation based on a weighted sliding window is used for filling. Z-Score standardization was performed on core body temperature, surface body temperature, and ambient temperature. The standardized body temperature features, environmental features, surgical scenario features, and individual basic information were then concatenated into a temporal feature vector set with a fixed time step and feature dimensions.

3. The deep learning-based surgical patient body temperature early warning method according to claim 1, characterized in that, The feature extraction subnetwork includes three one-dimensional convolutional layers, a batch normalization layer, and a max pooling layer, as well as two bidirectional long short-term memory network layers. The trend prediction subnetwork includes a gated recurrent unit network that outputs a body temperature prediction sequence for the next 30 minutes during the intraoperative phase and a body temperature prediction sequence for the next 2 hours before and after the operation.

4. The deep learning-based surgical patient body temperature early warning method according to claim 1, characterized in that, The dynamic thresholds include the intraoperative core body temperature range of 36.0℃-37.5℃ and the postoperative body temperature range of 36.2℃-38.0℃. The probability distribution model is constructed using the kernel density estimation method of Gaussian kernel, with a probability threshold of 0.03 during the intraoperative stage and a probability threshold of 0.05 during the postoperative stage. When the predicted body temperature exceeds the normal range for the corresponding stage or its probability in the probability distribution model is lower than the probability threshold, a low body temperature warning or an infectious fever warning is triggered.

5. The deep learning-based surgical patient body temperature early warning method according to claim 1, characterized in that, The method further includes a step of training the feature extraction subnetwork and the trend prediction subnetwork, including: Historical time-series feature data containing surgical patients’ body temperature characteristics, environmental characteristics, scene parameters and individual basic information are divided into training set, validation set and test set according to a preset ratio; The training set is input into the feature extraction subnetwork composed of the convolutional neural network and the bidirectional long short-term memory network, and the fused feature vector output by the feature extraction subnetwork is input into the trend prediction subnetwork composed of the gated recurrent units to calculate the corresponding body temperature prediction value. A hybrid loss function is constructed based on the mean squared error loss used for body temperature prediction error and the cross-entropy loss used for anomaly prediction accuracy. An adaptive moment estimation optimizer is then used to iteratively optimize the model parameters of the feature extraction subnetwork and the trend prediction subnetwork. During training, a learning rate decay strategy, Dropout regularization, and L2 regularization are used to suppress overfitting, and an early stopping strategy is triggered to terminate training when the validation set loss does not decrease in several consecutive training rounds. After training, the body temperature prediction performance and abnormal warning accuracy of the feature extraction subnetwork and the trend prediction subnetwork are evaluated using the test set, and the qualified model is used as the deployment model of the body temperature warning method.

6. The deep learning-based surgical patient body temperature early warning method according to claim 1, characterized in that, The method for constructing an individual's normal body temperature probability distribution model based on the kernel density estimation algorithm includes: The patient's historical normal body temperature data were obtained at different stages before, during and after the operation, and the corresponding body temperature probability distribution was established. The historical normal body temperature data were density estimated using a Gaussian kernel function to obtain a first probability distribution model for the intraoperative stage and a second probability distribution model for the postoperative stage. Based on the body temperature prediction sequence output by the trend prediction subnetwork, the corresponding probability value is calculated and compared with the dynamic threshold to determine whether the patient has entered the intraoperative low body temperature risk zone or the postoperative infectious fever risk zone.

7. The deep learning-based surgical patient body temperature early warning method according to claim 1, characterized in that, When an abnormal body temperature warning is triggered, the system outputs the warning target identifier, current body temperature, future body temperature prediction curve, and analysis of the cause of the abnormality, including: An early warning level is generated based on the changing trend of the predicted body temperature sequence, wherein the early warning level includes a mild abnormality warning and a severe abnormality warning. When the predicted body temperature sequence shows a continuous upward trend and exceeds the upper limit of the dynamic threshold, an infectious fever risk warning is generated. When the predicted body temperature sequence shows a rapid downward trend and falls below the lower limit of the dynamic threshold, an intraoperative hypothermia risk warning is generated.

8. A deep learning-based surgical patient body temperature early warning system, characterized in that, include: The data acquisition unit is used to collect multi-source body temperature data, environmental and scene data, and individual basic information of surgical patients to form raw monitoring data; The construction unit is used to perform data cleaning, missing value imputation and standardization on the original monitoring data, and to fuse the cleaned body temperature data, environmental data, scene parameters and individual basic information to construct a time-series feature dataset. The fusion unit is used to input the temporal feature dataset into the feature extraction subnetwork composed of a convolutional neural network and a bidirectional long short-term memory network to extract local temperature change features and long temporal dependence features of the patient and generate a fused feature vector. The prediction unit is used to input the fused feature vector into the trend prediction subnetwork composed of gated recurrent units to obtain the predicted body temperature sequence of the surgical patient within the future prediction step. The judgment unit is used to construct an individual normal body temperature probability distribution model based on the kernel density estimation algorithm, and substitute the body temperature prediction sequence into the probability distribution model. Based on whether the predicted value exceeds the dynamic threshold range or whether its probability value is lower than the set probability threshold, it determines whether to trigger an abnormal body temperature warning. The early warning unit is used to output the early warning object identifier, current body temperature, future body temperature prediction curve and abnormal cause analysis when it is determined that an abnormal body temperature warning has been triggered, and to feed the early warning results back to the medical institution's monitoring terminal.

9. An electronic device, comprising: The memory and processor are characterized in that the processor, when executing a computer program stored in the memory, implements the steps of the deep learning-based surgical patient body temperature early warning method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the deep learning-based surgical patient body temperature early warning method as described in any one of claims 1-7.

Images