Newborn abnormal temperature prediction method and device and storage medium

Abstract

The present application relates to the technical field of body temperature abnormality prediction, in particular to a neonatal body temperature abnormality prediction method, device and storage medium. It comprises the following steps: real-time acquisition of the body temperature and physiological signals of the neonate and pretreatment; based on fast Fourier transform, the time sequence characteristics reflecting the body temperature change in the body temperature and physiological signals are extracted to generate a characteristic vector, and the body temperature and physiological signals are processed through phase coupling analysis and frequency domain mask suppression mechanism to dynamically identify the low-frequency noise interference interval caused by irregular breathing and skin conductive disturbance of the neonate; based on the characteristic vector, the long short-term memory network is used to real-time predict the body temperature abnormality risk probability, and the gating weight of the long short-term memory network is dynamically regulated through the body temperature change rate and physiological state label double factors to optimize the feature fusion mechanism. Through the multi-modal physiological signal fusion and dynamic feature weighting mechanism, the accuracy and real-time performance of the neonatal body temperature abnormality prediction are significantly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of body temperature abnormality prediction, in particular to a neonatal body temperature abnormality prediction method, device and storage medium. BACKGROUND

[0002] Newborns are extremely sensitive to changes in external environmental temperature due to their incomplete development of the thermoregulatory system, especially for premature and low birth weight infants, and are prone to body temperature abnormalities, including hypothermia and hyperthermia. Traditional body temperature monitoring methods mainly rely on periodic measurement of body temperature, which lacks real-time monitoring and analysis capabilities for the trend of neonatal body temperature changes, making it difficult to detect potential risks and take effective intervention measures at the initial stage of body temperature abnormalities. In addition, the changes in the body temperature of newborns are not only influenced by environmental factors, but also closely related to physiological parameters such as heart rate, respiratory rate, and blood oxygen saturation. Single body temperature data cannot fully reflect the health status of newborns. In the face of complex and variable situations such as body temperature abnormalities caused by infection, dehydration, or central nervous system diseases, other physiological signal changes often accompany, increasing the difficulty and uncertainty of accurate diagnosis. In order to improve the quality and efficiency of neonatal care and reduce the occurrence of complications, there is an urgent need for a technology that can automatically and accurately predict body temperature abnormalities. Therefore, a neonatal body temperature abnormality prediction method, device and storage medium are provided. SUMMARY

[0003] The present application aims to provide a neonatal body temperature abnormality prediction method, device and storage medium to solve the problems of existing neonatal body temperature monitoring methods that cannot achieve real-time continuous monitoring, cannot accurately predict body temperature abnormality risks, and lack multi-modal physiological parameter collaborative analysis capabilities.

[0004] To achieve the above-mentioned purpose, the present application aims to provide a neonatal body temperature abnormality prediction method, comprising the following steps:

[0005] S1, real-time acquisition of the body temperature and physiological signals of the newborn and pre-processing;

[0006] S2, extraction of time sequence features reflecting body temperature changes in the body temperature and physiological signals based on fast Fourier transform, generation of a feature vector, and processing of the body temperature and physiological signals through phase coupling analysis and frequency domain mask suppression mechanism to dynamically identify low-frequency noise interference intervals caused by irregular breathing and skin conductive disturbance of the newborn;

[0007] S3, real-time prediction of body temperature abnormality risk probability based on the feature vector using a long short-term memory network, and dynamic regulation of the gating weight of the long short-term memory network through the body temperature change rate and physiological state label double factors to optimize the feature fusion mechanism;

[0008] S4. Risk classification judgment is made based on the probability of abnormal risks and combined with risk classification rules.

[0009] As a further improvement to this technical solution, step S2, extracting temporal features reflecting changes in body temperature from body temperature and physiological signals, includes the following steps:

[0010] S2.1. A fixed time window is used to divide the continuous body temperature and physiological signal time series into multiple analysis segments, forming a sliding window structure;

[0011] S2.2 Calculate the time-domain characteristics of body temperature and various physiological signals within each sliding window;

[0012] S2.3. For the body temperature and physiological signal sequences within each sliding window, the frequency domain features are extracted using Fast Fourier Transform to be used for periodic fluctuations in body temperature and physiological signals.

[0013] S2.4. Using the cross-modal analysis method, based on the body temperature and physiological signal sequences of the sliding window, calculate the dynamic coupling characteristics to reflect the coordinated fluctuation relationship between body temperature changes and other physiological activities.

[0014] S2.5. Use a feature selection algorithm to filter out time-domain features, frequency-domain features, and dynamic coupling features that reflect changes in body temperature, and combine the filtered features into a unified feature vector. , In the formula, The first feature vector is the first feature vector. One characteristic, The first eigenvector in the original feature vector One characteristic, The final number of features selected. For feature index, The feature index is used for filtering.

[0015] As a further improvement to this technical solution, in S2.3, the frequency domain features of the body temperature and physiological signal sequences within each sliding window are extracted using Fast Fourier Transform, including the following steps:

[0016] S2.31. By processing the body temperature and physiological signal sequences through phase coupling analysis and frequency domain masking suppression mechanism, the low-frequency noise interference range caused by irregular breathing and skin conductivity disturbance is dynamically identified, and an enhanced time domain signal is generated.

[0017] S2.32. Apply Fast Fourier Transform to the enhanced time-domain signal to convert the enhanced time-domain signal into a frequency-domain complex spectrum.

[0018] S2.33. Calculate the corresponding amplitude spectrum for each frequency domain complex spectrum, select the amplitude of the main frequency components in the frequency band, and construct the frequency domain feature vector.

[0019] S2.34, normalizing each frequency domain feature vector;

[0020] S2.35, splicing and fusing all normalized frequency domain feature vectors to form a joint frequency domain feature vector.

[0021] As a further improvement of the technical solution, in S2.31, the body temperature and physiological signal sequence are processed by phase coupling analysis and frequency domain mask suppression mechanism to dynamically identify the low frequency noise interference interval caused by irregular respiration and skin conductive disturbance, including the following steps:

[0022] S2.311, extract RR interval sequence from physiological signal, calculate instantaneous heart rate variability, synchronize to derive respiratory signal, use Hilbert transform to obtain instantaneous phase of instantaneous heart rate variability and respiratory signal, and calculate the phase difference between them, when the phase difference is greater than the upper threshold a, mark the x interval centered at the current time as respiratory interference interval;

[0023] S2.312, calculate the second derivative of body temperature, if the absolute value of the second derivative is greater than the derivative threshold , then expand the fixed length window centered at the current time point as the skin conductive interference interval;

[0024] S2.313, extract the reference noise signal of the respiratory interference interval and the skin conductive interference interval, window the reference noise signal and perform fast Fourier transform to obtain the complex spectrum, and take the module length of the complex spectrum to obtain the interference spectrum template;

[0025] S2.314, divide the frequency band, calculate the interference energy proportion according to the frequency band, and construct the frequency domain mask based on the interference spectrum template;

[0026] S2.315, apply a window function to the original body temperature and physiological signal, perform fast Fourier transform, multiply the complex spectrum by the constructed frequency domain mask, and restore the processed complex spectrum to the enhanced time domain signal through inverse Fourier transform.

[0027] As a further improvement of the technical solution, in S3, based on the feature vector, the long short term memory network is used to predict the body temperature abnormal risk probability in real time, including the following steps:

[0028] S3.1, organize the feature vectors of each time period constructed in a sliding window manner into a time sequence input sequence in time sequence;

[0029] S3.2, input the time sequence input sequence into the multi-layer stacked long short-term memory network, model the time dependence between the features through the gating mechanism thereof, extract deep time sequence features representing the body temperature change trend, and dynamically adjust the feature weight of each modality based on the body temperature trend and the physiological state;

[0030] S3.3, input the feature vector output by the adjusted long short-term memory network to the subsequent fully connected neural network to perform discriminant modeling of the body temperature abnormality risk;

[0031] S3.4, generate the body temperature abnormality risk probability value at the current time through the output layer of the fully connected network.

[0032] As a further improvement of the technical solution, in S3.2, dynamically adjusting the feature weight of each modality based on the body temperature trend and the physiological state comprises the following steps:

[0033] S3.21, extract a scalar value representing the body temperature trend from the long short-term memory network hidden state ;

[0034] S3.22, input the scalar value representing the body temperature trend to the lightweight state classification network to perform physiological state classification and output the newborn physiological state label at the current time;

[0035] S3.23, according to the physiological state label, query the corresponding pre-trained static gating vector, and combine the current body temperature change rate to dynamically adjust the weight to form the final gating weight vector;

[0036] S3.24, apply the gating weight vector to the feature vector output by the long short-term memory network element by element to scale to form the feature vector output by the adjusted long short-term memory network .

[0037] As a further improvement of the technical solution, in S3.23, combining the current body temperature change rate to dynamically adjust the weight to form the final gating weight vector comprises the following steps:

[0038] S3.231, use the scalar value representing the body temperature trend as an indicator of the rate of change of body temperature per unit time to represent the direction and strength of the current body temperature trend;

[0039] S3.232, construct a corresponding dynamic adjustment amount according to the direction and absolute value of the slope;

[0040] S3.233, combine the static gating vector and the dynamic adjustment amount to calculate the final gating weight vector.

[0041] As a further improvement of the technical solution, in the S4, the risk classification judgment is performed based on the abnormal risk probability and in combination with the risk classification rule, and includes the following steps:

[0042] S4.1, receiving the body temperature abnormal risk probability value output by the step S3;

[0043] S4.2, dividing the body temperature risk level based on the body temperature abnormal risk probability value;

[0044] S4.3, judging the risk level of the current body temperature state according to the interval in which the body temperature abnormal risk probability value is located;

[0045] S4.4, outputting the body temperature risk level result at the current time.

[0046] On the other hand, the present application provides a neonatal body temperature abnormality prediction device, comprising:

[0047] The acquisition module is used for real-time acquisition of the body temperature and physiological signals of the neonate and pre-processing;

[0048] The feature extraction module is used for extracting the time sequence features reflecting the body temperature change in the body temperature and physiological signals by using the fast Fourier transform, generating a feature vector, and processing the body temperature and physiological signals through the phase coupling analysis and the frequency domain mask suppression mechanism to dynamically identify the low-frequency noise interference interval caused by the irregular breathing and skin conductive disturbance of the neonate;

[0049] The body temperature abnormality prediction module is used for real-time prediction of the body temperature abnormality risk probability by using the long short-term memory network, and dynamically regulating the gating weight of the long short-term memory network through the body temperature change rate and physiological state label double factors to optimize the feature fusion mechanism;

[0050] The risk classification module is used for risk classification judgment by using the abnormal risk probability and in combination with the risk classification rule.

[0051] On the other hand, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of the neonatal body temperature abnormality prediction method of any one of the above.

[0052] Compared with the prior art, the present application has the following beneficial effects:

[0053] 1. The new-born body temperature abnormality prediction method, device and storage medium provided by the present application, through multi-modal physiological signal fusion and dynamic feature weighting mechanism, significantly improve the accuracy and real-time performance of new-born body temperature abnormality prediction. The long short-term memory network (LSTM) is combined with the body temperature change rate and physiological state label double-factor regulated gate weight to realize the adaptive fusion of multi-source information such as heart rate, respiration, and blood oxygen saturation, effectively enhance the adaptability of the model under different clinical conditions, and accurately identify the potential body temperature abnormality trend in complex physiological fluctuations, which is helpful for early intervention and clinical decision support.

[0054] 2. The new-born body temperature abnormality prediction method, device and storage medium provided by the present application, introduce the noise suppression strategy based on phase coupling analysis and frequency domain mask suppression mechanism, which can dynamically identify and suppress the low-frequency interference caused by irregular respiration and skin conductive disturbance, significantly improve the signal-to-noise ratio of body temperature and physiological signal. Combined with fast Fourier transform and time sequence feature extraction, the model's ability to capture the periodicity and stability of body temperature changes is further enhanced, providing a more stable and reliable physiological data basis for early warning of new-born body temperature abnormalities, and improving the robustness and clinical applicability of the system. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 The overall method flowchart of the present application. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0057] Embodiment 1: Please refer to Figure 1 The present embodiment provides a new-born body temperature abnormality prediction method, which comprises the following steps:

[0058] S1, real-time acquisition of the body temperature and physiological signals of the new-born and pretreatment;

[0059] In the present embodiment, the body temperature includes core body temperature (rectal temperature / ear temperature) and peripheral body temperature (axillary temperature, skin temperature); the physiological signals include heart rate, respiratory rate, and blood oxygen saturation; the sensors used for acquisition include thermal resistance, infrared thermometer, PPG sensor, and thoracic impedance band, wherein the pretreatment includes time alignment and interpolation completion, abnormal value detection and elimination;

[0060] S2, extracting the time sequence features reflecting the body temperature change in the body temperature and physiological signals based on the fast Fourier transform, generating a feature vector, and processing the body temperature and physiological signals through the phase coupling analysis and the frequency domain mask suppression mechanism, dynamically identifying the low-frequency noise interference interval caused by the irregular breathing and the skin conductive disturbance of the newborn;

[0061] In this embodiment, the time sequence features reflecting the body temperature change are extracted from the body temperature and physiological signals, including the following steps:

[0062] S2.1, the continuous body temperature and physiological signal time sequence is divided into multiple analysis segments by using a fixed time window (5 minutes or 10 minutes) to form a sliding window structure;

[0063] S2.2, in each sliding window, the time domain features of the body temperature and various physiological signals are calculated, including mean, variance, skewness, kurtosis, extreme amplitude, fluctuation range, coefficient of variation, etc., to represent the local change trend and stability;

[0064] S2.3, the frequency domain features of the body temperature and physiological signal sequence in each sliding window are extracted by using the fast Fourier transform, including the frequency spectrum distribution, the main frequency component, and the energy concentration degree, which are used for the periodic fluctuation in the body temperature and physiological signals;

[0065] Among them, the frequency domain features of the body temperature and physiological signal sequence in each sliding window are extracted by using the fast Fourier transform, including the following steps:

[0066] S2.31, the body temperature and physiological signal sequence are processed through the phase coupling analysis and the frequency domain mask suppression mechanism, the low-frequency noise interference interval caused by the irregular breathing and the skin conductive disturbance is dynamically identified, the enhanced time domain signal (body temperature and physiological signal sequence) is generated, and the quality and stability of the real physiological oscillation component in the frequency spectrum are improved;

[0067] Further, the phase coupling analysis and frequency domain mask suppression mechanism are used to solve the problem of low frequency noise interference caused by irregular breathing and skin conductive disturbance in neonatal body temperature monitoring. Irregular breathing (periodic breathing or apnea) can cause phase desynchronization between heart rate variability (HRV) signal and respiratory rhythm, which is manifested as abnormal energy fluctuation in low frequency band (0.05-0.15 Hz) in frequency domain. Skin conductive disturbance (limb movement, electrode contact change) can cause mutation (second derivative sudden increase) of skin temperature signal, forming artifacts in time domain, and leaking as wideband noise (especially affecting 0.15-2 Hz physiological rhythm band) in frequency domain. Traditional methods (fixed frequency band filtering or time domain smoothing) are difficult to dynamically distinguish such noise from real body temperature regulation physiological oscillation (vasomotor fluctuation), leading to distortion of feature extraction, and thus reducing the accuracy of body temperature abnormality prediction. The phase coupling analysis method (by detecting the instantaneous phase difference between HRV and respiratory signal) is used to identify the respiratory interference interval in real time, avoiding the identification inaccuracy caused by individual differences of fixed threshold method. At the same time, combined with the second derivative mutation detection of body temperature signal, the skin conductive disturbance period is accurately captured, overcoming the misjudgment problem existing in traditional amplitude threshold method. On this basis, a reference noise template is constructed and a frequency domain mask is generated, and the interference frequency band is dynamically suppressed according to the energy proportion of frequency band (low / middle / high frequency), realizing targeted spectrum purification and maximizing the retention of heart rate variability rhythm and other real physiological oscillation components. Further, through interference interval signal reconstruction (FFT to mask to inverse FFT), the stability and signal-to-noise ratio of time domain signal are enhanced, and the reliability of spectral features such as main frequency and energy concentration is improved. This mechanism effectively eliminates the pseudo-physiological fluctuations caused by breathing and skin disturbance, avoids the misjudgment of noise as body temperature regulation abnormality, ensures that the frequency domain features truly reflect the synergistic mechanism between body temperature and physiological signals, and thus provides solid support for the dynamic gate weight adjustment in LSTM model, finally significantly improves the accuracy and robustness of body temperature abnormality risk classification.

[0068] By phase coupling analysis and frequency domain mask suppression mechanism, the body temperature and physiological signal sequence are processed, and the low frequency noise interference interval caused by irregular breathing and skin conductive disturbance is dynamically identified, including the following steps:

[0069] S2.311, extract RR interval sequence from the physiological signal (the RR interval sequence refers to a sequence of time intervals between R waves (the most prominent wave peak in electrocardiogram) between consecutive heartbeats, which is used to reflect heart rate variability, and the physiological signal specifically refers to a heart rate signal), calculate instantaneous heart rate variability (the calculation of instantaneous heart rate variability specifically includes: interpolation and resampling of the RR interval sequence to obtain a heart rate variability signal with equal time intervals; then, using a sliding window to calculate the local standard deviation or frequency domain feature (LF / HF ratio) at each time, thereby obtaining the instantaneous heart rate variability index at the time), synchronize to derive the respiration signal (preferably using thoracic impedance band data; if not available, derive respiration from PPG (photoplethysmogram) (extracted by respiration-induced modulation RIM method or PPG envelope waveform)), use Hilbert transform to obtain the instantaneous phase of the instantaneous heart rate variability and the respiration signal, and calculate the phase difference between the two, when the phase difference is greater than the upper threshold value a (in this embodiment, the value of the upper threshold value a is 0.5π radian), mark the x interval centered at the current time as a respiratory disturbance interval (in this embodiment, x is 200ms);

[0070] wherein the instantaneous phase of the instantaneous heart rate variability is:

[0071] ;

[0072] the instantaneous phase of the respiration signal is:

[0073] ;

[0074] wherein, is a Hilbert transform operator, is time, is the instantaneous heart rate variability signal at time , represents the phase angle of a complex number, i.e. the instantaneous phase of the analytic signal, is the respiration signal at time ;

[0075] the phase difference between the two is:

[0076] ;

[0077] S2.312, calculate the second derivative of body temperature, if the absolute value of the second derivative is greater than the derivative threshold value (in this embodiment, , is degrees Celsius per square second), then expand a fixed length window centered at the current time point as a skin conductive disturbance interval;

[0078] Wherein, the second derivative of the body temperature (skin temperature) is:

[0079] ;

[0080] In the formula, is the value of the skin temperature at time , represents the second derivative of time , that is, the acceleration, is the acceleration component of the skin temperature change, reflecting the trend of the temperature change;

[0081] S2.313, extract the reference noise signal of the respiratory disturbance interval and the skin conductive disturbance interval (respiratory disturbance: separate the abnormal section (interval of respiratory amplitude < baseline 30%) from the respiratory signal; skin disturbance: extract the mutation section (interval of the absolute value of the second derivative of the skin temperature greater than the skin temperature threshold e) from the skin temperature signal, in the embodiment, the empirical value of the skin temperature threshold e is 0.05°C / s 2 ), the reference noise signal only takes the abnormal time section in the interference window, not the entire sliding window), windowing (Hanning window) and fast Fourier transform are performed on the reference noise signal to obtain the complex spectrum, and the modulus (that is, the amplitude) of the complex spectrum is obtained to obtain the interference spectrum template;

[0082] S2.314, divide the frequency band (low frequency 0.05-0.15Hz, medium frequency 0.15-0.5Hz, high frequency 0.5-2Hz), calculate the interference energy proportion according to the frequency band, and construct a frequency domain mask based on the interference spectrum template (based on the spectrum of the reference noise signal, calculate the energy proportion according to the frequency band, and then construct the mask) for suppressing noise energy leakage;

[0083] S2.315, apply a window function (Hanning window) to the original body temperature and physiological signals, perform fast Fourier transform, multiply the complex spectrum by the constructed frequency domain mask, suppress the energy of the interference frequency band, and restore the processed complex spectrum to the enhanced time domain signal through inverse Fourier transform, realize sequence reconstruction after interference suppression;

[0084] S2.32, apply fast Fourier transform (FFT) to the enhanced time domain signal to convert the enhanced time domain signal to a frequency domain complex spectrum;

[0085] S2.33, calculate the corresponding amplitude spectrum of each frequency domain complex spectrum, select the main frequency component amplitude in the frequency band (the main frequency is 0.05-2Hz, covering the body temperature regulation and heart rate variation physiological activity rhythm), and constitute a frequency domain feature vector;

[0086] S2.34, normalize each frequency domain feature vector (total amplitude normalization);

[0087] S2.35, splice and fuse all normalized frequency domain feature vectors (including body temperature, heart rate, respiratory rate, and blood oxygen saturation) to form a joint frequency domain feature vector;

[0088] S2.4, using a cross-modal analysis method, based on the body temperature and physiological signal sequence of the sliding window, calculate the dynamic coupling features, including correlation coefficient, mutual information, phase synchronization index, cross power spectral density (the specific steps are: first, pair the body temperature sequence and each type of physiological signal (heart rate, respiratory rate, and blood oxygen saturation) in each sliding window, and calculate the Pearson correlation coefficient, mutual information value, phase synchronization index, and cross power spectral density therebetween), reflecting the collaborative fluctuation relationship between body temperature change and other physiological activities;

[0089] S2.5, filter out the time domain features, frequency domain features, and dynamic coupling features reflecting the body temperature change through a feature selection algorithm (recursive feature elimination) (specifically: first, use the performance index (classification accuracy or loss function) of the body temperature anomaly prediction model as the evaluation standard, iteratively train the model and gradually eliminate the features with the smallest contribution to prediction, until a subset of features that are most representative and sensitive to body temperature change is retained), and combine the filtered features into a unified feature vector , , wherein, is the i-th feature of the original feature vector, that is, the complete feature set spliced from body temperature, physiological signals, frequency domain features, and coupling features, is the i-th feature in the original feature vector, , is the final selected feature number, , is the final selected feature number, is the feature index, indicating the index sequence number of the i-th feature retained after feature selection in the original feature vector.

[0090] S3, based on the feature vector, use the long short-term memory network to predict the body temperature anomaly risk probability in real time, and dynamically regulate the gating weight of the long short-term memory network through the body temperature change rate and the physiological state label double factors, optimize the feature fusion mechanism;

[0091] In this embodiment, based on the feature vector, the long short-term memory network is used to predict the body temperature anomaly risk probability in real time, including the following steps:

[0092] S3.1, organize the feature vectors of each time period constructed in the sliding window manner into a time sequence input sequence in time sequence;

[0093] ​S3.2, input the time series input sequence into a multi-layer stacked long short-term memory network (LSTM) to model the time dependence between features through its gating mechanism, extract deep time series features representing the trend of body temperature change, and dynamically adjust the weight of each modality feature based on the trend of body temperature and physiological state, wherein the basic architecture of the long short-term memory network (LSTM) includes an input layer, one or more LSTM hidden layers, and an output layer. The input layer receives a sequence of features organized by time steps, and the input of each time step is a feature vector; the LSTM hidden layer is composed of multiple units, each unit contains an input gate, a forget gate and an output gate to control the writing, retention and output of information, and captures long-term dependencies through an internal memory unit; the hidden state vector is transmitted over time to extract the time series pattern of body temperature and physiological signals; finally, the output of the LSTM is sent to the subsequent fully connected layer (multilayer perceptron) for body temperature abnormality probability prediction or classification judgment;

[0094] Wherein, in the body temperature rising / falling stage (rapid heating period or recovery period), different physiological signals (heart rate, respiration, blood oxygen) have significantly different indicative weights for body temperature abnormalities (heart rate signal is more sensitive in the early stage of fever, and respiration signal is more critical in the high fever period); the same body temperature value has different pathological meanings in different physiological states (normal, low fever, high fever), and needs to be differentially fused with multi-modal features (the weight of blood oxygen saturation needs to be higher than that in the normal state); the traditional static weighted model (fixed weight fusion or single LSTM gate) cannot adapt to such dynamic correlation, resulting in inaccurate feature fusion and reduced prediction specificity; dynamically adjusting the weight of each modality feature based on the trend of body temperature and physiological state drives the basic gating vector through the physiological state label, solidifies the expert experience weight in different states (including automatically increasing the channel weight of the respiration feature in the high fever state), generates a dynamic compensation based on the rate of change of body temperature, and real-time strengthens the modality features that are strongly related to the current trend (amplifies the sensitivity of heart rate variation features when the body temperature rises rapidly); through complete real-time sensitivity adjustment under state constraints and avoid overfitting risk of pure data-driven models; at the output end of the LSTM, the channel-level reweighting of each modality feature is performed , breaking through the traditional LSTM coarse-grained fusion limit, so that the model can preferentially focus on sudden indicators such as heart rate during rapid body temperature changes, and automatically focus on high-discriminative modalities (coordinated changes of respiration and blood oxygen) in abnormal states such as high fever, significantly improving the generalization ability in complex clinical scenarios;

[0095] Dynamically adjusting the weight of each modality feature based on the trend of body temperature and physiological state includes the following steps:

[0096] S3.21, extracting a scalar value representing the trend of body temperature from the long short-term memory network (LSTM) hidden state (Concretely: input the hidden state vector into a fully connected layer, and output a scalar value as the estimate of the temperature trend (the unit is temperature / time), while retaining the hidden state vector for subsequent steps);

[0097] S3.22, input the scalar value representing the temperature trend into a lightweight state classification network (the lightweight state classification network is composed of a 3-layer convolutional neural network, and the number of channels in each layer is 16 / 32 / 4, and the kernel size is 3), to perform physiological state classification, and output the newborn physiological state label (including normal, low fever, and high fever) at the current time;

[0098] S3.23, according to the physiological state label, query the corresponding pre-trained static gating vector (the static gating vector is generated by statistical analysis of the multi-modal contribution of historical samples in similar states, indicating the importance distribution of each modal feature (including heart rate, respiratory rate, and blood oxygen saturation) to temperature prediction in the current state), while combining the current temperature change rate to dynamically adjust the weight, forming the final gating weight vector;

[0099] Further, combine the current temperature change rate to dynamically adjust the weight, form the final gating weight vector, including the following steps:

[0100] S3.231, use the scalar value representing the temperature trend as the speed indicator of the temperature change per unit time, to represent the change direction and intensity of the current temperature trend (including rising, falling, or fluctuation amplitude);

[0101] S3.232, according to the direction (positive / negative) and absolute value size (speed strength) of the slope, construct the corresponding dynamic adjustment amount , which is used to reflect the sensitivity compensation of each modal information under different temperature change rates, wherein , wherein is a learnable scaling factor, is a learnable weight vector (dimension ), that is, through the training process of the neural network model, using a large amount of newborn temperature and physiological data, through an optimization algorithm (such as gradient descent), automatically learned and adjusted;

[0102] S3.233, combine the static gating vector and the dynamic adjustment amount to calculate the final gating weight vector (which is a weight vector after nonlinear normalization, with a value range of [0, 1]), , wherein is an S-shaped nonlinear activation function, is the current physiological state The generated basic gating vector Current physiological state;

[0103] S3.24. Feature vectors output by Long Short-Term Memory (LSTM) networks Element-wise scaling with gated weight vectors is applied to achieve dynamic weighted fusion of features from different modalities, forming the adjusted feature vector output by the Long Short-Term Memory (LSTM) network. , In the formula, This involves Hadamard product (element-wise multiplication) to achieve adaptive reweighting of feature channels across modalities; specifically, a static gating vector and a dynamic adjustment vector. and the final gating weight vector All dimensions are The dimension d is consistent with that of the feature vector output by the Long Short-Term Memory network.

[0104] S3.3, The feature vector output by the adjusted Long Short-Term Memory (LSTM) network. The input is fed into a fully connected neural network (Multilayer Perceptron MLP) to perform discriminative modeling of abnormal body temperature risk. The basic architecture of the fully connected neural network (Multilayer Perceptron MLP) includes an input layer, one or more hidden layers, and an output layer. The input layer receives feature vectors from the output of LSTM or other modules. The hidden layer consists of several fully connected neurons, with an activation function (ReLU) to introduce nonlinearity. The number of layers and the number of neurons per layer can be adjusted according to the task complexity. The final output layer uses the Sigmoid activation function to compress the result to the 0~1 range, which is used to output the probability of abnormal body temperature risk at the current time.

[0105] S3.4. Generate the probability value of abnormal body temperature at the current moment through the output layer of the fully connected network;

[0106] The probability value of abnormal body temperature is:

[0107] ;

[0108] In the formula, For the current moment The probability value of abnormal body temperature (ranging from 0 to 1, representing the probability of abnormal body temperature). This is the Sigmoid function, used to compress a linear output to the range of 0 to 1. This is the weight matrix of the output layer of a fully connected neural network (i.e., a multilayer perceptron, MLP), with dimension 1. ,in It is the dimension of the input features. This is a bias term, a scalar.

[0109] S4, risk classification is determined based on the abnormal risk probability and in combination with a risk grading rule;

[0110] In this embodiment, the risk classification is determined based on the abnormal risk probability and in combination with a risk grading rule, including the following steps:

[0111] S4.1, receiving the body temperature abnormal risk probability value output in step S3;

[0112] S4.2, dividing the body temperature risk level based on the body temperature abnormal risk probability value, the body temperature risk level including low risk (normal), medium risk (suspicious), and high risk (abnormal);

[0113] Low risk (normal): ;

[0114] Medium risk (suspicious): ;

[0115] High risk (abnormal): ;

[0116] wherein, is a body temperature risk threshold value, is a body temperature risk upper threshold value;

[0117] S4.3, determining the risk level of the current body temperature state according to the interval in which the body temperature abnormal risk probability value is located;

[0118] S4.4, outputting the body temperature risk level result at the current time.

[0119] Embodiment 2: The present embodiment provides a new-born baby body temperature abnormality prediction device, comprising:

[0120] a collection module for collecting and pre-processing the body temperature and physiological signals of the new-born baby in real time;

[0121] a feature extraction module for extracting time sequence features reflecting body temperature changes in the body temperature and physiological signals by using fast Fourier transform, generating a feature vector, and processing the body temperature and physiological signals by using phase coupling analysis and frequency domain mask suppression mechanism to dynamically identify low-frequency noise interference intervals caused by irregular breathing and skin conductive disturbance of the new-born baby;

[0122] a body temperature abnormality prediction module for predicting the body temperature abnormality risk probability in real time by using a long short-term memory network, and dynamically adjusting the gating weight of the long short-term memory network by using the body temperature change rate and physiological state label double factors to optimize the feature fusion mechanism;

[0123] a risk classification module for determining the risk classification by using the abnormal risk probability and in combination with a risk grading rule.

[0124] The embodiment also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the new-born infant abnormal body temperature prediction method.

[0125] The basic principles, main features and advantages of the present application are shown and described above. It should be understood by those skilled in the art that the present application is not limited by the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A device for predicting abnormal body temperature in newborns, characterized in that, include: The data acquisition module is used to collect and preprocess the newborn's body temperature and physiological signals in real time. The feature extraction module is used to extract the temporal features reflecting changes in body temperature from body temperature and physiological signals using fast Fourier transform, generate feature vectors, and process body temperature and physiological signals through phase coupling analysis and frequency domain masking suppression mechanism to dynamically identify the low-frequency noise interference range caused by irregular breathing and skin conductivity disturbances in newborns. The abnormal body temperature prediction module is used to predict the probability of abnormal body temperature in real time using a long short-term memory network. It also optimizes the feature fusion mechanism by dynamically adjusting the gating weights of the long short-term memory network through a dual factor of body temperature change rate and physiological state label. The risk classification module is used to classify risks by utilizing the probability of abnormal risks and combining them with risk grading rules.

2. The neonatal body temperature abnormality prediction device according to claim 1, characterized in that: The operation of the device includes the following steps: S1. Real-time acquisition and preprocessing of newborns' body temperature and physiological signals; S2. Based on Fast Fourier Transform, extract the temporal features reflecting changes in body temperature from body temperature and physiological signals, generate feature vectors, and process body temperature and physiological signals through phase coupling analysis and frequency domain masking suppression mechanism to dynamically identify the low-frequency noise interference range caused by irregular breathing and skin conductivity disturbances in newborns. S3. Based on feature vectors, the Long Short-Term Memory (LSTM) network is used to predict the probability of abnormal body temperature in real time. The gating weights of the LSTM network are dynamically adjusted by two factors: the rate of change of body temperature and physiological state labels, thereby optimizing the feature fusion mechanism. S4. Risk classification judgment is made based on the probability of abnormal risks and combined with risk classification rules.

3. The neonatal body temperature abnormality prediction device according to claim 2, characterized in that: In step S2, extracting temporal features reflecting changes in body temperature from body temperature and physiological signals includes the following steps: S2.

1. A fixed time window is used to divide the continuous body temperature and physiological signal time series into multiple analysis segments, forming a sliding window structure; S2.2 Calculate the time-domain characteristics of body temperature and various physiological signals within each sliding window; S2.

3. For the body temperature and physiological signal sequences within each sliding window, the frequency domain features are extracted using Fast Fourier Transform to be used for periodic fluctuations in body temperature and physiological signals. S2.

4. Using the cross-modal analysis method, based on the body temperature and physiological signal sequences of the sliding window, calculate the dynamic coupling characteristics to reflect the coordinated fluctuation relationship between body temperature changes and other physiological activities. S2.

5. Use a feature selection algorithm to filter out time-domain features, frequency-domain features, and dynamic coupling features that reflect changes in body temperature, and combine the filtered features into a unified feature vector. , In the formula, The first feature vector is the first feature vector. One characteristic, The first eigenvector in the original feature vector One characteristic, The final number of features selected. For feature index, The feature index is used for filtering.

4. The neonatal body temperature abnormality prediction device according to claim 3, characterized in that: In S2.3, the frequency domain features are extracted using Fast Fourier Transform for the body temperature and physiological signal sequences within each sliding window, including the following steps: S2.

31. By processing the body temperature and physiological signal sequences through phase coupling analysis and frequency domain masking suppression mechanism, the low-frequency noise interference range caused by irregular breathing and skin conductivity disturbance is dynamically identified, and an enhanced time domain signal is generated. S2.

32. Apply Fast Fourier Transform to the enhanced time-domain signal to convert the enhanced time-domain signal into a frequency-domain complex spectrum. S2.

33. Calculate the corresponding amplitude spectrum for each frequency domain complex spectrum, select the amplitude of the main frequency components in the frequency band, and construct the frequency domain feature vector. S2.

34. Normalize each frequency domain feature vector; S2.

35. All normalized frequency domain feature vectors are concatenated and fused to form a joint frequency domain feature vector.

5. The neonatal body temperature abnormality prediction device according to claim 4, characterized in that: In step S2.31, the body temperature and physiological signal sequences are processed through phase coupling analysis and frequency domain masking suppression mechanisms to dynamically identify low-frequency noise interference ranges caused by irregular breathing and skin conductivity disturbances, including the following steps: S2.311 Extract the RR interval sequence from the physiological signal, calculate the instantaneous heart rate variability, synchronously export the respiratory signal, use Hilbert transform to obtain the instantaneous phase of the instantaneous heart rate variability and the respiratory signal, and calculate the phase difference between the two. When the phase difference is greater than the upper limit threshold a, mark the x interval centered at the current time as the respiratory interference interval. S2.

312. Calculate the second derivative of body temperature. If the absolute value of the second derivative is greater than the derivative threshold... When the current time point is used as the center, a fixed-length window is extended as the skin conductivity interference range; S2.

313. Extract the reference noise signal from the breathing interference region and the skin conduction interference region, window the reference noise signal and perform a fast Fourier transform to obtain the complex spectrum, and take the modulus of the complex spectrum to obtain the interference spectrum template. S2.

314. Divide the frequency bands, calculate the proportion of interference energy according to the frequency bands, and construct a frequency domain mask based on the interference spectrum template; S2.

315. Apply a window function to the original body temperature and physiological signals, perform a fast Fourier transform, multiply the complex spectrum by the constructed frequency domain mask, and restore the processed complex spectrum to the enhanced time domain signal through an inverse Fourier transform.

6. The neonatal body temperature abnormality prediction device according to claim 2, characterized in that: In step S3, the probability of abnormal body temperature risk is predicted in real time using a long short-term memory network based on feature vectors, including the following steps: S3.1 Organize the feature vectors of each time period constructed in the sliding window method into a time-series input sequence in chronological order; S3.2 Input the temporal input sequence into a multi-layer stacked long short-term memory network, model the temporal dependency between features through its gating mechanism, extract deep temporal features representing the trend of body temperature change, and dynamically adjust the weights of each modality feature based on the body temperature trend and physiological state. S3.3, The feature vector output by the adjusted Long Short-Term Memory network The input is fed into a fully connected neural network to perform discriminative modeling of the risk of abnormal body temperature. S3.

4. Generate the probability value of abnormal body temperature at the current moment through the output layer of the fully connected network.

7. The neonatal body temperature abnormality prediction device according to claim 6, characterized in that: In step S3.2, the dynamic adjustment of the modal feature weights based on body temperature trends and physiological states includes the following steps: S3.21 Extract a scalar value representing the body temperature trend from the hidden states of the Long Short-Term Memory network. ; S3.22, the scalar value representing the trend of body temperature. Input a lightweight state classification network to classify physiological states and output the current physiological state label of the newborn. S3.

23. Based on the physiological state label, query the corresponding pre-trained static gating vector, and dynamically adjust the weights in combination with the current body temperature change rate to form the final gating weight vector. S3.

24. Feature vectors output by the Long Short-Term Memory network Element-wise scaling with a gated weight vector yields the adjusted feature vector output by the Long Short-Term Memory (LSTM) network. .

8. The neonatal body temperature abnormality prediction device according to claim 7, characterized in that: In step S3.23, the weights are dynamically adjusted based on the current rate of change in body temperature to form the final gating weight vector, including the following steps: S3.231, Use scalar values ​​that represent the trend of body temperature. As an indicator of the rate of change in body temperature per unit time, it is used to represent the direction and intensity of the current trend in body temperature. S3.

232. Based on the direction and absolute value of the slope, construct the corresponding dynamic adjustment amount; S3.

233. Combine the static gating vector with the dynamic adjustment amount to calculate the final gating weight vector.

9. The neonatal body temperature abnormality prediction device according to claim 2, characterized in that: In step S4, risk classification is performed based on the probability of abnormal risks and combined with risk grading rules, including the following steps: S4.1 Receive the probability value of abnormal body temperature output in step S3; S4.2 Classify body temperature risk levels based on the probability value of abnormal body temperature risk; S4.

3. Determine the risk level of the current body temperature status based on the range of the probability value of abnormal body temperature. S4.4 Output the current body temperature risk level result.

Images