Patient vital sign intelligent management system based on wearable device

Abstract

The invention discloses a patient vital sign intelligent management system based on a wearable device, and relates to the technical field of patient vital sign intelligent management, and the system comprises the following steps: obtaining a lung audio signal and an electrocardiosignal based on the wearable device worn on the chest of a lung cancer patient, and constructing a heart sound trigger point sequence; extracting lung sound signal segments in a preset heartbeat period to obtain a plurality of first signals; performing feature screening on the first signals with the same heartbeat cycle according to a plurality of preset feature dimensions to obtain a plurality of second signals, and extracting a power spectral density curve to obtain a plurality of residual interference energy sequences; and calculating variable coefficients of the residual interference energy sequence in a plurality of adjacent heartbeat cycles, selecting a second signal with the minimum variable coefficient as an optimal lung audio signal, and finally applying the optimal lung audio signal to intelligent vital sign management, thereby solving the problem that in a traditional method, lung sound and heart sound cannot be effectively distinguished. And the problem that lung sound characteristics are weakened by mistake or heart sound is left is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent management of patient vital signs, and more specifically, to an intelligent management system for patient vital signs based on wearable devices. Background Art

[0002] Lung cancer is one of the most common malignant tumors with the highest morbidity and mortality rates worldwide. Its early symptoms are often atypical, resulting in most patients being diagnosed in the advanced stages. Lung auscultation, a crucial diagnostic tool for lung cancer and its complications (such as obstructive pneumonia and pulmonary consolidation), has garnered widespread attention in recent years through wearable device technology.

[0003] With the rapid development of medical information technology and intelligent technologies, the use of wearable devices to monitor and manage patient vital signs has become an important research and application direction. In particular, in chronic disease management and telemedicine, wearable devices, with their real-time, non-invasive, and portable nature, have significantly improved the continuity and efficiency of medical services. Existing vital sign monitoring systems primarily rely on traditional parameters such as heart rate, body temperature, blood pressure, and blood oxygen saturation, supplemented by simple early warning mechanisms. However, such systems often overlook the dynamic assessment of respiratory-related signs, especially in the remote management of patients with lung diseases such as lung cancer and chronic obstructive pulmonary disease (COPD), which lack the accurate extraction and analysis of key indicators such as lung sounds.

[0004] However, lung sounds and heart sounds have a high degree of frequency overlap. Rhonchi, in particular, have a frequency range (100-150 Hz) similar to heart sounds. Bandpass filters or wavelet denoising methods used in wearable devices cannot effectively distinguish between the two, potentially leading to falsely weakened lung sound features or residual heart sounds. To address these issues, the present invention proposes a solution. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an intelligent patient vital signs management system based on wearable devices. By performing feature screening on signals with the same heartbeat cycle, the optimal lung audio signal is obtained to solve the problem that traditional methods cannot effectively distinguish between lung sounds and heart sounds, resulting in the lung sound features being mistakenly weakened or heart sounds remaining.

[0006] To achieve the above object, the present invention provides the following technical solutions: A wearable device-based intelligent patient vital signs management system includes a data acquisition module, a signal extraction module, a signal screening module, a signal evaluation module, and a vital signs management module, and there are connections between the modules: the data acquisition module is used to acquire lung audio signals and electrocardiogram signals based on the wearable device worn on the chest of a lung cancer patient; the signal extraction module is used to construct a heart sound trigger point sequence based on the electrocardiogram signal, and extract lung sound signal segments within a preset heartbeat cycle to obtain several first signals; the signal screening module is used to perform feature screening on the first signals of the same heartbeat cycle according to several preset feature dimensions to obtain several second signals, and extract power spectral density curves to obtain several residual interference energy sequences; the signal evaluation module is used to calculate the coefficient of variation of the residual interference energy sequence within several adjacent heartbeat cycles, and select the second signal with the smallest coefficient of variation as the optimal lung audio signal for the current heartbeat cycle; the vital signs management module is used to input the optimal lung audio signal into a preset neural network model to identify the lung sound type, and perform intelligent vital signs management in combination with the vital sign parameters of the lung cancer patient.

[0007] In a preferred embodiment, the construction of a heart sound trigger point sequence based on the ECG signal is specifically as follows: preprocessing the ECG signal to obtain a denoised ECG signal; detecting the R wave peak point of the denoised ECG signal, and dividing the heartbeat cycle based on the R wave peak point; within each heartbeat cycle, extracting the starting points of the first heart sound and the second heart sound as trigger points based on the time-frequency characteristics of the heart sound signal, and marking their timestamps; arranging the trigger points of several consecutive heartbeat cycles in chronological order to obtain a heart sound trigger point sequence.

[0008] In a preferred embodiment, the extraction of lung sound signal segments within a preset heartbeat cycle to obtain a plurality of first signals is specifically as follows: based on the starting point of the heart sound trigger point sequence, the lung sound signal is intercepted within a preset time window, where the time window covers the diastole between the first heart sound and the second heart sound and the systole between the second heart sound and the first heart sound of the next cycle; and an adaptive threshold method is used to separate the intercepted lung sound signal from the ambient noise to obtain the first signal.

[0009] In a preferred embodiment, the feature screening is performed according to several preset feature dimensions to obtain the second signal, specifically: time domain features are extracted from the first signal, and the time domain features include signal amplitude mean, variance and zero-crossing rate; frequency domain features are extracted from the first signal, and the frequency domain features include Mel-frequency cepstral coefficient components and energy entropy; frequency domain features are fused with frequency domain features to obtain a fused feature vector; principal component analysis is used to reduce the dimension of the fused feature vector to obtain a principal component reconstructed signal; and the second signal is generated based on the principal component reconstructed signal.

[0010] In a preferred embodiment, the power spectral density curve is extracted to obtain several residual interference energy sequences, specifically: the second signal is subjected to wavelet decomposition to separate sub-signals of different frequency bands; the power spectral density curve of each sub-signal is calculated, and abnormal peaks in the curve are detected; the interference signal corresponding to the abnormal peak is removed by a matching pursuit algorithm to obtain a residual signal; the residual signal is segmented according to the heartbeat cycle, the energy integral of each segment is calculated, and a residual interference energy sequence is generated.

[0011] In a preferred embodiment, the calculation of the coefficient of variation of the residual interference energy sequence within adjacent heartbeat cycles is specifically as follows: selecting the residual interference energy sequence of N consecutive heartbeat cycles, calculating the mean and standard deviation of each sequence, and taking the ratio of the standard deviation to the mean as the coefficient of variation; dynamically updating the coefficient of variation through a sliding window method, and recording the heartbeat cycle index corresponding to the minimum coefficient of variation.

[0012] In a preferred embodiment, the second signal with the smallest coefficient of variation is selected as the optimal lung audio signal, specifically: locating the corresponding second signal according to the minimum coefficient of variation index; performing time domain alignment and phase correction on the located second signal to generate the optimal lung audio signal.

[0013] In a preferred embodiment, the intelligent management of vital signs is performed in combination with vital sign parameters, specifically: a risk level score is generated based on the probability distribution of lung sound types; if moist rales or pleural friction sounds are detected and the respiratory rate exceeds a preset threshold, an early warning signal is triggered and a bronchodilator use plan is recommended; if the blood oxygen saturation is continuously lower than the critical value and the cough frequency is abnormal, the telemedicine platform is activated to transmit data to the doctor's terminal in real time; and the management strategy is dynamically adjusted, including medication reminders, respiratory training guidance, and emergency referral recommendations, to form a closed-loop management process.

[0014] The technical effects and advantages of the patient vital signs intelligent management system based on wearable devices of the present invention are as follows: 1. This invention uses a data acquisition module to acquire lung audio signals and ECG signals in real time from a wearable device worn on the patient's chest, ensuring efficient data collection. Next, a signal extraction module constructs a sequence of heart sound trigger points based on the ECG signals and extracts lung sound signal segments, laying the foundation for subsequent signal screening and processing. Through multi-dimensional feature screening and principal component analysis, the system extracts time-domain and frequency-domain features from the lung sound signals, reducing data redundancy and enhancing signal validity, ensuring the accuracy of subsequent analysis.

[0015] Furthermore, through wavelet decomposition and power spectral density curve calculation, the system effectively removes interfering noise from the signal and extracts the residual interference energy sequence, making the analysis of lung sound signals purer and more reliable. The coefficient of variation calculation method further improves the accuracy of signal evaluation, dynamically selecting the optimal lung audio signal and reducing errors caused by environmental interference or other factors.

[0016] In terms of intelligent management of vital signs, the system can generate a risk level score based on the probability distribution of lung sound types, and combine it with the patient's vital sign parameters (such as blood oxygen saturation, respiratory rate, etc.) to conduct dynamic health monitoring and management. When abnormal conditions such as moist rales or pleural friction sounds are detected, the system will trigger an early warning signal, recommend the use of bronchodilators, and transmit data to the doctor's terminal in real time through the telemedicine platform when necessary, further improving the patient's opportunity for timely intervention. By dynamically adjusting management strategies, the system can provide patients with personalized medication reminders, breathing training guidance, emergency referral recommendations, etc., forming a closed-loop management process, significantly improving the patient's quality of life and health management results. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a structural diagram of the patient vital signs intelligent management system based on wearable devices of the present invention. DETAILED DESCRIPTION

[0018] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] Example 1, Figure 1 The present invention provides an intelligent management system for patient vital signs based on wearable devices, including a data acquisition module, a signal extraction module, a signal screening module, a signal evaluation module, and a vital sign management module: A data acquisition module is used to acquire lung audio signals and electrocardiogram signals based on a wearable device worn on the chest of a lung cancer patient; In this example, the wearable device is: The wearable device includes a multi-channel MEMS micro-microphone array, a three-lead ECG electrode module, a low-power signal processing chip, and a synchronous clock control module, as follows: Multi-channel MEMS micro-microphone array: Attached to the second to fifth intercostal spaces on the patient's chest, covering the sound source area on the front of the lungs, responsible for collecting lung sound signals during breathing; Three-lead ECG electrode module: includes right arm, left arm, and chest leads, embedded in clothing patches or fixed to the skin surface via medical patches, for real-time ECG signal acquisition; Low-power signal processing chip: This chip synchronizes lung sound and ECG signal sampling (sampling rates are 8kHz for lung sounds and 500Hz for ECG), and wirelessly transmits the data to a mobile terminal or remote server via a Bluetooth Low Energy (BLE) module. Synchronous clock control module: used to synchronize the timing of lung sound and ECG channels to ensure that the two signals are aligned at the millisecond level.

[0020] It should be noted that after the user wears the device, the system starts the automatic calibration program to detect the electrode contact impedance and the working status of the microphone; the electrocardiogram module begins to collect ECG signals and performs real-time pre-detection of the R wave to mark the boundary of the heart cycle; the microphone array is started synchronously to collect the patient's lung audio signals during breathing; all signal data has a unified timestamp to ensure that the subsequent algorithm module can accurately extract lung sound segments within the heart cycle; the collected data is uploaded to the server for subsequent processing, including steps such as heart sound trigger point extraction, lung sound interception, feature screening and physical sign assessment.

[0021] A signal extraction module is used to construct a heart sound trigger point sequence based on the electrocardiogram signal and extract lung sound signal segments within a preset heartbeat cycle to obtain a plurality of first signals; In this example, a heart sound trigger point sequence is constructed based on the ECG signal, specifically: Preprocess the ECG signal to obtain a denoised ECG signal; Detect the R-wave peak point of the denoised ECG signal and divide the heartbeat cycle based on the R-wave peak point; In each heartbeat cycle, the starting points of the first and second heart sounds are extracted as trigger points based on the time-frequency characteristics of the heart sound signal, and their timestamps are marked; The trigger points of several consecutive heartbeat cycles are arranged in chronological order to obtain a heart sound trigger point sequence.

[0022] It should be noted that a heart sound trigger point sequence is a sequence of specific time points (trigger points) identified by analyzing the ECG and heart sound signals. These points correspond to the primary acoustic events (first and second heart sounds) produced by the heart during each cardiac cycle. This sequence, arranged in chronological order, accurately reflects the synchronization between the mechanical motion and electrophysiological activity of the heart during pumping. Its construction relies on detecting the peak of the R wave in the ECG signal. The R wave is the most prominent part of the QRS complex in the ECG and represents ventricular depolarization, the moment when the ventricles begin to contract. Due to its distinct peak characteristic, the R wave is often used as the reference point for demarcating each cardiac cycle. Within each demarcated cardiac cycle, the time-frequency characteristics of the heart sound signal are further combined to identify the onsets of the first and second heart sounds. The first heart sound typically occurs shortly after the R wave and corresponds to the sound of the atrial valve closing at the onset of ventricular contraction; the second heart sound occurs at the beginning of ventricular diastole and corresponds to the closing of the aortic and pulmonary valves. By extracting and marking the timestamps of each first and second heart sound, a complete, time-ordered sequence of heart sound trigger points is formed in multiple consecutive heartbeat cycles, which can be used to subsequently accurately capture lung sound signal fragments and improve the time alignment precision and accuracy of signal analysis.

[0023] In this example, lung sound signal segments within a preset heartbeat cycle are extracted to obtain several first signals, specifically: Based on the starting point of the heart sound trigger point sequence, the lung sound signal is intercepted within a preset time window, where the time window covers the systolic period between the first heart sound and the second heart sound, and the diastolic period between the second heart sound and the first heart sound of the next cycle; An adaptive threshold method is used to separate the intercepted lung sound signal from the environmental noise to obtain a first signal.

[0024] It should be noted that the period between the first and second heart sounds is called ventricular systole, which represents the process from the onset of the heart's contraction to the ejection of blood from the body. The period between the second heart sound and the next first heart sound is called ventricular diastole, which represents the process from the closing of the aortic valve to the relaxation of the heart and the refilling of blood. Strictly limiting the capture of lung sound signals to the systolic period between the first and second heart sounds and the diastolic period between the second and next first heart sounds enables accurate coverage of lung sounds throughout the entire cardiac cycle. This not only ensures high synchronization between lung sound segments and cardiac activity, but also effectively avoids heart sound interference and the incorporation of non-respiratory noise. This time-window positioning method based on heart sound trigger points is more personalized and dynamically adaptable than traditional fixed time window methods, and can improve the extraction quality and analysis accuracy of lung sound signals, providing more stable and reliable raw data support for subsequent lung sound classification, vital sign assessment, and intelligent early warning.

[0025] Furthermore, the present invention significantly improves the accuracy and practicality of lung sound acquisition by extracting lung sound signal segments within a preset time window based on a sequence of heart sound trigger points. Compared with the traditional method of intercepting lung sounds at fixed time intervals, the first and second heart sounds are used as reference points to clearly intercept lung sound data in the two key stages of ventricular systole and diastole, so that the extracted signal is more in line with the physiological rhythm, effectively avoiding heart sound interference and the mixing of irrelevant background noise. At the same time, the intercepted lung sound signal is noise-stripped by combining the adaptive threshold method, and the discrimination criteria are dynamically adjusted to adapt to different environments and individual differences, further enhancing the signal-to-noise ratio and recognition stability. This solution not only improves the quality of the original signal, but also provides a more accurate and reliable data basis for subsequent lung sound feature extraction, classification and recognition, and personalized vital signs monitoring, thereby achieving more efficient and intelligent lung disease early warning and intervention in the intelligent health management system.

[0026] A signal screening module is used to perform feature screening on the first signal of the same heartbeat cycle according to a plurality of preset feature dimensions to obtain a plurality of second signals, and extract power spectrum density curves to obtain a plurality of residual interference energy sequences; In this example, feature screening is performed according to several preset feature dimensions to obtain the second signal, specifically: Extracting time domain features from the first signal, wherein the time domain features include signal amplitude mean, variance, and zero-crossing rate; Extracting frequency domain features from the first signal, wherein the frequency domain features include Mel-frequency cepstral coefficient components and energy entropy; Fuse the frequency domain features and the frequency domain features to obtain a fused feature vector; Principal component analysis is used to reduce the dimension of the fused feature vector and obtain the principal component reconstruction signal; The signal is reconstructed based on the principal components to generate a second signal.

[0027] It is important to note that time-domain features reflecting the overall signal variation trend (such as amplitude mean, variance, and zero-crossing rate) and frequency-domain features reflecting the spectral distribution pattern (such as Mel-frequency cepstral coefficients and energy entropy) are extracted from the first signal simultaneously. These two complementary features comprehensively characterize the structure and complexity of lung sounds. By fusing the time-domain and frequency-domain features to construct a high-dimensional feature vector, the representation of lung sound characteristics is further enhanced. Dimensionality reduction of the fused vector using principal component analysis (PCA) not only reduces computational complexity and redundant information, but also effectively preserves the principal feature dimensions most valuable for classification. Finally, the signal reconstructed based on the principal components serves as the "second signal." This preserves the key characteristics of lung sounds while removing invalid or noisy interference, thereby improving the accuracy and stability of subsequent signal analysis, recognition, and evaluation. This method exhibits strong robustness and generalization capabilities, laying a solid foundation for high-quality screening and intelligent processing of lung audio data.

[0028] In this example, the power spectrum density curve is extracted to obtain several residual interference energy sequences, specifically: performing wavelet decomposition on the second signal to separate sub-signals of different frequency bands; Calculate the power spectral density curve of each sub-signal and detect abnormal peaks in the curve; The interference signal corresponding to the abnormal peak is removed by the matching pursuit algorithm to obtain the residual signal; The residual signal is segmented according to the heartbeat cycle, the energy integral of each segment is calculated and the residual interference energy sequence is generated.

[0029] Exemplarily, wavelet decomposition is performed on the patient's second signal. Wavelet decomposition decomposes the signal into multiple frequency band sub-signals. For example, the signal is decomposed into low-frequency (e.g., 0-10 Hz), mid-frequency (e.g., 10-30 Hz), and high-frequency (e.g., 30-60 Hz) sub-signals.

[0030] Furthermore, for each sub-signal, we calculate its power spectral density (PSD). For example, the power spectral density of sub-signal 1 (low frequency) is 5dB, the power spectral density of sub-signal 2 (mid-frequency) is 8dB, and the power spectral density of sub-signal 3 (high frequency) is 10dB.

[0031] It should be noted that the sliding peak-valley detection algorithm is used to locate abnormal peaks in the curve whose amplitude exceeds a dynamic threshold. The dynamic threshold is adaptively adjusted based on the energy distribution of the current sub-band and the historical noise statistical model.

[0032] In this example, the matching pursuit algorithm is used to remove the interference signal corresponding to the abnormal peak and obtain the residual signal, which is: A complete atomic library is constructed, wherein the atomic library includes Gaussian frequency modulated pulse, Morlet wavelet and respiratory rhythm matching basis functions; Screening the local optimal atomic set based on the time-frequency characteristics of the sub-signals; A sparse representation iterative approximation strategy is used to remove the interference components corresponding to abnormal peaks layer by layer: In each iteration step, the inner product of the residual signal and the basis function in the atom library is calculated, and the atom with the maximum correlation is selected as the matching atom; Subtract the contribution of the matching atom from the residual signal through a projection operation and update the residual signal; The iteration is repeated until a termination condition is met, wherein the termination condition includes that the residual energy is lower than a preset error limit or the number of matching atoms reaches an upper limit of a sparsity constraint.

[0033] Furthermore, the present invention further enhances the purity and analytical value of lung sound signals by performing multi-level spectral analysis and interference removal on the second signal. The specific method includes wavelet decomposition, power spectral density (PSD) calculation, outlier peak detection, and matching pursuit processing, achieving high-precision extraction of residual interference energy. First, wavelet decomposition is used to decompose the second signal into multiple sub-signals with different frequency bands. This enables the system to capture the frequency components and local variations hidden in lung sounds at multiple scales, enhancing its ability to handle non-stationary signals. Subsequently, power spectral density curves are calculated for each sub-signal to identify outlier frequency peaks that are highly correlated with environmental noise or equipment interference, providing a clear basis for interference identification. The introduction of a matching pursuit algorithm further improves the accuracy of interference removal. This algorithm gradually removes outlier components based on signal structural characteristics, retaining only core components closely related to lung physiological activity, effectively preventing the loss of useful lung sound information. Furthermore, the residual signal after interference removal is segmented according to the cardiac cycle, and the energy integral of each segment is calculated. Ultimately, a residual interference energy sequence is generated, providing a quantitative basis for subsequent variability analysis. This method not only improves the signal-to-noise ratio but also enhances the ability to perceive subtle changes in lung sound signals. It is particularly suitable for monitoring subtle symptoms and identifying abnormalities in patients with chronic lung diseases in the early stages. Compared with traditional filtering methods, this solution has higher adaptability, selectivity, and fidelity. While maintaining signal integrity, it significantly reduces the impact of external interference on intelligent analysis results. This provides high-quality input data for subsequent lung sound optimization and vital sign management, improving the accuracy and robustness of the entire system.

[0034] a signal evaluation module, configured to calculate a coefficient of variation of the residual interference energy sequence within a number of adjacent heartbeat cycles, and select the second signal with the smallest coefficient of variation as the optimal lung audio signal for the current heartbeat cycle; In this example, the coefficient of variation of the residual interference energy sequence within adjacent heartbeat cycles is calculated as follows: Select the residual interference energy sequence of N consecutive heartbeat cycles, calculate the mean and standard deviation of each sequence, and use the ratio of the standard deviation to the mean as the coefficient of variation; The coefficient of variation is dynamically updated through the sliding window method, and the heartbeat cycle index corresponding to the minimum coefficient of variation is recorded.

[0035] It is important to note that, over N consecutive cardiac cycles, the mean and standard deviation of each residual interference energy sequence are calculated, and the ratio of the standard deviation to the mean is used as the coefficient of variation (CV) of the sequence, thereby quantifying signal stability and fluctuation. As a dimensionless metric, the CV effectively eliminates the influence of absolute energy differences on the evaluation results, making the system more versatile and adaptable to individual needs. Furthermore, a sliding window method is used to dynamically update the CV. Specifically, as data is continuously input, the system calculates multiple overlapping cardiac cycle groups using a sliding time window of a certain length, continuously evaluating the energy stability of each signal segment. This allows for more sensitive response to subtle changes in signal quality. This strategy not only improves the real-time performance of signal quality assessment but also avoids misjudgments caused by single cycle anomalies, enhancing the robustness of the overall algorithm. By recording the cardiac cycle index with the smallest CV, the system can automatically identify the most stable and least disturbed lung sound signal at the time, providing optimal data input for the subsequent lung sound recognition model. Compared to traditional signal selection methods that rely on fixed thresholds or static strategies, this method offers greater dynamic adaptability, adjusting the evaluation window and selection criteria in real time based on actual data changes, while maintaining high accuracy in noisy and volatile scenarios. This signal selection mechanism, based on statistical stability, provides strong support for highly reliable and physiologically relevant lung sound monitoring in complex environments for patients with chronic diseases like lung cancer, and is a key component in improving the intelligent diagnostic performance of wearable devices.

[0036] In this example, the second signal with the smallest coefficient of variation is selected as the optimal lung audio signal, specifically: Locating the corresponding second signal according to the minimum coefficient of variation index; The localized second signal is time-domain aligned and phase-corrected to generate an optimal lung audio signal.

[0037] It should be noted that the present invention significantly improves the quality and accuracy of lung sound signals by selecting the second signal with the smallest coefficient of variation as the optimal lung sound signal. The coefficient of variation, a key indicator of signal stability, is dynamically calculated and uses sliding window technology to effectively identify the most stable and least interfering signal segments during multi-cycle acquisition. After determining the heartbeat cycle corresponding to the smallest coefficient of variation, the system automatically locates the associated second signal. This signal represents the purest lung sound data, providing high-quality input for subsequent lung sound analysis. To further optimize signal quality, the system employs time domain alignment and phase correction techniques to meticulously process the selected signal. This process eliminates the effects of time offset or phase differences during signal acquisition, ensuring signal consistency and synchronization in the time domain. Time domain alignment aligns signals from different heartbeat cycles onto a unified time scale, while phase correction compensates for phase deviations in the heart sound signal, making the lung sound signal more stable and coherent across multiple cycles. Ultimately, the signal after time domain alignment and phase correction is determined as the optimal lung sound signal, serving as the foundation for subsequent lung sound type identification and vital sign management. Compared with traditional signal selection and optimization technologies, this method can more accurately extract core information representing lung health status, reduce environmental interference and unnecessary noise, and thus provide more reliable data support for intelligent health management systems.

[0038] The vital signs management module is used to input the optimal lung audio signal into a preset neural network model to identify the lung sound type, and to perform intelligent vital signs management based on the vital sign parameters of lung cancer patients.

[0039] In this example, the preset neural network model is: The preset neural network model includes input layer, feature extraction layer, fusion layer and classification layer, as follows: Input layer: Receives the time-frequency spectrum of the optimal lung audio signal and the vital signs of lung cancer patients, including blood oxygen saturation, respiratory rate, and cough frequency; Feature extraction layer: A parallel convolutional neural network (CNN) branch and a long short-term memory network (LSTM) branch are used to extract spatial features and temporal features respectively; Fusion layer: concatenates the spatial features output by CNN and the temporal features output by LSTM into tensors, and dynamically assigns weights through the attention mechanism; Classification layer: Outputs the probability of lung sound type based on fusion features, including normal breath sounds, moist rales, dry rales, and pleural friction sounds.

[0040] In this example, intelligent management of vital signs is performed in combination with vital sign parameters, specifically: Generate a risk level score based on the probability distribution of lung sound types; If crackles or pleural friction rubs are detected and the respiratory rate exceeds a preset threshold, an early warning signal is triggered and a bronchodilator regimen is recommended; If the blood oxygen saturation remains below the critical value and the cough frequency is abnormal, the telemedicine platform will be activated to transmit data to the doctor's terminal in real time; Dynamically adjust management strategies, including medication reminders, respiratory training guidance, and emergency referral recommendations, to form a closed-loop management process.

[0041] It should be noted that a risk level score is generated based on the probability distribution of lung sound types, thereby effectively assessing the patient's lung health status and identifying potential disease risks in real time. This risk scoring system based on lung sound characteristics and physical sign data can not only objectively reflect the patient's health status, but also provide doctors with a more accurate diagnosis basis. In particular, when moist rales or pleural friction sounds are detected and the respiratory rate exceeds the preset threshold, the system can automatically trigger an early warning signal and recommend the use of bronchodilators, promptly reminding patients to take necessary drug interventions to effectively prevent further deterioration of the disease.

[0042] Furthermore, when the system detects that blood oxygen saturation remains below a critical value and coughing frequency is abnormal, it can activate a telemedicine platform and transmit the patient's health data to the doctor's terminal in real time, allowing the doctor to keep abreast of changes in the patient's condition and quickly take countermeasures. This real-time data transmission and remote intervention mechanism not only enhances interaction between patients and doctors, but also improves the timeliness and effectiveness of medical services. The system also has the ability to dynamically adjust management strategies, automatically adjusting management plans including medication reminders, breathing training guidance, and emergency referral recommendations based on different risk levels and patient needs, ensuring that patients receive personalized and continuously optimized health management services, ultimately forming a closed-loop management process. This intelligent management approach is of great significance in disease prevention, early intervention, and long-term health monitoring. It not only improves patients' quality of life but also provides innovative solutions for telemedicine and intelligent health management.

[0043] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0044] The above embodiments may be implemented in whole or in part through software, hardware, firmware or any other combination. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0045] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0046] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0047] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0048] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An intelligent management system for patient vital signs based on wearable devices, characterized in that: It includes data acquisition module, signal extraction module, signal screening module, signal evaluation module and vital sign management module. There are connections between modules: A data acquisition module is used to acquire lung audio signals and electrocardiogram signals based on a wearable device worn on the chest of a lung cancer patient; A signal extraction module is used to construct a heart sound trigger point sequence based on the electrocardiogram signal and extract lung sound signal segments within a preset heartbeat cycle to obtain a plurality of first signals; A signal screening module is used to perform feature screening on the first signal of the same heartbeat cycle according to a plurality of preset feature dimensions to obtain a plurality of second signals, and extract power spectrum density curves to obtain a plurality of residual interference energy sequences; a signal evaluation module, configured to calculate a coefficient of variation of the residual interference energy sequence within a number of adjacent heartbeat cycles, and select the second signal with the smallest coefficient of variation as the optimal lung audio signal for the current heartbeat cycle; The vital signs management module is used to input the optimal lung audio signal into a preset neural network model to identify the lung sound type, and to perform intelligent vital signs management based on the vital sign parameters of lung cancer patients.

2. The patient vital signs intelligent management system based on wearable devices according to claim 1 is characterized in that: The heart sound trigger point sequence is constructed based on the electrocardiogram signal, specifically: Preprocess the ECG signal to obtain a denoised ECG signal; Detect the R-wave peak point of the denoised ECG signal and divide the heartbeat cycle based on the R-wave peak point; In each heartbeat cycle, the starting points of the first and second heart sounds are extracted as trigger points based on the time-frequency characteristics of the heart sound signal, and timestamps are marked; The trigger points of several consecutive heartbeat cycles are arranged in chronological order to obtain a heart sound trigger point sequence.

3. The patient vital signs intelligent management system based on wearable devices according to claim 2 is characterized in that: The lung sound signal segments within the preset heartbeat cycle are extracted to obtain a plurality of first signals, specifically: Based on the starting point of the heart sound trigger point sequence, the lung sound signal is intercepted within a preset time window, where the time window covers the systolic period between the first heart sound and the second heart sound, and the diastolic period between the second heart sound and the first heart sound of the next cycle; An adaptive threshold method is used to separate the intercepted lung sound signal from the environmental noise to obtain a first signal.

4. The patient vital signs intelligent management system based on wearable devices according to claim 3 is characterized in that: The feature screening is performed according to a plurality of preset feature dimensions to obtain the second signal, specifically: Extracting time domain features from the first signal, wherein the time domain features include signal amplitude mean, variance, and zero-crossing rate; Extracting frequency domain features from the first signal, wherein the frequency domain features include Mel-frequency cepstral coefficient components and energy entropy; Fuse the frequency domain features and the frequency domain features to obtain a fused feature vector; Principal component analysis is used to reduce the dimension of the fused feature vector and obtain the principal component reconstruction signal; The signal is reconstructed based on the principal components to generate a second signal.

5. The patient vital signs intelligent management system based on wearable devices according to claim 4 is characterized in that: The power spectrum density curve is extracted to obtain several residual interference energy sequences, specifically: performing wavelet decomposition on the second signal to separate sub-signals of different frequency bands; Calculate the power spectral density curve of each sub-signal and detect abnormal peaks in the curve; The interference signal corresponding to the abnormal peak is removed by the matching pursuit algorithm to obtain the residual signal; The residual signal is segmented according to the heartbeat cycle, the energy integral of each segment is calculated and the residual interference energy sequence is generated.

6. The patient vital signs intelligent management system based on wearable devices according to claim 5 is characterized in that: The calculation of the coefficient of variation of the residual interference energy sequence within adjacent heartbeat cycles is specifically as follows: Select the residual interference energy sequence of N consecutive heartbeat cycles, calculate the mean and standard deviation of each sequence, and use the ratio of the standard deviation to the mean as the coefficient of variation; The coefficient of variation is dynamically updated through the sliding window method, and the heartbeat cycle index corresponding to the minimum coefficient of variation is recorded.

7. The patient vital signs intelligent management system based on wearable devices according to claim 6 is characterized in that: The second signal with the smallest coefficient of variation is selected as the optimal lung audio signal, specifically: Locating the corresponding second signal according to the minimum coefficient of variation index; The localized second signal is time-domain aligned and phase-corrected to generate an optimal lung audio signal.

8. The patient vital signs intelligent management system based on wearable devices according to claim 7 is characterized in that: The intelligent management of vital signs by combining vital sign parameters is specifically as follows: Generate a risk level score based on the probability distribution of lung sound types; If crackles or pleural friction rubs are detected and the respiratory rate exceeds a preset threshold, an early warning signal is triggered and a bronchodilator regimen is recommended; If the blood oxygen saturation remains below the critical value and the cough frequency is abnormal, the telemedicine platform will be activated to transmit data to the doctor's terminal in real time; Dynamically adjust management strategies, including medication reminders, respiratory training guidance, and emergency referral recommendations, to form a closed-loop management process.

Images