Intelligent night respiration monitoring system and method for critical patient

Abstract

The invention discloses a critical patient night respiration intelligent monitoring system and method, and the method comprises the steps: firstly carrying out the independent signal internal quality evaluation of each path of infrared, radar and acoustic signals, and innovatively introducing the cross-modal signal consistency mutual verification, thereby generating a more robust quality index to achieve the dynamic weight fusion. Thus, interfered low-quality signals can be intelligently discriminated and restrained in real time, meanwhile, high-quality signals are preferentially collected, it is ensured that stable and reliable fused breathing parameters are output in a complex monitoring environment, finally, high-precision and high-reliability continuous monitoring of the breathing state of a critical patient at night is achieved, the false alarm rate and the missing report rate are effectively reduced, and the working efficiency is improved. The monitoring efficiency and the patient safety are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent monitoring, and more particularly, to a night respiratory intelligent monitoring system and method for critical patients. BACKGROUND

[0002] For patients in the intensive care field, as a key vital sign, the continuous and accurate monitoring of the state of respiration is crucial for early warning of disease deterioration and evaluation of treatment effect. Especially at night, achieving uninterrupted automatic monitoring of patients without disturbance can effectively make up for the gaps in human patrol and timely detect respiratory abnormal events. Therefore, building an intelligent night respiratory monitoring scheme for critical patients aims to obtain reliable respiratory parameters through non-contact means and conduct automatic analysis and alarm, which has become an important development direction to improve the quality and efficiency of monitoring.

[0003] To cope with complex monitoring environments, existing technologies often use infrared, radar or acoustic sensors and other non-contact sensors to achieve redundancy and complementarity. However, these technical solutions face significant challenges in practical application. On the one hand, the monitoring method of a single modality is easily disturbed by specific environmental factors or changes in patient status, resulting in a decline in signal quality or even failure, and insufficient monitoring stability. On the other hand, although some solutions attempt to fuse multi-modal data, they usually use fixed weighting strategies or simple signal averaging methods. This static fusion mechanism cannot adapt to the dynamic changes in the quality of signals from various sources, and when the signal of a sensor is distorted due to interference, the mechanism cannot effectively identify and reduce its weight, but instead pollutes the final fusion result. Therefore, the existing technologies generally fail to perform real-time quality assessment and adaptive dynamic fusion of multi-source signals, resulting in low accuracy and robustness of the monitoring results, which cannot meet the high reliability requirements of intensive care.

[0004] Therefore, an optimized night respiratory intelligent monitoring scheme for critical patients is expected. SUMMARY

[0005] To solve the above technical problems, the present application is proposed. The embodiments of the present application provide a night respiratory intelligent monitoring system and method for critical patients, which first performs independent signal-in quality assessment on infrared, radar and acoustic signals and other signals, and innovatively introduces cross-modal signal inter-consistency mutual verification, thereby generating a more robust quality index to realize dynamic weighted fusion. In this way, the low-quality signals disturbed can be identified and suppressed in real time and intelligently, while high-quality signals are preferentially acquired, ensuring the output of stable and reliable fusion respiratory parameters in complex monitoring environments, and ultimately achieving high-precision and high-reliability continuous monitoring of the night respiratory state of critical patients, effectively reducing the false alarm and missed alarm rates, and improving the monitoring efficiency and patient safety.

[0006] According to an aspect of the present application, there is provided a method for intelligent monitoring of night-time respiration of a critical patient, comprising: preprocessing the acquired raw infrared video stream, raw radar data stream and raw acoustic audio stream to obtain infrared respiration waveform, radar respiration waveform and acoustic respiration waveform; performing signal quality assessment on the infrared respiration waveform, radar respiration waveform and acoustic respiration waveform to obtain double-assessed infrared signal quality index, double-assessed radar signal quality index and double-assessed acoustic signal quality index; based on the double-assessed infrared signal quality index, double-assessed radar signal quality index and double-assessed acoustic signal quality index, performing dynamic weighted fusion and core respiration parameter calculation on the infrared respiration waveform, radar respiration waveform and acoustic respiration waveform to obtain final respiration rate and final respiration amplitude variability; performing abnormal pattern recognition and alarm decision based on the final respiration rate and final respiration amplitude variability to generate an alarm signal.

[0007] According to another aspect of the present application, there is provided a system for intelligent monitoring of night-time respiration of a critical patient, comprising: a data preprocessing module for preprocessing the acquired raw infrared video stream, raw radar data stream and raw acoustic audio stream to obtain infrared respiration waveform, radar respiration waveform and acoustic respiration waveform; a signal quality assessment module for performing signal quality assessment on the infrared respiration waveform, radar respiration waveform and acoustic respiration waveform to obtain double-assessed infrared signal quality index, double-assessed radar signal quality index and double-assessed acoustic signal quality index; a dynamic weighted fusion and respiration parameter calculation module for performing dynamic weighted fusion and core respiration parameter calculation on the infrared respiration waveform, radar respiration waveform and acoustic respiration waveform based on the double-assessed infrared signal quality index, double-assessed radar signal quality index and double-assessed acoustic signal quality index to obtain final respiration rate and final respiration amplitude variability; an abnormal pattern recognition and alarm decision module for performing abnormal pattern recognition and alarm decision based on the final respiration rate and final respiration amplitude variability to generate an alarm signal.

[0008] Compared with the prior art, the application provides a severe patient night respiratory intelligent monitoring system and method, which firstly performs independent signal-in quality evaluation on infrared, radar and acoustic signals and other signals, and innovatively introduces cross-modal signal inter-consistency mutual verification, so as to generate a more robust quality index to realize dynamic weighted fusion. In this way, the low-quality signal disturbed can be intelligently discriminated and inhibited in real time, and the high-quality signal is preferentially acquired, so as to ensure that stable and reliable fusion respiratory parameters are output in a complex monitoring environment, and finally high-precision and high-reliability continuous monitoring of the night respiratory state of a severe patient is realized, the false alarm and missed alarm rates are effectively reduced, and the monitoring efficiency and patient safety are improved. BRIEF DESCRIPTION OF DRAWINGS

[0009] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description of embodiments of the present application taken in conjunction with the accompanying drawings. The drawings provided in the appended hereto are for purposes of illustration and description only and in no way limit the application, which is set forth in the appended claims. In the drawings, like reference numerals indicate like elements or steps among different drawings.

[0010] Figure 1 A flow chart of a severe patient night respiratory intelligent monitoring method according to an embodiment of the present application; Figure 2 A data flow schematic diagram of a severe patient night respiratory intelligent monitoring method according to an embodiment of the present application; Figure 3 A block diagram of a severe patient night respiratory intelligent monitoring system according to an embodiment of the present application. DETAILED DESCRIPTION

[0011] Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, and are not all the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein.

[0012] As shown in the present application and claims, unless the context clearly indicates otherwise, the words "one", "an", "a", and / or "the" do not mean to specify a single number, but can include a plurality or singularity. Generally, the terms "comprise" and "include" only indicate the inclusion of the steps and elements explicitly identified, and these steps and elements do not constitute an exclusive list, and the method or device can also include other steps or elements.

[0013] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0014] Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in sequence. Instead, various steps can be processed in reverse order or simultaneously, as needed. Meanwhile, other operations can also be added to these processes, or one or more steps of operation can be removed from these processes.

[0015] In the following, the example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, and are not all the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein.

[0016] In the technical solutions of the present application, an intelligent monitoring method for night breathing of a critical patient is proposed. Figure 1 A flowchart of the intelligent monitoring method for night breathing of a critical patient according to the embodiments of the present application is shown. Figure 2 A system architecture diagram of the intelligent monitoring method for night breathing of a critical patient according to the embodiments of the present application is shown. Figure 1 and Figure 2 As shown in the intelligent monitoring method for night breathing of a critical patient according to the embodiments of the present application, the method comprises the following steps: S1, preprocessing the obtained original infrared video stream, original radar data stream and original acoustic audio stream to obtain infrared breathing waveform, radar breathing waveform and acoustic breathing waveform; S2, performing signal quality evaluation on the infrared breathing waveform, radar breathing waveform and acoustic breathing waveform to obtain double-evaluation infrared signal quality index, double-evaluation radar signal quality index and double-evaluation acoustic signal quality index; S3, based on the double-evaluation infrared signal quality index, double-evaluation radar signal quality index and double-evaluation acoustic signal quality index, performing dynamic weighted fusion and core breathing parameter calculation on the infrared breathing waveform, radar breathing waveform and acoustic breathing waveform to obtain final respiratory rate and final respiratory amplitude variability; S4, performing abnormal pattern recognition and alarm decision based on the final respiratory rate and final respiratory amplitude variability to generate an alarm signal.

[0017] In particular, the S1 preprocesses the acquired raw infrared video stream, raw radar data stream and raw acoustic audio stream to obtain infrared respiratory waveform, radar respiratory waveform and acoustic respiratory waveform. It should be understood that the raw sensor data itself contains a large amount of noise and redundant information irrelevant to respiration, and the physical meaning, data format and numerical range of each data stream are different, and cannot be directly compared or fused. For example, the infrared video is a sequence of images, the radar data is the phase information of electromagnetic waves, and the acoustic data is the sound pressure signal. Therefore, in the technical solution of the present application, the acquired raw infrared video stream, raw radar data stream and raw acoustic audio stream are preprocessed to extract time series signals that can represent the core features of respiratory motion, and place them in a unified framework. In the intensive care scene, the introduction of multi-modal data is the key to ensuring the robustness of monitoring. Specifically, the raw infrared video stream can capture the temperature changes of the patient's chest and abdomen or the oral and nasal regions, and can non-contact detect the subtle heat exchange caused by respiration, which is particularly effective in the patient's static state; the raw radar data stream uses the Doppler effect and can penetrate thin coverings such as bedclothes to accurately detect the chest displacement caused by respiration, thereby enhancing the adaptability to the shielding condition; and the raw acoustic audio stream can collect respiratory airflow sound, snoring sound, etc., thereby providing direct evidence for whether the respiratory tract is unobstructed and whether there is abnormal respiratory sound. It is worth mentioning that the raw infrared video stream can be continuously acquired by the infrared thermal imager deployed above the patient's bed at a preset frame rate (such as 10 frames / second); the raw radar data stream can be obtained by the millimeter wave radar module placed near the patient (such as the bedside cabinet) to continuously emit and receive electromagnetic waves; and the raw acoustic audio stream can be recorded in real time by the directional high-fidelity microphone installed near the patient's head.

[0018] In implementation, first, the original infrared video stream, the original radar data stream and the original acoustic audio stream are subjected to independent modal extraction of respiratory feature signals to obtain infrared respiratory feature time series signals, radar respiratory feature time series signals and acoustic respiratory feature time series signals. That is, for the three kinds of original heterogeneous data streams, the respective algorithms are used to separate and extract the feature time series signals capable of reflecting the respiratory rhythm and depth from the respective modal. Specifically, for the original infrared video stream, the system will first define one or more regions of interest (ROI) in the picture thereof, for example, the regions covering the chest, abdomen or mouth and nose of the patient, and then calculate and track the changes of the average brightness value or temperature value of the pixels in the ROI in the time dimension, thereby forming an infrared respiratory feature time series signal reflecting the respiratory movement; for the original radar data stream, the I / Q (in-phase quadrature) baseband signal thereof is usually processed, and the small distance changes between the radar and the patient caused by the displacement of the chest are calculated by the phase demodulation technology (such as the arctangent demodulation method), and the change sequence constitutes the radar respiratory feature time series signal; for the original acoustic audio stream, the digital signal processing technology is applied, for example, the environmental noise is filtered out by a band-pass filter, and the specific frequency band related to the respiratory airflow sound is retained, and then the energy envelope of the filtered audio signal is calculated, and the locus of the envelope changing over time forms the acoustic respiratory feature time series signal. Further, the infrared respiratory feature time series signals, the radar respiratory feature time series signals and the acoustic respiratory feature time series signals are subjected to signal filtering and normalization to obtain infrared respiratory waveforms, radar respiratory waveforms and acoustic respiratory waveforms. Among them, the signal filtering is mainly to eliminate the noise components not conforming to the physiological rules, and a band-pass filter is usually used, the passband range of which is set according to the respiratory frequency of normal adults, thereby filtering out the high-frequency noise caused by the involuntary shaking of the body, the electromagnetic interference of the equipment and the like, and the low-frequency interference such as baseline drift caused by the slow posture change of the body; the signal normalization is to eliminate the dimensional differences of the amplitudes of different sensor signals, so that the processed waveforms can fluctuate within a unified numerical interval. A commonly used method is the minimum-maximum normalization, which linearly scales the amplitude of each signal to a preset range, for example, [-1, 1] or. Through the filtering and normalization processing, the final output is the infrared respiratory waveform, the radar respiratory waveform and the acoustic respiratory waveform which are morphologically regular, amplitude-unified and capable of clearly reflecting the respiratory process.

[0019] In particular, the S2 performs signal quality evaluation on the infrared respiratory waveform, the radar respiratory waveform and the acoustic respiratory waveform to obtain a double-evaluated infrared signal quality index, a double-evaluated radar signal quality index and a double-evaluated acoustic signal quality index. It should be understood that in a real intensive care clinical environment, signal quality is dynamic and unpredictable. Any single type of non-contact sensor has its inherent physical limitations, and its signal quality is easily disturbed by various internal and external factors. For example, the patient turns over or pulls up the quilt in sleep, which directly blocks the field of view of the infrared thermal imager, causing the infrared signal quality to drop sharply or even fail completely; medical staff at the bedside for treatment, nursing or talking with the patient, their activities will introduce a lot of body motion artifacts and voice noise, thereby seriously interfering with the accuracy of radar signals and acoustic signals. Without an effective real-time quality evaluation mechanism, the system will not be able to distinguish the accuracy and reliability of the signal. The static weighting mechanism presets the fixed importance weight of each sensor in any scene, which is seriously inconsistent with the clinical reality. When a high-quality signal is polluted, the static fusion mechanism not only cannot identify and reduce its weight, but also allows the distorted data to pollute the final fusion result, which directly leads to serious deviation of the output respiratory parameters, and even may trigger false clinical alarms or miss real respiratory abnormal events. Therefore, in the technical solution of the present application, the infrared respiratory waveform, the radar respiratory waveform and the acoustic respiratory waveform are evaluated for signal quality, a more intelligent and robust monitoring system that can dynamically adapt to the dramatic changes in signal quality in the clinical monitoring environment is constructed, and only the most reliable information can enter the decision-making loop, thereby fundamentally improving the accuracy, stability and anti-interference ability of the entire monitoring system.

[0020] In specific implementation, first, the infrared respiratory waveform, the radar respiratory waveform and the acoustic respiratory waveform are evaluated for signal internal quality index to obtain an infrared signal internal quality index, a radar signal internal quality index and an acoustic signal internal quality index. In this process, first, the periodicity score, the spectral purity score and the waveform stability score of the infrared respiratory waveform are calculated. It should be understood that a single indicator is insufficient to comprehensively describe the quality of a respiratory waveform. As a physiological activity, the signal of respiration simultaneously has three characteristics in an ideal state: first, periodicity; second, the signal energy should be concentrated on the respiratory main frequency, i.e. spectral purity; and finally, the respiratory rhythm and depth should be relatively stable in a short time, i.e. waveform stability. Therefore, by calculating the periodicity score, the spectral purity score and the waveform stability score respectively, the system can finely describe the signal quality from three orthogonal dimensions of time regularity, frequency purity and shape consistency, thereby constructing a more robust and comprehensive quality evaluation model than a single indicator. Specifically, the periodicity score, the spectral purity score and the waveform stability score of the infrared respiratory waveform are calculated as follows: First, the periodicity score of the infrared respiration waveform is calculated. Specifically, first, time-domain feature extraction is performed on the infrared respiration waveform to obtain an infrared waveform autocorrelation function sequence, an infrared respiration waveform peak set, and an infrared respiration waveform trough set; then, frequency-domain feature extraction is performed on the infrared respiration waveform to obtain an infrared respiration waveform power spectral density function; further, based on the infrared waveform autocorrelation function sequence, the periodicity score of the infrared respiration waveform is calculated. The autocorrelation function is a mathematical tool for measuring the similarity of a signal to itself at different time delays. For a periodic signal, its autocorrelation function will present a peak at the delay point equal to an integer multiple of its period. The specific method of calculating the periodicity score is that the system finds the maximum peak in the autocorrelation function sequence within the physiologically reasonable respiratory period range. The normalized amplitude of this peak (usually the autocorrelation value at a delay of 0 is normalized to 1) is defined as the periodicity score. A score close to 1 indicates that the respiration waveform has very strong and clear repetition rules, which is an important indicator of high-quality signals; this process is represented by the formula: wherein, is the periodicity score of the infrared respiration waveform, is the normalized autocorrelation function, is a preset effective respiratory period time range; Then, based on the infrared respiration waveform power spectral density function, the spectral purity score of the infrared respiration waveform is calculated. The infrared respiration waveform power spectral density function (PSD) describes the distribution of signal power in the frequency domain. A pure respiration signal should have most of its energy concentrated around a single respiration main frequency. The specific method of calculating the spectral purity score is that the system first finds the frequency point with the highest energy within the effective respiration frequency range of the PSD graph, i.e., the main respiration frequency; then, the total energy within an extremely narrow frequency band centered on the main respiration frequency is calculated and compared with the total energy within the entire effective respiration frequency range. The ratio of these two energies is the spectral purity score. The higher the score, the more concentrated the signal energy, and the less likely it is to be affected by other frequency noise or interference; this process is represented by the formula: wherein, is the spectral purity score of the infrared respiration waveform, is the infrared respiration waveform power spectral density function, is the main respiration frequency, is a defined narrow bandwidth, is an effective respiration frequency range; Further, based on the set of infrared respiratory waveform peaks and the set of infrared respiratory waveform troughs, a waveform stability score of the infrared respiratory waveform is calculated. The waveform stability score aims to quantify the consistency of consecutive respiratory cycles in morphology, mainly including the stability of cycle length and the stability of respiratory amplitude. The specific calculation process includes: first, using the set of peaks, the instantaneous cycle length of each respiratory cycle (i.e. the time difference between adjacent peaks) is calculated to form a cycle length sequence; then, using the set of peaks and the set of troughs, the instantaneous respiratory amplitude of each respiratory cycle (i.e. the amplitude difference between a peak and the trough immediately following it) is calculated to form a respiratory amplitude sequence; then, the coefficient of variation (CV) of the two sequences is calculated respectively, i.e. the standard deviation divided by the mean; the smaller the coefficient of variation, the higher the stability of the sequence. The waveform stability score is finally defined as a function negatively related to the two coefficients of variation, for example, by weighted combination of their corresponding stability measures (1-CV); the process is represented by the formula: wherein, is the waveform stability score of the infrared respiratory waveform, and are the coefficients of variation of the cycle length sequence and the respiratory amplitude sequence, respectively, and are weight terms; Then, the weighted sum of the periodicity score, the spectral purity score and the waveform stability score of the infrared respiratory waveform is calculated to obtain the infrared signal internal quality index. It should be understood that the three scores of periodicity, spectral purity and waveform stability produced in the previous steps, although they finely depict specific aspects of signal quality, directly using multiple indicators in subsequent dynamic fusion decision-making will make the algorithm model abnormally complex and difficult to optimize. Therefore, in the technical solution of the present application, by fusing the three scores into a unified signal internal quality index system, an intuitive and quantitative evaluation of the overall credibility of the current signal can be obtained, and this single numerical value can be directly used as the reference input of the subsequent verification link and provide a clear and unambiguous weight basis for the final dynamic weighted fusion. Specifically, the periodicity score, the spectral purity score and the waveform stability score can be multiplied by the respective preset weight coefficients and the respective products can be added to obtain the infrared signal internal quality index; the process is represented by the formula: wherein, is the infrared signal internal quality index, , and The weight coefficients of the periodicity score, the spectral purity score, and the waveform stability score, respectively, are pre-determined based on a large amount of clinical data analysis and expert knowledge base, and these weights reflect the relative importance of different quality dimensions in judging whether a respiratory signal is real and reliable in a specific application scenario (such as intensive care). For example, in an environment susceptible to periodic mechanical noise (such as a ventilator fan), the weight of the spectral purity score may be assigned a higher value. Generally, to ensure that the final signal-in quality index is normalized to the same numerical interval (such as 0 to 1) as each sub-score, the sum of these weight coefficients is set to 1. By adjusting these weights, the quality evaluation algorithm can be fine-tuned to adapt to different sensor characteristics and specific needs of clinical monitoring.

[0021] Further, based on the infrared respiratory waveform, the radar respiratory waveform, and the acoustic respiratory waveform, the infrared signal-in quality index, the radar signal-in quality index, and the acoustic signal-in quality index are verified for inter-signal consistency and modified to obtain a double-evaluation infrared signal quality index, a double-evaluation radar signal quality index, and a double-evaluation acoustic signal quality index. In this process, first, based on the infrared respiratory waveform, the radar respiratory waveform, and the acoustic respiratory waveform, a consistency matrix is calculated. It should be understood that in a multi-modal signal consistency evaluation mechanism, when frequency consistency and amplitude consistency are fused to form a comprehensive judgment, a static weighted fusion strategy is used, i.e., the frequency and amplitude dimensions are pre-assigned fixed importance weights. The inherent defect of this design is that it assumes that the reliability of frequency information and amplitude information is constant in any monitoring scenario. However, in a real intensive care clinical environment, signal quality is dynamic. For example, when a non-contact sensor is affected by a stable periodic interference source (such as a device fan), the extracted respiratory frequency information is stable, but it is completely wrong; on the contrary, when a patient has irregular shallow and rapid breathing, the respiratory amplitude sequence may lose structure due to too low signal-to-noise ratio, making amplitude correlation calculation meaningless. The static weighting mechanism cannot identify and adapt to the dynamic changes in the reliability of specific dimension information, and thus may assign too high a weight to the consistency score based on unreliable data, or too low a weight to the consistency score based on reliable data, ultimately polluting the accuracy of the consistency evaluation and leading to a misjudgment of signal quality.

[0022] To overcome the above technical defects, an adaptive weighted fusion technology based on information theory is proposed, the core of which is not to rely on fixed prior weights, but to dynamically generate fusion weights by real-time quantifying the reliability of each consistency measure. Specifically, based on the infrared respiratory waveform, the radar respiratory waveform, and the acoustic respiratory waveform, the consistency matrix is calculated in the following steps: Firstly, cross-modal comparative parameter extraction is performed on the infrared respiratory waveform, radar respiratory waveform, and acoustic respiratory waveform to obtain a set of instantaneous respiratory rates and a set of per-cycle respiratory amplitude sequences. In this process, the system first needs to extract standardized physiological parameters that can be directly compared from the three respiratory waveforms. Specifically, the system will analyze each waveform to obtain two core time series: one is a set of instantaneous respiratory rates, which is a sequence of respiratory frequencies calculated for each respiratory cycle within the analysis window; the other is a set of per-cycle respiratory amplitude sequences, which is a sequence composed of the amplitudes of each respiratory cycle (such as the difference between the peak and the trough); Then, based on the set of instantaneous respiratory rates and the set of per-cycle respiratory amplitude sequences, the basic frequency consistency score matrix and the basic amplitude consistency score matrix are calculated. It should be understood that before performing a deeper confidence assessment, the basic measure of similarity between signals must first be obtained. Specifically, for any pair of signals, their frequency consistency score and amplitude consistency score are calculated respectively. Frequency consistency quantifies the difference in instantaneous respiratory rates between two signals through an exponential decay function. The smaller the difference, the higher the score value. Its calculation formula is: wherein, is the i-th instantaneous respiratory rate in the set of instantaneous respiratory rates, is a preset consistency calculation parameter.

[0023] Amplitude consistency measures the linear similarity in shape between the two respiratory amplitude sequences by calculating the Pearson correlation coefficient of the two respiratory amplitude sequences. Its calculation formula is: wherein, and are the i-th and j-th per-cycle respiratory amplitude sequences in the set of per-cycle respiratory amplitude sequences. In this way, two unweighted, independent, and original values reflecting the degree of similarity in frequency and amplitude are obtained respectively; ​​​Then, a frequency consistency confidence matrix is calculated based on the power spectral density functions of the infrared, radar and acoustic respiratory waveforms, and an amplitude consistency confidence matrix is calculated based on the respiratory amplitude sequences of the infrared, radar and acoustic respiratory waveforms. It can be understood that, in view of the technical defect, a mechanism is needed to evaluate the reliability of the basic scores calculated in the first step. Specifically, the concept of information theory is introduced to quantify the certainty of the frequency information and the amplitude information respectively. For the frequency dimension, spectral entropy is used to measure the randomness or uncertainty of the power spectrum of the signal. A power spectrum with a sharp unimodal peak (indicating a clear dominant frequency) has a low entropy value, while a noise spectrum with a flat or multi-peak shape has a high entropy value. The spectral entropy of the power spectral density function PSD(f) is The calculation formula is: In this technical scenario, the spectral entropy measures the uncertainty of the question of "where does the signal energy concentrate". Accordingly, a confidence score for frequency consistency can be constructed , which is inversely proportional to the average spectral entropy of the two signals. For the amplitude dimension, mutual information is used to measure the deeper statistical dependence between the two amplitude sequences, which can capture linear and nonlinear relationships. The mutual information between two amplitude sequences is calculated as follows: , where and are the entropies of the two periodic respiratory amplitude sequences; In the monitoring scenario, the mutual information quantifies the degree to which the uncertainty about the amplitude change of one signal can be reduced by observing the amplitude change of another signal, i.e. the amount of information they share. The amplitude consistency confidence score constructed accordingly is proportional to the normalized mutual information of the two signals. In this way, a dynamic and objective confidence score is generated for the frequency and amplitude consistency dimensions respectively. The execution effect is that two new metrics and are output, which no longer focus on whether the signals themselves are consistent, but evaluate the quality of the evidence itself for judging consistency; ​Subsequently, based on the base frequency consistency score matrix, the base amplitude consistency score matrix, the frequency consistency confidence matrix and the amplitude consistency confidence matrix, the consistency matrix is determined. That is, the confidence score obtained in the previous step is converted into an actual fusion weight, thereby completing the correction of the static weighted defects. The execution process is that two confidence scores and are mapped into a set of dynamic weights whose sum is 1 and . The Softmax function can amplify the difference between the confidences, so that the party with higher confidence obtains exponentially enhanced weight. The calculation formula of the dynamic weight is: Here, the parameters control the decisiveness of the decision, and the larger the value of β, the more the weight distribution tends to "winner takes all". Subsequently, the base consistency scores calculated in the first step are weighted and summed using the pair of dynamically generated weights to obtain the final comprehensive consistency score: In this way, a fundamental change from static weight to dynamic weight is completed, and intelligent fusion driven by the quality of the data itself is realized. The final consistency score has a high degree of scene adaptability. When the spectral morphology of the signal is clear and the main frequency is clear, the system will automatically increase the weight of the frequency consistency; and when the amplitude sequence of the signal exhibits a highly synchronized complex pattern, the system will instead trust the amplitude consistency evaluation more, thereby ensuring the robustness and accuracy of the comprehensive evaluation result.

[0024] In summary, this optimization mechanism constructs a more intelligent and robust multimodal signal consistency assessment system capable of dynamically adapting to changes in signal quality within the clinical monitoring environment. By abandoning static weights and adopting an information theory-based adaptive weighting mechanism, this technique achieves significant technical effects: it can determine the reliability of frequency and amplitude information in real time and dynamically adjust their weight in the final consistency assessment accordingly. This makes the entire assessment system more immune to interference. For example, it can automatically identify and reduce the frequency consistency weight of a signal contaminated by periodic artifacts, while increasing its importance in amplitude consistency with other signals, thus avoiding overall assessment failure due to single-dimensional information contamination. Ultimately, this intelligent assessment mechanism can more accurately identify high-quality signal sources, providing more reliable input for subsequent signal fusion and clinical decision-making. Its technical effects are directly reflected in improved accuracy and stability of the entire intensive care system, effectively reducing false alarms and missed alarms caused by sensor interference or changes in patient condition, ultimately serving the fundamental technical objectives of improving patient safety and reducing "alarm fatigue" among medical staff. Furthermore, based on the consistency matrix, the quality indices of the infrared signal, radar signal, and acoustic signal are corrected to obtain the dual-evaluation infrared signal quality index, dual-evaluation radar signal quality index, and dual-evaluation acoustic signal quality index. It should be understood that while the signal quality index objectively evaluates the morphological quality of the waveform itself, it cannot perceive whether the signal is synchronized with other signal sources at the macroscopic physiological information level; the consistency matrix precisely provides this cross-modal external reference frame. By performing the correction steps, this invention establishes a penalty and incentive mechanism: for a signal that is highly inconsistent with other signals, even if its waveform is very regular, its initial high quality score must be significantly penalized because it is very likely an artifact; conversely, for a signal that is highly consistent with other signals, its credibility has been externally verified, and its quality score should be consolidated or improved. In this way, it is possible to ensure that the entire monitoring system can effectively resist false signal interference.

[0025] Specifically, firstly, for any signal to be corrected (taking an infrared signal as an example, index i), a comprehensive consistency score needs to be calculated from the consistency matrix. This score represents the average degree of consistency between the signal and all other signals in the monitoring network. The calculation involves extracting all off-diagonal elements from the rows (or columns) related to the signal in the consistency matrix and calculating their mean; this process is expressed by the formula: in, It is the overall consistency score of the i-th signal. The total number of signal channels (N=3). Let be the element in the i-th row and j-th column of the consistency matrix; Secondly, after obtaining the overall consistency score, the system will employ a penalized correction function to combine this score with the signal's intra-signal quality index to calculate the final dual-evaluation signal quality index. A direct and efficient implementation is to multiply the two, as the consistency score itself lies between 0 and 1 and can serve as a correction factor; this process is expressed by the formula: in, To dual-evaluate the signal quality index, This is the signal quality index. According to this formula, if a signal has low average consistency with all other signals (…), then… If the initial signal quality index is close to 0, then no matter how high its initial signal quality index is, its final dual evaluation quality index will be significantly reduced, approaching 0; conversely, only when a signal has both excellent inherent waveform ( High), and can also be highly synchronized with other signals ( Only when the score is high can its final quality score remain high.

[0026] This process is applied sequentially to the infrared, radar, and acoustic signals, ultimately yielding their respective dual-evaluation infrared signal quality index, dual-evaluation radar signal quality index, and dual-evaluation acoustic signal quality index.

[0027] Specifically, in step S3, based on the dual-assessment infrared signal quality index, dual-assessment radar signal quality index, and dual-assessment acoustic signal quality index, dynamic weighted fusion and core respiratory parameter calculation are performed on the infrared respiratory waveform, radar respiratory waveform, and acoustic respiratory waveform to obtain the final respiratory rate and final respiratory amplitude variability. It should be understood that the dual-assessment infrared signal quality index, dual-assessment radar signal quality index, and dual-assessment acoustic signal quality index obtained in the preceding steps have already provided authoritative scores for the real-time quality of each signal, but these scores themselves are not the final clinical parameters. By directly using the dual-assessment signal quality index as a weighting coefficient, this application constructs an intelligent fusion system that can adapt in real time. At any given moment, the signal evaluated as having the highest quality will dominate the fusion result, while the influence of signals with poor quality will be correspondingly weakened. For example, when a patient covers themselves with a blanket, causing a sharp drop in infrared signal quality, its extremely low dual-assessment quality index will make its contribution to the weighted sum negligible, thereby ensuring that the fused signal mainly reflects information from other high-quality sources such as radar or acoustics. This mechanism ensures that regardless of changes in the external environment or the patient's condition, the system can intelligently select highly reliable signals to generate a fused respiratory waveform that closely approximates the actual physiological state at any given time. Finally, core respiratory parameters are calculated from the fused single optimal waveform. It should be understood that a waveform obtained through intelligent fusion has a signal-to-noise ratio and anti-interference capability far exceeding any single original waveform. Based on this, periodic feature extraction and parameter calculation are performed, greatly ensuring the accuracy and stability of the results. The final respiratory rate reflects the speed of the respiratory rhythm, and the final respiratory amplitude variability reflects the stability of respiratory depth. The combination of these two provides strong quantitative evidence for the identification of abnormal respiratory patterns.

[0028] In practice, firstly, based on the dual-evaluation infrared signal quality index, dual-evaluation radar signal quality index, and dual-evaluation acoustic signal quality index, the infrared breathing waveform, radar breathing waveform, and acoustic breathing waveform are dynamically weighted and fused to obtain the fused breathing waveform. During this process, at each time point, the system performs a weighted average calculation, summing the amplitudes of the three waveforms at that moment according to their corresponding quality indices, thereby generating the amplitude of the fused breathing waveform at that moment. This process is expressed by the formula: in, To fuse the amplitude of the breathing waveform at time point t, , and These represent the amplitudes of the three original waveforms at time point t. , and These are the corresponding dual-evaluation signal quality indices; by repeating this operation for all time points within the analysis window, a complete, smooth, and robust fused breathing waveform can be obtained. Next, periodic feature extraction is performed on the fused respiratory waveform to obtain the final respiratory rate and a sequence containing the amplitude of each respiratory cycle within the current window. During this process, the system applies signal processing algorithms such as peak-valley detection to the fused respiratory waveform to accurately identify the peaks and troughs of each respiratory cycle. Based on these feature points, the final respiratory rate can be calculated; for example, by calculating the time interval sequence between consecutive peaks and taking its average, the final respiratory rate can be expressed by the formula: At the same time, the system will also obtain a sequence containing the amplitude of each respiratory cycle within the current window, where each element is obtained by calculating the amplitude difference between the k-th peak and the immediately following trough; Then, based on the sequence containing the amplitude of each respiratory cycle within the current window, the final respiratory amplitude variability is calculated. That is, based on the respiratory cycle amplitude sequence obtained in the previous step, its coefficient of variation is calculated, and this value is defined as the final respiratory amplitude variability; specifically, the final respiratory amplitude variability is calculated using the following formula: in, The standard deviation of the sequence containing the amplitude of each respiratory cycle within the current window. It is the mean of the sequence containing the amplitude of each respiratory cycle within the current window.

[0029] Specifically, S4 involves abnormal pattern recognition and alarm decision-making based on the final respiratory rate and final respiratory amplitude variability to generate an alarm signal. By simultaneously analyzing both respiratory rate and respiratory amplitude variability, this invention can distinguish more diverse respiratory patterns with different clinical significance, thereby significantly improving the specificity of the alarm and effectively avoiding frequent false alarms caused by simple threshold settings.

[0030] In practice, the first step is to identify abnormal patterns based on the variability of final respiratory rate and final respiratory amplitude. During this process, the variability of final respiratory rate and final respiratory amplitude is matched in real time against a predefined rule base based on clinical guidelines and expert knowledge. This rule base maps different combinations of parameters to specific respiratory patterns. The rules can be defined as follows: Tachypnea pattern recognition: When the respiratory rate is consistently higher than the preset upper limit threshold and the respiratory amplitude is relatively stable, it is recognized as tachypnea. Bradyspnea pattern recognition: When the respiratory rate is consistently below a preset lower threshold, it is identified as bradyspnea; Irregular breathing pattern recognition: When the variability of breathing amplitude is consistently higher than the preset instability threshold, it can be identified as irregular breathing regardless of the breathing rate; Therefore, after identifying potential abnormal patterns, the system does not immediately generate an alarm, but instead enters a deliberate decision-making process. This process includes the following steps: Duration verification: The system starts a timer for each identified abnormal pattern. Only when the pattern persists for more than a preset alarm confirmation duration will the system confirm that it is a real, clinically significant event, rather than a transient fluctuation. Severity rating and alarm strategy: Different abnormal patterns correspond to different risk levels. The system will determine the alarm level based on the type of pattern and the degree to which the parameters deviate from the normal range, such as "Attention" (yellow alarm), "Warning" (orange alarm), or "Critical" (red alarm). Alarm Signal Generation and Transmission: Once the alarm decision is finalized, the system generates a structured alarm signal. This signal is not merely a trigger for audible and visual cues; it is a data packet containing rich contextual information, including: event type, event timestamp, event duration, respiratory rate and respiratory amplitude variability at the time of alarm triggering, and alarm level. This signal is then sent to various output units of the system, such as the bedside monitor display, the alarm list at the central monitoring station, and the patient's electronic medical record system for subsequent review and analysis.

[0031] In summary, the intelligent nighttime respiratory monitoring method for critically ill patients according to the embodiments of this application is explained. It first performs independent signal quality assessments on various signals, including infrared, radar, and acoustic signals, and innovatively introduces cross-modal signal consistency cross-verification to generate a more robust quality index for dynamic weighted fusion. This allows for real-time and intelligent identification and suppression of interfering low-quality signals, while prioritizing the use of high-quality signals. This ensures stable and reliable fused respiratory parameters are output even in complex monitoring environments, ultimately achieving high-precision and high-reliability continuous monitoring of the nighttime respiratory status of critically ill patients, effectively reducing false alarms and missed alarms, and improving monitoring efficiency and patient safety.

[0032] Furthermore, a nighttime intelligent respiratory monitoring system for critically ill patients is also provided.

[0033] Figure 3 This is a block diagram of a nighttime intelligent respiratory monitoring system for critically ill patients according to an embodiment of this application. Figure 3As shown, the intelligent nighttime respiratory monitoring system 300 for critically ill patients according to an embodiment of this application includes: a data preprocessing module 310, used to preprocess the acquired raw infrared video stream, raw radar data stream, and raw acoustic audio stream to obtain infrared respiratory waveforms, radar respiratory waveforms, and acoustic respiratory waveforms; a signal quality assessment module 320, used to assess the signal quality of the infrared respiratory waveforms, radar respiratory waveforms, and acoustic respiratory waveforms to obtain a dual-assessment infrared signal quality index, a dual-assessment radar signal quality index, and a dual-assessment acoustic signal quality index; a dynamic weighted fusion and respiratory parameter calculation module 330, used to perform dynamic weighted fusion and core respiratory parameter calculation on the infrared respiratory waveforms, radar respiratory waveforms, and acoustic respiratory waveforms based on the dual-assessment infrared signal quality index, dual-assessment radar signal quality index, and dual-assessment acoustic signal quality index to obtain the final respiratory rate and the final respiratory amplitude variability; and an abnormal pattern recognition and alarm decision module 340, used to perform abnormal pattern recognition and alarm decision based on the final respiratory rate and the final respiratory amplitude variability to generate an alarm signal.

[0034] As described above, the intelligent nighttime respiratory monitoring system 300 for critically ill patients according to embodiments of this application can be implemented in various wireless terminals, such as servers with intelligent nighttime respiratory monitoring algorithms for critically ill patients. In one possible implementation, the intelligent nighttime respiratory monitoring system 300 for critically ill patients according to embodiments of this application can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the intelligent nighttime respiratory monitoring system 300 for critically ill patients can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the intelligent nighttime respiratory monitoring system 300 for critically ill patients can also be one of many hardware modules of the wireless terminal.

[0035] Alternatively, in another example, the critical care patient nighttime respiratory intelligent monitoring system 300 and the wireless terminal can also be separate devices, and the critical care patient nighttime respiratory intelligent monitoring system 300 can be connected to the wireless terminal via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0036] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for intelligent monitoring of respiration of a critically ill patient during night, characterized in that, The method comprises the following steps: preprocessing the obtained original infrared video stream, original radar data stream and original acoustic audio stream to obtain infrared respiratory waveform, radar respiratory waveform and acoustic respiratory waveform; signal quality evaluation is performed on the infrared respiratory waveform, radar respiratory waveform and acoustic respiratory waveform to obtain double-evaluation infrared signal quality index, double-evaluation radar signal quality index and double-evaluation acoustic signal quality index; based on the double-evaluation infrared signal quality index, double-evaluation radar signal quality index and double-evaluation acoustic signal quality index, dynamic weighted fusion and core respiratory parameter calculation are performed on the infrared respiratory waveform, radar respiratory waveform and acoustic respiratory waveform to obtain final respiratory rate and final respiratory amplitude variability; based on the final respiratory rate and final respiratory amplitude variability, abnormal pattern recognition and alarm decision are performed to generate an alarm signal.

2. The intelligent monitoring of a critically ill patient's respiration during night hours method according to claim 1, characterized in that, The preprocessing of the obtained original infrared video stream, original radar data stream and original acoustic audio stream to obtain infrared respiratory waveform, radar respiratory waveform and acoustic respiratory waveform comprises the following steps: respiratory feature signal independent modal extraction is performed on the original infrared video stream, original radar data stream and original acoustic audio stream to obtain infrared respiratory feature time sequence signal, radar respiratory feature time sequence signal and acoustic respiratory feature time sequence signal; signal filtering and normalization are performed on the infrared respiratory feature time sequence signal, radar respiratory feature time sequence signal and acoustic respiratory feature time sequence signal to obtain the infrared respiratory waveform, radar respiratory waveform and acoustic respiratory waveform.

3. The intelligent monitoring of a critically ill patient's respiration during night hours method according to claim 1, characterized in that, The signal quality evaluation on the infrared respiratory waveform, radar respiratory waveform and acoustic respiratory waveform to obtain double-evaluation infrared signal quality index, double-evaluation radar signal quality index and double-evaluation acoustic signal quality index comprises the following steps: signal intra-quality index evaluation is performed on the infrared respiratory waveform, radar respiratory waveform and acoustic respiratory waveform to obtain infrared signal intra-quality index, radar signal intra-quality index and acoustic signal intra-quality index; based on the infrared respiratory waveform, radar respiratory waveform and acoustic respiratory waveform, signal inter-consistency check and quality index correction are performed on the infrared signal intra-quality index, radar signal intra-quality index and acoustic signal intra-quality index to obtain double-evaluation infrared signal quality index, double-evaluation radar signal quality index and double-evaluation acoustic signal quality index.

4. The intelligent monitoring of a critically ill patient's respiration during night hours method according to claim 3, characterized in that, The signal intra-quality index evaluation on the infrared respiratory waveform, radar respiratory waveform and acoustic respiratory waveform to obtain infrared signal intra-quality index, radar signal intra-quality index and acoustic signal intra-quality index comprises the following steps: the periodicity score, spectral purity score and waveform stability score of the infrared respiratory waveform are calculated; the weighted sum of the periodicity score, spectral purity score and waveform stability score of the infrared respiratory waveform is calculated to obtain the infrared signal intra-quality index.

5. The intelligent monitoring of a critically ill patient's respiration during night hours method according to claim 4, characterized in that, The calculation of the periodicity score, spectral purity score and waveform stability score of the infrared respiratory waveform comprises the following steps: time domain feature quantity extraction is performed on the infrared respiratory waveform to obtain infrared waveform autocorrelation function sequence, infrared respiratory waveform peak set and infrared respiratory waveform valley set; frequency domain feature quantity extraction is performed on the infrared respiratory waveform to obtain infrared respiratory waveform power spectral density function; calculating a periodicity score of the infrared respiratory waveform based on a sequence of autocorrelation functions of the infrared waveform; calculating a spectral purity score of the infrared respiratory waveform based on a power spectral density function of the infrared respiratory waveform; calculating a waveform stability score of the infrared respiratory waveform based on a set of infrared respiratory waveform peaks and a set of infrared respiratory waveform troughs.

6. The intelligent monitoring of a critically ill patient's respiration during night hours method according to claim 3, characterized in that, performing inter-signal consistency check and quality index correction on the infrared signal internal quality index, the radar signal internal quality index and the acoustic signal internal quality index to obtain the double-evaluated infrared signal quality index, the double-evaluated radar signal quality index and the double-evaluated acoustic signal quality index based on the infrared respiratory waveform, the radar respiratory waveform and the acoustic respiratory waveform, including: calculating a consistency matrix based on the infrared respiratory waveform, the radar respiratory waveform and the acoustic respiratory waveform; correcting the infrared signal internal quality index, the radar signal internal quality index and the acoustic signal internal quality index based on the consistency matrix to obtain the double-evaluated infrared signal quality index, the double-evaluated radar signal quality index and the double-evaluated acoustic signal quality index.

7. The intelligent monitoring of a critically ill patient's respiration during night hours method according to claim 6, characterized in that, calculating a consistency matrix based on the infrared respiratory waveform, the radar respiratory waveform and the acoustic respiratory waveform, including: extracting cross-modal comparison parameters from the infrared respiratory waveform, the radar respiratory waveform and the acoustic respiratory waveform to obtain a set of instantaneous respiratory rates and a set of per-cycle respiratory amplitude sequences; calculating a fundamental frequency consistency score matrix and a fundamental amplitude consistency score matrix based on the set of instantaneous respiratory rates and the set of per-cycle respiratory amplitude sequences; calculating a frequency consistency confidence matrix based on power spectral density functions of the infrared respiratory waveform, the radar respiratory waveform and the acoustic respiratory waveform; calculating an amplitude consistency confidence matrix based on respiratory amplitude sequences of the infrared respiratory waveform, the radar respiratory waveform and the acoustic respiratory waveform; determining the consistency matrix based on the fundamental frequency consistency score matrix, the fundamental amplitude consistency score matrix, the frequency consistency confidence matrix and the amplitude consistency confidence matrix.

8. The intelligent monitoring of a critically ill patient's nocturnal respiration method according to claim 1, wherein, performing dynamic weighted fusion and core respiratory parameter calculation on the infrared respiratory waveform, the radar respiratory waveform and the acoustic respiratory waveform based on the double-evaluated infrared signal quality index, the double-evaluated radar signal quality index and the double-evaluated acoustic signal quality index to obtain a final respiratory rate and a final respiratory amplitude variability, including: performing dynamic weighted fusion on the infrared respiratory waveform, the radar respiratory waveform and the acoustic respiratory waveform based on the double-evaluated infrared signal quality index, the double-evaluated radar signal quality index and the double-evaluated acoustic signal quality index to obtain a fused respiratory waveform; extracting periodicity features from the fused respiratory waveform to obtain the final respiratory rate and a sequence containing the amplitude of each respiratory cycle in the current window; calculating the final respiratory amplitude variability based on the sequence containing the amplitude of each respiratory cycle in the current window.

9. The intelligent monitoring of a critically ill patient's respiration during night hours method according to claim 8, characterized in that, calculating the final respiratory amplitude variability based on the sequence containing the amplitude of each respiratory cycle in the current window, including: calculating the final respiratory amplitude variability with the following formula, the formula being: ， wherein, is the standard deviation of the sequence comprising the amplitudes of each breath cycle within the current window, is the mean of the sequence comprising the amplitudes of each breath cycle within the current window.

10. A system for intelligent monitoring of respiration of a critically ill patient during night, characterized in that, including: a data preprocessing module configured to preprocess the acquired raw infrared video stream, raw radar data stream, and raw acoustic audio stream to obtain an infrared respiration waveform, a radar respiration waveform, and an acoustic respiration waveform; a signal quality assessment module configured to assess the signal quality of the infrared respiration waveform, the radar respiration waveform, and the acoustic respiration waveform to obtain a double-assessed infrared signal quality index, a double-assessed radar signal quality index, and a double-assessed acoustic signal quality index; a dynamic weighted fusion and respiration parameter solving module configured to perform dynamic weighted fusion and core respiration parameter solving on the infrared respiration waveform, the radar respiration waveform, and the acoustic respiration waveform based on the double-assessed infrared signal quality index, the double-assessed radar signal quality index, and the double-assessed acoustic signal quality index to obtain a final respiration rate and a final respiration amplitude variability; an abnormal pattern recognition and alarm decision module configured to perform abnormal pattern recognition and alarm decision based on the final respiration rate and the final respiration amplitude variability to generate an alarm signal.

Images