Real-time monitoring method for respiratory state of PACU patient based on 60ghz millimeter wave radar

Abstract

The present application relates to a PACU patient respiratory state real-time monitoring method based on 60GHz millimeter wave radar, which solves the defects of insufficient stability of contact monitoring under high-frequency nursing intervention and body movement conditions, insufficient identification ability of existing radar methods for non-stationary abnormal respiratory state, and probability burr and alarm flicker prone to occur in bedside continuous early warning compared with the prior art.The present application comprises the following steps: PACU patient radar signal acquisition;radar signal preprocessing;construction of respiratory state recognition network;training of respiratory state recognition network;acquisition of PACU patient radar real-time signal and preprocessing;obtaining of monitoring results.The present application applies 60GHz millimeter wave radar to the PACU extubation recovery period patient respiratory state monitoring scene, collects I / Q radar echo signals of the chest and abdominal regions of the patient through a non-contact mode, can continuously acquire radar signals reflecting chest and abdominal micro-movements without increasing the burden of contact sensors on the patient, and is suitable for the monitoring environment of frequent PACU bedside nursing operations and frequent patient body position changes.

Description

Technical Field

[0001] This invention relates to the fields of non-contact vital sign monitoring, millimeter-wave radar signal processing, and perioperative intelligent monitoring, specifically a method for real-time monitoring of respiratory status in PACU patients based on 60GHz millimeter-wave radar. Background Technology

[0002] The post-anesthesia care unit (PACU) is a location where perioperative respiratory risks are highly concentrated. During the extubation recovery phase, patients typically undergo a dynamic transition from mechanical ventilation to spontaneous breathing, during which they may experience patient-ventilator asynchrony, agitation, struggling, respiratory arrest, and target departure. Failure to promptly identify these conditions can increase the risk of airway events, inadequate ventilation, and delayed bedside intervention.

[0003] Current PACU respiratory monitoring primarily relies on ventilator waveforms, pulse oxygen saturation, end-tidal carbon dioxide, and bedside observation by healthcare workers. While these methods are valuable in clinical practice, continuous and stable monitoring is easily affected by patient movement, nursing procedures, changes in tubing coupling, and equipment disconnection. Millimeter-wave radar, with its non-contact nature, minimal privacy exposure, continuous data acquisition, and sensitivity to subtle chest and abdominal movements, is suitable for bedside auxiliary monitoring.

[0004] However, existing vital sign monitoring based on millimeter-wave radar mainly focuses on respiratory rate estimation or periodic rhythm analysis under relatively stable conditions. For short-term, sudden, and non-stationary micro-motion distortions that occur during the PACU extubation recovery period, simply relying on respiratory rate estimation or single-frame model output is insufficient to obtain stable and reliable bedside early warning results. Furthermore, the real PACU environment is subject to interference from personnel movement, bed vibration, multipath reflection, patient position changes, and nursing interventions, while the proportion of abnormal events is relatively low. This leads to challenges in modeling, such as local distortion extraction, long-term time-series dependent modeling, and continuous early warning anti-shake. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies, such as insufficient stability of contact monitoring under high-frequency nursing intervention and physical movement conditions, insufficient ability of existing radar methods to identify non-stable abnormal respiratory states, and the tendency of continuous bedside warnings to exhibit probability spikes and alarm flashing. This invention provides a real-time respiratory status monitoring method for PACU patients based on 60GHz millimeter-wave radar to solve the above problems.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] A method for real-time monitoring of respiratory status in PACU patients based on 60GHz millimeter-wave radar includes the following steps:

[0008] The acquisition of radar signals from PACU patients was carried out using a 60GHz millimeter-wave radar to collect I / Q radar echo signals from the patient's chest and abdominal region, and target range units were selected from the I / Q radar echo signals.

[0009] Preprocessing of radar signals;

[0010] Construct a respiratory state recognition network;

[0011] Training the breathing state recognition network: Input the sliding window sample into the breathing state recognition network, output the probability of abnormal breathing state, and train the breathing state recognition network.

[0012] Acquire real-time radar signals from PACU patients and perform preprocessing;

[0013] Obtaining monitoring results: The pre-processed real-time radar signal of the PACU patient is input into the trained respiratory status recognition network to obtain the probability of abnormal respiratory status; after the respiratory status recognition network, the abnormal respiratory status probability is stabilized by the anti-shake decision module according to the physical gating rule, probability smoothing rule, hysteresis threshold rule and evidence accumulation rule, and the bedside respiratory status is output.

[0014] The preprocessing of the radar signal includes the following steps:

[0015] Phase signals are extracted from the I / Q radar echo signals corresponding to the target range cell;

[0016] The phase signal is unwrapped to eliminate phase jumps;

[0017] The unwound phase signal is filtered to suppress environmental noise and non-breathing motion interference.

[0018] The filtered signal is normalized to obtain the radar micro-motion signal reflecting the slight movement of the chest and abdomen.

[0019] A sliding window sample is constructed for the radar micro-motion signal according to a preset window length. The sliding window sample is used to cover multiple respiratory cycles and retain short-term abnormal micro-motion distortions during the PACU extubation recovery period.

[0020] The construction of the respiratory state recognition network includes the following steps:

[0021] The breathing state recognition network is designed to include a multi-scale one-dimensional convolutional branch, a channel attention module, a bidirectional temporal modeling module, and a classification output module. The multi-scale one-dimensional convolutional branch is used to extract breathing trend features and short-term local mutation features. The channel attention module is used to enhance the channel response related to breathing state discrimination. The bidirectional temporal modeling module is used to fuse forward and backward contextual information. Finally, the classification output module outputs the probability of abnormal breathing state.

[0022] A multi-scale one-dimensional convolution branch is set up to extract features from sliding window samples through one-dimensional convolution with different kernel scales, respectively obtaining respiratory trend features and short-term local mutation features, and then fusing features of different scales;

[0023] A channel attention module is set up to recalibrate the channels of the fusion features output by multi-scale one-dimensional convolution branches, thereby enhancing the channel response related to respiratory state discrimination.

[0024] A bidirectional temporal modeling module is set up to perform forward and backward context modeling on the temporal features processed by the channel attention module, so as to obtain temporal features that reflect the process of respiratory state changes.

[0025] A classification output module is set to output the classification probabilities of stable tolerance state and abnormal resistance state based on the time-series characteristics, and to obtain the probability of abnormal respiratory state based on the classification probabilities.

[0026] The real-time monitoring system for training breathing status includes the following steps:

[0027] Simultaneously collect radar signals, ventilator pressure waveforms, flow waveforms, volume waveforms, and clinical observation records of PACU patients;

[0028] The radar signal is preprocessed and a sliding window sample is constructed;

[0029] Based on the ventilator pressure waveform, flow waveform, volume waveform, and clinical observation records, the ranges of stable tolerance state and abnormal resistance state are determined, and the state ranges are mapped to the corresponding sliding window samples to obtain training labels;

[0030] The sliding window samples with training labels are input into the multi-scale one-dimensional convolution branch. Respiratory trend features and short-term local mutation features are extracted through one-dimensional convolution with different kernel scales. The features of different scales are then fused to obtain multi-scale fused features.

[0031] The multi-scale fusion features are input into the channel attention module, and the multi-scale fusion features are recalibrated to obtain attention-enhanced features related to respiratory state discrimination.

[0032] The attention enhancement features are input into the bidirectional temporal modeling module, and forward and backward contextual information are fused to obtain temporal features that reflect the process of respiratory state changes.

[0033] The time-series features are input into the classification output module, which outputs the classification probabilities of stable tolerance state and abnormal resistance state, and obtains the probability of abnormal breathing state based on the classification probabilities.

[0034] Based on the difference between the abnormal breathing state probability and the training label, the parameters of the multi-scale one-dimensional convolutional branch, the channel attention module, the bidirectional temporal modeling module, and the classification output module are updated to obtain the trained breathing state recognition network.

[0035] A computer-readable storage medium storing a computer program, which, when executed by a processor, enables a method for real-time monitoring of the respiratory status of PACU patients based on 60GHz millimeter-wave radar.

[0036] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, which, when executed by the processor, enables a method for real-time monitoring of the respiratory status of PACU patients based on 60GHz millimeter-wave radar.

[0037] Beneficial effects

[0038] The present invention provides a real-time respiratory status monitoring method for PACU patients based on 60GHz millimeter-wave radar. Compared with the prior art, this method applies 60GHz millimeter-wave radar to the respiratory status monitoring scenario of PACU patients in the extubation recovery period. It collects I / Q radar echo signals of the patient's chest and abdomen region in a non-contact manner, and can continuously acquire radar signals reflecting chest and abdominal micro-movements without increasing the burden on the patient's contact sensors. It is suitable for monitoring environments in PACU where bedside nursing operations are frequent and patient positions change frequently.

[0039] This invention converts radar chest and abdominal micromotion signals into time-series samples that can be used for respiratory status identification based on target range cell selection, phase extraction, phase unwinding, filtering, normalization, and sliding window sample construction. This helps to preserve short-term, sudden, and non-stationary abnormal micromotion changes during the PACU extubation recovery period.

[0040] This invention analyzes sliding window samples through a respiratory state recognition network and outputs the probability of abnormal respiratory states, enabling the system to identify stable tolerance states and abnormal resistance states from the time-series characteristics of chest and abdominal micro-movements, which is different from the traditional radar monitoring method that mainly relies on periodic respiratory rate estimation.

[0041] This invention sets up an anti-shake decision module after the respiratory state recognition network. Through physical gating rules, probability smoothing rules, hysteresis threshold rules, and evidence accumulation rules, it stabilizes the probability of abnormal respiratory states, which can reduce false alarms, frequent state jumps, and alarm flashing caused by the probability fluctuation of a single sliding window sample, and improve the stability of bedside continuous monitoring results.

[0042] This invention can output bedside respiratory statuses such as stable tolerance, abnormal resistance, risk of apnea, empty bed or target detachment, expanding radar monitoring results from single physiological parameter estimation to state-based monitoring results for PACU clinical nursing scenarios, facilitating continuous auxiliary observation of patients' respiratory status during extubation recovery by medical staff. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the PACU bedside millimeter-wave radar deployment in an embodiment of the present invention;

[0044] Figure 2 This is a schematic diagram of the overall process of the real-time monitoring method for respiratory status of PACU patients based on 60GHz millimeter-wave radar according to the present invention.

[0045] Figure 3 This is a schematic diagram of the respiratory state recognition network structure in this invention;

[0046] Figure 4 This is a schematic diagram of the anti-shake decision-making process in this invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following embodiments are used to explain this invention and are not intended to limit the scope of protection of this invention.

[0048] like Figure 1 As shown, in one embodiment, a 60GHz millimeter-wave radar is installed beside the PACU bed, with the radar's detection direction covering the patient's chest and abdomen area. The 60GHz millimeter-wave radar can be mounted on an adjustable bracket, which is used to adjust the radar's height, horizontal distance, pitch angle, and yaw angle relative to the patient's chest and abdomen area, thereby adapting to the PACU bedside space, patient position, and nursing operating environment. The radar continuously acquires I / Q radar echo signals from the patient's chest and abdomen area and selects target distance units corresponding to the patient's chest and abdomen area from the I / Q radar echo signals as the data source for subsequent respiratory status analysis.

[0049] like Figure 2As shown, the real-time monitoring method for respiratory status of PACU patients based on 60GHz millimeter-wave radar of the present invention includes the following process: First, I / Q radar echo signals of the chest and abdomen region of PACU patients are acquired using 60GHz millimeter-wave radar, and target range cells are selected; then, the radar signals corresponding to the target range cells are subjected to phase extraction, phase dewinding, filtering, and normalization processing to obtain radar micro-motion signals reflecting the micro-movements of the patient's chest and abdomen; subsequently, a sliding window sample is constructed according to a preset window length; then, the sliding window sample is input into a respiratory status recognition network to obtain the probability of abnormal respiratory status; finally, after the respiratory status recognition network, the probability of abnormal respiratory status is stabilized by an anti-shake decision module, and the bedside respiratory status is output.

[0050] During radar signal preprocessing, phase signals are extracted based on the I / Q radar echo signals corresponding to the target range unit. The phase signals are then unwrapped to eliminate phase jumps. The unwrapped phase signals are filtered to suppress environmental noise and non-respiratory body motion interference. The filtered signals are then normalized to obtain radar micro-motion signals reflecting the patient's chest and abdominal micro-movements. Subsequently, a sliding window sample is constructed from the radar micro-motion signals according to a preset window length, ensuring that the sliding window sample covers multiple respiratory cycles and preserves short-term abnormal micro-motion distortions that may occur during the PACU extubation recovery period.

[0051] like Figure 3 As shown, the respiratory state recognition network includes a multi-scale one-dimensional convolutional branch, a channel attention module, a bidirectional temporal modeling module, and a classification output module. The multi-scale one-dimensional convolutional branch extracts features from sliding window samples through one-dimensional convolutions at different scales, obtaining respiratory trend features and short-term local mutation features, and then fuses features from different scales. The channel attention module recalibrates the channels of the fused features, enhancing channel responses relevant to respiratory state discrimination. The bidirectional temporal modeling module fuses forward and backward contextual information to obtain temporal features reflecting the respiratory state change process. The classification output module outputs classification probabilities for stable tolerance states and abnormal adversarial states based on the temporal features, and obtains the probability of abnormal respiratory states based on the classification probabilities.

[0052] During the training phase, radar signals, ventilator pressure waveforms, flow waveforms, volume waveforms, and clinical observation records of PACU patients were simultaneously acquired. Based on the ventilator pressure waveforms, flow waveforms, volume waveforms, and clinical observation records, stable tolerance state and abnormal resistance state intervals were determined, and these state intervals were mapped to corresponding sliding window samples to obtain training labels. The sliding window samples with training labels were input into the respiratory state recognition network. Based on the difference between the network output probability and the training labels, the parameters of the multi-scale one-dimensional convolutional branch, channel attention module, bidirectional temporal modeling module, and classification output module were updated to obtain the trained respiratory state recognition network.

[0053] During the real-time monitoring phase, the system continuously acquires real-time radar signals from PACU patients and constructs real-time sliding window samples using the same preprocessing method as in the training phase. These real-time sliding window samples are then input into the trained respiratory state recognition network to obtain the probability of abnormal respiratory states corresponding to the current sliding window sample.

[0054] like Figure 4 As shown, a stabilization decision module is set after the breathing state recognition network. This stabilization decision module stabilizes the probability of abnormal breathing states based on physical gating rules, probability smoothing rules, hysteresis threshold rules, and evidence accumulation rules. Specifically, the physical gating rule is used to determine whether there is an empty bed, target detachment, or significant risk of apnea based on signal energy, signal amplitude, or target distance range; the probability smoothing rule is used to reduce the impact of short-term probability fluctuations on the output results; the hysteresis threshold rule is used to avoid frequent switching of bedside breathing states near critical thresholds; and the evidence accumulation rule is used to output the abnormal state after abnormal evidence continuously meets preset conditions. After the above stabilization decision processing, the system outputs the bedside breathing state, which includes a stable tolerance state, an abnormal resistance state, an apnea risk state, and an empty bed or target detachment state.

[0055] Through the above methods, the present invention can perform non-contact continuous monitoring of the patient's chest and abdominal micromovements during the PACU extubation recovery period, and obtain a stable bedside respiratory status output through a respiratory status recognition network and anti-shake decision module, thereby reducing false alarms and alarm flashing caused by the probability fluctuation of a single sliding window sample.

[0056] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for real-time monitoring of respiratory status in PACU patients based on 60GHz millimeter-wave radar, characterized in that, Includes the following steps: 11) Acquisition of radar signals for PACU patients: I / Q radar echo signals of the patient's chest and abdomen region were acquired using a 60GHz millimeter-wave radar, and target range units were selected from the I / Q radar echo signals. 12) Radar signal preprocessing; 13) Construct a respiratory state recognition network; 14) Training the breathing state recognition network: Input the sliding window sample into the breathing state recognition network, output the probability of abnormal breathing state, and train the breathing state recognition network. 15) Acquire real-time radar signals from PACU patients and perform preprocessing; 16) Obtaining monitoring results: The pre-processed real-time radar signal of the PACU patient is input into the trained respiratory status recognition network to obtain the probability of abnormal respiratory status; after the respiratory status recognition network, the abnormal respiratory status probability is stabilized by the anti-shake decision module according to the physical gating rule, probability smoothing rule, hysteresis threshold rule and evidence accumulation rule, and the bedside respiratory status is output.

2. The method for real-time monitoring of respiratory status of PACU patients based on 60GHz millimeter-wave radar according to claim 1, characterized in that, The preprocessing of the radar signal includes the following steps: 21) Extract the phase signal based on the I / Q radar echo signal corresponding to the target range cell; 22) Perform phase unwrapping on the phase signal to eliminate phase jumps; 23) Filter the unwound phase signal to suppress environmental noise and non-breathing motion interference; 24) The filtered signal is normalized to obtain the radar micro-motion signal reflecting the micro-motion of the chest and abdomen; 25) Construct a sliding window sample for the radar micro-motion signal according to a preset window length. The sliding window sample is used to cover multiple respiratory cycles and retain short-term abnormal micro-motion distortion during the PACU extubation recovery period.

3. The method for real-time monitoring of respiratory status of PACU patients based on 60GHz millimeter-wave radar according to claim 1, characterized in that, The construction of the respiratory state recognition network includes the following steps: 31) The breathing state recognition network is set to include a multi-scale one-dimensional convolutional branch, a channel attention module, a bidirectional temporal modeling module, and a classification output module. The multi-scale one-dimensional convolutional branch is used to extract breathing trend features and short-term local mutation features. The channel attention module is used to enhance the channel response related to breathing state discrimination. The bidirectional temporal modeling module is used to fuse forward and backward context information. The classification output module outputs the probability of abnormal breathing state. 32) Set up a multi-scale one-dimensional convolution branch to extract features from sliding window samples through one-dimensional convolution with different kernel scales, obtain respiratory trend features and short-term local mutation features respectively, and fuse features of different scales; 33) Set up a channel attention module to perform channel recalibration on the fusion features of the multi-scale one-dimensional convolution branch outputs, and enhance the channel response related to respiratory state discrimination; 34) A bidirectional temporal modeling module is set up to perform forward and backward context modeling on the temporal features processed by the channel attention module to obtain temporal features that reflect the process of respiratory state changes; 35) Set a classification output module to output the classification probabilities of stable tolerance state and abnormal resistance state according to the time sequence characteristics, and obtain the probability of abnormal breathing state according to the classification probabilities.

4. The method for real-time monitoring of respiratory status of PACU patients based on 60GHz millimeter-wave radar according to claim 1, characterized in that, The real-time monitoring system for training breathing status includes the following steps: 41) Simultaneously collect radar signals, ventilator pressure waveforms, flow waveforms, volume waveforms, and clinical observation records of PACU patients; 42) Preprocess the radar signal and construct a sliding window sample; 43) Determine the intervals between stable tolerance state and abnormal resistance state based on the ventilator pressure waveform, flow waveform, volume waveform and clinical observation records, and map the state intervals to the corresponding sliding window samples to obtain training labels; 44) Input the sliding window samples with training labels into the multi-scale one-dimensional convolution branch, extract the breathing trend features and short-term local mutation features through one-dimensional convolution with different kernel scales, and fuse the features of different scales to obtain multi-scale fused features. 45) Input the multi-scale fusion features into the channel attention module, perform channel recalibration on the multi-scale fusion features, and obtain attention enhancement features related to respiratory state discrimination; 46) Input the attention enhancement features into the bidirectional temporal modeling module, fuse the forward and backward contextual information, and obtain temporal features that reflect the process of respiratory state changes; 47) Input the time-series features into the classification output module, output the classification probabilities of stable tolerance state and abnormal resistance state, and obtain the probability of abnormal breathing state based on the classification probabilities; 48) Based on the difference between the abnormal breathing state probability and the training label, update the parameters of the multi-scale one-dimensional convolutional branch, the channel attention module, the bidirectional temporal modeling module, and the classification output module to obtain the trained breathing state recognition network.

5. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, enables the real-time monitoring method for the respiratory status of PACU patients based on 60GHz millimeter-wave radar as described in any one of claims 1-4.

6. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it can realize the method for real-time monitoring of the respiratory status of PACU patients based on 60GHz millimeter-wave radar as described in any one of claims 1-4.

Images