Newborn intelligent monitoring bed and dynamic identification method for abnormal vital signs

Abstract

The application discloses a neonatal intelligent monitoring bed and a vital sign abnormality dynamic identification method, and relates to the technical field of neonatal monitoring beds.The neonatal intelligent monitoring bed comprises a bed frame, a flexible mattress, a bed plate driving unit, a main control unit and a warning unit, is provided with a multi-section adjustable flexible bed plate and a multi-modal sensing unit, and is characterized in that the multi-section adjustable flexible bed plate is divided into a head section bed plate, a trunk section bed plate, a left side turning bed plate and a right side turning bed plate, each section of the bed plate can be independently adjusted in inclination, the bed plate driving unit is in transmission connection with the multi-section adjustable flexible bed plate and is used for driving each section of the bed plate to adjust the inclination.The application realizes reflux preposition prevention and control, accurately captures the precursor pathological characteristics 10-30 seconds before reflux occurs, completely changes the lag mode of the prior art that alarms only after milk overflow occurs, avoids the risk of aspiration and suffocation from the root, and fills the technical blank of pre-prevention and control of neonatal reflux.

Description

Technical Field

[0001] This application relates to the field of medical monitoring technology, and in particular to a newborn intelligent monitoring bed and a method for dynamic identification of abnormal vital signs. Background Technology

[0002] Newborns, especially premature infants with a corrected gestational age of less than 37 weeks, have a clinical incidence of gastroesophageal reflux as high as 80% due to the incomplete development of the lower esophageal sphincter and the horizontal position of the stomach. Aspiration and asphyxia caused by reflux are one of the core causes of sudden neonatal death syndrome and a key challenge in neonatal safety management in NICUs and mother-infant rooming-in settings. Currently, clinical prevention and control of neonatal reflux mainly relies on regular manual rounds and post-incident management by medical staff, which cannot achieve 24-hour continuous monitoring. This increases the risk of adverse events during breaks in rounds and significantly increases the manpower costs of clinical nursing.

[0003] Among the existing technologies for neonatal reflux prevention, one type is reflux monitoring equipment, which mostly uses contact electrodes and photoelectric sensors to collect vital signs. These devices need to be in direct contact with the newborn's skin, which can easily cause pressure damage to the newborn's delicate skin. They are also easily affected by swaddling and the newborn's position, and cannot operate stably in routine care scenarios. The other type is anti-reflux monitoring beds, which mostly adopt a static bed design with a fixed angle. They can only achieve passive protection by raising the head of the bed in a pre-set manner. They cannot dynamically adjust the position according to the newborn's real-time physiological state. Excessive elevation of the position can easily cause side effects such as the newborn sliding down and increased intracranial pressure, and the protective effect is very limited.

[0004] More importantly, existing reflux monitoring solutions are all reactive alarms that occur after spitting up or vomiting. By this time, the refluxed material has already entered the newborn's mouth and nose, and the risk of aspiration and suffocation has already materialized. These solutions cannot capture the warning signs that precede reflux and provide early warnings. In addition, existing identification solutions mostly use general fixed threshold algorithms, which are not adapted to the physiological characteristics of newborns of different corrected gestational ages. This results in serious problems of high false alarms and high false alarms. Frequent false alarms cause noise pollution in the NICU, which can affect the neurological development of premature infants and also reduce the response efficiency of medical staff to alarms.

[0005] Currently, there is no complete technical solution in this field that can simultaneously achieve early identification of reflux precursors, non-contact continuous monitoring, and adaptive closed-loop intervention, which cannot meet the core needs of clinical neonatal reflux safety prevention and control.

[0006] Therefore, we propose a method for intelligent neonatal monitoring beds and dynamic identification of abnormal vital signs. Summary of the Invention

[0007] This application aims to at least partially solve one of the technical problems in the aforementioned technologies.

[0008] To achieve the above objectives, the first aspect of this application proposes a newborn intelligent monitoring bed, including a bed frame, a flexible mattress, a bed board drive unit, a main control unit and an early warning unit, and equipped with a multi-segment adjustable flexible bed board and a multi-modal sensing unit;

[0009] The multi-segment adjustable flexible bed board is divided into a head section bed board, a torso section bed board, a left folding bed board, and a right folding bed board. Each section bed board can be independently adjusted in tilt angle. The bed board drive unit is connected to the multi-segment adjustable flexible bed board for driving each section bed board to adjust its tilt angle.

[0010] The multimodal sensing unit includes a distributed flexible pressure sensing array and a dual-channel 24GHz micro millimeter-wave sensing module. The distributed flexible pressure sensing array is embedded in the flexible mattress and is used to collect the newborn's body position distribution, chest and abdominal respiratory movements, abdominal pressure changes and body movement signals. The dual-channel 24GHz micro millimeter-wave sensing module is symmetrically installed on both sides of the bed frame headboard and is used to collect the newborn's oral and nasal cavity micro-flow respiratory signals and chest and abdominal displacement respiratory wave signals.

[0011] The main control unit is electrically connected to the multimodal sensing unit, the bed board driving unit, and the early warning unit, respectively. It is used to extract the early signs of neonatal gastroesophageal reflux based on the signals collected by the multimodal sensing unit, identify the reflux risk level, and control the early warning unit to output the corresponding early warning signal and control the bed board driving unit to adjust the posture of the multi-segment adjustable flexible bed board according to the risk level.

[0012] This technical solution utilizes dual-modal non-contact acquisition to fundamentally avoid damage to the delicate skin of newborns caused by contact sensors. It is also unaffected by swaddling or body position obstruction, enabling continuous and stable monitoring for 24 hours. This allows the bed structure to be adapted to clinical needs and enables precise dynamic adjustment of the anti-reverse fluid position. Unlike the passive protection of a static bed with a fixed angle, this solution significantly improves the protective effect and safety.

[0013] The second aspect of this application proposes a method for dynamic identification of abnormal vital signs in newborns, including the following steps:

[0014] S1 signal synchronous acquisition: Through the same clock source of the main control unit, the distributed flexible pressure sensing array and the dual 24GHz micro millimeter wave sensing module are controlled to synchronously acquire signals and complete the hard time alignment of the two signals.

[0015] S2 signal preprocessing: The two acquired signals are filtered and denoised, decomposed into frequency bands, and invalid windows are removed to obtain effective vital signs signals;

[0016] S3 Precursor Feature Extraction: Extract chest and abdominal movement linkage deviation features, oral and nasal microflow respiratory abnormality features, and body position-abdominal pressure abnormality features from effective vital signs signals to construct a gastroesophageal reflux precursor feature set;

[0017] S4 Risk Classification Identification: Input the feature set of gastroesophageal reflux precursors into a pre-trained neonatal reflux risk classification model and output the corresponding reflux risk level;

[0018] S5 graded linkage intervention: Trigger the corresponding level of warning according to the risk level of reflux, and at the same time control the multi-segment adjustable flexible bed board to adjust to the corresponding posture. After the intervention is completed, continuously monitor the changes in vital signs, and control the bed board to return to its original position after the vital signs return to normal.

[0019] This technical solution enables early warning and intervention for gastroesophageal reflux, avoiding the risk of aspiration and suffocation caused by reflux at its source. All steps are completed locally on the bed, without cloud dependence, and the real-time performance and safety meet the requirements of medical equipment.

[0020] In addition, the neonatal intelligent monitoring bed proposed in this application may also have the following additional technical features:

[0021] As a further description of the above technical solution:

[0022] The bed board drive unit uses a medical silent stepper motor, and the adjustment speed of a single bed board is 1° / s; the lifting range of the head section bed board is 0-15°, and the lifting range of the left and right side bed boards is 0-20°. The bed adjustment process of this technical solution is stress-free and interference-free, which not only ensures the anti-backflow effect of body position adjustment, but also does not affect the stability of the monitoring signal, thus solving the problem of mutual interference between bed adjustment and monitoring in the existing technology.

[0023] As a further description of the above technical solution:

[0024] The distributed flexible pressure sensor array is divided into four acquisition zones: head, chest, abdomen, and pelvis. The sensor point density in the chest and abdomen zones is twice that in the head and pelvis zones. The range of this sensor array is 0-10 kPa, with a sampling rate of 100 Hz. The dual-channel 24 GHz micro millimeter-wave sensor module has a sampling rate of 200 Hz and a distance resolution of no less than 1 mm. This technical solution can accurately capture the weak vital signs of newborns, ensuring recognition accuracy from the acquisition end. The dual-modal sensor parameters are matched to each other, providing a standardized data source for subsequent feature extraction and signal alignment, thereby reducing false alarm and false negative rates from the source.

[0025] As a further description of the above technical solution:

[0026] The main control unit uses an STM32H7 series MCU paired with an ESP32-S3AI acceleration unit. All signal processing and recognition operations are completed on the local embedded end of the bed. The main control unit is equipped with RS485 and Ethernet communication interfaces for connecting to the hospital's central monitoring system at the nursing station. This technical solution can guarantee a delay of ≤300ms from signal acquisition to intervention action execution, fully meeting the real-time requirements of medical monitoring. The local operation mode complies with medical data security standards, eliminating the risk of leakage of patient privacy data.

[0027] As a further description of the above technical solution:

[0028] The warning unit adopts a three-level warning mechanism. When the risk is low, it triggers a local silent soft light LED prompt on the bed. When the risk is medium, it triggers a pop-up prompt on the central monitoring system at the nurse station. When the risk is high, it simultaneously triggers a low-decibel audible and visual alarm on the bed, an emergency audible and visual alarm at the nurse station, and a warning push notification on the medical staff's handheld terminal. This technical solution completely solves the problem of noise pollution in the NICU caused by the common audible and visual alarms in existing technologies, reduces the adverse effects of noise on the neurological development of premature infants, meets the special requirements of the clinical environment, and the graded warning mechanism can effectively reduce alarm fatigue of medical staff, improve the response efficiency of high-risk emergency events, and the warning mode is accurately matched with the risk level, so as not to miss high-risk events or cause unnecessary clinical interference.

[0029] In addition, the method for dynamic identification of abnormal vital signs in newborns proposed in this application may also have the following additional technical features:

[0030] As a further description of the above technical solution:

[0031] The specific signal preprocessing method in step S2 is as follows: For the pressure sensing signal, a sliding window mid-range filter is used to remove baseline drift; an adaptive notch filter is used to filter out 50Hz power frequency interference; then, a third-order Butterworth filter is used to decompose the signal into low-frequency postural abdominal pressure components, mid-frequency respiratory motion components, and high-frequency body motion interference components. When the high-frequency body motion interference component exceeds a preset threshold, the current window is marked as an invalid window and feature extraction is paused. For the millimeter-wave sensing signal, an FFT transform is used to lock the distance gate between the newborn's mouth / nose and chest / abdomen regions. After phase demodulation to extract the effective respiratory signal, a method using pressure... The force sensing signal is denoised using the same filtering rules, and its time axis is aligned with that of the pressure sensing signal. This technical solution can effectively suppress interference from the NICU environment and the newborn's own movements, significantly improving the purity of the effective signal and reducing the false alarm rate of the identification process from the source. The preprocessing process is fully compatible with the signal characteristics of the dual-modal sensing, maximizing the retention of effective features related to the precursors of reflux while eliminating invalid interference. This solves the problems of poor anti-interference ability and easy signal distortion in existing technologies. The processing process is standardized and can run stably on embedded devices without complex calculations, thus avoiding recognition delay.

[0032] As a further description of the above technical solution:

[0033] In step S3, the characteristics of chest and abdominal movement linkage deviation include the mean phase difference of chest and abdominal respiratory waves, the mean linkage coefficient, and the linkage variability. The characteristics of micro-flow respiratory abnormalities in the oral and nasal cavities include the mean inspiratory-expiratory ratio, the peak-to-peak variability, and the proportion of micro-breathing pauses. The characteristics of abnormal body position and abdominal pressure include the trunk-head tilt angle difference, the increase in abdominal pressure, and the variability of abdominal pressure. All characteristics are calculated based on the effective vital signs signals of five consecutive respiratory cycles. This technical solution can accurately capture early warning signals before reflux occurs, truly achieving early warning. The multi-dimensional characteristics verify each other, avoiding high false alarms caused by single characteristics. It can improve the identification accuracy of pathological reflux to over 95% and control the false alarm rate to within 3%.

[0034] As a further description of the above technical solution:

[0035] The specific rules for the graded linkage intervention in step S5 are as follows: under low-risk conditions, only the corresponding warning is triggered and the characteristic changes are continuously monitored, without adjusting the bed board posture;

[0036] In the medium-risk state, simultaneously raise the head section of the bed board to 10° and the right side of the bed board to 12°. After adjustment, continuously monitor for 3 respiratory cycles. If the characteristics return to normal, maintain the current position for 30 minutes and then automatically reset.

[0037] In high-risk situations, the highest level of warning is triggered simultaneously, raising the head section of the bed board to 15° and the right side of the bed board to 15°, continuously locking and monitoring vital signs data until medical staff manually take over or the vital signs return to normal.

[0038] This technical solution enables adaptive dynamic intervention based on real-time risk, unlike the static passive protection of existing technologies with fixed angles. It can accurately block the progression of reflux at different levels, with an intervention effectiveness rate of over 90%. All intervention actions strictly follow neonatal clinical nursing guidelines, ensuring the anti-reflux effect while completely avoiding position-related safety risks.

[0039] Advantages of this invention:

[0040] The newborn intelligent monitoring bed and the dynamic identification method for abnormal vital signs according to this application overturn the traditional post-event identification logic, realize the prevention and control of reflux in advance, accurately capture the pathological characteristics of the precursor 10-30 seconds before the occurrence of reflux, completely change the lagging mode of existing technology that only alarms after milk regurgitation occurs, avoid the risk of aspiration and suffocation from the root, and fill the technical gap in the prevention and control of newborn reflux in advance.

[0041] The dual-modal contactless sensing architecture balances safety and accuracy. It adopts a complementary pressure and millimeter wave acquisition design, which is completely contactless and does not need to be attached to the newborn's skin, avoiding the risk of skin damage from contact devices. It can penetrate the swaddling clothes to stably capture the weak vital signs of newborns, providing a highly reliable data source for accurate identification.

[0042] The monitoring and intervention system features a closed-loop design with strong clinical adaptability and feasibility. It addresses the industry pain point of the disconnect between monitoring and intervention, automatically matching graded postural interventions after risk identification, achieving a reflux relief rate of over 90%, adapting to newborns of different gestational ages, with an accuracy rate of ≥95% and a false alarm rate of ≤3%, and local operations comply with medical data security requirements.

[0043] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0044] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0045] Figure 1 This is a schematic diagram of the module connections of a neonatal intelligent monitoring bed and a method for dynamic identification of abnormal vital signs according to an embodiment of this application;

[0046] Figure 2 This is a schematic flowchart of a method for identifying abnormal vital signs in a newborn intelligent monitoring bed according to an embodiment of this application;

[0047] Figure 3 This is a bar chart comparing the core performance indicators of a neonatal intelligent monitoring bed and a dynamic identification method for abnormal vital signs according to an embodiment of this application;

[0048] Figure 4 This is a comparison diagram of the backflow warning advance time of a neonatal intelligent monitoring bed and a dynamic identification method for abnormal vital signs according to an embodiment of this application;

[0049] Figure 5 This is a multi-dimensional comprehensive comparison radar chart of a neonatal intelligent monitoring bed and a method for dynamic identification of abnormal vital signs according to an embodiment of this application. Detailed Implementation

[0050] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0051] The following description, in conjunction with the accompanying drawings, illustrates an embodiment of the intelligent neonatal monitoring bed and a method for dynamically identifying abnormal vital signs according to this application.

[0052] like Figure 1 As shown, the neonatal intelligent monitoring bed of Embodiment 1 of this application may include a bed frame, a flexible mattress, a bed board drive unit, a main control unit and an early warning unit, a multi-segment adjustable flexible bed board and a multi-modal sensing unit;

[0053] The multi-segment adjustable flexible bed board is divided into a head section bed board, a trunk section bed board, a left side turning bed board, and a right side turning bed board. Each section of the bed board can be independently adjusted in tilt. The bed board drive unit is connected to the multi-segment adjustable flexible bed board and is used to drive each section of the bed board to adjust its tilt. This embodiment can accurately adapt to the clinical positioning requirements for neonatal anti-reflux, unlike the defects of the one-piece bed board that cannot take into account both head height and side turning.

[0054] The multimodal sensing unit includes a distributed flexible pressure sensor array and a dual-channel 24GHz micro millimeter-wave sensor module. The distributed flexible pressure sensor array is embedded in the flexible mattress and is used to collect the newborn's body position distribution, chest and abdominal respiratory movements, abdominal pressure changes, and body movement signals. The dual-channel 24GHz micro millimeter-wave sensor module is symmetrically installed on both sides of the bed frame and headboard to collect the newborn's oral and nasal micro-flow respiratory signals and chest and abdominal displacement respiratory wave signals. In this embodiment, the pressure sensor captures chest and abdominal mechanical movements, body position, and abdominal pressure changes, while the millimeter-wave sensor penetrates the blanket to capture the oral and nasal micro-breathing airflow. The two signals verify each other, and the core vital signs can be collected without contacting the newborn's skin.

[0055] The main control unit is electrically connected to the multimodal sensing unit, the bed board drive unit, and the early warning unit. It is used to extract the early signs of neonatal gastroesophageal reflux based on the signals collected by the multimodal sensing unit, identify the reflux risk level, and control the early warning unit to output the corresponding early warning signal and control the bed board drive unit to adjust the posture of the multi-segment adjustable flexible bed board according to the risk level, thus deeply coupling the monitoring function with the intervention function.

[0056] like Figure 2 As shown, the method for dynamic identification of abnormal vital signs in newborns according to Embodiment 1 of this application includes the following steps:

[0057] S1 signal synchronous acquisition: Through the same clock source of the main control unit, the distributed flexible pressure sensing array and the dual 24GHz micro millimeter wave sensing module are controlled to synchronously acquire signals and complete the hard time alignment of the two signals. The hard time alignment using the same clock source ensures that the time axis error of the pressure sensing and millimeter wave sensing signals is ≤1ms, avoiding feature extraction deviation due to signal timing misalignment, and ensuring accurate calculation of the linkage features of chest and abdominal movement and respiratory airflow.

[0058] S2 signal preprocessing: The two acquired signals are filtered and denoised, decomposed into frequency bands and removed from invalid windows to obtain valid vital signs signals. The valid signals are purified through preprocessing, and invalid data caused by body movement and environmental interference are removed to prevent invalid signals from entering the subsequent recognition process.

[0059] S3 Precursor Feature Extraction: Extract chest and abdominal movement linkage deviation features, oral and nasal microflow respiratory abnormality features, and postural-abdominal pressure abnormality features from effective vital signs signals to construct a gastroesophageal reflux precursor feature set. Specifically extract reflux precursor features, rather than post-emergence features after spitting up, to capture the pathophysiological changes 10-30 seconds before reflux occurs. Then, complete the risk classification through a pre-trained model and finally match the corresponding intervention action.

[0060] S4 Risk Classification Identification: Input the feature set of gastroesophageal reflux precursors into a pre-trained neonatal reflux risk classification model and output the corresponding reflux risk level;

[0061] S5 graded linkage intervention: Trigger the corresponding level of warning based on the risk level of reflux, and at the same time control the multi-segment adjustable flexible bed board to adjust to the corresponding posture. After the intervention is completed, continuously monitor the changes in vital signs, and control the bed board to return to its original position after the vital signs return to normal.

[0062] like Figure 1 As shown:

[0063] The bed board drive unit uses a medical-grade silent stepper motor, adjusting the single-segment bed board at a speed of 1° / s. This prevents rapid bed movements from startling the newborn and avoids rapid changes in position that could cause the newborn to move and interfere with the acquisition of vital signs. The head section of the bed board can be raised within a range of 0-15 degrees, while the left and right side sections can be raised within a range of 0-20 degrees. This avoids side effects such as the newborn slipping, increased intracranial pressure, and spinal developmental abnormalities caused by excessive elevation. Furthermore, both sides of the bed board must not be raised simultaneously to prevent the resulting cramped position from compressing the newborn's chest and abdomen, affecting normal breathing, and eliminating position-related safety risks.

[0064] like Figure 1 As shown:

[0065] The distributed flexible pressure sensor array is divided into four acquisition zones: head, chest, abdomen, and pelvis. The sensor point density in the chest and abdomen zones is twice that of the head and pelvis zones. This higher sensor density can accurately capture minute displacement changes in chest and abdominal breathing, avoiding the omission of core features. The sensor array has a range of 0-10 kPa and a sampling rate of 100 Hz, which fully covers the respiratory frequency of newborns (40-60 breaths / min). It can completely preserve the effective features of the respiratory signal and avoid signal distortion caused by undersampling. The dual-channel 24 GHz micro millimeter-wave sensor module has a sampling rate of 200 Hz and a distance resolution of no less than 1 mm. It can accurately capture the micro-flow respiratory airflow in the newborn's mouth and nose and the micron-level respiratory displacement in the chest and abdomen, and capture minute respiratory pattern abnormalities caused by the precursor of regurgitation.

[0066] like Figure 1 As shown:

[0067] The main control unit uses an STM32H7 series MCU paired with an ESP32-S3 AI acceleration unit. The MCU is responsible for bed driving, signal acquisition and peripheral control, while the AI ​​acceleration unit is responsible for lightweight model calculation and feature extraction. All signal processing and recognition calculations are completed on the local embedded end of the bed. The main control unit is equipped with RS485 and Ethernet communication interfaces for connecting to the hospital's central monitoring system at the nursing station.

[0068] like Figure 1 As shown:

[0069] The early warning unit adopts a three-level early warning mechanism. When the risk is low, a local silent soft light LED prompt is triggered on the bed. When the risk is medium, a pop-up prompt is triggered on the central monitoring system at the nurse station. When the risk is high, a low-decibel audible and visual alarm is triggered simultaneously on the bed, an emergency audible and visual alarm at the nurse station, and an early warning push notification on the medical staff's handheld terminal. The corresponding graded early warning mode means that the lower the risk level, the smaller the warning range and the lower the noise. Only high-risk events trigger audible and visual alarms across all channels to avoid unnecessary alarm noise. Low-risk events only use a local silent soft light prompt on the bed, which is silent and flicker-free, will not stimulate the newborn's retina, and will not produce noise pollution, which is in line with the protection requirements for the neurological development of premature infants in the NICU. Medium-risk events only push a pop-up prompt to the nurse station and do not trigger the bed's audible and visual alarms, which ensures that medical staff are aware of the risk without causing noise in the NICU environment and avoiding alarm fatigue caused by frequent alarms.

[0070] like Figure 1 As shown:

[0071] The specific method of signal preprocessing in step S2 is as follows: the pressure sensing signal is filtered by sliding window mid-range filtering to remove baseline drift, and 50Hz power frequency interference is filtered out by adaptive notch filter. Then, a third-order Butterworth filter is used to decompose it into low-frequency postural abdominal pressure component, mid-frequency respiratory motion component and high-frequency body motion interference component. When the high-frequency body motion interference component exceeds the preset threshold, the current window is marked as an invalid window and feature extraction is paused. The pressure sensing signal is decomposed by frequency band filtering to split the mixed signal into three independent components: postural abdominal pressure, respiratory motion and body motion interference. This can accurately separate the body motion interference caused by neonatal startle reflex, crying and nursing operations. When the body motion component exceeds the standard, the invalid window is marked to avoid interference signals from entering the subsequent recognition process.

[0072] The distance gate between the newborn's mouth / nose and chest / abdomen regions is locked using FFT transform on the millimeter-wave sensor signal. After phase demodulation to extract the effective respiratory signal, noise reduction is performed using the same filtering rules as the pressure sensor signal, and the time axis is aligned with the pressure sensor signal. The distance gate for the millimeter-wave sensor signal is locked using FFT transform, which can filter static and dynamic interference such as bed structure, movement of medical staff, and environmental objects, retaining only the effective signals from the newborn's mouth / nose and chest / abdomen regions. The two signals are aligned with the time axis using the same filtering rules to ensure that the timing of the two signals is completely matched during subsequent feature extraction, avoiding feature calculation errors caused by timing misalignment.

[0073] like Figure 2 As shown:

[0074] In step S3, the characteristics of chest-abdominal motor linkage deviation include the mean phase difference of chest-abdominal respiratory waves, the mean linkage coefficient, and the linkage variability. The characteristics of abnormal micro-flow breathing in the oropharynx include the mean inspiratory-expiratory ratio, the peak-to-peak variability, and the proportion of micro-breathing pauses. The characteristics of abnormal body position-abdominal pressure include the trunk-head tilt angle difference, the magnitude of abdominal pressure increase, and the abdominal pressure variability. Among these, chest-abdominal motor linkage deviation is the earliest signal of reflux stimulating the vagus nerve, appearing 15-30 seconds earlier than the regurgitation event. Oral-nasal breathing abnormality is a characteristic of reflux stimulating the pharynx, appearing 10-20 seconds earlier than the regurgitation event. Abnormal body position-abdominal pressure is the core pre-inducing factor of reflux. The three characteristics verify each other, forming a complete system of prodromal features. All features are calculated based on effective vital signs signals from 5 consecutive respiratory cycles, adapted to the respiratory rate of 40-60 breaths / min in newborns. This ensures sufficient sample size to avoid occasional errors and avoids warning lag due to excessively long sampling cycles.

[0075] like Figure 2 As shown:

[0076] The specific rules for the graded linkage intervention in step S5 are as follows: under low-risk conditions, only the corresponding warning is triggered and the characteristic changes are continuously monitored, without adjusting the bed board posture;

[0077] In the medium-risk state, simultaneously raise the head section of the bed board to 10° and the right side of the bed board to 12°. After adjustment, continuously monitor for 3 respiratory cycles. If the characteristics return to normal, maintain the current position for 30 minutes and then automatically reset.

[0078] In high-risk situations, the highest level of warning is triggered simultaneously, raising the head section of the bed board to 15° and the right side of the bed board to 15°, continuously locking and monitoring vital signs data until medical staff manually take over or the vital signs return to normal.

[0079] In the above process, for low-risk cases, only monitoring and no intervention are required to avoid unnecessary changes in body position that could cause stress to the newborn. For medium-risk cases, the clinically recommended 10° head elevation and 12° right lateral decubitus position are used, which can effectively promote gastric emptying and block the progression of reflux without causing position-related side effects. For high-risk cases, the maximum anti-reflux position within the safety threshold is used, and an emergency alarm is triggered to minimize the risk of aspiration.

[0080] Example 2, the following is a specific case for illustration:

[0081] This embodiment is applied in the neonatal intensive care unit of a hospital, and the monitored subjects are premature infants with a corrected gestational age of 32 weeks, a birth weight of 1.8 kg, and a clinical history of frequent physiological gastroesophageal reflux, requiring 24-hour continuous monitoring:

[0082] The hardware in this embodiment uses a 4-segment adjustable flexible bed board, with the head section bed board being 300mm long, the torso section bed board being 400mm long, and the width of each left / right folding bed board segment being 200mm. The drive unit uses four medical-grade silent 42-stepper motors with a microstepping accuracy of 16, and the adjustment speed of each bed board segment is fixed at 1° / s. It is equipped with mechanical hard limits and software dual protection, with the head section bed board raising range of 0-15° and the side folding bed board raising range of 0-20° on one side, prohibiting the simultaneous raising of both side folding bed boards. It is equipped with a 12V 10Ah medical backup lithium battery, which can maintain full-function operation of the system for ≥6 hours after a power outage.

[0083] The distributed flexible pressure sensor array uses a medical-grade capacitive flexible sensing film embedded in a 50mm thick medical silicone mattress. It is fixedly divided into 4 acquisition zones: head zone (8×4 sensing points), chest zone (12×8 sensing points), abdomen zone (12×8 sensing points), and pelvic zone (8×4 sensing points). The measurement range is 0-10kPa, the resolution is 0.5Pa, and the sampling rate is 100Hz.

[0084] Dual-channel 24GHz millimeter-wave sensor module: The Texas Instruments IWR1443 single-chip millimeter-wave radar is used. The two modules are symmetrically installed on both sides of the bed rails at the head of the bed. The installation height is 600mm from the mattress plane and tilted downwards at 15°, aiming at the newborn's mouth, nose and chest and abdomen area. The sampling rate is 200Hz and the distance resolution is 1mm. The dual-channel differential acquisition mode is used to eliminate environmental static interference.

[0085] The main control unit uses an STM32H743IIT6 MCU paired with an ESP32-S3 AI acceleration unit. The MCU is responsible for signal acquisition, motor drive, and peripheral control, while the ESP32-S3 is responsible for feature extraction and model inference. All calculations are completed locally on the bed, without cloud dependency. It is equipped with RS485 and Ethernet standard interfaces to directly connect to the NICU central monitoring system. The early warning unit adopts a three-level fixed mechanism: low risk triggers a soft warm light LED prompt at the bedside; medium risk triggers a pop-up prompt from the central monitoring system; and high risk simultaneously triggers a 55dB low-decibel audible and visual alarm on the bedside, an emergency alarm from the central monitoring system, and an early warning push notification from the medical staff's handheld PDA.

[0086] This embodiment uses a fixed 10-second acquisition window (covering approximately 8 respiratory cycles of a newborn, adapted to their physiological respiratory rate of 40-60 breaths / min), and the entire process is executed cyclically according to a fixed time sequence. The specific steps are as follows:

[0087] 1) Using the same 10MHz clock source from the main control unit, the pressure sensing array and millimeter-wave module are triggered to synchronously acquire data. Hard time alignment is performed on the two signals, with a time alignment error ≤1ms. The alignment formula is as follows:

[0088]

[0089] In the formula: The aligned millimeter-wave signal sequence, This is the original millimeter-wave signal sequence. The time difference between the two signals is calculated by cross-correlation of the peak values ​​of the respiratory waves in the two signals.

[0090] 2) A 50-point sliding window mid-range filter is used to remove signal baseline drift, and a 50Hz adaptive notch filter is used to filter out power frequency interference. The filter transfer function is:

[0091]

[0092] In the formula: It is the power frequency angular frequency. For pressure sensing sampling rate, The radius of the filter poles;

[0093] A third-order Butterworth filter is used to decompose the signal into frequency bands: low-frequency components below 0.1Hz are used to extract body position and abdominal pressure baseline, mid-frequency components from 0.2-2Hz are used to extract chest and abdominal respiratory waves, and high-frequency components above 2Hz are used to extract body motion interference; when the peak value of the high-frequency component exceeds three times the peak value of the mid-frequency component, the current window is marked as an invalid window and feature extraction is paused.

[0094] A 256-point FFT transformation was performed on the dual-channel millimeter-wave intermediate frequency signals to extract distance dimension information. Fixed distance gates were locked for the newborn's mouth and nose (distance gate 3), chest region (distance gate 5), and abdominal region (distance gate 6) to filter out bed and environmental interference. Micro-flow respiratory signals from the mouth and nose and chest and abdominal displacement respiratory waves were extracted through phase demodulation. After denoising using the same filtering rules as the pressure sensing signals, the time axis was aligned with the pressure signals.

[0095] 3) For the preprocessed effective signal, based on 5 consecutive respiratory cycles, 3 core features are extracted, with no redundant features:

[0096] Chest wall respiratory wave sequences were extracted from pressure sensor signals in the chest and abdominal regions, respectively. Abdominal wall respiratory wave sequence Calculate the normalized cross-correlation function of the two to obtain the chest-abdomen linkage coefficient. :

[0097]

[0098]

[0099] In the formula This is the amount of time delay. The number of sampling points per respiratory cycle; The closer it is to 1, the better the synchronization of chest and abdominal movements;

[0100] Simultaneous calculation of the phase difference corresponding to the cross-correlation peak And the mean phase difference and the correlation variability over five consecutive periods (variability = standard deviation / mean);

[0101] Identifying the inspiratory time of a single respiratory cycle from millimeter-wave oral and nasal respiratory signals. ;

[0102] Calculate the peak value variability of respiratory peak :

[0103]

[0104] In the formula The standard deviation of the peak values ​​over five consecutive respiratory cycles. This is the average of the peak-to-peak values;

[0105] The percentage of micro-breathing pauses (breathing signal amplitude below baseline and duration ≥2s) was simultaneously counted.

[0106] The trunk-head tilt difference is calculated by taking the average static pressure values ​​of the head and pelvic regions. :

[0107]

[0108] In the formula The average pressure in the pelvic region. The average pressure in the head region, The average density of a newborn's body. It is the acceleration due to gravity. The length of a newborn's body from head to pelvis;

[0109] Simultaneously calculate the continuous increase in baseline abdominal pressure and the variability of abdominal pressure.

[0110] 4) Input the above feature set into the pre-trained lightweight LightGBM binary classification model. The model is trained based on clinical reflux labeled data of 1200 newborns with different corrected gestational ages. In this embodiment, the model weights are adapted to a corrected gestational age of 32 weeks, and the output is a fixed three-level risk level:

[0111] Low risk: Only one abnormal characteristic, lasting for <5 respiratory cycles, with a regurgitation probability of <30%;

[0112] Medium risk: Abnormalities in two core features: chest-abdominal coordination and abnormal breathing, lasting for 5-10 respiratory cycles, with a regurgitation probability of 30%-70%;

[0113] High risk: All three characteristics are abnormal, lasting for more than 10 respiratory cycles, and the probability of regurgitation is greater than 70%.

[0114] 5) During continuous monitoring in this embodiment, the model identified a medium-risk state: the chest-abdomen linkage coefficient dropped to 0.42, the phase difference reached 72°, and the inspiratory-to-expiratory ratio dropped to 1:2.8. The fixed medium-risk intervention procedure was immediately executed.

[0115] The headboard is raised to 10° at a speed of 1° / s, and the right-side tilting board is raised to 12° simultaneously. During the adjustment process, adaptive filtering is activated to eliminate motion interference, and monitoring and feature extraction are not interrupted.

[0116] After the adjustment was completed, three respiratory cycles were continuously monitored. The chest-abdomen linkage coefficient recovered to 0.85, the inspiratory-to-expiratory ratio recovered to 1:1.8, and the characteristics returned to the normal range.

[0117] After maintaining the current position for 30 minutes, the bed board will automatically return to the supine position.

[0118] In summary, as Figures 3-5As shown, according to the neonatal intelligent monitoring bed and dynamic identification method for abnormal vital signs in Embodiment 2 of this application, during a continuous 72-hour monitoring process, a total of 12 reflux precursor events were identified. Warnings and interventions were completed 15-28 seconds in advance for all of them. No regurgitation or aspiration events occurred after any intervention. The accuracy rate of identifying pathological reflux precursors was 96.2%, the false alarm rate was 2.1%, and the reflux relief rate of intervention in medium-risk positions was 100%, which fully meets the safety and performance requirements of NICU clinical neonatal monitoring.

[0119] In the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0120] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0121] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A neonatal intelligent monitoring bed, comprising a bed frame, a flexible mattress, a bed board drive unit, a main control unit, and an early warning unit, characterized in that, It is equipped with a multi-segment adjustable flexible bed board and a multi-modal sensing unit; The multi-segment adjustable flexible bed board is divided into a head section bed board, a torso section bed board, a left folding bed board, and a right folding bed board. Each section bed board can be independently adjusted in tilt angle. The bed board drive unit is connected to the multi-segment adjustable flexible bed board for driving each section bed board to adjust its tilt angle. The multimodal sensing unit includes a distributed flexible pressure sensing array and a dual-channel 24GHz micro millimeter-wave sensing module. The distributed flexible pressure sensing array is embedded in the flexible mattress and is used to collect the newborn's body position distribution, chest and abdominal respiratory movements, abdominal pressure changes and body movement signals. The dual-channel 24GHz micro millimeter-wave sensing module is symmetrically installed on both sides of the bed frame headboard and is used to collect the newborn's oral and nasal cavity micro-flow respiratory signals and chest and abdominal displacement respiratory wave signals. The main control unit is electrically connected to the multimodal sensing unit, the bed board driving unit, and the early warning unit, respectively. It is used to extract the early signs of neonatal gastroesophageal reflux based on the signals collected by the multimodal sensing unit, identify the reflux risk level, and control the early warning unit to output the corresponding early warning signal and control the bed board driving unit to adjust the posture of the multi-segment adjustable flexible bed board according to the risk level.

2. The method for dynamic identification of abnormal vital signs in newborns according to claim 1, characterized in that, Includes the following steps: S1 signal synchronous acquisition: Through the same clock source of the main control unit, the distributed flexible pressure sensing array and the dual 24GHz micro millimeter wave sensing module are controlled to synchronously acquire signals and complete the hard time alignment of the two signals. S2 signal preprocessing: The two acquired signals are filtered and denoised, decomposed into frequency bands, and invalid windows are removed to obtain effective vital signs signals; S3 Precursor Feature Extraction: Extract chest and abdominal movement linkage deviation features, oral and nasal microflow respiratory abnormality features, and body position-abdominal pressure abnormality features from effective vital signs signals to construct a gastroesophageal reflux precursor feature set; S4 Risk Classification Identification: Input the feature set of gastroesophageal reflux precursors into a pre-trained neonatal reflux risk classification model and output the corresponding reflux risk level; S5 graded linkage intervention: Trigger the corresponding level of warning based on the risk level of reflux, and at the same time control the multi-segment adjustable flexible bed board to adjust to the corresponding posture. After the intervention is completed, continuously monitor the changes in vital signs, and control the bed board to return to its original position after the vital signs return to normal.

3. The intelligent neonatal monitoring bed according to claim 1, characterized in that, The bed board drive unit uses a medical silent stepper motor, and the adjustment speed of a single bed board is 1° / s; the lifting range of the head section bed board is 0-15°, and the lifting range of the left and right side flipping bed boards is 0-20°.

4. The intelligent neonatal monitoring bed according to claim 1, characterized in that, The distributed flexible pressure sensing array is divided into four acquisition zones: head zone, chest zone, abdomen zone, and pelvic zone. The sensing point density in the chest and abdomen zones is twice that in the head and pelvic zones. The range of the sensing array is 0-10 kPa, the sampling rate is 100 Hz, the sampling rate of the dual 24 GHz micro millimeter wave sensing module is 200 Hz, and the distance resolution is not less than 1 mm.

5. The intelligent neonatal monitoring bed according to claim 1, characterized in that, The main control unit uses an STM32H7 series MCU paired with an ESP32-S3AI acceleration unit. All signal processing and recognition operations are completed on the local embedded end of the bed. The main control unit is equipped with RS485 and Ethernet communication interfaces for connecting to the hospital's central monitoring system at the nursing station.

6. The intelligent neonatal monitoring bed according to claim 1, characterized in that, The warning unit adopts a three-level warning mechanism. When the risk is low, it triggers a local silent soft light LED prompt on the bed. When the risk is medium, it triggers a pop-up prompt from the central monitoring system at the nurse station. When the risk is high, it simultaneously triggers a low-decibel sound and light alarm on the bed, an emergency sound and light alarm at the nurse station, and a warning push notification from the medical staff's handheld terminal.

7. The method for dynamic identification of abnormal vital signs in newborns according to claim 2, characterized in that, The specific method of signal preprocessing in step S2 is as follows: the pressure sensing signal is filtered by sliding window mid-range filtering to remove baseline drift, the 50Hz power frequency interference is filtered out by adaptive notch filter, and then the third-order Butterworth filter is used to decompose it into low-frequency body position abdominal pressure component, mid-frequency respiratory motion component and high-frequency body motion interference component. When the high-frequency body motion interference component exceeds the preset threshold, the current window is marked as an invalid window and feature extraction is paused. The distance gate between the newborn's mouth and nose and chest and abdomen regions is locked by FFT transformation of the millimeter wave sensing signal. After the effective respiratory signal is extracted by phase demodulation, the same filtering rule as the pressure sensing signal is used to complete the noise reduction and the time axis is aligned with the pressure sensing signal.

8. The method for dynamic identification of abnormal vital signs in newborns according to claim 2, characterized in that, In step S3, the characteristics of chest and abdominal movement linkage deviation include the mean phase difference of chest and abdominal respiratory waves, the mean linkage coefficient, and the linkage variability. The characteristics of micro-flow respiratory abnormalities in the oral and nasal cavities include the mean inspiratory-expiratory ratio, the peak-to-peak value variability, and the proportion of microbreathing pauses. The characteristics of abnormal body position-abdominal pressure include the trunk-head tilt angle difference, the increase in abdominal pressure, and the variability of abdominal pressure. All characteristics are calculated based on the effective vital signs signals of five consecutive respiratory cycles.

9. The method for dynamic identification of abnormal vital signs in newborns according to claim 2, characterized in that, The specific rules for the graded linkage intervention in step S5 are as follows: under low-risk conditions, only the corresponding warning is triggered and the characteristic changes are continuously monitored, without adjusting the bed board posture; In the medium-risk state, simultaneously raise the head section of the bed board to 10° and the right side of the bed board to 12°. After adjustment, continuously monitor for 3 respiratory cycles. If the characteristics return to normal, maintain the current position for 30 minutes and then automatically reset. In high-risk situations, the highest level of warning is triggered simultaneously, raising the head section of the bed board to 15° and the right side of the bed board to 15°, continuously monitoring vital signs data until medical staff manually take over or the vital signs return to normal.

Images