EP4704686A2 - Contactless continuous blood pressure estimation - Google Patents

Abstract

The present disclosure relates to a contactless system for monitoring a blood pressure of a subject (e.g., human patient, non-human patient, etc.) based on vibration signals of a structure supporting the subject. The system can include a sensor coupled to the structure. A computing device in data communication with the sensor can obtain realtime sensor data from the sensor, the real-time sensor data may include seismic signal segments recorded at a predetermined sampling rate. The computing device can further filter the sensor data to remove seismic signal segments not related to vital activities and analyze the filtered sensor data to determine continuous and real-time blood pressure measurements of the subject, which include the estimated systolic blood pressure values and estimated diastolic blood pressure values. A user interface can display of at least the blood pressure measurements.

Description

CONTACTLESS CONTINUOUS BLOOD PRESSURE ESTIMATION

CROSS REFERENCE TO RELATED CASES

[0001] This application claims the benefit of and priority to U.S. Patent Application No. 63/500,403 filed on 5 May 2023, entitled “CONTACTLESS CONTINUOUS BLOOD PRESSURE ESTIMATION,” the contents of which are incorporated by reference in their entirety herein.

BACKGROUND

[0002] Several studies highlight the strong links between systolic blood pressure (SBP), diastolic blood pressure (DBP), sleep, aging, and the prevalence of cardiovascular and vascular diseases. As of 2016, there were approximately 47.8 million individuals 65 years and older in the United States, with 26% living alone at home and 18% residing in senior healthcare facilities according to the U.S. Census Bureau. The growing aging population will only exacerbate the impact of uncontrolled hypertension on the health and well-being of society, as well as contribute to skyrocketing healthcare costs.

BRIEF DESCRIPTION OF THE DRAWINGS

[0003] The foregoing will be apparent from the following more particular description of example embodiments of the present disclosure, as illustrated in the accompanying drawings in which like reference characters refer to the same parts throughout the different views. The drawings are not necessarily to scale, emphasis instead being placed upon illustrating embodiments of the present disclosure.

[0004] FIG. 1 illustrates a network environment according to various embodiments of the present disclosure. [0005] FIG. 2 illustrates a flowchart illustrating an example of functionality implemented as portions of an application executed in a computing environment in the network environment of FIG. 1 according to various embodiments of the present disclosure.

[0006] FIG. 3 illustrates a user display of vital signs being monitored according to various embodiments of the present disclosure.

[0007] FIG. 4 illustrates a data structure of seismic signals and blood pressure according to various embodiments of the present disclosure.

[0008] FIG. 5 illustrates recorded vibrational data according to various embodiments of the present disclosure.

[0009] FIG. 6 illustrates a flow chart of the blood pressure estimation workflow according to various embodiments of the present disclosure.

[0010] FIG. 7 illustrates an architecture of a Convolutional Neural Network - Long Short Term Memory (CNN-LSTM) hybrid model according to various embodiments of the present disclosure.

[0011] FIGS. 8A and 8B illustrate histograms for SBP (FIG. 8A) and DBP (FIG. 8B) according to various embodiments of the present disclosure.

[0012] FIGS. 9A and 9B illustrate error histograms for SBP (FIG. 9A) and DBP (FIG. 9B) according to various embodiments of the present disclosure.

[0013] FIGS. 10A and 10B illustrate Bland Altman plots for the CNN-LSTM hybrid model and the Vital Temporal Convolutional Networks (VTCN) model, respectively, for SBP (FIG. 10A) and DBP (FIG. 10B) according to various embodiments of the present disclosure.

[0014] FIGS. 11A and 11 B illustrate regression plots for the CNN-LSTM hybrid model and the VTCN model, respectively, for SBP (FIG. 11 A) and DBP (FIG. 11 B) according to various embodiments of the present disclosure.

[0015] FIG. 12 illustrates convergence curves for the CNN-LSTM hybrid model on both the training and validation sets for SBP (top) and DBP (bottom) according to various embodiments of the present disclosure.

[0016] FIG. 13 illustrates convergence curves for the VTCN model on both the training and validation sets for SBP (top) and DBP (bottom) according to various embodiments of the present disclosure.

DETAILED DESCRIPTION

[0017] The present disclosure relates to estimating blood pressure of subjects (e.g., human patients, non-human patients, such as, animals, etc.), in real-time, during sleep cycles using vibration signals from a structure supporting the subject (e.g., a bed, a chair, etc.). Monitoring blood pressure during sleep is important for those suffering from hypertension related diseases such as heart attacks, strokes, heart failure, and kidney disease. However, standard sleep clinic technology, polysomnography, is both costly and inaccessible to families. Most commercial alternatives, such as wearable sensors and wrist devices, require bodily contact and can be intrusive. The contact-free and nonintrusive solution to monitor blood pressure disclosed herein provides a bed-mounted continuous blood pressure monitoring sensor system for contact-free continuous estimating blood pressure and assessing cardiovascular health of subjects including, human patients and non-human patients (e.g., animals). The contact-free and bedmounted continuous blood pressure monitoring system of the present disclosure eliminates the need for external wearable devices and physical contact, while avoiding privacy or radiation concerns associated with other technologies such as cameras or radars.

[0018] Most previous sleep monitoring systems utilize instrumented sensors installed in the mattress or bed frame to assess body behaviors during sleep. For example, ultrasound transmitters and receivers were used in to characterize mattress movements, while force sensors installed on bed legs were used in to classify body movements. However, these systems do not focus on monitoring vital signs such as heart rate, respiration rate, and blood pressure. Load cells placed at bed comers were used in to detect body movements, but their system is only suitable for short-term analysis and does not detect specific sleep postures. Infrared sensor arrays were used in to analyze body vibration and respiratory activities, while conductive fiber sensors were used in to detect body position, heart rate, and respiration rate. However, these devices are expensive and not practical for home use. For example, in one study, a wearable skin-like optoelectronic system for cuff-less continuous blood pressure monitoring was introduced with a model using virtual work to connect pulse transit time and continuous non-invasive arterial pressure. In another study, a wireless, skin-interfaced devices for pediatric critical care was developed with a proposed combination of linear form and non-linear inverse square relation to estimating blood pressure by heart rate and pulse arrival time. Moreover, the healthcare market is among the rapidly growing markets for Wi-Fi and other wireless LAN technologies. Wireless vital sign monitors such as Vital-Jacket® and Vital Signs Systems were developed in recent years to monitor the health and well-being of people, especially for those patients suffering from chronic diseases. Moreover, continuous monitoring works such as pressure sensors in the mattress and temperature sensors were used in two studies, but none of these solutions is suitable for affordable and non-intrusive homebased sleep monitoring and vital sign estimation. One real-time and contact-free sensor system developed an envelope-based estimation method for heart rate and respiratory rate but without blood pressure estimations.

[0019] Seismic signal-based sleep monitoring has gained popularity due to its contact- free nature. Previous studies have utilized seismic signals to monitor heart rate during sleep for one or two subjects. More recently, estimated both heart rate and respiration rate using a vertical vibration sensor. Biomedical micro-vibrations are typically measured through ballistocardiogram (BCG) and SCG. BCG reflects the cardiac ejection of blood into the vasculature while SCG represents the local vibrations of the chest wall resulting from vital sign changes. BCG is related to heart movements, while SCG waveform is a combination of the mechanical pulse response of the vasculature and body to the cardiac ejection of blood. While both SCG and BCG signals contain information for blood pressure estimation, SCG is more appropriate for real-time monitoring as it describes the micro chest accelerations, whereas BCG represents mass displacements.

[0020] In this context, the concepts described herein are directed to methods and systems for contact-free blood pressure monitoring via bed-mounted sensors to detect micro-vibrations of a subject (e.g., human patients, non-human patients, such as, animals, etc.), on the bed or similar structure. According to various embodiments, the contact-free and bed-mounted continuous blood pressure monitoring system of the present disclosure comprises a seismometer sensor in data communication with a computing device. The seismometer sensor can be attached to a structure (e.g., bed frame) that is associated with a subject. The seismometer sensor is attached to the structure and is not in direct contact with the subject. The computing device is configured to obtain seismic data from the seismometer sensor and analyze the seismic data to identify characteristics of a subject. The characteristics can include blood pressure (both SBP and DBP), as well as, heart rate, respiratory rate, and/or other features.

[0021] FIG. 1 illustrates an example of the contact-free blood pressure monitoring system 100 system according to various embodiments of the present disclosure. As described herein, one or more sensors 103 can be attached to a structure 106, such as a bed frame, which is non-intrusive and not in contact with any portion of a body of a subject being monitored. A monitoring device 109 is in data communication with the sensor 103 (e.g., seismometer) for real-time data collection and/or processing. In some embodiments, the monitoring device 109 can be in data communication with the sensor 103 directly or via a communication network 113. In some embodiments, the sensor 103 and/or monitoring device 109 can also be in data communication with a computing environment 116 and/or a client device 119.

[0022] The system 100 of the present disclosure is designed to continuously monitor micro-vibrations via one or more sensors 103 on a structure 106, such as a bed frame, as seismic signals are detected by an attached sensor 103. In particular, FIG. 1 illustrates a sensor 103 attached to structure 106 and coupled to a monitoring device 109. In various embodiments, the sensor 103 can comprise a seismometer or other type of device capable of detecting micro-vibrations or seismic signals. FIG. 1 illustrates a nonlimiting example showing the structure 106 as bed, with the sensor 103 coupled to the frame of the bed. For example, FIG. 1 illustrates an example of the sensor 103 mounted to the underside of the structure 106 or bed. In another example, the sensor 103 can be mounted on another portion of the structure 106, such as, for example, the top side of the bed or other type of structure 106. In some embodiments, the monitoring device 109 can be in data communication with the sensor 103 directly or via a communication network 113. In some embodiments, the monitoring device 109 comprises the sensor 103.

[0023] As illustrated in FIG. 1 , the contact-free blood pressure monitoring system 100 system can include a computing environment 116 in data communication with the sensor 103 and/or monitoring device 109 via a communication network 113. The monitoring device 109 can include a communication interface (not shown) that allows the monitoring device 109 to communicatively couple with other communication devices. The communication interfaces may include one or more wireless connection(s) such as, e.g., Bluetooth or other radio frequency (RF) connection and/or one or more wired connection(s). In some examples, the monitoring device 109 further can be data communication with one or more computing devices over a network 113. For example, as shown in FIG. 1 , the monitoring device 109 can be in data communication with a computing environment 116 and one or more client devices 119.

[0024] The monitoring device 109 comprises at least one computing device comprising a processor and a memory. A sensor monitoring application 123 may be executed by the processor in the monitoring device 109 according to various embodiments. Also, various data is stored in a data store 126 that is accessible to the monitoring device 109. The data store 126 may be representative of a plurality of data stores 126 as can be appreciated. The data stored in the data store 126 for example, is associated with the operation of the various applications and/or functional entities described below. For example, the seismic sensor data 129 received from the sensors 103 can be stored in the data store 126. In various embodiments, the monitoring device 109 can comprise the sensor 103. In various embodiments, the monitoring device 109 can comprise an analog to digital converter to sample the seismic signals detected by the sensor 103 at a predetermined sampling rate. In various embodiments, the monitoring device 109 can extract high-frequency time series data from the seismic signals. In various embodiments, the monitoring device 109 can comprise additional applications, executable by the processor, to further analyze the seismic data collected from the one or more sensors 103. For example, the monitoring device 109 can include one or more applications, when executed, analyze the sensor data 129 to determine continuous and real-time measurements of blood pressure, heart rate, respiratory rate, movement, posture, and fall off bed of a subject positioned on the bed.

[0025] In various examples, the computing environment 116 can comprise, for example, a server computer or any other system providing computing capability. Alternatively, the computing environment 116 may employ a plurality of computing devices that may be arranged, for example, in one or more server banks or computer banks or other arrangements. Such computing devices may be located in a single installation or may be distributed among many different geographical locations. For example, the computing environment 116 may include a plurality of computing devices that together may comprise a hosted computing resource, a grid computing resource and/or any other distributed computing arrangement. In some cases, the computing environment 116 may correspond to an elastic computing resource where the allotted capacity of processing, network, storage, or other computing-related resources may vary over time.

[0026] Various applications and/or other functionality may be executed in the computing environment 116 according to various embodiments. For example, a blood pressure estimator 133 may be executed by the processor in the computing environment according to various embodiments. Also, various data is stored in a data store 136 that is accessible to the computing environment 116. The data store 136 may be representative of a plurality of data stores 136 as can be appreciated. The data stored in the data store 136, for example seismic data 129, blood pressure data 139, and patient data 143, is associated with the operation of the various applications and/or functional entities described below. For example, the seismic data 129 obtained from the sensors 103 can also be stored in a data store 136 in the computing environment 116.

[0027] The components executed on the computing environment 116, for example, include list of applications, and other applications, services, processes, systems, engines, or functionality not discussed in detail herein. In some examples, the blood pressure estimator 133 comprises a neural network application to estimate the systolic and diastolic blood pressure of the subject being monitored. The neural network application can comprise a convolutional neural network (CNN). In one example, a convolutional neural network (CNN) and bidirectional long short-term memory (LSTM) components can be used to model the temporal dependencies, as described further herein. In another example, a modified Temporal Convolutional Network (TCN) can be used, as described further herein.

[0028] The raw seismic data can be processed with data analytics models of the blood pressure estimator 133 to estimate systolic and diastolic blood pressure. For example, the data analytics models can include CNN-LSTM, TCN models, and/or other types of data analytics models. The data analytics models can be any digital signal processing (DSP) or artificial intelligence (Al) models. For example, the blood pressure estimator 133 can include a decision tree classifier, a gradient boost classifier, a Gaussian naive Bayes classifier, a reinforcement learning algorithm, a logistic regression classifier, a random forest classifier, a decision tree classifier, a multi-layer perceptron classifier, a recurrent neural network, a neural network, a label-specific attention network, an ensemble model, and/or any other type of trained model as can be appreciated.

[0029] In various embodiments, the monitoring device 109 and/or the computing environment 116 can further include applications to analyze the sensor data 129 to determine continuous and real-time measurements of other vital activities, such as heart rate, respiratory rate, movement, posture, and fall off bed of a subject positioned on the bed. Accordingly, the data store 136 can also store patient data 143 comprising data related to at least one of heart rate, respiratory rate, movement, posture, and fall off bed of a subject positioned on the bed, as well as other information regarding the patient (e.g., subject) or patient history.

[0030] The computing environment 116 can be in data communication with one or more client devices 119. The client device 119 is representative of a plurality of client devices that can be coupled to the network 113. The client device 119 can include a processorbased system such as a computer system. Such a computer system can be embodied in the form of a personal computer (e.g., a desktop computer, a laptop computer, or similar device), a mobile computing device (e.g., personal digital assistants, cellular telephones, smartphones, web pads, tablet computer systems, music players, portable game consoles, electronic book readers, and similar devices), media playback devices (e.g., media streaming devices, BluRay® players, digital video disc (DVD) players, set-top boxes, and similar devices), a videogame console, or other devices with like capability. The client device 119 can include one or more displays, such as liquid crystal displays (LCDs), gas plasma-based flat panel displays, organic light emitting diode (OLED) displays, electrophoretic ink (“E-ink”) displays, projectors, or other types of display devices. In some instances, the display can be a component of the client device 119 or can be connected to the client device 119 through a wired or wireless connection.

[0031] The client device 119 can be configured to execute various applications such as a monitoring application 146 or other applications. The monitoring application 146 can be executed in a client device 119 to access network content served up by the computing environment 116 or other servers, thereby rendering a user interface 149 on the display. To this end, the monitoring application 146 can include a browser, a dedicated application, or other executable, and the user interface 149 can include a network page, an application screen, or other user mechanism for obtaining user input. The client device 119 can be configured to execute applications beyond the monitoring application 146 such as email applications, social networking applications, word processors, spreadsheets, or other applications.

[0032] For example, the blood pressure estimator 133 in the computing environment 116 can generate a user interface 149 that can include a display of at least the blood pressure measurements, and render the user interface 149 via a display on the client device 119. In some examples, the real-time monitoring data may be transmitted to the client device 119 and rendered on a user interface 149 associated with the monitoring application 146. Additionally, the blood pressure estimator 133 in the computing environment 116 can transmit the notification signal to the client device 119 to alert the user via the monitoring application 146 that one or more of the vital activities being monitored is determined to be outside a predefined range.

[0033] Referring next to FIG. 2, shown is a flowchart that provides one example of the operation of a portion of the blood pressure estimator 133 according to various embodiments. It is understood that the flowchart of FIG. 2 provides merely an example of the many different types of functional arrangements that may be employed to implement the operation of the portion of the blood pressure estimator 133 as described herein. As an alternative, the flowchart of FIG. 2 may be viewed as depicting an example of elements of a method implemented in the network environment 100 (FIG. 1 ) according to one or more embodiments.

[0034] At box 202, the blood pressure estimator 133 obtains, by a monitoring device 109 in data communication with the sensor 103 coupled to the structure supporting a subject being monitored, real-time sensor data including vibration signals detected by the sensor 103. In various examples, the monitoring device 109 can comprise the sensor 103. In various examples, the monitoring device 109 can be in data communication with the sensor 103 directly or via a network 113. In various embodiments, one or more sensors 103 can be coupled to a structure 106 supporting a subject being monitored. The sensors 103 can detect micro-vibrations on the surface of the structure 106. For example, the sensor 103 can comprise a seismometer or other device to detect microvibrations. In some examples, high-frequency time series data can be extracted from the real-time sensor data comprising seismic signal segments recorded at a predetermined sampling rate.

[0035] At box 204, the blood pressure estimator 133 filters the sensor data to remove seismic signal segments not related to vital activities. In some examples, filtering the sensor data to remove seismic signal segments is based on at least one of a spectral distance, an energy standard deviation, and a standard deviation. In some examples, both the spectral distance and/or energy standard deviation are evaluated to remove white noise. For example, white noise can be filtered by evaluating the spectral distance, and removing seismic signal segments that are less than a spectral distance threshold. In another example, white noise can be filtered by evaluating the energy standard deviation, and removing seismic signal segments that are less than an energy standard deviation threshold. To remove seismic signal segments related to movement, the standard deviation can be evaluated, and seismic signal segments removed that are greater than a standard deviation threshold. Additionally, noise components with isodominant frequencies can be suppressed, for example by using a notch filter after applying a spectrum scanning method to the sensor data.

[0036] At box 206, the blood pressure estimator 133 analyzes the sensor data to determine continuous and real-time blood pressure measurements of the subject. The real-time blood pressure measurements include estimated systolic blood pressure values and estimated diastolic blood pressure values. In some examples, the estimated systolic blood pressure values and the estimated diastolic blood pressure values are based at least in part on a Convolutional Neural Network - Long Short Term Memory (CNN-LSTM) hybrid model. The CNN-LSTM hybrid model can comprise a plurality of temporal convolutional layers, a plurality of max pooling layers, and a plurality of bidirectional LSTM layers, as further described herein. In some examples, the estimated systolic blood pressure values and the estimated diastolic blood pressure values are based at least in part on a modified Temporal Convolutional Networks (TCN) model. The modified TCM model can comprise an additional batch normalization layer and rectified linear unity (ReLU) activation function in each block of the TCN model, and a linear layer for regression. The sensor data can also be further analyzed to determine continuous and real-time measurements of heart rate, respiratory rate, movement, posture, and fall off bed of a subject positioned on the bed. In various examples, the blood pressure estimator 133 can further detect an event based at least in part on at least one of the heart rate, the respiratory rate, or blood pressure of the subject. The event can include at least one of the systolic blood pressure or the diastolic blood pressure being outside a predefined range.

[0037] At box 208, the blood pressure estimator 133 generates a user interface comprising a display of at least the blood pressure measurements. In various examples, the user interface can be rendered via a display of a client device 119. In various examples, the blood pressure estimator 133 can update the user interface to include at least one of the heart rate, the respiratory rate, or blood pressure of the subject. In some examples, the blood pressure estimator 133 generates an alert in response to the detected event. The alert or notification can be at least one of an auditory or visual or vibratory alert. In some examples, at least one computing device is in communication with a smart device configured to communicate with a third party, and generating the alert further comprises instructing the smart device to send a communication with the third party. [0038] According to various embodiments, a contact-free blood pressure monitoring system is used that can continuously monitor sleep and vital signs without the need for sensors or instrumentation on the body of a subject (e.g., human patients, non-human patients, such as, animals, etc.) being monitored. According to various embodiments, a contact-free blood pressure monitoring system uses a seismic sensor mounted on the bed to collect real-time bedseismogram (BSG) signals, which measure micro-vibrations on the bed frame that are induced by heart movement, respiration, and blood pressure changes. This technology of monitoring seismic signals has been widely used in geophysical and civil engineering fields for many years, and more recently, in analyzing living being health conditions. For example, the seismocardiogram (SCG) signals created by chest vibration have been used to extract heartbeat and implement user authentication on mobile phones. Based on a bed-mounted geophone, researchers monitored heart rate during sleep. More recently, a vertical geophone with a 2.5 kHz sampling frequency was used to implement both heart rate and respiration rate monitoring. However, there has been limited research on real-time blood pressure monitoring using SCG signals.

[0039] In various examples, the contact-free blood pressure monitoring system 100 can be easily set up by the sensor 106 attaching it to a bed frame. For example, the system can include a monitoring device 106 comprising of a Raspberry Pi 3 and an application delivery controller (ADC) that connects to the sensor 103 and saves raw data to a local database (e.g., data store 126). In addition to monitoring on/off bed detection, sleep posture, and heart and respiration rates, the contact-free blood pressure monitoring system 100 of the present disclosure is capable of monitoring blood pressure during sleep, making it a unique and effective solution. An example user interface 149 that can be shown to user monitoring the subject is illustrated in FIG. 3. The system disclosed herein provides a cost-effective and contact-free system for blood pressure monitoring during sleep using BSG signals.

[0040] In various examples, the contact-free blood pressure monitoring system 100 of the present disclosure can be powered by advanced Al algorithms that can accurately estimate real-time blood pressure. The system disclosed herein provides two reliable methods for blood pressure estimation, both of which use time-series features extracted from BSG signals. The first method is a convolution-recurrent hybrid neural network model, which leverages convolutional neural network (CNN) components to learn representations from the input high-dimensional time-series features, and bidirectional long short-term memory (LSTM) components to model the temporal dependencies. The second method is called Vital Temporal Convolutional Networks (VTCN) and is based on a modified causal convolutional architecture. Unlike existing vital sign estimation methods, which require a single data analysis window of 30-120 seconds, the contact- free blood pressure monitoring system 100 of the present disclosure continuously monitors changes in SBP/DBP using a seismometer with a 100 Hz sampling frequency and collects the BSG signal caused by bed vibrations, allowing for data segments of 10 seconds. This makes the system more suitable for real-time blood pressure monitoring during sleep and even clinical studies.

[0041] The algorithms disclosed herein are the first successful attempt to estimate continuous blood pressure during sleep without physical contact. In various examples, a sensor 103 of the contact-free blood pressure monitoring system 100 of the present disclosure is used to measure micro-vibrations on a structure 106 that are induced by heart movement, respiration, blood pressure changes, and/or other physiological changes of the subject. With a high sampling frequency, the system can generate realtime time-series features and estimate blood pressure in real time. This innovative approach offers a contact-free method of monitoring blood pressure, which could be beneficial for those with sleep disorders or other medical conditions that require continuous monitoring.

[0042] The blood pressure estimation methods disclosed herein are based on advanced Al algorithms that meet the criteria for evaluating commercial blood pressure measuring devices. The performance of these algorithms is then compared to state-of- the-art deep learning models for time-series analysis. Numerical results show that the Al models disclosed herein are superior to the existing models. This demonstrates the potential of Al algorithms to accurately measure blood pressure with high levels of accuracy.

EXAMPLE

Data collection

[0043] In various embodiments, the sensor(s) 103 of the present disclosure can be utilized to detect mini-vibrations on the structure 106 the sensor(s) 103 are attached to. In various examples, the sampling rate is 100 Hz. In this example, the data was collected in a controlled setting at the University of Georgia Clinical and Translational Research Unit (CTRU) using a standard hospital bed and a family bed. Two sensors 103 were installed on each bed. The main difference between the two types of beds is the bed frame and mattress, which could affect the propagation of vibration waves. Each data sample consisted of a 10-second recorded BSG signal. The data structure is illustrated in FIG. 4.

Data quality control

[0044] The BSG signal collected by the sensor 103 is composed of various vibration sources, such as background white noise, home environment noises from appliances (e.g., air conditioner, washer, dryer), body movements during sleep, and vibrations caused by vital activities like heartbeat, respiration, and blood pressure changes. To accurately estimate blood pressure, vibrations that do not reflect vital activities are filtered out. An automatic signal processing algorithm is used to eliminate signals with “bad quality". For each 10-second segment, the spectral distance, energy standard deviation, and standard deviation are measured. If one of the following three thresholds is triggered, the signal segment is labeled as “bad quality" and filtered out: spectral distance is less than 7, energy standard deviation is less than 18, and the standard deviation is greater than 5. Small spectral distance or energy standard deviation indicates behavior similar to white noise and usually lacks meaningful information on vital activities. On the other hand, a signal segment with a large standard deviation is usually dominated by a body movement like changing sleep posture, which is much stronger than common vital activities. These thresholds are based on experience with BSG signal processing. To further filter out nuisance information, the frequency domain is checked. FIG. 5 shows graphical representations of a five-minute recorded vibration data. The available frequency range is 0-50 Hz, based on the seismometer’s 100 Hz sampling rate and the Nyquist sampling theorem. The frame 503 represents the background noise with varying dominant frequencies, while frame 506 denotes machine vibrations from a dryer with an 11 Hz dominant frequency and harmonics at 22 Hz and 33 Hz. To remove these machine noises, a spectrum scanning method can be applied, then use notch filters to suppress the noise components with iso-dominant frequencies.

[0045] In addition, the data with irregular pulse pressure (the difference between SBP and DBP) is removed. A high pulse pressure, e.g., greater than 60 mmHg (millimeters of mercury), is usually considered as an indicator of an immediate heart problem. On the other hand, a low pulse pressure, say beneath 20 mmHg, may be a sign of poor heart function. In this example, only quality control of the higher bound is considered. So, the upper bound of 60 mmHg is the conservative limit to filter out outliers that may inaccurately represent blood pressure anomalies. The flow chart of the proposed data collection, preprocessing, and estimation procedure is summarized in FIG. 6. In particular, the signals are collected, and then processed using motion filtering and signal filtering. Once processed, any low quality data that is detected is dropped and the remaining data is aligned. The blood pressure is then estimated using SBP and DBP.

BLOOD PRESSURE ESTIMATION METHODS

[0046] According to various examples, two deep-learning methods for estimating blood pressures (SBP and DBP) from the collected 100 Hz BSG signals are introduced: a hybrid model that combines a CNN and an LSTM network, and a VTCN model. Implementation details are provided and hyperparameter selection are discussed for both methods.

CNN-LSTM hybrid model

[0047] A CNN-LSTM hybrid model is presented, as illustrated in FIG. 7. The CNN component is a feedforward neural network that leverages local receptive fields, shared weights, and pooling to significantly reduce the number of parameters in fully connected neural networks, enabling dimensionality reduction and feature extraction. 1 D-CNNs have been widely used in electrocardiogram anomaly detection and are well-suited for real-time, low-cost applications due to their low computational requirements. The model features three temporal convolutional layers, each followed by a max pooling layer, with a receptive field size of 3 and rectified Linear Unit (ReLU) activation. Each convolutional layer has 128 feature maps.

[0048] To be specific, the convolution process can be expressed as process can be described as follows: where x_k ^l represents the input, b_k ^l is a scalar bias of the Zc-th neuron at layer I, and s-^-1 is the output of the i-th neuron at layer Z-1 . Moreover, w-_k ¹ is the kernel from the i-th neuron at layer Z-1 to the Zc-th neuron at layer Z. Therefore, y_k, the output of the neuron, can be expressed by the activation function of x_k, as shown below yi, = /■«)■ where f is the ReLU activation function.

[0049] Each pooling layer utilizes a 1 -D max-pooling with a window size of 8 and a stride length of 2. Max-pooling reduces the dimensions of the input representation by taking the maximum value within the defined pool size window. This helps to downsample the output of the convolutional layer and prevent overfitting. The max pooling operation can be described as follows: sk^l = yk^l ^ ss, where s_k is the output of the maximum pooling layer and J, ss represents the downsampling operation conducted by a maximum pooling layer. [0050] The model also incorporates three bidirectional LSTM layers after the convolutional section. Bidirectional LSTMs access the data in both forward and backward directions by combining a forward-fed layer and a reverse-fed layer. This allows the LSTMs to learn from values both in the past and future within the sequence. Both directions are performed similarly to a standard LSTM, and their outputs are then concatenated to provide the overall result. The mathematical process can be described as follows: c_t = ft ® -i + it ® , and hf = Of tanh(Cf), where W denotes the self-updating weights and b denotes the bias vector. In addition, cr(.) and tanhf) represent the sigmoid and hyperbolic tangent functions, respectively. The operator ® indicates element-wise multiplication. In the experiments, the Adam optimization algorithm was used with a learning rate of 0.001 to update the values of W and b.

[0051] The final layer is a fully connected dense layer, which provides the final output of the network. During training, 90% of the training set was used for model training and 10% was set aside as the validation set. The training and testing of this network structure were performed separately for SBP and DBP.

VTCN model [0052] The temporal Convolution Model (TCN) is a member of the Convolutional Neural Network (CNN) family. Unlike traditional CNNs, TCNs employ a causal convolution, meaning that the convolution process at time t only depends on observations prior to time t, thus avoiding the issue of "information leakage." Furthermore, TCNs use dilated convolutions, which enables the model to capture information over a longer history with an exponentially larger receptive field compared to non-dilated causal convolutions. In essence, TCNs can be considered as the combination of 1 D fully-convolutional networks and causal convolutions.

[0053] To be more specific, for a 1 D sequence input X e H and a filter f {0, 1, ..., k - 1} IR, the dilated convolution operation F on elements s of the sequence is defined as follows:

[0054] where d is the dilation factor, k is the filter size, and s-d represents the direction of the past.

[0055] The VTCN model consists of a series of blocks, with each block containing a sequence of convolutional layers. The number of blocks is denoted by j and the number of layers within a block by I. The dilation rate d increases by a factor of 21 for consecutive layers. The activations in the Z-th layer and j-th block are represented as X^j’¹ e m^Fx7’, where F is the number of filters and T is the number of corresponding time steps. The input for each block SF is the output from the previous block S^F , with the exception of the first block, which receives the input data.

[0056] A modified TCN model, referred to as the Vital TCN or VTCN, includes an additional batch normalization layer and ReLU activation function in each block of the classical TCN model. Additionally, a linear layer is added at the end of the layers to make the model suitable for regression.

Implementation and hyperparameter selection

[0057] In each of the experiments, the observations in a BSG signal as a time series are considered, with their collection time serving as the index. The first 70% of these observations are distributed to the training set and the remaining 30% to the test set. To facilitate the tuning of hyperparameters, a validation set is used, which is comprised of 10% of the training sample. This validation set is selected after sorting the daily observed samples by A. Before fitting the CNN-LSTM hybrid model, the training set is normalized. The testing set is also normalized, using the mean and standard deviation computed from the training set. Note that this normalization process does not make use of any information from the testing set. So, there is no data leakage issue in the experiments. The VTCN model does not require normalization.

[0058] The numerical implementation of the CNN-LSTM hybrid model and the VTCN model were both based on popular deep learning frameworks. The CNN-LSTM hybrid model was based on Keras, a Python deep learning API, while the VTCN model was based on PyTorch, a machine learning framework developed by Meta Al. Both models were trained and validated over 150 epochs with mean absolute error (MAE) as the loss function. The weights of the nodes were optimized in each epoch to minimize the MAE on the validation set. After the training and validation process, the optimal weights that produced the lowest MAE on the validation set were used to predict the testing set.

EXPERIMENTS AND EVALUATIONS

[0059] The performance of the contact-free blood pressure monitoring system 100 of the present disclosure was evaluated in real-time, contact-free blood pressure estimation. The quality control measures were followed to produce a preprocessed dataset of 33,982 observations collected from more than 75 participants over 43 days. True systolic blood pressure (SBP) and diastolic blood pressure (DBP) measurements were obtained using a United States Food and Drug Administration (FDA) approved clinical blood pressure monitor, following established clinical techniques. The histograms of SBP and DBP are shown in FIGS. 8A and 8B, with sample means of 120 mmHg and 70 mmHg, respectively. The dataset was divided into two parts: first 70% of the daily data was used for training, and the remaining 30% was used for testing. So there is no overlapping and data leakage issues between training and testing sets. Both the CNN-LSTM hybrid model and the VTCN model were trained on the training set, following the model training and hyperparameter selection methods discussed herein. The trained models were then used to predict SBP and DBP on the testing set. The prediction error was evaluated using mean absolute error (MAE), standard deviation (SD), and mean absolute percentage difference (MAPD).

[0060] In addition, the performance of the two proposed Al models with several state- of-the-art deep learning methods for time series data were compared. The accuracy of the blood pressure estimation was evaluated using three widely accepted criteria: the British Hypertension Society (BHS) protocol, the Association for the Advancement of Medical Instrumentation (AAMI) standard, and ISO 81060-2-2018 from the U.S. FDA.

Analysis of experiment results

[0061] In Table 1 , shown are the results of MAE, SD, 95% confidence interval of MAE, and the correlation coefficient between the predicted and true values. The results indicate that both the CNN-LSTM hybrid model and the VTCN model exhibit low MAE and SD for both systolic and diastolic blood pressure estimations. Additionally, the predictions made by both models have high correlations with the true values. The CNN-LSTM hybrid model slightly outperforms the VTCN model as it has lower MAE and SD for both systolic and diastolic blood pressure. On average, the estimation errors for diastolic blood pressure are lower than those for systolic blood pressure as diastolic blood pressure has a smaller scale. FIGS. 9A and 9B display error histograms that visualize the distribution of prediction errors. The histograms indicate that the prediction errors for both models are symmetric and tightly concentrated around 0. Moreover, the tails of the error distributions decay rapidly to zero, indicating good prediction performance. In line with the findings in Table 1 , the peak of the error histogram for the CNN-LSTM hybrid model is higher than that for the VTCN model, suggesting that the CNN-LSTM hybrid model slightly outperforms the VTCN model.

[0062] Table 1 : Summary of blood pressure estimation performance

[0063] In the Bland-Altman plot (also known as the Tukey mean-difference plot), the agreement between two arrays of data are evaluated, such as the predicted values and the ground truth. FIGS. 10A and 10B present the Bland-Altman plots for SBP and DBP predictions. The dense cluster of points near the “difference = 0" line indicates a strong agreement between the predicted values and the truth. The dotted lines 1003 (e.g., 1003a, 1003b) and dotted lines 1006 (e.g., 1006a, 1006b) represent the limits of agreement for the CNN-LSTM hybrid model and VTCN model, respectively. For the CNN- LSTM hybrid model, 93.82% and 94.26% of points fall within the limits of agreement for SBP and DBP, respectively. For the VTCN model, 94.37% and 95.20% of points fall within the limits of agreement for SBP and DBP, respectively. These results suggest a high level of agreement between the true blood pressures and the predictions made by the models.

[0064] In FIGS. 11 A and 11 B, scatter plots are used to demonstrate the agreement level between the sorted true blood pressures and their corresponding predictions made by the proposed models. The line 1100 (e.g., 1100a, 1100b) indicates the theoretical “perfect" correlation, while the dashed line 1103 (e.g., 1103a, 1103b) and dashed lined 1106 (e.g., 1106a, 1106b) depict the actual correlation coefficients (as listed in Table 1) of the CNN-LSTM hybrid model and the VTCN model, respectively. The scatter plots reveal a strong positive correlation between the true blood pressures and the predictions generated by the models. The calculated correlation lines are close to the “perfect" correlation line in the area where the majority of data points are concentrated.

[0065] Based on the analysis, there is strong evidence to support the accuracy of the CNN-LSTM hybrid model and VTCN model in estimating SBP and DBP using the BSG signals collected by the contact-free blood pressure monitoring system 100 of the present disclosure. The strong correlation and high level of agreement between the predictions and the truth demonstrate the effectiveness of the system for contact-free blood pressure monitoring during sleep.

Convergence analysis

[0066] In this section, the numerical stability and convergence of the CNN-LSTM hybrid model and the VTCN model are evaluated, which both use MAE as their loss function. To ensure consistency, the loss is calculated as the normalized MAE after normalization. By examining the convergence of the CNN-LSTM hybrid model on the training and validation sets. FIG. 12 displays the convergence curves of the model in terms of loss versus epochs for both sets. The loss decreases gradually over the first 100 epochs in the training set, which suggests overfitting. Conversely, in the validation set, the loss decreases rapidly in the first 20 epochs before leveling off. Next, the convergence of the VTCN model on both the training and validation sets is examined, as shown in FIG. 13. The convergence curves for both sets indicate that the VTCN model is also numerically stable and converges efficiently. Evaluation metrics

[0067] Three well-established criteria used to evaluate and grade blood pressure monitoring devices are evaluated based on their estimation accuracy. The American Association for the Advancement of Medical Instrumentation (AAMI) published a monograph in 1987, which became a national standard for the evaluation of sphygmomanometers. This monograph included a standard for evaluating the accuracy of blood pressure monitoring devices, which are generated through a consensus process by committees of experts in research, development, and design from user, industry, and government communities. According to the AAMI standardl , a blood pressure monitoring device is considered to "pass" if its blood pressure estimation satisfies MAE 5 mmHg and SD 8 mmHg.

[0068] The British and Irish Hypertension Society (BHS) protocol released in 1990 is used to assess the accuracy of commercial blood pressure measurement devices. The BHS protocol grades devices based on the criteria stated in Table 2. The grades are determined by the percentage of error readings within 5 mmHg, 10 mmHg, and 15 mmHg, where A denotes the best agreement with the mercury standard, and D denotes the least agreement. For a device to be recommended for commercialization, it must achieve a grade of at least B for both systolic and diastolic pressures.

[0069] The U.S. Food and Drug Administration (FDA) evaluation criteria suggest that the mean of the errors of the paired determinations of the sphygmomanometer-under- test for all subjects should be within or equal to 5.0 mmHg, with an experimental standard deviation no greater than 8.0 mmHg. [0070] Additionally, the mean absolute percentage difference (MAPD) as a reference in accordance with the IEEE standard for device accuracy. This provides a measure of the stability of the predictions.

[0071] Table 2: British Hypertension Society grading criteria

Comparison with existing methods

[0072] The CNN-LSTM and VTCN models are compared to several state-of-the-art deep learning methods for time series analysis, including the Multilevel Wavelet Decomposition Network (mWDN), 1 D-ResCNN (ResCNN), Recurrent Neural Network and Fully Convolutional Network (RNN-FCN), Explainable Convolutional Neural Network (XCM), and Time-series Transformer (TST). All competing methods and the models followed the same training, validation, and testing protocols.

[0073] Table 3 shows the evaluation results of the models and the five competing methods using the AAMI standard and the FDA standard. The results demonstrate that none of the five algorithms meet the AAMI standard for SBP estimation. Only the RNN- FCN algorithm meets the AAMI standard for DBP estimation. In contrast, both the CNN- LSTM and VTCN models meet the AAMI standard for both SBP and DBP estimations and have the smallest mean absolute errors for SBP and DBP, respectively.

[0074] Table 3: Comparison of SBP and DBP estimations based on the AAMI standard and FDA standard.

[0075] Table 4 evaluates the five competitive methods and the two models using the BHS protocol. Only the RNN-FCN algorithm achieves a grade A for both SBP and DBP estimations among the competitive methods. However, both the CNNLSTM and VTCN models achieve a grade A and have higher percentages of absolute differences smaller than or equal to 5, 10, and 15 mmHg compared to RNN-FCN. In particular, the CNNLSTM model has the highest cumulative percentage of readings for SBP with absolute differences smaller or equal to 5 mmHg, making it the best for SBP estimation. Meanwhile, the VTCN algorithm has the highest cumulative percentage of readings for DBP with absolute differences smaller or equal to 5 mmHg, making it the best for DBP estimation.

[0076] Table 4: Comparison of SBP and DBP estimations based on the BHS protocol.

[0077] In conclusion, both the CNN-LSTM and VTCN models pass the AAMI standard, FDA standard, and BHS protocol with outstanding performance. They outperform all five competing methods in terms of high estimation accuracy and low variability. These results provide strong evidence that the contact-free blood pressure monitoring system 100 of the present disclosure can be considered an affordable and contact-free sleep monitoring solution with excellent real-time blood pressure estimation capabilities.

[0078] The continuous blood pressure monitoring disclosed herein obtains data in a contactless fashion. The study employs a contact-free and bed-mounted seismic sensor (e.g., sensor 103) and two advanced Al models that make use of high-frequency time series data extracted from the BSG signal to estimate blood pressure. Through extensive experiments in controlled hospital environments and real-world scenarios, the accuracy and robustness of the algorithm in estimating systolic and diastolic blood pressures in real time is demonstrated. In addition, the models meet three well-established criteria used to evaluate and grade blood pressure monitoring devices. Furthermore, the Al models are compared with five state-of-the-art deep learning-based time series models, and the results show that the models outperform all five competing models in terms of accuracy and low variability.

[0079] It should be emphasized that the above-described embodiments of the present disclosure are merely possible examples of implementations set forth for a clear understanding of the principles of the disclosure. Many variations and modifications may be made to the above-described embodiment(s) without departing substantially from the spirit and principles of the disclosure. All such modifications and variations are intended to be included herein within the scope of this disclosure and protected by the following claims.

[0080] A number of software components previously discussed are stored in the memory of the respective computing devices and are executable by the processor of the respective computing devices. In this respect, the term "executable" means a program file that is in a form that can ultimately be run by the processor. Examples of executable programs can be a compiled program that can be translated into machine code in a format that can be loaded into a random access portion of the memory and run by the processor, source code that can be expressed in proper format such as object code that is capable of being loaded into a random access portion of the memory and executed by the processor, or source code that can be interpreted by another executable program to generate instructions in a random access portion of the memory to be executed by the processor. An executable program can be stored in any portion or component of the memory, including random access memory (RAM), read-only memory (ROM), hard drive, solid-state drive, Universal Serial Bus (USB) flash drive, memory card, optical disc such as compact disc (CD) or digital versatile disc (DVD), floppy disk, magnetic tape, or other memory components.

[0081] The memory includes both volatile and nonvolatile memory and data storage components. Volatile components are those that do not retain data values upon loss of power. Nonvolatile components are those that retain data upon a loss of power. Thus, the memory can include random access memory (RAM), read-only memory (ROM), hard disk drives, solid-state drives, USB flash drives, memory cards accessed via a memory card reader, floppy disks accessed via an associated floppy disk drive, optical discs accessed via an optical disc drive, magnetic tapes accessed via an appropriate tape drive, or other memory components, or a combination of any two or more of these memory components. In addition, the RAM can include static random access memory (SRAM), dynamic random access memory (DRAM), or magnetic random access memory (MRAM) and other such devices. The ROM can include a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or other like memory device.

[0082] Although the applications and systems described herein can be embodied in software or code executed by general purpose hardware as discussed above, as an alternative the same can also be embodied in dedicated hardware or a combination of software/general purpose hardware and dedicated hardware. If embodied in dedicated hardware, each can be implemented as a circuit or state machine that employs any one of or a combination of a number of technologies. These technologies can include, but are not limited to, discrete logic circuits having logic gates for implementing various logic functions upon an application of one or more data signals, application specific integrated circuits (ASICs) having appropriate logic gates, field-programmable gate arrays (FPGAs), or other components, etc. Such technologies are generally well known by those skilled in the art and, consequently, are not described in detail herein. [0083] The flowchart shows the functionality and operation of an implementation of portions of the various embodiments of the present disclosure. If embodied in software, each block can represent a module, segment, or portion of code that includes program instructions to implement the specified logical function(s). The program instructions can be embodied in the form of source code that includes human-readable statements written in a programming language or machine code that includes numerical instructions recognizable by a suitable execution system such as a processor in a computer system. The machine code can be converted from the source code through various processes. For example, the machine code can be generated from the source code with a compiler prior to execution of the corresponding application. As another example, the machine code can be generated from the source code concurrently with execution with an interpreter. Other approaches can also be used. If embodied in hardware, each block can represent a circuit or a number of interconnected circuits to implement the specified logical function or functions.

[0084] Although the flowchart in FIG. 2 shows a specific order of execution, it is understood that the order of execution can differ from that which is depicted. For example, the order of execution of two or more blocks can be scrambled relative to the order shown. Also, two or more blocks shown in succession can be executed concurrently or with partial concurrence. Further, in some embodiments, one or more of the blocks shown in the flowchart can be skipped or omitted. In addition, any number of counters, state variables, warning semaphores, or messages might be added to the logical flow described herein, for purposes of enhanced utility, accounting, performance measurement, or providing troubleshooting aids, etc. It is understood that all such variations are within the scope of the present disclosure.

[0085] Also, any logic or application described herein that includes software or code can be embodied in any non-transitory computer-readable medium for use by or in connection with an instruction execution system such as a processor in a computer system or other system. In this sense, the logic can include statements including instructions and declarations that can be fetched from the computer-readable medium and executed by the instruction execution system. In the context of the present disclosure, a "computer-readable medium" can be any medium that can contain, store, or maintain the logic or application described herein for use by or in connection with the instruction execution system. Moreover, a collection of distributed computer-readable media located across a plurality of computing devices (e.g, storage area networks or distributed or clustered filesystems or databases) may also be collectively considered as a single non-transitory computer-readable medium.

[0086] The computer-readable medium can include any one of many physical media such as magnetic, optical, or semiconductor media. More specific examples of a suitable computer-readable medium would include, but are not limited to, magnetic tapes, magnetic floppy diskettes, magnetic hard drives, memory cards, solid-state drives, USB flash drives, or optical discs. Also, the computer-readable medium can be a random access memory (RAM) including static random access memory (SRAM) and dynamic random access memory (DRAM), or magnetic random access memory (MRAM). In addition, the computer-readable medium can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or other type of memory device.

[0087] Further, any logic or application described herein can be implemented and structured in a variety of ways. For example, one or more applications described can be implemented as modules or components of a single application. Further, one or more applications described herein can be executed in shared or separate computing devices or a combination thereof. For example, a plurality of the applications described herein can execute in the same computing device, or in multiple computing devices in the same computing environment 103.

[0088] In addition to the foregoing, the various embodiments of the present disclosure include, but are not limited to, the embodiments set forth in the following clauses.

[0089] Clause 1. A system for monitoring a blood pressure of a subject based on vibration signals of a structure supporting the subject, the system comprising: a sensor coupled to the structure; at least one computing device in data communication with the sensor; and an application executable in the at least one computing device, wherein when executed, the application causes the at least one computing device to at least: obtain real-time sensor data from the sensor, the real-time sensor data comprising seismic signal segments recorded at a predetermined sampling rate; filter the sensor data to remove seismic signal segments not related to vital activities; analyze the filtered sensor data to determine continuous and real-time blood pressure measurements of the subject, the real-time blood pressure measurements comprising estimated systolic blood pressure values and estimated diastolic blood pressure values; generate a user interface comprising a display of at least the blood pressure measurements; and render the user interface via a display. [0090] Clause 2. The system of clause 1 , wherein the estimated systolic blood pressure values and the estimated diastolic blood pressure values are based at least in part on a Convolutional Neural Network - Long Short Term Memory (CNN-LSTM) hybrid model.

[0091] Clause 3. The system of clause 2, wherein the CNN-LSTM hybrid model comprise a plurality of temporal convolutional layers, a plurality of max pooling layers, and a plurality of bidirectional LSTM layers.

[0092] Clause 4. The system of clause 1 , wherein the estimated systolic blood pressure values and the estimated diastolic blood pressure values are based at least in part on a modified Temporal Convolutional Networks (TCN) model.

[0093] Clause 5. The system of clause 4, wherein the modified TCM model comprises an additional batch normalization layer and rectified linear unity (ReLU) activation function in each block of the TCN model, and a linear layer for regression.

[0094] Clause 6. The system of any one of clauses 1 to 5, wherein, when executed, the application further causes the at least one computing device to extract high-frequency time series data from the real-time sensor data.

[0095] Clause 7. The system of any one of clauses 1 to 6, wherein to filter the sensor data to remove seismic signal segments not related to vital activities, the application further causes the at least one computing device to at least: filter the sensor data to remove seismic signal segments related to movement of the subject on the structure; and filter the sensor data to remove seismic signal segments related to noise.

[0096] Clause 8. The system of any one of clauses 1 to 7, wherein to filter the sensor data to remove seismic signal segments not related to vital activities, the application further causes the at least one computing device to at least filter the sensor data to remove seismic signal segments based on at least one of a spectral distance, an energy standard deviation, and a standard deviation.

[0097] Clause 9. The system of clause 8, wherein the seismic signal segments are removed when at least one of: the spectral distance is less than a spectral distance threshold; the energy standard deviation is less than an energy standard deviation threshold; or the standard deviation is greater than a standard deviation threshold.

[0098] Clause 10. The system of any one of clauses 1 to 9, wherein, when executed, the application further causes the at least one computing device to at least suppress noise components with iso-dominant frequencies.

[0099] Clause 11 . The system of any one of clauses 1 to 10, wherein, when executed, the application further causes the at least one computing device to analyze the sensor data to determine continuous and real-time measurements of heart rate, respiratory rate, movement, posture, and fall off bed of a subject positioned on the bed.

[0100] Clause 12. The system of clause 11 , wherein, when executed, the application further causes the at least one computing device to at least update the user interface to include at least one of the heart rate, the respiratory rate, or blood pressure of the subject. [0101] Clause 13. The system of clause 11 or clause 12, wherein, when executed, the application further causes the at least one computing device to at least detect an event based at least in part on at least one of the heart rate, the respiratory rate, or blood pressure of the subject.

[0102] Clause 14. The system of clause 13, wherein the event comprises at least one of the systolic blood pressure being outside a predefined range or the diastolic blood pressure being outside the predefined range.

[0103] Clause 15. The system of clause 13 or clause 14, wherein, when executed, the application further causes the at least one computing device to at least generate an alert in response to the detected event.

[0104] Clause 16. The system of clause 15, wherein the alert is at least one of an auditory or visual or vibratory alert.

[0105] Clause 17. The system of clause 15 or clause 16, wherein the at least one computing device is in communication with a smart device configured to communicate with a third party, and generating the alert further comprises instructing the smart device to send a communication with the third party.

[0106] Clause 18. The system of any one of clauses 1 to 17, wherein the sensor is not in direct contact with the subject.

[0107] Clause 19. A method for monitoring a blood pressure of a subject, the method comprising: obtaining, by a computing device in data communication with a sensor, realtime sensor data comprising vibration signals detected by the sensor; filtering, by the computing device, the sensor data to remove seismic signal segments not related to vital activities; and analyzing, by the computing device, the sensor data to determine continuous and real-time blood pressure measurements of the subject, the real-time blood pressure measurements comprising estimated systolic blood pressure values and estimated diastolic blood pressure values; generating a user interface comprising a display of at least the blood pressure measurements; and rendering the user interface via a display.

[0108] Clause 20. The method of clause 19, wherein the estimated systolic blood pressure values and the estimated diastolic blood pressure values are based at least in part on a Convolutional Neural Network - Long Short Term Memory (CNN-LSTM) hybrid model.

[0109] Clause 21. The method of clause 20, wherein the CNN-LSTM hybrid model comprise a plurality of temporal convolutional layers, a plurality of max pooling layers, and a plurality of bidirectional LSTM layers.

[0110] Clause 22. The method of clause 19, wherein the estimated systolic blood pressure values and the estimated diastolic blood pressure values are based at least in part on a modified Temporal Convolutional Networks (TCN) model.

[0111] Clause 23. The method of clause 22, wherein the modified TCM model comprises an additional batch normalization layer and rectified linear unity (ReLU) activation function in each block of the TCN model, and a linear layer for regression.

[0112] Clause 24. The method of any one of clauses 19 to 23, further comprising: extracting high-frequency time series data from the real-time sensor data comprising seismic signal segments recorded at a predetermined sampling rate; and filtering the sensor data to remove seismic signal segments based on at least one of a spectral distance, an energy standard deviation, and a standard deviation.

[0113] Clause 25. The method of any one of clauses 19 to 24, further comprising applying a spectrum scanning method to the sensor data; and suppressing noise components with iso-dominant frequencies using a notch filter.

[0114] Clause 26. The method of any one of clauses 19 to 25, further comprising analyzing the sensor data to determine continuous and real-time measurements of heart rate, respiratory rate, movement, posture, and fall off bed of a subject positioned on the bed; and updating the user interface to include at least one of the heart rate, the respiratory rate, or blood pressure of the subject.

[0115] Clause 27. The method of clause 26, further comprising detecting an event based at least in part on at least one of the heart rate, the respiratory rate, or blood pressure of the subject.

[0116] Clause 28. The method of clause 27, wherein the event comprises at least one of the systolic blood pressure or the diastolic blood pressure being outside a predefined range.

[0117] Clause 29. The method of clause 27 or clause 28, further comprising generating an alert in response to the detected event.

[0118] Clause 30. The method of clause 29, wherein the alert is at least one of an auditory or visual or vibratory alert.

[0119] Clause 31. The method of clause 29 or clause 30, wherein the at least one computing device is in communication with a smart device configured to communicate with a third party, and generating the alert further comprises instructing the smart device to send a communication with the third party.

[0120] It should be noted that ratios, concentrations, amounts, and other numerical data may be expressed herein in a range format. It is to be understood that such a range format is used for convenience and brevity, and thus, should be interpreted in a flexible manner to include not only the numerical values explicitly recited as the limits of the range, but also to include all the individual numerical values or sub-ranges encompassed within that range as if each numerical value and sub-range is explicitly recited. To illustrate, a concentration range of “about 0.1 % to about 5%” should be interpreted to include not only the explicitly recited concentration of about 0.1 wt% to about 5 wt%, but also include individual concentrations (e.g., 1 %, 2%, 3%, and 4%) and the sub-ranges (e.g., 0.5%, 1.1 %, 2.2%, 3.3%, and 4.4%) within the indicated range. The term “about” can include traditional rounding according to significant figures of numerical values. In addition, the phrase “about ‘x’ to ‘y’” includes “about ‘x’ to about ‘y’”.

Claims

CLAIMS We claim:

1. A system for monitoring a blood pressure of a subject based on vibration signals of a structure supporting the subject, the system comprising: a sensor coupled to the structure; at least one computing device in data communication with the sensor; and an application executable in the at least one computing device, wherein when executed, the application causes the at least one computing device to at least: obtain real-time sensor data from the sensor, the real-time sensor data comprising seismic signal segments recorded at a predetermined sampling rate; filter the sensor data to remove seismic signal segments not related to vital activities; analyze the filtered sensor data to determine continuous and realtime blood pressure measurements of the subject, the real-time blood pressure measurements comprising estimated systolic blood pressure values and estimated diastolic blood pressure values; generate a user interface comprising a display of at least the blood pressure measurements; and render the user interface via a display.

2. The system of claim 1 , wherein the estimated systolic blood pressure values and the estimated diastolic blood pressure values are based at least in part on a Convolutional Neural Network - Long Short Term Memory (CNN-LSTM) hybrid model.

3. The system of claim 2, wherein the CNN-LSTM hybrid model comprise a plurality of temporal convolutional layers, a plurality of max pooling layers, and a plurality of bidirectional LSTM layers.

4. The system of claim 1 , wherein the estimated systolic blood pressure values and the estimated diastolic blood pressure values are based at least in part on a modified Temporal Convolutional Networks (TCN) model.

5. The system of claim 4, wherein the modified TCM model comprises an additional batch normalization layer and rectified linear unity (ReLU) activation function in each block of the TCN model, and a linear layer for regression.

6. The system of any one of claims 1 to 5, wherein, when executed, the application further causes the at least one computing device to extract high-frequency time series data from the real-time sensor data.

7. The system of any one of claims 1 to 6, wherein to filter the sensor data to remove seismic signal segments not related to vital activities, the application further causes the at least one computing device to at least: filter the sensor data to remove seismic signal segments related to movement of the subject on the structure; and filter the sensor data to remove seismic signal segments related to noise.

8. The system of any one of claims 1 to 7, wherein to filter the sensor data to remove seismic signal segments not related to vital activities, the application further causes the at least one computing device to at least filter the sensor data to remove seismic signal segments based on at least one of a spectral distance, an energy standard deviation, and a standard deviation.

9. The system of claim 8, wherein the seismic signal segments are removed when at least one of: the spectral distance is less than a spectral distance threshold; the energy standard deviation is less than an energy standard deviation threshold; or the standard deviation is greater than a standard deviation threshold.

10. The system of any one of claims 1 to 9, wherein, when executed, the application further causes the at least one computing device to at least suppress noise components with iso-dominant frequencies.

11. The system of any one of claims 1 to 10, wherein, when executed, the application further causes the at least one computing device to analyze the sensor data to determine continuous and real-time measurements of heart rate, respiratory rate, movement, posture, and fall off bed of a subject positioned on the bed.

12. The system of claim 11 , wherein, when executed, the application further causes the at least one computing device to at least update the user interface to include at least one of the heart rate, the respiratory rate, or blood pressure of the subject.

13. The system of claim 11 or claim 12, wherein, when executed, the application further causes the at least one computing device to at least detect an event based at least in part on at least one of the heart rate, the respiratory rate, or blood pressure of the subject.

14. The system of claim 13, wherein the event comprises at least one of the systolic blood pressure being outside a predefined range or the diastolic blood pressure being outside the predefined range.

15. The system of claim 13 or claim 14, wherein, when executed, the application further causes the at least one computing device to at least generate an alert in response to the detected event.

16. The system of claim 15, wherein the alert is at least one of an auditory or visual or vibratory alert.

17. The system of claim 15 or claim 16, wherein the at least one computing device is in communication with a smart device configured to communicate with a third party, and generating the alert further comprises instructing the smart device to send a communication with the third party.

18. The system of any one of claims 1 to 17, wherein the sensor is not in direct contact with the subject.

19. A method for monitoring a blood pressure of a subject, the method comprising: obtaining, by a computing device in data communication with a sensor, realtime sensor data comprising vibration signals detected by the sensor; filtering, by the computing device, the sensor data to remove seismic signal segments not related to vital activities; and analyzing, by the computing device, the sensor data to determine continuous and real-time blood pressure measurements of the subject, the real-time blood pressure measurements comprising estimated systolic blood pressure values and estimated diastolic blood pressure values; generating a user interface comprising a display of at least the blood pressure measurements; and rendering the user interface via a display.

20. The method of claim 19, wherein the estimated systolic blood pressure values and the estimated diastolic blood pressure values are based at least in part on a

Convolutional Neural Network - Long Short Term Memory (CNN-LSTM) hybrid model.

21. The method of claim 20, wherein the CNN-LSTM hybrid model comprise a plurality of temporal convolutional layers, a plurality of max pooling layers, and a plurality of bidirectional LSTM layers.

22. The method of claim 19, wherein the estimated systolic blood pressure values and the estimated diastolic blood pressure values are based at least in part on a modified Temporal Convolutional Networks (TCN) model.

23. The method of claim 22, wherein the modified TCM model comprises an additional batch normalization layer and rectified linear unity (ReLU) activation function in each block of the TCN model, and a linear layer for regression.

24. The method of any one of claims 19 to 23, further comprising: extracting high-frequency time series data from the real-time sensor data comprising seismic signal segments recorded at a predetermined sampling rate; and filtering the sensor data to remove seismic signal segments based on at least one of a spectral distance, an energy standard deviation, and a standard deviation.

25. The method of any one of claims 19 to 24, further comprising applying a spectrum scanning method to the sensor data; and suppressing noise components with iso-dominant frequencies using a notch filter.

26. The method of any one of claims 19 to 25, further comprising analyzing the sensor data to determine continuous and real-time measurements of heart rate, respiratory rate, movement, posture, and fall off bed of a subject positioned on the bed; and updating the user interface to include at least one of the heart rate, the respiratory rate, or blood pressure of the subject.

27. The method of claim 26, further comprising detecting an event based at least in part on at least one of the heart rate, the respiratory rate, or blood pressure of the subject.

28. The method of claim 27, wherein the event comprises at least one of the systolic blood pressure or the diastolic blood pressure being outside a predefined range.

29. The method of claim 27 or claim 28, further comprising generating an alert in response to the detected event.

30. The method of claim 29, wherein the alert is at least one of an auditory or visual or vibratory alert.

31. The method of claim 29 or claim 30, wherein the at least one computing device is in communication with a smart device configured to communicate with a third party, and generating the alert further comprises instructing the smart device to send a communication with the third party.