Device for machine-learning real-time monitoring of patient vital signs

Abstract

A device for machine learning-dependent real-time monitoring of patients' vital functions, consisting of a sensor array for recording physiological data, including at least one of the following: heart rate, respiratory rate, temperature, blood pressure; an electronic data processor connected to the sensor array, which receives and processes the physiological data; a data processing unit that includes an integrated machine learning module and has several pre-trained models for real-time detection, classification, and prediction of anomalies in physiological data; a communication interface for sending output data and notifications either out-of-band / offline to external devices or monitoring systems; and an interactive user interface that provides real-time insight into vital signs and alerts generated by the machine learning module, with the user interface configured to allow any medical personnel to interact with the device to adjust monitoring parameters or thresholds.

Description

Field of the invention

[0001] The present invention relates to a device for real-time monitoring of patient vital signs using advanced machine learning techniques. Background of the invention

[0002] Patient monitoring is essential in the current healthcare paradigm for rapid and accurate diagnosis and interventions. Most conventional monitoring systems are manually entered and have limited capabilities for analyzing enormous amounts of physiological data in a short period of time. The development of machine learning algorithms in patient monitoring devices could therefore improve the detection of vital sign abnormalities and assist healthcare professionals in making timely decisions through real-time data analysis.

[0003] Continuous monitoring of patient vital signs is becoming increasingly important in the evolving healthcare landscape to enable early diagnosis and effective treatment. Although necessary, traditional patient monitoring systems have limitations that often compromise their effectiveness. Current solutions are generally device-based and target individual vital signs (HR / BP / temperature) in isolation, without integrating these measurements into a holistic monitoring system. Cardiac monitors, for example, can monitor heart rate and rhythm but may not be able to measure other physiological parameters simultaneously. This fragmented, siloed care leaves physicians lacking the comprehensive overview of patients' health status needed for true understanding.As a result, healthcare providers may not be able to detect important changes in a patient's physical condition in a timely manner, leading to delayed and consequently withheld interventions.

[0004] Furthermore, many current systems use simplified algorithms for data analysis—arbitrarily set alert levels that inform clinicians about acute patient conditions. However, sometimes a reactive approach is too little, too late, and does not adequately support edge cases or failover scenarios where constant monitoring combined with proactive decision-making is required. For example, a patient may exhibit fluctuating acceptable heart rates—but not at the alarming lows that ultimately indicate something more problematic—and without the advanced analytics required to transform these statistically anomalous measurements into actionable information, opportunities for earlier intervention may be missed.Furthermore, these traditional systems are often rigid and don't offer providers meaningful customization of the monitoring parameters a provider can establish per patient. This is because the inflexibility associated with thresholds and alert criteria can lead to a false sense of security, as alerts are always either red or green, potentially desensitizing healthcare workers and leading to alarm fatigue.

[0005] Wearable electronics and telemedicine help address these limitations and promise better patient monitoring. Furthermore, most wearable devices are designed for general health monitoring rather than for a specific medical use case. Situational data can provide inaccurate measurements and limited information for clinical decision-making.

[0006] In addition, telehealth solutions have also been developed recently in the wake of the global health crisis. These systems allow providers to remotely monitor patients' vital signs and symptoms, reducing the dependence on in-person visits. However, current telehealth platforms often require patients to self-report their symptoms or manually measure their vital signs, making the information prone to inaccuracies and delays in data transmission. Furthermore, most telehealth systems are little more than isolated pieces of software that cannot be integrated with electronic health records or the typical hospital information system. This separation makes it even more difficult for healthcare providers to access and analyze patient data in a consistent manner.

[0007] The healthcare sector is seeing the first signs of machine learning and artificial intelligence (AI), which could represent a potential advance in patient monitoring. Existing solutions use machine learning algorithms to process large amounts of data and make predictions about patient outcomes. However, there are several concerns regarding data quality and integration. Machine learning models require a lot of good data, and this data typically needs to be diverse for these models to work well. However, because many healthcare systems are often siloed, patient data is more isolated and cannot be comprehensively linked to other sources.Even when moles are digitized using computer vision models—for example, by a patient taking and uploading photos of their skin to be analyzed by algorithms—the lack of standardization in medical data formats represents a significant barrier to integrating machine learning innovations into current workflows.

[0008] The introduction of machine learning-based monitoring solutions presents additional challenges regarding training and the deployment of such implementations in clinical settings. Machine learning results should be accompanied by a high degree of care in training healthcare professionals so they can understand the results and integrate them into their clinical reasoning. Without proper preparation and training, there is a very real risk that healthcare professionals will become overly reliant on automated systems, compromising their own clinical judgment.

[0009] Furthermore, the ethical implications and regulatory aspects of machine learning in healthcare must be considered. Before these almost friendly robots can be fully deployed to monitor patients, issues of data privacy, data security, and even algorithmic bias still need to be addressed. Since machine learning solutions often require access to highly sensitive health information, the handling of patient data must be handled carefully and ethically. Regulatory compliance and oversight: Regulatory restrictions further increase the complexity of developing or deploying machine learning-based monitoring systems. Summary of the invention

[0010] The invention comprises a state-of-the-art, machine learning-based real-time monitoring system for patient vital signs. The device contains a sensor cluster that can record various physiological data signals such as heart rate, respiratory rate and saturation, temperature, or blood pressure. Sensor types include a photoplethysmography (PPG) sensor for heart rate and blood oxygen saturation, a piezoelectric sensor for monitoring respiratory rate, including chest movement, a thermistor for continuous temperature measurements, and a non-invasive, time-averaged real-time blood pressure sensor based on oscillometric technology.

[0011] The sensor array is operatively connected to a data processing unit designed to receive, process, and analyze physiological data. The data storage module in this unit enables the preservation of historical data and metadata for retrospective analysis, as well as the training of LRP models. A real-time data processing engine in the data processor uses statistical methods to minimize noise in the raw sensor values, which are fed into the machine learning module. The machine learning module is located in the data processing unit and contains numerous pre-trained models for real-time detection, classification, and prediction of abnormalities in physiological data. The module uses a supervised learning technique that requires training data consisting of labeled examples grouping normal and abnormal vital signs from different patient populations.It also includes a form of adaptive learning so the module can continuously adapt its models over time based on current data from health sensors and feedback from medical personnel, improving predictions.

[0012] The device also features a protocol-independent communication interface that supports both protocols. It connects to mobile devices via Bluetooth Low Energy (BLE) for local communication, to the cloud via Wi-Fi for remote monitoring, and to hospital information systems directly via a wired interface, ensuring interoperability with existing medical infrastructure.

[0013] The user interface is optimized for interaction and features a multi-touch display that shows healthcare providers interactive charts and vital sign trends over time. Medical staff can also set customizable alerts and notifications to establish thresholds for individual patients, enabling patient-centered monitoring.

[0014] It is equipped with a rechargeable lithium battery that lasts for 48 hours with normal monitoring.

[0015] The present invention relates to the development of systems and methods that enhance existing monitoring of patient vital signs by incorporating machine learning technology to improve the accuracy, efficiency, and usability of such monitoring processes. A first objective of the invention is to describe a complete device that enables a wide range of physiological data (heart and respiratory rate, temperature, or blood pressure) to be collected in real time. This holistic approach is crucial for giving healthcare providers a 360-degree view of the patient's health status, allowing them to intervene before established markers are lost and to make clear clinical judgments.

[0016] The innovation also included significant attention to the user interface design, focusing on creating an intuitive and engaging ecosystem for clinicians. The device features multi-touch displays and offers customizable alerts and interactive charts that show trends over time, providing healthcare professionals with accessible access to high-priority information.

[0017] Defining this UX spectrum shortens the learning curve for healthcare professionals, allowing them to quickly navigate the system when alerts triggered by patient data occur.

[0018] Finally, a primary purpose of the invention is to enable energy-optimized operation of this device by utilizing a rechargeable lithium-ion battery with ultra-low power consumption for long operating life cycles with minimal energy loss when idle. The sustainability features not only enable continuous monitoring but also minimize device recharging, which in turn benefits usability in diverse healthcare environments. Overall, the new device aims to transform patient vital sign monitoring by combining innovative technological methods with design principles that consider the human factor to improve the quality of care and patient outcomes. Short description of the drawing Fig. Figure 1 shows a block diagram of a device for machine learning-controlled real-time monitoring of patient vital signs. Detailed description of the device

[0019] In Fig.Figure 1 shows a block diagram of a device for machine-learning-driven real-time monitoring of patient vital signs. The device 100 includes a sensor array 102 capable of sensing physiological data, including at least a patient's heart rate, respiratory rate, temperature, and blood pressure; an electronic data processor 104 connected to the sensor array that receives and processes the physiological data; a data processing unit 106 including a machine learning module integrated therein and having a plurality of pre-trained models for real-time detection, classification, and prediction of abnormalities in the physiological data; a communication interface 108 for sending the output data and notifications either out-of-band / offline to external devices or monitoring systems.and an interactive user interface 110 to provide insight into the vital signs and alerts generated in real time by the machine learning module, the user interface being configured to allow medical personnel to interact with the device to adjust monitoring parameters or thresholds;

[0020] In one embodiment, the sensor array 102 includes: a photoplethysmography sensor for measuring heart rate and blood oxygen saturation, a piezoelectric sensor for monitoring respiratory rate based on chest movements, a thermistor for measuring body temperature and continuous temperature readings, and a non-invasive blood pressure sensor that uses oscillometric technology to provide real-time blood pressure readings.

[0021] In one embodiment, the data processing unit 106 includes: a data storage module that stores historical vital sign data and metadata for future retrospective analysis and model training; a real-time data processing engine that applies statistical techniques to remove noise from the raw sensor measurements before feeding the data into the machine learning module.

[0022] In one embodiment, the communication interface 108 provides a connection interface with at least one of the following protocols: Bluetooth Low Energy protocol for short-range connections with the smartphone device; Wi-Fi protocol for linking to the cloud and remote monitoring channels; and wired connection for direct access to hospital systems and integration with other medical devices.

[0023] In one embodiment, user interface 110 includes a multi-touch screen capable of displaying interactive, complex charts and trends of vital signs, with dynamic alerts and notifications sent to caregivers when a certain threshold for the patient's measurements is set.

[0024] In one embodiment, the device 100 is powered by a rechargeable lithium-ion battery that operates for at least 48 hours at full load in continuous mode, with a consumption mode activated during periods of inactivity.

[0025] According to one embodiment, the user interfaces 110 include: a multi-touch display (110(a)) capable of providing touch graphs and trend data on vital signs over time, as well as custom alerts / alarms that allow healthcare providers to set individual thresholds for individual treatment for each patient.

[0026] In one example, the device uses a rechargeable lithium-ion battery with an expected operating time of continuous monitoring of at least 48 hours and has a power saving mode that turns on when there is no activity.

[0027] Details of the real-time patient vital signs monitor for machine learning application: The proposed device is a compact and portable structure that can be used for continuous monitoring in many locations, such as hospitals, clinics, or at home. The sensor unit is attached to a flexible housing with an ergonomic form factor that can be worn on or near the patient. This ensures high-quality data acquisition while minimizing disruption to daily routine.

[0028] The housing houses a data processing hub that delivers exceptional processing speeds for rapid analysis and real-time feedback to healthcare providers. With its embedded communications interface, it enables all types of connectivity options, whether long-distance data transmission or short-distance communication. The ability for remote monitoring and immediate medical response makes this feature so important.

[0029] The intuitively designed system gives healthcare professionals instant access to vital sign trends, alerts, and historical data. The multi-touch display allows healthcare providers to easily navigate and adjust settings to quickly change monitoring parameters.

[0030] According to one aspect of the present invention, a system for real-time monitoring of patient vital signs using machine learning is provided. The sensor array is the primary human-machine interface for patients. The sensor array is designed to capture a wide range of biosignals. This is an example where a photoplethysmography (PPG) sensor, based on light absorption, can measure heart rate and SpO2 by detecting changes in blood volume. Because it is non-invasive, patients can be monitored comfortably and for extended periods, allowing the device to be worn continuously if necessary in various settings, including hospital or home care settings.Additionally, a piezoelectric respiratory rate sensor records observed chest movements, while a thermistor continuously measures body temperature to provide important information about the patient's health. The overall functionality is further enhanced by the integration of an oscillometric, MEMS-based, non-invasive blood pressure sensor capable of providing real-time blood pressure results without invasive intervention. Taken together, this data can provide a comprehensive overview of the patient's health, allowing us to intervene quickly and effectively.

[0031] In one embodiment, the data processing unit is capable of receiving and processing the acquired data from the sensor array. The compute unit is powerful enough and designed for high-performance computing, as these sensors will generate a significant amount of data. A key part of the data processing block is a storage module used to store historical vital sign measurements and metadata. This is crucial for post-mortem analysis of protocols and training the machine learning models on your device. For example, if a patient is continuously monitored over an extended period of time and recordings of their vital signs are also available, this can be used to identify patterns that precede clinical events (such as cardiac arrest), potentially improving the predictive model.In addition, the real-time data processing engine uses statistical methods to eliminate noise and artifacts from the sensor's raw data. The engine uses algorithms such as Kalman filters and moving averages to ensure the machine learning module receives accurate and clean data, allowing it to increase its analysis accuracy.

[0032] According to one embodiment, the interpretive machine learning module is part of a data processing unit that uses physiological data obtained from a sensor array. Training the module uses a supervised learning approach on a large-scale dataset with labeled normal and abnormal vital signs. A healthcare training dataset may include heart rate values ​​labeled either as normal or as a possible arrhythmia / tachycardia detected for various patient demographics. An adaptive learning mechanism allows the machine learning module to continuously update its models as new data is received along with feedback from healthcare professionals. The system continues to iterate, learning over time how to make the best predictions in a way that remains understandable, yet is timely and relevant to changing dynamics within patients.This means that if a person's heart rate spikes during a monitored period, this data is used by the machine learning module to more accurately predict such future events.

[0033] In one aspect, the communications interface was designed to allow easy data transfer between the monitoring device and multiple external systems. The interface is compatible with multiple communication protocols, including Bluetooth Low Energy (BLE), allowing healthcare providers to access patient data via mobile devices even away from the bedside. This could mean that a nurse would be notified of out-of-range patient heart rates on their mobile device as soon as they occur. Additionally, the device includes Wi-Fi for cloud integration, providing remote access to real-time data for centralized hospital information systems. This feature is critical in larger healthcare facilities for keeping track of multiple patients simultaneously.A wired implant interface allows the device to connect directly to existing hospital systems, conceptually integrating the brain chip (and by extension, a part of your body) that natively speaks Internet Protocol and is functionally useful medically into technologies that support many more parts per person.

[0034] The user interface is designed as an intuitive multi-touch screen, simplifying interaction with the system for healthcare providers. It is engineered to display patient vital signs in real time and to make trends and fluctuations clear and easy to monitor. This makes it easier for a healthcare provider to spot trends in patient data (e.g., heart rate over the past 24 hours as a graph) and identify if something is amiss. The customizable alerts and notifications allow medical staff to set thresholds tailored to each patient's individual health profile. For example, an alert can be set to trigger when a patient's blood pressure exceeds certain limits. This strategy promotes personalized care for each patient.This requires a healthcare team to customize the parameters for their patients. However, this level of customization has another advantage: It allows healthcare teams to respond much more quickly and adapt to specific needs at any time.

[0035] In one example, the device includes a power management system designed to prioritize the device's energy efficiency. A rechargeable lithium-ion battery powers the device, providing 48 hours or more of continuous monitoring operation. Such long battery life is important for patient comfort and system acceptance, as recharging breaks occur more frequently during longer periods of play. The device also features a low-power consumption level that is activated when idle. For example, if a device detects that no one is watching it, the company says it can decide to enter a low-speed sleep mode while still being able to wake up very quickly.This energy-efficient design ensures that the device remains reliable and operational in a variety of healthcare environments, enabling 24 / 7 patient monitoring without placing undue strain on already overworked healthcare professionals. REFERENCE 100 Block diagram of a device for real-time monitoring of patient vital signs using machine learning. 102 Sensor arrangement 104 Electronic data processor 106 Data processing unit 108 Communication interface 110 User interface

Claims

[1] A device for machine learning-dependent real-time monitoring of patients’ vital functions, consisting of a sensor array for collecting physiological data, including at least one of the following: heart rate, respiratory rate, temperature, blood pressure; an electronic data processor connected to the sensor array that receives and processes the physiological data; a data processing unit including an integrated machine learning module with multiple pre-trained models for real-time detection, classification, and prediction of anomalies in the physiological data; a communication interface for sending the output data and notifications either out-of-band / offline to external devices or monitoring systems; and an interactive user interface that provides insight into real-time vital signs and alerts generated by the machine learning module, with the user interface configured to allow any medical personnel to interact with the device to adjust monitoring parameters or thresholds. [2] The device of claim 1, wherein the sensor array comprises: a photoplethysmography sensor for measuring heart rate and blood oxygen saturation; a piezoelectric sensor to monitor respiratory rate based on chest movement; a thermistor for measuring body temperature, continuous temperature display; and a non-invasive blood pressure sensor that uses oscillometric technology to provide real-time blood pressure readings. [3] The device according to claim 1, wherein the data processing unit comprises: a data storage module that stores historical vital data and metadata for future retrospective analyses and model training; and a real-time data processing engine that applies statistical techniques to remove noise from the raw sensor measurements before the data is fed to the machine learning module. [4] The device of claim 1, wherein the interactive user interface provides a connection interface with at least one of the following protocols: Bluetooth Low Energy protocol for short-range connections with the smartphone device; Wi-Fi protocol for connecting to the cloud and Remote monitoring channels; and wired connection for direct access to hospital systems and integration with other medical devices. [5] The device of claim 1, wherein the user interface comprises a multi-touch screen capable of displaying interactive, complex charts and trends of vital signs, with dynamic alerts and notifications sent to caregivers who set a specific threshold for the patient's measurements. [6] A device according to claim 1, wherein the device is powered by a rechargeable lithium-ion battery capable of operating continuously for at least 48 hours at full load, with a consumption mode activated during periods of inactivity.

Images