SK2212023U1 - A method of processing and analyzing a photoplethysmographic signal for non-invasive determination of the filling pressure of the left ventricle of the heart and a device for performing this method - Google Patents

Abstract

A method of processing and analyzing the photoplethysmographic signal for the non-invasive determination of the filling pressure of the left ventricle of the heart using a sensor of a photoplethysmographic signal connected or connectable to a device containing a processor for executing program instructions for processing the sensed photoplethysmographic signal and a user interface for displaying or interpreting the results of the processing and analyzing the photoplethysmographic signal via a user interface, where the photoplethysmographic signal is measured when the user's position changes, while this signal is processed to analyze the parameters of the photoplethysmographic signal curve by statistical methods or advanced machine learning methods, which classify the change in the parameters of the photoplethysmographic signal when the user's position changes.

Description

The technical solution relates to the method of processing and analyzing the photoplethysmographic signal for non-invasive determination of the filling pressure of the left ventricle of the heart and the device for performing this method.

Current state of the art

0.4 - 2.2% of the population in developed countries suffer from heart failure (from English Heart failure - HF) - more than 10 million patients in the EU and 6 million in the USA. 0.5-0.6 million cases are diagnosed each year with increasing incidence. Due to the frequent hospitalizations of these patients, 1-2% of the total health care expenditure is spent on HF. The cost of HF in the US alone is expected to reach at least $160 billion per year by 2030. Patients hospitalized with HF have a 50% risk of rehospitalization within 6 months, HF symptoms recur in 65% of patients within 3 months, and patients are limited by symptoms that significantly impair their quality of life. The concept of telemonitoring for patients with HF was introduced to reduce the number of decompensations and hospitalizations for HF, reduce mortality, total costs per patient, improve treatment personalization, patient quality of life and efficiency for an overburdened healthcare system.

The first telemonitoring systems were based on monitoring weight, vital signs and symptoms perceived by patients, for example through telephone checks. Although results from smaller studies have been promising, this approach has failed to demonstrate benefit in large, randomized trials, Stevenson, Lynne Warner, et al. “Remote Monitoring for Heart Failure Management at Home.” Journal of the American College of Cardiology 81.23 (2023): 2272-2291. However, a recent meta-analysis of 29 studies with 10,981 patients showed a modest but statistically significant improvement in quality of life and a reduction in hospitalizations for patients with HF, Takeda A, Martin N, Taylor RS, et al. Disease management interventions for heart failure. Cochrane Database Syst Rev. 2019;1:CD002752.s. On the other hand, symptoms in these patients are often not recognized until shortly before hospitalization, and even patients who perceive this change are often unsure of the significance of the symptoms and delay seeking medical attention, Altice NF, Madigan EA. Factors associated with delayed careseeking in hospitalized patients with heart failure. Heart Lung. 2012;41:244-254. Additionally, weight measurement has been shown to be a poor signal with significant variation due to bowel habits and caloric intake with very little sensitivity (10 to 20%) for detecting worsening HF, Adamson PB. Pathophysiology of the transition from chronic compensated and acute decompensated heart failure: new insights from continuous monitoring devices. Curr Heart Fail Rep. 2009;6(4):287-292. Several digital health care systems are currently in clinical trials that combine weight with other variables such as heart rate, blood pressure, patient-observed symptoms, or exercise data, Stevenson, Lynne Warner, et al. “Remote Monitoring for Heart Failure Management at Home.” Journal of the American College of Cardiology 81.23 (2023): 2272-2291.

Although symptoms of HF, such as edema or dyspnea, develop relatively suddenly, they are preceded by complex pathophysiological processes. These include progressive increases in cardiac filling pressures, neurohormonal and hemodynamic changes, and volume redistribution.

Several different monitoring systems based on monitoring these pathophysiological changes have been developed and studied in order to improve outcomes.

Pulmonary congestion (accumulation of fluid in the lung tissue) is the cause of one of the most significant symptoms of heart failure. Several methods are currently available to measure the level of pulmonary congestion. Most lung tissue is filled with air, which is a poor electrical conductor. The accumulation of conductive fluid in the lungs leads to a decrease in electrical impedance. Modern implantable devices for HF patients (CRT - cardioresynchronization therapy, ICD - implantable cardioverter - defibrillator, CRTD cardiac resynchronization therapy with defibrillator) can monitor changes in electrical impedance between the lead electrodes in the heart and the pulse generator on the chest wall. These changes in bioimpedance can be detected as early as 2 to 3 weeks before the HF patient is hospitalized, with a return to baseline usually a week or more after discharge. An example of such a solution is CorVue from St. Jude Medical (St. Paul, MN, USA) or OptiVol from Medtronic (Minneapolis, MN, USA). A correlation between reduced thoracic impedance and worsening of HF symptoms has been demonstrated in several small studies. However, a meta-analysis and meta-regression of several randomized trials did not show an overall benefit compared to usual management, Hajduczok AG, Muallem SN, Nudy MS, et al. Letter to the editor to update the article "Remote monitoring for heart failure using implantable devices: a systematic review, meta-analysis, and metaregression of randomized controlled trials". Heart Fail Rev. 2022;27(3):985-987.

Several studies, including REM-HF, Morgan JM, Kitt S, Gill J, et al. Remote management of heart failure using

SK 221-2023 U1 implantable electronic devices. Eur Heart J. 2017;38:2352-2360, CLEPSYDRA, Auricchio A, Gold MR, Brugada J, et al. Long-term effectiveness of the combined minute ventilation and patient activity sensors as a predictor of heart failure events: the CLEPSYDRA study. Eur J Heart Fail. 2014;16:663-670, MultiSENSE Trial, Boehmer JP, Hariharan R, Devecchi FG, et al. A multisensor algorithm predicts heart failure events in patients with implanted devices: results from the MultiSENSE study and Gardner RS, Singh JP, Stancak B, et al. Heart-Logic multisensor algorithm identifies patients during periods of significantly increased risk of heart failure events: results from the MultiSENSE study. Circ Heart Fail. 2018;11(7):e004669. J Am Coll Cardiol HF. 2017;5:216-225, SELENE-HF, D'Onofrio A, Solimene F, Calo L, et al. Combining home monitoring temporal trends from implanted defibrillators and baseline patient risk profile to predict heart failure hospitalizations: results from the SELENE HF study. Europe. 2022;24:234-244, or TRIAGE-HF, Virani SA, Sharma V, McCann M, et al. Prospective evaluation of integrated device diagnostics for heart failure management: results of the TRIAGEHFstudy. ESC Heart Fail. 2018;5:809-817 and Adamson, Philip B. "Pathophysiology of the transition from chronic compensated and acute decompensated heart failure: new insights from continuous monitoring devices." Current heart failure reports 6.4 (2009): 287-292, investigated the possibilities of improving the diagnostic properties of pulmonary impedance measurements by combining them with other physiological parameters such as heart rate, heart rate variability, activity or heart sounds. For example, the MultiSENSE study, which examined Boston Scientific's (Marlborough, MA, USA) HeartLogic CRTD-based monitoring system with the ability to measure impedance and monitor heart and respiratory rates, activity, and cardiac echoes, demonstrated a sensitivity of 70% and a specificity of 85.7% for predicting worsening of HF with a median of 34 days from notification to event. On the other hand, the REM-HF trial investigating the clinical and cost-effectiveness of remote monitoring of patients with CRT/CRT-D or ICD using HF risk scores based on thoracic impedance, activity and heart rate (nocturnal heart rate, rhythm and heart rate variability) failed to show any impact on all-cause mortality or unplanned hospitalizations of patients with HF.

Thoracic impedance can also be used to estimate left ventricular ejection volume and cardiac filling pressures based on volume changes of blood in the chest. However, this method, called impedance cardiography, showed only modest and weak correlations with cardiac output and pulmonary artery pressure and did not correlate with subsequent death or hospitalization within 6 months, Kamath SA, Drazer MH, Tassisa, et al. Correlation of impedance cardiography with invasive hemodynamic measurements in patients with advanced heart failure: the bioimpedance cardiography (BIG) substudy of the ESCAPE trial. Am Heart J. 2009;158:217-223.

Another method for assessing the level of pulmonary congestion is remote dielectric sensing (ReDS) based on the analysis of a focused electromagnetic signal with a transmitter and sensor embedded in the chest belt. This technology has been approved by the Food and Drug Administration (FDA) and has also received European CE certification for monitoring the amount of fluid in the lungs. ReDS correlated with pulmonary wedge pressure, was able to detect pulmonary edema and accurately differentiate patients with HF and pulmonary congestion (defined by chest computed tomography - CT) from normal subjects. The SMILE randomized trial of 268 patients showed a 48% reduction in HF hospitalizations during a 6-month follow-up, Abraham WT, Anker S, Burkhoff D, et al. Primary results of the sensible medical innovations lung fluid status monitor allow reducing readmission rate of heart failure patients (SMILE) trial. J Card Fail. 2019;25(11):938.

The MicroCore System from ZOLL Medical (Chelmsford, MA, USA) is a radiofrequency sensor patch that is applied to the lateral chest wall. Although data are very limited, the BMAD study demonstrated a reduction in HF hospitalizations as well as a composite outcome - reduction in HF hospitalizations, deaths, and emergency room visits, Boehmer J, et al. Impact of Heart Failure Management Using Thoracic Fluid Monitoring From a Novel Wearab/e Sensor: Results of the Benefits of Microcor (pCor™) in Ambulatory Decompensated Heart Failure (BMAD) Trial. Presented as Late Breaking Clinical Trial at the 2023 American College of Cardiology Annual Scientific Session, March 6, 2023.

Smaller study Amir, Offer, etal. "Remote speech analysis in the evaluation ofhospitalized patients with acute decompensated heart failure." Heart Failure 10.1 (2022): 41-49 on 40 adult patients demonstrated that the level of congestion can also be measured by analyzing voice changes using automated speech analysis technology. This methodology is based on the influence of changes in lung volumes and swelling of the soft tissues of the vocal tract during congestion on the spectral parameters of speech.

Another option for monitoring HF status is based on direct invasive measurement of cardiac pressures. Cardiac filling pressures were shown to be persistently elevated 3 weeks before hospitalization for HF in patients with HF with preserved ejection fraction (HFpEF) and 4 weeks before hospitalization for HF with reduced ejection fraction (HFrEF), Zile, Michael R., et al. "Transition from chronic compensated to acute decompensated heart failure: pathophysiological insights obtained from continuous monitoring of intracardiac pressures." Circulation 118.14 (2008): 1433-1441.

SK 221-2023 U1

CardioMEMS is a pulmonary artery (PA) branch implanted pressure monitoring sensor developed by Abbott (Abbott Park, IL, USA). The CHAMPION randomized trial demonstrated a 37% reduction in hospitalizations for HF, Abraham, William T., et al. "Wireless pulmonary artery haemodynamic monitoring in chronic heart failure: a randomized controlled trial. The Lancet 377.9766 (2011): 658-666. The US Food and Drug Administration (FDA)-approved GUIDE-HF study showed a significant reduction in hospitalizations for HF that was also comparable between patients with HFrEF and HFpEF, Žile, Michael R., et al. "Hemodynamically-guided management of heart failure across the ejection fraction spectrum: the GUIDE-HF trial. Heart Failure 10.12 (2022): 931-944. Other studies have shown a benefit in reducing PA pressure and HF hospitalizations at 6, 12, and 24 months, Stevenson, Lynne Warner, et al. "Remote Monitoring for Heart Failure Management at Home." Journal of the American College of Cardiology 81.23 (2023): 2272-229.

The HeartPod system developed by Savacor, later acquired by St. Jude Medical (St. Paul, MN, USA) was a small implantable device capable of monitoring left atrial (LA) pressures. Although the LAPTOP study was terminated early due to complications with insertion of the device through the atrial septum, analysis of interim results showed a reduction in hospitalizations for HF comparable to those of the CHAMPION study, Abraham, William T., et al. "Hemodynamic monitoring in advanced heart failure: results from the LAPTOP-HF trial. Journal of Cardiac Failure 22.11 (2016): 940. In addition, a meta-analysis of the LAPTOP, GUIDE-HF, and CHAMPION trials showed a significant reduction in mortality at two years.

Another direct LA pressure monitoring device called the V-LAP from Vectorious Medical Technologies (Tel Aviv, Israel) was successfully implanted in the atrial septum in 24 patients. There were no device-related complications and LA pressure correlated well with wedge pressure at 3 months, Perl L, Meerkin D, D'Amario D, et al. The V-LAP system for remote left atrial pressure monitoring of patients with heart failure: remote left atrial pressure monitoring. J Cardiac Fail. 2022;28:963-972.

A study of 33 individuals before planned cardiac catheterization by Silber et al. demonstrated that measurement of the pulse amplitude ratio of the PPG signal during the Valsalva maneuver reflects left ventricular filling pressures, Slber, Harry A., et al. Fingerphotoplethysmography during the Valsalva maneuver reflects left ventricular filling pressure. American Journal of Physiology-Heart and Circulatory Physiology 302.10 (2012): H2043H2047. These results were subsequently confirmed by other studies. This system is based on the measurement of heart pressure changes induced by the Valsalva maneuver.

Document EP 2 706 907 A2, which describes an automated device and method for non-invasive detection of heart filling pressure, is related to the mentioned system. The device includes a computer-readable medium programmed to analyze the pulse amplitude of the photoplethysmographic signal. The analysis is performed for the purpose of determining the moment in time when the pulse amplitude of the photoplethysmographic signal does not change by a predetermined value during the first predetermined time period. The method also includes displaying instructions to the user for preparing to initiate forced exhalation against resistance, so-called Valsalva maneuver. This expiratory effort performed by the user is compared to a predetermined resulting range for expiratory effort, and an indicator of the user's expiratory effort relative to the predetermined resulting range for expiratory effort is displayed on the user interface.

The mentioned solution is based on the use of the so-called Valsalva maneuver, to induce a change in pressures in the chest, which induce changes in the circulatory system. However, this maneuver is difficult to reproduce, so users cannot repeat this maneuver correctly. When exhaling forcefully against resistance, the user often shakes, which is reflected in more artifacts. With this solution, there is a need for an additional device for measuring expiratory effort in order to standardize the measurement. In the event that the Valsalva maneuver cannot be performed, this document describes a device capable of supplying the user with air under pressure.

The goal of the proposed solution is to substantially eliminate the shortcomings of the current state of the art.

The essence of the technical solution

The stated goal is achieved by processing and analyzing the photoplethysmographic signal, hereinafter referred to as the "PPG signal, for the non-invasive determination of the filling pressure of the left ventricle of the heart, according to this technical solution. The method according to the technical solution uses a PPG signal sensor connected or connectable to a device containing a processor for executing program instructions for processing and analyzing the captured PPG signal and a user interface for displaying or interpreting the results of processing and analyzing the PPG signal through the user interface, or a PPG signal sensor connected or connectable to a device containing means for transmitting the scanned PPG signal to a remote device containing a processor for executing program instructions for processing and analyzing the PPG signal, means for receiving processing results, and analysis of the PPG signal from a remote device containing

SK 221-2023 U1 processor, and a user interface for displaying or interpreting the results of PPG signal processing and analysis through the user interface.

In the method according to this technical solution, the PPG signal is detected by a PPG signal sensor or device for detecting the PPG signal during a passive or active change in the position of the user, which leads to an increased venous return of blood to the heart. The PPG signal is recorded continuously before, during and after changing the user's position, or separately in the first position of the user and separately in the second position of the user after changing his position.

The captured PPG signal is subjected to denoising and filtering to specific frequencies.

The denoised and filtered PPG signal is evaluated using one, more or a combination of signal quality indices selected from the group - asymmetry coefficient, spiciness coefficient, signal-to-noise ratio, entropy, rate of crossings through the zero axis, relative spectral power, peak value of the autocorrelation function, crossing the zero axis of the autocorrelation function, the spectral entropy, all of which are calculated on the time series as a whole, and then on individual pulses and cumulatively to determine the quality photoplethysmographic signal sufficient to determine the filling pressure of the left ventricle of the heart.

The PPG signal is processed in the following steps in the given or different order: identification of artifacts; identification of individual heartbeats; identifying parts of the PPG signal curve to obtain one, more or a combination of parameters selected from the group or groups below:

Physiological parameters - blood oxygen saturation, heart rate calculated from individual pulses or from Empirical decomposition into modes, respiratory rate calculated from individual pulses - XB123EF1FM1 algorithm or from empirical decomposition into modes;

Pulse wave characteristics - systolic peak position, systolic peak value, diastolic peak position, diastolic peak value, position of the lowest point on the wave, value of the lowest point on the wave, amplitude (value of the highest point on the wave), position of the highest point on the wave, maximum slope, systole length expressed in percentiles for percentiles [0.1, 0.25, 0.33, 0.5, 0.66], index B/A, where B is the value of the minimum of the second derivative of the PPG signal and A is the value of the maximum of the second derivative of the PPG signal, diastole length expressed in percentiles for the percentiles [0.1, 0.25, 0, 33, 0.5, 0.66], average, minimum and maximum NN intervals, minimum and maximum differences in NN intervals, standard deviation of NN intervals, average of standard deviations of all NN intervals for each 5-minute segment in a long recording, standard deviation of the averages of NN intervals for each 5-minute segment in a long recording, root mean square value of differences in NN intervals, number of NN intervals over 20, resp. 50 ms, spectral power for very low frequency up to 0.04 Hz, spectral power for low frequency 0.04 to 0.15 Hz, spectral power for high frequency 0.15 to 0.4 Hz, properties of Poincaré shape - standard deviation of small and the large base of the Poincaré ellipse, their ratio and content of the Poincaré ellipse, entropy of NN intervals, alpha value of short-term fluctuations in the detrended fluctuation analysis, alpha value of long-term fluctuations in detrended fluctuation analysis, area under the curve of one pulse;

Time series properties, catch22 - mode of z-scored distribution, histogram with 5 bins, mode of scored distribution, histogram with 10 bins, longest period of consecutive values above the average of the entire time series, time interval between consecutive extremes above the average, time interval between consecutive extremes below the average, the time of the first transition below the value of 1/e in the autocorrelation function, the first minimum of the autocorrelation functions, the total power of the lowest 5 frequencies in the Fourier spectrum, the centroid of the complete Fourier spectrum, the average error when predicting the next value of the time series from the average of the previous 3 values, the time reversibility index, automutual information - the value of the mutual information between the time series and its copy, the first minimum of the function of auto-mutual information, the ratio of differences of consecutive points that are greater than 4% of the standard deviation of the entire series, the longest period in a row successive reductions in signal values, the entropy of two consecutive letters in the 3-letter symbolization of the time series, the difference in the correlation length between the original time series and the rescaled time series, which arises as the differences between each pair of consecutive values, the exponential curve fit of consecutive of distances in the 2-dimensional nesting space, which is created by inserting points according to the time delay, ratio of slow to faster fluctuations, estimated using linear detrending in each window, where the time series itself has been downsampled to 50% of the original sampling frequency, ratio of slow to faster fluctuations, estimated using linear detrending in each window in the original time series, trace covariance matrix of transitions in 3-letter time series symbolization, periodicity index.

The PPG signal is further analyzed by statistical methods and/or advanced machine learning methods, which classify the change of the above obtained parameters, one, more or their combination, to increased or non-increased filling pressures, and/or estimate, by regression, the value of the filling pressure based on this change or changes. The classification and regression capability is obtained by training on a large volume of data from users with normal and abnormal left ventricular filling pressures.

SK 221-2023 U1

The parameters are analyzed during the entire measurement and/or individually within one and the second position of the user and their change between the first and second position of the user.

Finally, information about the filling pressure of the left ventricle of the heart, from the analyzed photoplethysmographic signal, is displayed or interpreted to the user via the user interface of the device.

This technical solution also provides a device for measuring, processing and analyzing the PPG signal for non-invasively determining the filling pressure of the left ventricle of the heart comprising a PPG signal sensor connected or connectable to a device containing a processor for executing program instructions for processing and analyzing the detected PPG signal in a manner according to this technical solution. and a user interface for displaying or interpreting the results of signal processing and analysis through said user interface.

The PPG signal sensor, the device containing the processor for executing program instructions for processing and analyzing the captured PPG signal in the manner according to this technical solution and the user interface can be physically arranged in different assemblies, as one compact device containing all the mentioned members, or as an assembly of interconnected and/ or attachable listed members.

A device for measuring, processing and analyzing a PPG signal for non-invasively determining the filling pressure of the left ventricle of the heart can also be advantageously provided in an arrangement that includes a PPG signal sensor connected or connectable to a device containing means for transmitting the PPG signal detected by the sensor to a device containing a processor for executing instructions a program for processing and analyzing a PPG signal in a manner according to this technical solution, means for receiving the results of processing and analyzing a PPG signal from a device containing a processor, and a user interface for displaying or interpreting these results of PPG signal processing and analysis.

For all possible arrangements of the device according to this technical solution, the above-mentioned processor is connected to a computer-readable medium which is programmed with the steps for carrying out the above-mentioned method and the device is adapted to carry out the above-mentioned method in the steps programmed on the computer-readable medium. The solution is based on the different change of the PPG signal in users with normal and abnormal filling pressure of the left ventricle of the heart, which allows the use of commonly available devices without the need for any specialized equipment. From the user's point of view, only a very simple and undemanding maneuver is used to induce a change in blood volumes, which is easily reproducible, without inter-individual and intra-individual variability.

Overview of images on drawings

The technical solution is also clarified with the help of images in the accompanying drawings, in which

fig. 1 graphically illustrates the effect of active repositioning on hemodynamics in a user with heart failure and normal volume status, and a user with heart failure and volume overload;

fig. 2 is an overview diagram of the measurement and processing of the PPG signal for the non-invasive determination of the filling pressure of the left ventricle of the heart, using this technical solution;

fig. 3 is a block diagram of an exemplary system for non-invasively assessing the left ventricular pressure of a user according to this technical solution;

fig. 4 shows an example of a raw PPG signal;

fig. 5 graphically illustrates an example of PPG signal processing steps;

fig. 6 graphically shows the course of the curve of the PPG signal processed according to this technical solution, before (left) and after (right) an active change of position in a user with heart failure with normal volume status;

fig. 7 graphically shows the course of the PPG signal curve processed according to this technical solution, before (left) and after (right) an active change of position in a user with heart failure with volume overload.

Implementation examples

The mentioned technical solution will be further described in detail on the implementation examples with reference to fig. 1 to 7 in the accompanying drawings.

For the purposes of this technical solution, it is possible to use any sensor 1 PPG signal, or device for measuring the PPG signal that detects or senses and records the PPG signal using either the transmittance or reflectance method. A PPG signal sensor is understood as an element whose function is only to detect a PPG signal, and a PPG signal measuring device is a complex device that includes a PPG sensor and, in addition to

SK 221-2023 PPG signal sensing U1 also has other functions such as PPG signal recording.

For the purposes of this technical solution, the user interface of the device is understood as hardware and software elements of the device enabling the interaction of the device with the user, such as e.g. display, speakers, microphone, physical or software controls, etc.

A change in the user's position that causes a hemodynamic change, or a change that leads to an increased venous return of blood to the heart is understood to include a total change in the user's position, i.e. a change in position, from standing to sitting, from standing to lying down, from sitting to lying down, a total change of position on the bed, etc., as well as a change in position, not the total one, i.e. a change in the position of, for example, only the user's limbs, e.g. raising an arm, leg, or just the torso, etc.

Changing the position of the user for the purposes of this technical solution can be either active, when the user changes the position by himself without help, or passive, when the user is positioned with external help.

In general, the effect of a change of position, for example an active change of position, on hemodynamics in a user with heart failure and normal volume status, and a user with heart failure and volume overload is graphically explained in fig. 1.

With reference to the block diagram shown in Fig. 3, an example of a system for non-invasive assessment of the user's left ventricle pressure according to this technical solution will be described.

The user puts on a measuring device, in this example a 1 PPG signal sensor. The PPG signal sensor 1 can be practically placed anywhere on the user's body, where the PPG signal can be detected. This represents a universal solution for recording the PPG signal from multiple locations.

After inserting the PPG signal sensor 1, the actual measurement of the PPG signal will begin. Advantageously, the user is informed about the start of the measurement. Information about the start of the measurement can be advantageously conveyed by written or spoken instructions, played through the user interface of the device to which the PPG sensor 1 is connected or connectable. Such a device is advantageously, for example, the smartphone 2 of the user, on which the corresponding mobile application can be run. The PPG signal sensor 1 can be connected wirelessly to the smartphone 2, e.g. via bluetooth, Wi-fi, etc., or cable.

Via the user interface of the smartphone 2, all other written or spoken instructions regarding the measurement of the PPG signal, for example regarding the time in which the user should actively change the position, end the measurement, etc., are advantageously conveyed.

After starting the measurement, the user is instructed to change the position mentioned above, which leads to an increased venous return of blood to the heart, while the PPG signal is read continuously before, during and after the change of the user's position, and/or separately in the first position of the user and separately in the second position user after changing location.

After the end of the measurement, the PPG signal obtained from said measurement is sent via smartphone 2 for processing, analysis and determination of the filling pressure of the left ventricle of the heart. Processing, analysis and determination of the filling pressure of the left ventricle of the heart in the embodiment shown in fig. 2 takes place advantageously through computing resources in the cloud 3. The relevant results are then sent to the user's smartphone 2, where they are interpreted via its user interface, i.e. e.g. typically displayed on the device's display, or communicated vocally through the device's speakers, or both.

In general, in the arrangement according to fig. 2, the system is designed so that the PPG signal sensor 1 is connectable to a device that includes a user interface. This device further includes means for transferring the scanned PPG signal to a device containing a processor for executing program instructions for processing and analyzing the PPG signal. Said device containing the user interface then also advantageously includes means for receiving the results of processing and analysis of the PPG signal from the device containing the processor, i.e. in this case from cloud 3. The user interface of the device then serves to display or interpret the results of signal processing and analysis to the user.

It is also possible, if the device with a user interface, i.e. for example the mentioned smartphone 2, has sufficient computing power, the PPG signal obtained from the sensor 1 can be processed and analyzed directly on this device with a user interface, while the results of the processing and analysis of the PPG signal are then respectively interpreted through the user interface of the device. With respect to the PPG signal sensor 1, embodiments are also possible where the PPG signal sensor 1 is connected to the device with the user interface or can be connected externally, or the PPG signal sensor 1 is directly part of the device with the user interface.

The PPG signal detected by sensor 1 is subjected to denoising and filtering to specific frequencies.

The denoised and filtered PPG signal is evaluated using one, more or a combination of signal quality indices selected from the group - asymmetry coefficient, spiciness coefficient, signal-to-noise ratio, entropy, rate of crossings through the zero axis, relative spectral power, peak value of the autocorrelation function, crossing the zero axis of the autocorrelation function, the spectral entropy, all of which are calculated on the time series as a whole, and then on individual pulses and cumulatively to determined whether the PPG signal quality

SK 221-2023 U1 sufficient for subsequent processing and analysis.

If the signal quality is not sufficient, the system will recommend the user to repeat the measurement, i.e. j. from the "START" position in fig. 3, as the results of poor quality signal processing could be incorrect - false positive or negative. Since this system does not contain any specialized equipment, and does not include maneuvers that are relatively difficult to repeat accurately, such as the Valsalva maneuver, restarting the measurement and performing the subsequent steps of the method is undemanding and unburdening to the user.

Subsequently, the processing of the PPG signal is carried out in the following steps, graphically represented in fig. 5, in the order below, or in a different order; the choice and order of the processing steps and their parameters are specific to the device - chosen by testing the optimal configuration for a specific measuring device: - identification of artifacts;

- identification of individual heartbeats;

- identification of parts of the PPG signal curve to obtain one, more or a combination of parameters selected from the group or groups below (for accuracy, some terms in brackets are assigned their English equivalent):

- Physiological parameters - blood oxygen saturation, heart rate calculated from individual pulses or from Empirical decomposition into modes, respiratory rate calculated from individual pulses - XB123EF1FM1 algorithm or from empirical decomposition into modes;

Pulse wave characteristics - systolic peak position, systolic peak value, diastolic peak position, diastolic peak value, position of the lowest point on the wave, value of the lowest point on the wave, amplitude (value of the highest point on the wave), position of the highest point on the wave (English Crest time), maximum slope, systole length expressed in percentiles for percentiles [0.1, 0.25, 0.33, 0.5, 0.66] (English Systolic length percentiles [0.1, 0.25, 0.33, 0.5, 0.66]), index B/A (English B over A) where B is the value of the minimum of the second derivative of the PPG signal and A is the value of the maximum of the second derivative of the PPG signal, diastole length expressed in percentiles for the percentiles [0.1, 0.25, 0.33, 0.5, 0.66] (diastolic length percentiles [0.1, 0.25, 0.33, 0.5, 0.66]), average, minimum and maximum NN intervals (eng. NN intervals - NN interval is the time between individual pulses, if the entire signal contains 100 pulses, we can count 99 NN intervals), minimum and maximum differences in NN intervals (eng. NN intervals difference - we calculate their difference from a series of NN intervals, if the entire signal contains 100 pulses, we have 99 NN intervals and 98 differences in NN intervals), standard deviation of NN intervals (English SDNN - standard deviation of NN interval), the average of the standard deviations of all NN intervals for each 5-minute segment in a long recording (English SDNN index), the standard deviation of the averages of NN intervals for each 5-minute segment in a long recording (English SDANN - standard deviation of the average NN interval), root mean square value of differences in NN intervals (English RMSNN - root mean square of NN interval differences), number of NN intervals over 20, resp. 50 ms, spectral power for very low frequency up to 0.04 Hz, spectral power for low frequency 0.04 to 0.15 Hz, spectral power for high frequency 0.15 to 0.4 Hz, Poincaré features of the Poincare plot) - standard deviation of the minor and major subaxis of the Poincaré ellipse, their ratio and content of the Poincaré ellipse, entropy of NN intervals (eng. sample entropy of the NN series), alpha value of the short-term fluctuations in detrended fluctuation analysis (English alpha value of the short-term fluctuations from DFA - detrended fluctuation analysis, alpha value of the long-term fluctuations in detrended fluctuation analysis -term fluctuations from DFA - detrended fluctuation analysis, area under the curve of one pulse;

Properties of the time series, catch22 - mode of z-scored distribution, histogram with 5 bins (English Mode of z-scored distribution, 5-bin histogram, mode of z-scored distribution, histogram with 10 bins , 10-bin histogram, longest period of consecutive values above the average of the entire time series, time interval between consecutive extremes above the average, time interval between successive extremes below the average, the time of the first transition below the value of 1/e in the autocorrelation function, the first minimum of the autocorrelation function, the total power of the lowest 5 frequencies in the Fourier spectrum, the centroid of the complete Fourier spectrum, the average error when predicting the next value of the time series from the average of the previous 3 values (English. Mean error from a rolling 3-sample mean forecasting), index of time reversibility - calculated as the average of the third powers of the differences of consecutive points (English: Time-reversibility statistic, automutual information (English: Automutual information) - the value of the mutual information between the time series and its copy, the first minimum of the automutual information function, the ratio of the differences of consecutive points that are greater than 4% of the standard deviation of the entire series (English Proportion of successive differences exceeding 0.04σ), the longest period of consecutive decreases in signal values, entropy of two consecutive letters in a 3-letter symbolization of the time series (Shannon entropy of two successive letters in equiprobable 3-letter symbolization), difference in the correlation length between the original time series and the rescaled time series, which will arise as the differences between each consecutive pair of continuous values (eng.

SK 221-2023 U1

Change in correlation length after iterative differentiating), exponential fit to successive distances in 2-d embedding space, which is created by temporal embedding of points according to time delay (exponential fit to successive distances in 2-d embedding space), ratio of slow fluctuations to faster ones, estimated using linear detrending in each window, where the time series itself was downsampled to 50% of the original sampling frequency (English. Proportion of slower timescale fluctuations that scale with DFA, 50% sampling), the ratio of slow fluctuations to faster ones, estimated using linear detrending in each window in the original time series (English. Proportion of slower timescale fluctuations that scale with linearly rescaled range fits), Trace of covariance of transition matrix between symbols in 3 letter alphabet), periodicity index according to Wang et al. 2007.

For completeness, the periodicity index of Wang et al. 2007 refers to Wang X, Wirth A, Wang L (2007) Structure-based statistical features and multivariate time series clustering. In: Proceedings —IEEE international conference on data mining, ICDM, pp 351-360. ISSN 15504786. https://doi. org/10.1109/ICDM. 2007.103

In fig. 5 shows the respective graphic representations of the successively processed PPG signal, where graph A represents the raw PPG signal, graph B represents the denoised signal and filtered to specific frequencies, graph C represents the signal with the detected artifact, graph D represents the signal with the artifact removed, graph E represents the signal with individual heartbeats detected, and graph F contains some selected signal parameters marked for analysis.

The PPG signal is further analyzed by statistical methods, both Bayesian and frequentist, and/or advanced machine learning methods, which classify the change of the previously obtained parameters, one, more or their combination, to increased or non-increased filling pressures, and/or estimate, by regression, the value of the filling pressure based on this change or changes. The classification and regression capability is obtained by training on a large volume of data from users with normal and abnormal left ventricular filling pressures.

The parameters are analyzed from the entire measurement and/or individually within one and the second position of the user, while a comparison between the first and second position is used to evaluate the filling pressure of the left ventricle of the heart, i.e. j. comparison between the above-mentioned PPG signal parameters obtained from the measurement of the PPG signal in the first position of the user, i.e. before the change of the user's position, and the above-mentioned parameters of the PPG signal obtained from the measurement of the PPG signal in the second position of the user, i.e. in the position after the change of the user's position?.

From the analysis of the mentioned parameters, the filling pressure of the left ventricle of the heart is determined - reduced, normal, increased or its specific value.

Information about the filling pressure of the heart's left ventricle is then displayed or otherwise interpreted, e.g. voice, to the user through the user interface. With regard to the information itself about the filling pressure of the left ventricle of the heart, this can be in different forms, e.g. as a specific numerical value, or as a relative figure, e.g. "reduced", "normal", "increased", etc.

The device for measuring, processing and analyzing the PPG signal for the non-invasive determination of the filling pressure of the left ventricle of the heart, according to this technical solution, i.e. for carrying out the method and steps described above, will in the most advantageous version for the user contain a sensor 1 of the PPG signal preferably wirelessly connectable to a smartphone 2. Sensor 1 PPG signal can also be connected to the smartphone 2 with a cable. From the smartphone 2, the PPG signal is transmitted typically via the Internet to the cloud 3, which contains computing means for the aforementioned processing and analysis of the captured PPG signal and sending the results of the processing and analysis of the PPG signal from the cloud 3 back to the user's smartphone 2. The results from the processing and analysis will then be displayed or otherwise interpreted via the user interface of the smartphone 2. The mentioned functions of the smartphone 2 will typically be provided by a dedicated mobile application.

A device for measuring, processing and analyzing the PPG signal for non-invasively determining the filling pressure of the left ventricle of the heart can be provided generally in an arrangement that includes a PPG signal sensor 1 connected or connectable to a device containing a processor for executing program instructions for processing and analyzing the detected PPG signal mentioned above in a manner according to this technical solution, and a device with a user interface for displaying or interpreting the results of signal processing and analysis through this user interface.

The sensor 1 of the PPG signal, the device containing the processor for executing program instructions for processing and analyzing the detected PPG signal in the manner according to this technical solution, and the device with the user interface can be physically arranged in different assemblies, as one compact device containing all the mentioned members, or as an assembly with each other attached and/or attachable said members.

Thus, in one embodiment, a device for measuring, processing and analyzing a PPG signal to non-invasively determine left ventricular filling pressure can be provided in an arrangement that includes a PPG sensor

SK 221-2023 U1 signal connected or connectable to a device containing means for transmitting a PPG signal detected by a sensor to a remote device containing a processor for executing program instructions for processing and analyzing a PPG signal in a manner according to this technical solution, means for receiving the results of processing and analysis of a PPG signal from a remote device comprising a processor, and a user interface for displaying or interpreting these results of PPG signal processing and analysis. Such a device is the most advantageous from the user's point of view, e.g. smartphone 2 and a remote device containing a processor, e.g. cloud 3, local PC or server, etc.

In another arrangement, a device containing a processor for executing program instructions for processing and analyzing a PPG signal in a manner according to this technical solution can also be a device that directly receives a PPG signal from sensor 1, while this device also contains a user interface for displaying or interpreting the results signal processing and analysis through this user interface, which can be, for example, a smartphone 2 itself, a tablet, a local PC, etc.

For all possible arrangements of the device according to this technical solution, the above-mentioned processor is connected to a computer-readable medium which is programmed with the steps for carrying out the above-mentioned method and the device is adapted to carry out the above-mentioned method in the steps programmed on the computer-readable medium.

Industrial applicability

The presented technical solution provides a widely available and non-invasive solution for monitoring left ventricular filling pressure in users with heart failure, both diagnosed and undiagnosed; thus, the technical solution enables diagnosing heart failure in undiagnosed users and monitoring the state of heart failure in users with already diagnosed heart failure. This technical solution can be used, for example, for home and clinical monitoring, therapy optimization, and also as a measuring device for further heart failure research in several areas, including physiology, pathophysiology, pharmacology, and clinical practice.

Claims (2)

1. A method of processing and analyzing a photoplethysmographic signal for non-invasive determination of the filling pressure of the left ventricle of the heart using a sensor (1) of a photoplethysmographic signal connected or connectable to a device containing a processor for executing program instructions for processing the detected photoplethysmographic signal and a user interface for displaying or interpreting the processing results and analysis of the photoplethysmographic signal via the user interface or using the sensor (1) of a photoplethysmographic signal connected or connectable to a device containing means for transmitting the captured photoplethysmographic signal to a remote device containing a processor for executing program instructions for processing and analyzing the photoplethysmographic signal, means for receiving the results of processing and analyzing the photoplethysmographic signal from a remote device containing a processor, and a user interface for displaying or interpreting the results of the processing and analysis of the photoplethysmographic signal through a user interface, characterized in that it includes the steps: (I) denoising and filtering the photoplethysmographic signal to specific frequencies obtained by the photoplethysmographic signal sensor or photoplethysmographic signal acquisition device during a passive or active change in the user's position that leads to increased venous return of blood to the heart, while the signal is continuously captured before, during and after changing the user's position, or separately in the first position of the user and separately in the second position of the user after the position change; (II) evaluation of the denoised and filtered photoplethysmographic signal using one, more or a combination of signal quality indices selected from the group including asymmetry coefficient, spiciness coefficient, signal-to-noise ratio, entropy, zero-crossing rate, relative spectral power, peak value of the autocorrelation function, the zero crossing of the autocorrelation function, the spectral entropy, all of which are calculated on the time series as a whole and then on the individual pulses, and cumulatively to determine the quality of the photoplethysmographic signal sufficient to determine the filling pressure of the left ventricle of the heart; (III) processing the photoplethysmographic signal, which includes the following steps in the above or different order: (a) identification of artifacts; (b) identification of individual heartbeats; (c) identification of parts of the photoplethysmographic signal curve to obtain one, more or a combination of parameters selected from the group comprising (i) physiological parameters: blood oxygen saturation, heart rate calculated from individual pulses or from empirical decomposition into modes, respiratory rate calculated from individual pulses - XB123EF1FM1 algorithm or from empirical mode decomposition, (ii) pulse wave characteristics: systolic peak position, value peak systolic, position of diastolic peak, value of diastolic peak, position of the lowest point on the wave, value of the lowest point on the wave, amplitude (value of the highest point on the wave), position of the highest point on the wave, maximum slope, systole length expressed in percentiles for percentiles [0 .1, 0.25, 0.33, 0.5, 0.66], index B/A, where B is the value minima of the second derivative of the PPG signal and A is the value of the maximum of the second derivative of the PPG signal, diastole length expressed in percentiles for percentiles [0.1, 0.25, 0.33, 0.5, 0.66], mean, minimum and maximum of NN intervals, minimum and maximum differences in NN intervals, standard deviation of NN intervals, average of standard deviations of all NN intervals for each 5-minute segment in a long recording, standard deviation of NN averages intervals for each 5-minute segment in a long recording, mean square value of differences in NN intervals, number of NN intervals over 20, resp. 50 ms, spectral power for very low frequency up to 0.04 Hz, spectral power for low frequency 0.04 to 0.15 Hz, spectral power for high frequency 0.15 to 0.4 Hz, properties of Poincaré shape - standard deviation of small and the large base of the Poincaré ellipse, their ratio and content of the Poincaré ellipse, entropy of NN intervals, alpha value of short-term fluctuations in the detrended fluctuation analysis, alpha value of long-term fluctuations in detrended fluctuation analysis, area under the curve of one pulse; (iii) properties of the time series, catch 22: mode of the scored distribution, histogram with 5 bins, mode of the z-scored distribution, histogram with 10 bins, the longest period of consecutive values above the average of the entire time series, the time interval between consecutive extremes above average, time interval between consecutive extremes below the average, time of the first transition below the value of 1/e in the autocorrelation function, first minimum of the autocorrelation function, the total power of the lowest 5 frequencies in the Fourier spectrum, the centroid of the complete Fourier spectrum, the average error when predicting the next value of the time series from the average of the previous 3 values, the time reversibility index, automutual information - the value of the mutual information between the time series and its copy, the first minimum auto-mutual information functions, the ratio of differences of consecutive points that are greater than 4% of the standard deviation of the entire series, the longest period successive reductions in signal values, the entropy of two consecutive letters in the 3-letter symbolization of the time series, the difference in the correlation length between the original time series and the rescaled time series, which arises as the differences between each pair of consecutive values, the exponential fit of the curve for consecutive distances in 2 SK 221-2023 U1 dimensional nesting space, which is created by temporally embedding points according to the time delay, the ratio of slow fluctuations to faster ones, estimated using linear detrending in each window, where the time series itself was downsampled to 50% of the original sampling frequency, the ratio of slow fluctuations to faster ones , estimated using linear detrending in each window in the original time series, the covariance trace matrix of transitions in 3-letter time series symbolization, periodicity index; (IV) analysis of the photoplethysmographic signal by statistical methods and/or advanced machine learning methods, which classify the change of the parameters obtained above, one, more or their combination, to increased or non-increased filling pressures, and/or estimate, by regression, the value of the filling pressure based on this changes or changes, while the ability of classification and regression is obtained by training on a large volume of data from users with normal and abnormal filling pressures of the left ventricle of the heart, while the parameters are analyzed from the entire measurement and/or individually within one and two positions of the user, while a comparison between the first and second position (V) is used to evaluate the filling pressure of the left ventricle of the heart, displaying or interpreting information about the filling pressure of the left ventricle of the heart, from the analyzed photoplethysmographic signal to the user through the user interface of the device. 2. Device for carrying out the method according to claim 1, characterized in that it contains a photoplethysmographic signal sensor (1) connected or connectable to a device containing a processor for executing program instructions for processing and analyzing the captured photoplethysmographic signal and a user interface for displaying or interpreting the processing results and analysis of the photoplethysmographic signal through the user interface, or a sensor (1) of the photoplethysmographic signal connected or connectable to a device containing means for transmission of the sensed photoplethysmographic signal to a remote device including a processor for executing program instructions for processing and analyzing the photoplethysmographic signal, means for receiving the results of processing and analyzing the photoplethysmographic signal from the remote device including the processor, and a user interface for displaying or interpreting the results of processing and analyzing the photoplethysmographic signal through a user interface interface.