System and method for respiratory condition monitoring

Abstract

The present invention relates to the field of physiological monitoring and health analytics, and more particularly to systems and methods for detecting and predicting respiratory conditions, such as exacerbations of a chronic obstructive pulmonary disease (COPD), using wearable sensor data and advanced signal analysis. As can be seen from the provided description, the claimed invention represents a system and method for respiratory condition monitoring that improve the relevant technological field by enabling continuous, multi-signal analysis to detect early signs of respiratory deterioration. By integrating dynamic physiological coupling metrics through advanced signal processing techniques and machine learning algorithms, the suggested system becomes capable of extracting clinically relevant information from physiological data - thereby increasing the reliability of exacerbation detection and enabling early warnings of potential respiratory decline.

Description

SYSTEM AND METHOD FOR RESPIRATORY CONDITION MONITORINGCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of priority of U.S. Provisional Patent Application No. 63 / 673,756, filed July 21, 2024, titled “System and Method for Evaluating Lung Related Conditions”, the contents of which are all incorporated herein by reference in their entirety.FIELD OF THE INVENTION

[0002] The present invention relates to the field of physiological monitoring and health analytics, and more particularly to systems and methods for detecting and predicting respiratory conditions, such as exacerbations of a chronic obstructive pulmonary disease (COPD), using wearable sensor data and advanced signal analysis.BACKGROUND OF THE INVENTION

[0003] Chronic obstructive pulmonary disease (COPD) is a progressive respiratory disorder characterized by airflow limitation and reduced pulmonary function. It is a leading cause of morbidity and mortality worldwide, with exacerbations contributing significantly to hospitalizations, healthcare costs, and patient deterioration. Despite its clinical significance, early detection of COPD exacerbations remains a major challenge in respiratory medicine.

[0004] Traditional approaches to monitoring COPD rely heavily on patient-reported symptoms, spirometry, or periodic clinical assessments. These methods are often reactive rather than proactive and lack the sensitivity to detect subtle physiological changes that precede exacerbations. Moreover, existing remote monitoring solutions typically focus on isolated parameters such as heart rate, oxygen saturation, or respiratory rate, without accounting for the complex interplay between different physiological systems.

[0005] Recent research has explored the synchronization between periodic physiological rhythms in humans - specifically, the coupling between breathing and walking patterns. This phenomenon, known as Locomotor-Respiratory Coupling, reflects the natural alignment between respiratory cycles and locomotor activity during movement. Studies have shown that this coupling is often disrupted in individuals with respiratory conditions such as COPD, and that the degree of misalignment may serve as an early indicator of physiological deterioration. These findings suggest that analyzing the dynamic relationship between respiratory and locomotor signals may provide valuable insights into respiratory health and disease progression.

[0006] It is further known in the art that electrocardiogram (ECG) signals may carry information indicative of respiratory activity. Techniques such as ECG-Derived Respiration (EDR) have been developed to extract respiratory dynamics from ECG data, leveraging physiological interactions between cardiac and respiratory systems. These methods have demonstrated the feasibility of indirectly assessing respiratory conditions without requiring dedicated respiratory sensors.

[0007] Recent advancements in wearable technology and biosensors have opened new possibilities for continuous monitoring of physiological signals. These technologies can potentially provide more comprehensive and real-time data on a patient's respiratory status. However, interpreting this wealth of data to derive meaningful insights about a patient's condition remains a complex task.

[0008] There is a growing interest in developing more sophisticated monitoring systems that can integrate multiple physiological signals and provide early warnings of potential respiratory deterioration.SUMMARY OF THE INVENTION

[0009] Accordingly, there is a need for a system and method for respiratory condition monitoring that would improve the relevant technological field by enabling continuous, multi-signal analysis to detect early signs of respiratory deterioration. By integrating dynamic physiological coupling metrics through advanced signal processing techniques and machine learning algorithms, such a system would be capable of extracting clinically relevant information from physiological data - thereby increasing the reliability of exacerbation detection and enabling early warnings of potential respiratory decline. These capabilities could significantly improve patient care by supporting timely interventions and reducing the frequency and severity of exacerbations. Furthermore, by leveraging indirect respiratory indicators such as ECG-derived and locomotor data, the system may reduce sensor complexity while maintaining diagnostic accuracy.

[0010] To address the aforementioned needs, the following is suggested.

[0011] In the general aspect, the present invention may be directed to a system for monitoring a respiratory condition of a subject. The system includes at least one non-transitory memory device, wherein modules of instruction code are stored; and at least one processor associated with said at least one memory device and configured to execute the modules of instruction code, whereupon execution of said modules of instruction code, the at least one processor may be configured to: receive, from one or more sensors, physiological signals of the subject, the physiological signals comprising: (i) an electrocardiogram (ECG) signal representing cardiac electrical activity of the subject; and (ii) a locomotion signal representing the subject’s physical movement; compute a setof recurrence quantification analysis (RQA) features, based on the ECG signal and the locomotion signal; evaluate one or more changes in the RQA features over time; and issue an alert indicative of a potential exacerbation of the respiratory condition based on the evaluated one or more changes in the RQA features.

[0012] In some embodiments, the processor may be configured to compute the set of RQA features as a set of cross-recurrence quantification analysis (CRQA) features by generating a crossrecurrence plot (CRP) based on the ECG signal and the locomotion signal, each reconstructed in a common phase space, and extracting the CRQA features as numerical features from the CRP that quantify temporal patterns of recurrence and similarity between the ECG signal and the locomotion signal.

[0013] In some embodiments, the set of CRQA features may include features representing diagonal structures of the CRP, selected from the list comprising: a determinism percentage score (%DET), representing a proportion of recurrent points that form diagonal lines in the CRP; a mean diagonal line length in the CRP, representing an average duration of coupled behavior between the ECG signal and the locomotion signal; a maximum diagonal line length in the CRP, representing a longest continuous period of coupled behavior, and an entropy of diagonal line length, quantifying diversity of diagonal structures in the CRP.

[0014] In some embodiments, the set of CRQA features may include features representing Vertical or Horizontal (V / H) structures of the CRP, selected from the list comprising: a laminarity value, representing a proportion of recurrent points that form V / H line structures in the CRP; a mean, non-diagonal line length, representing an average duration in which one of the ECG signal and the locomotion signal persists in a specific state concurrently while the other of the ECG signal and the locomotion signal experiences recurrence; a maximum V / H line length, representing a longest duration in which one of the ECG signal and the locomotion signal persists in a specific state concurrently while the other of the ECG signal and the locomotion signal experiences recurrence; and an entropy of V / H line lengths, quantifying diversity of V / H structures in the CRP.

[0015] In some embodiments, the set of CRQA features may include a weighted columns entropy, quantifying diversity of proximity profiles of the ECG signal and the locomotion signal across temporal points.

[0016] In some embodiments, the at least on processor may be configured to: compute said set of the CRQA features over one or more first periods of observation and one or more second periods of observation, said one or more first periods preceding said one or more second periods. The processor may be further configured to evaluate one or more changes in the CRQA features overtime by: applying, for each of the one or more second periods of observation, an anomaly detection machine-learning (ML)-based model on the respective set of CRQA features, said anomaly detection ML-based model being trained based on sets of CRQA features computed for said one or more first periods of observation, and being configured to determine, as an output, whether the respective second period is anomalous with respect to the one or more first periods; and evaluating said one or more changes in the CRQA features based on the output of the anomaly detection ML- based model.

[0017] In some embodiments, said at least one processor may be configured to apply, for each of the one or more second periods of observation, the anomaly detection ML-based model by: for each of the one or more second periods of observation: (a) projecting said computed set of the CRQA features from an original feature space into a latent feature space derived by applying Principal Component Analysis (PCA) to a training dataset comprising training samples each representing the set of CRQA features computed for said one or more first periods of observation; (b) reconstructing the projected set of the CRQA features into the original feature space; and (c) computing a reconstruction error, representing a measure of deviation between the computed set of the CRQA features and the respective reconstructed projected set of the CRQA features. The processor may be further configured to evaluate said one or more changes in the CRQA features based on the reconstruction error computed for said one or more second periods.

[0018] In some embodiments, the processor may evaluate said one or more changes in the CRQA features based on the reconstruction error by determining a novelty score for a target period of observation comprising said one or more second periods of observation. The novelty score may be a statistical metric derived from the distribution of values of said reconstruction error across said one or more second periods, and may be indicative of the overall magnitude or variability of said values of the reconstruction error. The processor may be further configured to issue the alert indicative of the potential exacerbation, based on the determined novelty score.

[0019] In some embodiments, the processor may compute the set of CRQA features in a continuous manner and evaluate changes in the CRQA features over time across successive periods of observation defined by a sliding time window of predetermined duration.

[0020] In some embodiments, the processor may be further configured to compute, based on the ECG signal, one or more ECG-derived metrics indicative of cardiac characteristics of the subject, evaluate changes in these metrics over time, and issue the alert based on both the changes in CRQA features and the changes in ECG-derived metrics.

[0021] In some embodiments, the ECG-derived metrics may include one or more ST interval morphology metrics selected from the group consisting of: (i) a ST interval gradient metric, computed from gradient values calculated between the maximum and minimum points within the ST interval of each R-R interval during a third predefined period of observation, using one or more statistical aggregation functions; (ii) a ST interval min-to-median amplitude difference metric, computed from amplitude differences within the ST interval of each R-R interval during a fourth predefined period of observation; and (iii) a ST interval morphology classification metric, computed by: (a) applying a trained neural network to classify ST intervals of each R-R interval during a fifth predefined period of observation into predefined morphology classes; and (b) determining statistical distribution of the ST intervals across the predefined morphology classes.

[0022] In some embodiments, the ECG-derived metrics may include an ECG stability index, computed by applying one or more statistical aggregation functions to standard deviations of segments of the ECG signal over a sixth predefined period of observation.

[0023] In some embodiments, the one or more ECG-derived metrics may include a heart rate metric, computed by applying one or more statistical aggregation functions to a plurality of heart rate measurements derived from the ECG signal over a seventh predefined period of observation.

[0024] In some embodiments, the processor may compute the one or more ECG-derived metrics over one or more first periods of observation and one or more second periods of observation, said one or more first periods preceding said one or more second periods. The processor may be further configured to evaluate said one or more changes in the ECG-derived metrics over time by, for each of the one or more second periods of observation, applying a Local Outlier Factor (LOF) algorithm trained on a dataset comprising samples of the ECG-derived metrics computed for said one or more first periods of observation, to determine whether the ECG-derived metrics for the second period represent an outlier relative to the baseline distribution established from the first periods. The processor may be further configured to issue the alert indicative of the potential exacerbation, based on determining whether the ECG-derived metrics for the second period represent an outlier.

[0025] In some embodiments, the processor may evaluate said one or more changes in the ECG- derived metrics by determining an anomaly score for a target period of observation comprising said one or more second periods of observation. The anomaly score may be a statistical metric derived from the proportion of second periods determined as outliers relative to the total number of said one or more second periods. The processor may be further configured to issue the alert indicative of the potential exacerbation, further based on the determined anomaly score.

[0026] In another general aspect, the present invention may be directed to a system for monitoring a respiratory condition of a subject. The system includes at least one non-transitory memory device, wherein modules of instruction code are stored; and at least one processor associated with said at least one memory device and configured to execute the modules of instruction code, whereupon execution of said modules of instruction code, the at least one processor is configured to: receive, from one or more sensors, physiological signals of the subject, the physiological signals comprising an electrocardiogram (ECG) signal representing cardiac electrical activity of the subject; compute, based on the ECG signal, one or more ST interval morphology metrics; evaluate one or more changes in the one or more ST interval morphology metrics over time; and issue an alert indicative of a potential exacerbation of the respiratory condition based on the evaluated one or more changes in the one or more ST interval morphology metrics. The ST interval morphology metrics may be selected from the group consisting of: (i) a ST interval gradient metric, computed from gradient values calculated between the maximum and minimum points within the ST interval of each R-R interval during a first predefined period of observation, using one or more statistical aggregation functions; (ii) a ST interval min-to-median amplitude difference metric, computed from amplitude differences within the ST interval of each R-R interval during a second predefined period of observation; and (iii) a ST interval morphology classification metric, computed by: (a) applying a trained neural network to classify ST intervals of each R-R interval during a third predefined period of observation into predefined morphology classes; and (b) determining statistical distribution of the ST interval across the predefined morphology classes.

[0027] In yet another general aspect, the present invention may be directed to a method for monitoring a respiratory condition of a subject. The method includes acquiring physiological signals from the subject using one or more sensors, the signals comprising an electrocardiogram (ECG) signal representing cardiac electrical activity of the subject and a locomotion signal representing the subject's physical movement. The method further includes computing a set of recurrence quantification analysis (RQA) features based on the ECG signal and the locomotion signal, evaluating one or more changes in the RQA features over time, and issuing an alert indicative of a potential exacerbation of the respiratory condition based on the evaluated changes in the RQA features.

[0028] In some embodiments, computing the set of RQA features may include generating a crossrecurrence plot (CRP) based on the ECG signal and the locomotion signal, each reconstructed in a common phase space, and extracting said set of RQA features as a set of cross-recurrence quantification analysis (CRQA) features, representing numerical features from the CRP thatquantify temporal patterns of recurrence and similarity between the ECG signal and the locomotion signal.

[0029] In some embodiments, the set of CRQA features may include features representing diagonal structures of the CRP, selected from the list comprising: a determinism percentage score (%DET), representing a proportion of recurrent points that form diagonal lines in the CRP; a mean diagonal line length in the CRP, representing an average duration of coupled behavior between the ECG signal and the locomotion signal; a maximum diagonal line length in the CRP, representing a longest continuous period of coupled behavior; and an entropy of diagonal line length, quantifying diversity of diagonal structures in the CRP.

[0030] In some embodiments, the set of CRQA features may include features representing Vertical or Horizontal (V / H) structures of the CRP, selected from the list comprising: a laminarity value, representing a proportion of recurrent points that form V / H line structures in the CRP; a mean, non-diagonal line length, representing an average duration in which one of the ECG signal and the locomotion signal persists in a specific state concurrently while the other of the ECG signal and the locomotion signal experiences recurrence; a maximum V / H line length, representing a longest duration in which one of the ECG signal and the locomotion signal persists in a specific state concurrently while the other of the ECG signal and the locomotion signal experiences recurrence; and an entropy of V / H line lengths, quantifying diversity of V / H structures in the CRP.

[0031] In some embodiments, the set of CRQA features may include a weighted columns entropy, quantifying diversity of proximity profiles of the ECG signal and the locomotion signal across temporal points.BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The subject matter regarded as the invention is particularly pointed out and distinctly claimed in the concluding portion of the specification. The invention, however, both as to organization and method of operation, together with objects, features, and advantages thereof, may best be understood by reference to the following detailed description when read with the accompanying drawings in which:

[0033] Fig. 1 is a block diagram providing a general representation of a system for monitoring a respiratory condition of a subject, according to some embodiments;

[0034] Fig. 2 is a block diagram providing a detailed representation of the system monitoring a respiratory condition of a subject, according to some embodiments;

[0035] Fig. 3 is an example of comparative analysis of physiological coupling between a COPD patient (left) and a healthy individual (right), based on cross-recurrence quantification analysis (CRQA), according to some embodiments of the present invention;

[0036] Fig. 4 is a schematic representation of an ECG signal, according to some embodiments of the present invention; and

[0037] Fig. 5 is a flow diagram, depicting a method for monitoring a respiratory condition of a subject, according to some embodiments.

[0038] It will be appreciated that for simplicity and clarity of illustration, elements shown in the figures have not necessarily been drawn to scale. For example, the dimensions of some of the elements may be exaggerated relative to other elements for clarity. Further, where considered appropriate, reference numerals may be repeated among the figures to indicate corresponding or analogous elements.DETAILED DESCRIPTION OF THE PRESENT INVENTION

[0039] One skilled in the art will realize the invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The foregoing embodiments are therefore to be considered in all respects illustrative rather than limiting of the invention described herein. Scope of the invention is thus indicated by the appended claims, rather than by the foregoing description, and all changes that come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

[0040] In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be understood by those skilled in the art that the present invention may be practiced without these specific details. In other instances, well-known methods, procedures, and components have not been described in detail so as not to obscure the present invention. Some features or elements described with respect to one embodiment may be combined with features or elements described with respect to other embodiments. For the sake of clarity, discussion of same or similar features or elements may not be repeated.

[0041] Although embodiments of the invention are not limited in this regard, discussions utilizing terms such as, for example, “processing,” “computing,” “calculating,” “determining,” “establishing”, “analyzing”, “checking”, “choosing”, “selecting”, “omitting”, “training”, “applying”, “forming”, “projecting”, “evaluating”, “issuing”, “generating”, “extracting”, “reconstructing”, or the like, may refer to operation(s) and / or process(es) of a computer, a computing platform, a computing system, or other electronic computing device, that manipulatesand / or transforms data represented as physical (e.g., electronic) quantities within the computer’s registers and / or memories into other data similarly represented as physical quantities within the computer’s registers and / or memories or other information non-transitory storage medium that may store instructions to perform operations and / or processes.

[0042] Although embodiments of the invention are not limited in this regard, the terms “plurality” and “a plurality” as used herein may include, for example, “multiple” or “two or more”. The terms “plurality” or “a plurality” may be used throughout the specification to describe two or more components, devices, elements, units, parameters, or the like. The term “set” when used herein may include one or more items.

[0043] In embodiments of the present invention, some steps of the claimed method may be performed using machine-learning (ML)-based models or may include actions performed on ML- based models. ML-based models may be configured or “trained” for a specific task, e.g., classification or regression.

[0044] In some embodiments, ML-based models may be artificial neural networks (ANN).

[0045] A neural network (NN) or an artificial neural network (ANN), e.g., a neural network implementing a machine learning (ML) or artificial intelligence (Al) function, may refer to an information processing paradigm that may include nodes, referred to as neurons, organized into layers, with links between the neurons. The links may transfer signals between neurons and may be associated with weights. A NN may be configured or trained for a specific task, e.g., pattern recognition or classification. Training a NN for the specific task may involve adjusting these weights based on examples. Each neuron of an intermediate or last layer may receive an input signal, e.g., a weighted sum of output signals from other neurons, and may process the input signal using a linear or nonlinear function (e.g., an activation function). The results of the input and intermediate layers may be transferred to other neurons and the results of the output layer may be provided as the output of the NN. Typically, the neurons and links within a NN are represented by mathematical constructs, such as activation functions and matrices of data elements and weights. A processor, e.g., CPUs or graphics processing units (GPUs), or a dedicated hardware device may perform the relevant calculations.

[0046] In some embodiments, ML-based models may be linear regression models. Linear regression is a statistical model that may be used for predicting continuous numerical outcomes based on one or more predictor variables. The linear regression model estimates the relationship between the dependent variable and the independent variables by fitting a linear equation to observed data. The model may be trained by adjusting the coefficients of the predictor variables tominimize the difference between the predicted values and the actual outcomes, typically using optimization techniques such as least squares or gradient descent. During training, the model iteratively updates its coefficients to reduce a loss function, such as mean squared error, which quantifies the prediction error across the training dataset. The training process continues until the model converges to a set of coefficients that best fit the data. A processor, such as a central processing unit (CPU), graphics processing unit (GPU), or a dedicated hardware device, may perform the relevant calculations.

[0047] In some embodiments, ML-based models may include dimensionality reduction techniques such as Principal Component Analysis (PCA). PCA is a statistical method used to transform a high-dimensional dataset into a lower-dimensional space while preserving as much variance as possible. It achieves this by identifying orthogonal axes - principal components - that capture the directions of maximum variance in the data. In the context of anomaly detection, PCA may be used to project input features into a latent space and reconstruct them back into the original space, with the reconstruction error serving as a measure of deviation. The model may be trained on baseline data representing normal physiological states, and deviations from this baseline may be quantified using statistical metrics derived from reconstruction errors. A processor, such as a central processing unit (CPU), graphics processing unit (GPU), or a dedicated hardware device, may perform the relevant matrix operations and eigenvalue computations.

[0048] In some embodiments, ML-based models may include density-based anomaly detection algorithms such as the Local Outlier Factor (LOF). LOF is a method that identifies anomalous data points by comparing the local density of each point to that of its neighbors. A data point is considered an outlier if its local density is significantly lower than that of surrounding points. The LOF algorithm may be trained on a baseline dataset representing typical physiological patterns, and subsequently applied to new data to assess whether it deviates from the learned distribution. The sensitivity of the LOF model may be tuned using parameters such as the number of neighbors (k) and contamination factor (c), which influence the granularity and threshold of outlier detection. A processor, such as a CPU, GPU, or dedicated hardware device, may execute the necessary computations to evaluate local densities and outlier scores.

[0049] It shall be understood that, depending on specific embodiments of the present invention, other known architectures of ML-based models may be used herein.

[0050] It should be obvious for the one ordinarily skilled in the art that various ML-based models can be implemented without departing from the essence of the present invention. It should also be understood that in some embodiments ML-based model may be a single ML-based model or a set(ensemble) of ML-based models realizing as a whole the same function as a single one. Hence, in view of the scope of the present invention, the abovementioned variants should be considered equivalent.

[0051] In various sections of the present disclosure, reference is made to temporal designations such as “first period,” “second period,” “third period,” and so forth. Unless explicitly defined otherwise within a specific embodiment, these terms are used for illustrative convenience and do not necessarily imply distinct or sequential time intervals. Depending on the implementation, such periods may refer to durations that are identical, similar, overlapping, or entirely distinct. Likewise, the sequence of these periods may vary: they may occur one after another, coincide in time, or follow any other temporal arrangement suitable for the embodiment. Accordingly, the use of these terms shall not be constmed as limiting the scope of the invention, and variations in the temporal structuring of data analysis or signal processing are considered within the purview of the disclosed embodiments.

[0052] It shall be understood that the analytical operations described throughout this disclosure (such as the computation of cross-recurrence quantification analysis (CRQA) features, principal component analysis (PCA), local outlier factor (LOF) scoring, and neural network-based classification of ECG segments) are computationally intensive and cannot be feasibly performed manually or with pen-and-paper methods. These operations involve high-dimensional data processing, matrix transformations, iterative optimization procedures, and statistical modeling techniques that require significant computational resources.

[0053] In particular, the system may be required to process large volumes of time-series physiological data (e.g., ECG and accelerometer signals sampled continuously over 24-hour periods), apply dimensionality reduction techniques, and execute machine learning models trained on individualized datasets. These tasks often involve parallel processing, floating-point arithmetic, and memory-intensive operations that are only practical when executed by a digital processing device, such as a microprocessor, digital signal processor (DSP), or graphics processing unit (GPU).

[0054] Accordingly, in all embodiments, the described methods and algorithms are intended to be implemented by a computing device - either as part of a distributed system or within a single wearable unit. The use of automated, processor-executed computation is not merely a convenience but a functional necessity for the invention to operate as intended.

[0055] Unless explicitly stated, the method embodiments described herein are not constrained to a particular order or sequence. Additionally, some of the described method embodiments orelements thereof can occur or be performed simultaneously, at the same point in time, concurrently, or iteratively and repeatedly.

[0056] The present invention addresses the critical need for early detection of exacerbations in chronic respiratory conditions such as COPD. As outlined in the background, conventional monitoring approaches are often reactive and limited in scope, relying on subjective symptom reporting or isolated physiological parameters. These methods fail to capture the complex, dynamic changes that precede clinical deterioration. While wearable technologies have enabled continuous physiological monitoring, the challenge remains in translating raw data into actionable clinical insights.

[0057] The invention provides a novel system and method that operationalizes recent research findings on physiological coupling - specifically, the synchronization between respiratory and locomotor rhythms, known as Locomotor-Respiratory Coupling. This coupling, which reflects the natural alignment between breathing and movement, has been shown to degrade in individuals with COPD and may serve as an early marker of disease progression. Furthermore, it is known that ECG signals can indirectly reflect respiratory activity, offering a practical means of assessing respiratory dynamics without requiring dedicated respiratory sensors.

[0058] Building on these insights, the inventors conducted extensive research to characterize the correlation between ECG and locomotion signals in both healthy individuals and COPD patients across varying levels of disease severity. This research revealed distinct patterns of coupling disruption that correlate with disease progression and exacerbation risk. These findings underpin the present invention and inform the design of the system’s analytical framework.

[0059] Specifically, the invention implements a set of recurrence quantification analysis (RQA) features that capture the dynamic relationship between ECG and locomotor signals. Depending on the specific embodiment, the RQA features may include cross-recurrence quantification analysis (CRQA) features - computed through the generation of a cross-recurrence plot (CRP) - or features derived using alternative techniques, such as those based on a joint recurrence plot (JRP). These features - derived from diagonal, vertical / horizontal, and column structures in cross-recurrence plots - quantify aspects such as determinism, laminarity, and entropy, which are indicative of physiological stability or disruption. In addition, the system incorporates ECG-only derived metrics, including ST interval morphology and stability indices, which have been identified as particularly informative for detecting early signs of respiratory deterioration.

[0060] Cross-Recurrence Quantification Analysis (CRQA) is a known computational technique used to analyze the dynamic relationship between two time-series signals. It is particularlyeffective for identifying patterns of synchronization, coupling, or shared dynamics between systems that evolve over time. CRQA operates by reconstructing the phase space of each signal - a mathematical representation of the system’s state over time - allowing for the comparison of their trajectories in a common multidimensional space. The phase space is typically reconstructed using time-delay embedding, where each point in the space represents a vector of delayed signal values. For example, the phase space of an ECG signal may be reconstructed by taking successive vectors of voltage values sampled at fixed time delays, capturing the underlying dynamics of cardiac activity. Similarly, for a locomotion signal such as acceleration data from a wearable sensor, the phase space may be reconstructed from the magnitude of acceleration over time, capturing the periodicity and variability of movement. Once both signals are embedded in their respective phase spaces, a cross-recurrence plot (CRP) is generated to visualize and quantify the temporal alignment of similar states across the two signals. From this plot, a variety of numerical features - such as determinism, laminarity, entropy, and line lengths - can be extracted to characterize the nature and stability of the coupling. In the context of the present invention, CRQA serves as a core analytical tool for quantifying the interaction between ECG and locomotor signals, enabling the system to detect subtle disruptions in physiological coupling that may indicate early signs of respiratory deterioration.

[0061] The invention provides a system that receives ECG and locomotion signals, computes CRQA features, evaluates changes in these features over time, and issues alerts indicative of potential exacerbations. This approach enables continuous, non-invasive monitoring that is sensitive to early physiological changes and integrates multiple signal sources into a unified analytical model. By leveraging advanced signal processing and machine learning techniques - such as principal component analysis (PCA) and local outlier factor (LOF) algorithms - the system can robustly detect deviations from a subject’s baseline state, even in the presence of noise or variability.

[0062] In this way, the invention translates foundational physiological research into a practical, clinically applicable solution. It improves upon the prior art by enabling earlier and more reliable detection of respiratory deterioration, supporting timely intervention, and reducing the burden on patients and healthcare systems.

[0063] Reference is now made to Fig. 1, which is a block diagram providing a general representation of system 100 for monitoring a respiratory condition of a subject, according to some embodiments.

[0064] In some embodiments, system 100 may include computing device 1.

[0065] Computing device 1 may include a processor or controller 2 that may be, for example, a central processing unit (CPU) processor, a chip or any suitable computing or computational device, an operating system 3, a memory device 4, instruction code 5, a storage system 6, input devices 7 and output devices 8. Processor 2 (or one or more controllers or processors, possibly across multiple units or devices) may be configured to carry out methods described herein, and / or to execute or act as the various modules, units, etc. More than one computing device 1 may be included in, and one or more computing devices 1 may act as the components of, system 100 according to embodiments of the invention.

[0066] In some embodiments, system 100 may be implemented as a cloud-based computing solution. In such cases, computing device 1 may not include local input or output devices, and may instead be configured to receive physiological signals - such as those acquired from sensors - via a network connection, such as the Internet. Accordingly, references throughout this disclosure to the processor being “configured to receive signals from input devices” shall be understood to encompass both direct and indirect connections. For example, the processor may receive data through one or more intermediary systems, including remote gateways, mobile devices, or cloudbased storage services. Furthermore, in certain embodiments, the physiological data acquired from sensors may first be stored in an external memory source (e.g., storage system 6) and subsequently transferred to the processor for analysis. This architectural flexibility allows the system to support both local and distributed implementations, including wearable device configurations and centralized cloud processing environments. Accordingly, the present invention shall not be considered limited in this regard.

[0067] Operating system 3 may be or may include any code segment (e.g., one similar to instruction code 5 described herein) designed and / or configured to perform tasks involving coordination, scheduling, arbitration, supervising, controlling or otherwise managing operation of computing device 1, for example, scheduling execution of software programs or tasks or enabling software programs or other modules or units to communicate. Operating system 3 may be a commercial operating system. It is noted that an operating system 3 may be an optional component, e.g., in some embodiments, a system may include a computing device that does not require or include an operating system 3.

[0068] Memory device 4 may be or may include, for example, a Random-Access Memory (RAM), a read only memory (ROM), a Dynamic RAM (DRAM), a Synchronous DRAM (SDRAM), a double data rate (DDR) memory chip, a Flash memory, a volatile memory, a non-volatile memory, a cache memory, a buffer, a short-term memory unit, a long-term memory unit, or othersuitable memory units or storage units. Memory device 4 may be or may include a plurality of possibly different memory units. Memory device 4 may be a computer or processor non-transitory readable medium, or a computer non-transitory storage medium, e.g., a RAM. In one embodiment, a non-transitory storage medium such as memory device 4, a hard disk drive, another storage device, etc. may store instructions or code which when executed by a processor may cause the processor to carry out methods as described herein.

[0069] Instruction code 5 may be any executable code, e.g., an application, a program, a process, task, or script. Instruction code 5 may be executed by processor or controller 2 possibly under control of operating system 3. For example, instruction code 5 may be a standalone application or an API module that may be configured to acquire, using one or more sensors, physiological signals from the subject, the physiological signals comprising an electrocardiogram (ECG) signal, and a locomotion signal; computing a set of cross-recurrence quantification analysis (CRQA) features based on the ECG signal and the locomotion signal; evaluating one or more changes in the CRQA features over time; and issuing an alert indicative of a potential exacerbation of the respiratory condition based on the evaluated one or more changes in the CRQA features, as described herein. Although, for the sake of clarity, a single item of instruction code 5 is shown in Fig. 2, a system according to some embodiments of the invention may include a plurality of executable code segments or modules similar to instruction code 5 that may be loaded into memory device 4 and cause processor 2 to carry out methods described herein. It shall further be understood that, in some embodiments, the modules of instruction code 5 may be distributed across multiple computing devices and executed independently by different processors 2. For example, one module of instruction code 5 may be executed by processor 2 of a wearable computing device, such as a mobile phone or smartwatch, while another module may be executed by processor 2 of a remote server implementing a cloud-based service. This distributed architecture allows for flexible deployment of the system across local and remote environments, enabling efficient processing and data sharing between devices.

[0070] Storage system 6 may be or may include, for example, a flash memory as known in the art, a memory that is internal to, or embedded in, a micro controller or chip as known in the art, a hard disk drive, a CD-Recordable (CD-R) drive, a Blu-ray disk (BD), a universal serial bus (USB) device or other suitable removable and / or fixed storage unit. Various types of input and output data may be stored in storage system 6 and may be loaded from storage system 6 into memory device 4 where it may be processed by processor or controller 2. In some embodiments, some of the components shown in Fig. 2 may be omitted. For example, memory device 4 may be a non-volatile memory having the storage capacity of storage system 6. Accordingly, although shown as a separate component, storage system 6 may be embedded or included in memory device 4.

[0071] Input devices 7 may be or may include any suitable input devices, components, or systems, e.g., wearable ECG monitors, motion controllers, accelerometers (e.g., Tri-axial Microelectromechanical Systems (MEMS) Accelerometers) etc. In some embodiments, input devices 7 may further include additional physiological sensors, such as heart rate monitors, skin conductance sensors, peripheral capillary oxygen saturation (SpCh) sensors etc., to provide more comprehensive data and to further analyze it for detecting COPD exacerbation. Input devices 7 may also include, e.g., a detachable keyboard or keypad, a mouse and the like. Output devices 8 may include one or more (possibly detachable) displays or monitors, headphones, speakers and / or any other suitable output devices. Various input and output devices (7 and 8) may be integrated into a single device, such as a smartphone or a digital fitness tracker. Any applicable input / output (VO) devices may be connected to computing device 1 as shown by blocks 7 and 8. For example, a wired or wireless network interface card (NIC), a universal serial bus (USB) device or external hard drive may be included in input devices 7 and / or output devices 8. It will be recognized that any suitable number of input devices 7 and output device 8 may be operatively connected to computing device 1 as shown by blocks 7 and 8.

[0072] System 100, according to some embodiments of the invention, may include components such as, but not limited to, a plurality of central processing units (CPU) or any other suitable multipurpose or specific processors or controllers (e.g., similar to element 2), a plurality of input units, a plurality of output units, a plurality of memory units, and a plurality of storage units.

[0073] Reference is now made to Fig. 2, which is a block diagram providing a detailed representation of system 100 for monitoring a respiratory condition of a subject, according to some embodiments.

[0074] According to some embodiments of the invention, system 100 may be implemented as a combination of software and hardware modules. For example, system 100 may be or may include computing devices such as element 1 of Fig. 1. Furthermore, system 100 may be adapted to execute one or more modules of instruction code (e.g., element 5 of Fig. 1) to request, receive, analyze, calculate and produce various data.

[0075] As further described in detail herein, system 100 may be adapted to execute one or more modules of instruction code (e.g., element 5 of Fig. 1) in order to perform steps of the claimed method.

[0076] As shown in Fig. 2, arrows may represent the flow of one or more data elements to and from system 100 and / or among modules or elements of system 100. Some arrows have been omitted in Fig. 2 for the purpose of clarity.

[0077] As shown in Fig. 2, in some embodiments, system 100 may include, or be connectable to one or more sensors that may be configured to measure physiological signals of the subject (operating as input devices 7 of Fig. 1). Said one or more sensors may include wearable ECG monitor 102. In some embodiments, said one or more sensors may further include accelerometer 101.

[0078] Accelerometer 101, such as Tri-axial Microelectromechanical Systems (MEMS) accelerometer, may be implemented as a wearable solution, e.g., integrated into a wristwatch (such as a smartwatch or a fitness tracker). In some embodiments, accelerometer 101 may be configured to generate a locomotion signal representing the subject’s physical movement (e.g., locomotion signal 101A), as commonly known in the art. E.g., locomotion signal 101A may include timeseries recordings of the subject’s acceleration along three orthogonal axes (X, Y, and Z), reflecting changes in inclination, orientation, and movement intensity over time. These signals can be used to infer walking patterns, gait variability, and other motion-related characteristics relevant to respiratory-locomotor or ECG-locomotor coupling analysis.

[0079] ECG monitor 102 may be likewise implemented as a wearable solution, e.g., including a chest-worn patch, a smart shirt, or a wrist-worn device equipped with dry or wet electrodes. In some embodiments, ECG monitor 102 may be configured to acquire an electrocardiogram (ECG) signal representing the subject’s cardiac electrical activity, e.g., accumulated over time (e.g., ECG signal 102A), as commonly known in the art. ECG signal 102A may be sampled continuously or intermittently and may include information such as R-R interval data, ST interval morphology, and heart rate variability. The acquired ECG signal may be transmitted to a local or remote processor for further analysis, including phase space reconstruction and cross-recurrence quantification analysis (CRQA), as described in further detail below.

[0080] In some embodiments, system 100 may further include bio-coupling module 110. Biocoupling module 110 may be configured to receive locomotion signal 101 A and ECG signal 102A.

[0081] In some embodiments, bio-coupling module 110 may be further configured to compute a set of recurrence quantification analysis (RQA) features (e.g., set of cross-recurrence quantification analysis (CRQA) features HOB), based on ECG signal 102A and the locomotion signal 101A. E.g., in some embodiments, module 110 may be configured to generate a cross-recurrence plot (CRP) (e.g., CRP 110A) based on ECG signal 102A and locomotion signal 101A, each reconstructed in a common phase space.

[0082] As known in the art, the generation of CRP 110A may involve reconstructing the phase space of each signal using time-delay embedding, where each point in the phase space represents a vector of delayed signal values. Once both signals are embedded, the CRP is formed by comparing all pairs of points from the two phase spaces and marking a recurrence whenever the distance between a pair of points falls below a predefined threshold. The resulting CRP 110A may be a two-dimensional matrix that visually and quantitatively captures the temporal alignment and shared dynamics between the two signals (e.g., ECG signal 102A and locomotion signal 101A), serving as the basis for extracting CRQA features 110B.

[0083] Bio-coupling module 110 may be further configured to extract CRQA features 110B as numerical features from CRP 110A that quantify temporal patterns of recurrence and similarity between the ECG signal and the locomotion signal.

[0084] The extraction of CRQA features 110B may involve analyzing the geometric structures formed within the CRP, such as diagonal lines, vertical or horizontal lines, and column-wise patterns. For example, diagonal line structures may be used to compute features such as determinism (the proportion of recurrent points forming diagonal lines), mean and maximum diagonal line lengths (indicating the duration of coupled behavior), and entropy of diagonal line lengths (reflecting the complexity of coupling). Similarly, vertical or horizontal line structures may yield features such as laminarity, mean and maximum line lengths, and entropy, which characterize the persistence of one signal in a stable state while the other recurs. Additionally, column-wise entropy may be computed to assess the diversity of proximity profiles across time. These features collectively provide a quantitative representation of the dynamic coupling between the ECG and locomotion signals and serve as the basis for detecting deviations associated with respiratory deterioration.

[0085] Based on the conducted research, certain CRQA features have been found highly indicative of COPD exacerbations.

[0086] Specifically, in some embodiments, the set of CRQA features 110B may include features representing diagonal structures of CRP 110A, said features representing diagonal structures being selected from the list comprising: a determinism percentage score (%DET), representing a proportion of recurrent points that form diagonal lines in CRP 110A; a mean diagonal line length in CRP 110A, representing an average duration of coupled behavior between ECG signal 102A and locomotion signal 101 A; a maximum diagonal line length in CRP 110 A, representing a longestcontinuous period of coupled behavior; and an entropy of diagonal line length, quantifying diversity of diagonal structures in CRP 110A.

[0087] Additionally or alternatively, in some embodiments, the set of CRQA features 110B may include features representing Vertical or Horizontal (V / H) structures of CRP 110A, said features representing V / H structures being selected from the list comprising: a laminarity value, representing a proportion of recurrent points that form V / H line structures in CRP 110A; a mean, non-diagonal line length, representing an average duration in which one of ECG signal 102A and locomotion signal 101 A persists in a specific state concurrently while the other of ECG signal 102A and locomotion signal 101 A experiences recurrence; a maximum V / H line length, representing a longest duration in which one of ECG signal 102A and locomotion signal 101 A persists in a specific state concurrently while the other of ECG signal 102A and locomotion signal 101 A experiences recurrence; and an entropy of V / H line lengths, quantifying diversity of V / H structures in CRP 110A.

[0088] Additionally or alternatively, in some embodiments, the set of CRQA features HOB may include a weighted columns entropy, quantifying diversity of proximity profiles of ECG signal 102A and locomotion signal 101A across temporal points.

[0089] It shall be understood that ECG signals and locomotion signals may exhibit significant variability between individual subjects. For the purposes of the present invention, such variability may further depend on the severity of the subject’s COPD condition. According to the concept of the present invention, the proposed solution is designed to accommodate this variability by enabling personalized analysis of respiratory condition exacerbation, while maintaining high reliability and robustness despite differences in the input data. This adaptability allows the system to provide clinically meaningful insights across a diverse patient population and varying disease states.

[0090] Therefore, in some embodiments, it is suggested that bio-coupling module 110 may be configured to compute said set of CRQA features HOB over one or more first periods of observation and one or more second periods of observation, said one or more first periods preceding said one or more second periods. For example, in some embodiments, module 110 may calculate CRQA features 110B for a series of one-minute intervals collected over a span of three to five days (the aforementioned “first” periods of observation), and then calculate CRQA features HOB for one-minute intervals collected during a subsequent day (the “second” period of observation). This temporal structuring enables the system to establish a personalized baselinefrom the first periods and to evaluate deviations or anomalies in the second period relative to that baseline.

[0091] System 100 may further include Principal Component Analysis (PCA) module 120. PCA module 120 may be configured to receive sets of CRQA features 110B, calculated over said “first” and “second” periods of observation. PCA module 120 may be further configured to evaluate one or more changes in CRQA features 110B over time by performing the following algorithm. For each of the one or more second periods of observation, module 120 may be configured to: (a) project respective set of CRQA features 110B from an original feature space (e.g., original feature space 120A) into a latent feature space (e.g., latent feature space 120B) derived by applying Principal Component Analysis (PCA) to a training dataset comprising training samples each representing the set of CRQA features HOB computed for said one or more first periods of observation; (b) reconstruct the projected set of CRQA features 110B into original feature space 120A; and (c) compute a reconstruction error (e.g., reconstruction error(s) 120C), representing a measure of deviation between the computed set of CRQA features HOB and the respective reconstructed projected set of CRQA features 110B.

[0092] As known in the art, Principal Component Analysis (PCA) is a dimensionality reduction technique that transforms a set of possibly correlated variables into a set of linearly uncorrelated variables called principal components. In the context of the present invention, PCA module 120 may first determine a latent feature space by applying PCA to a training dataset comprising CRQA feature vectors computed over the first periods of observation. This involves computing the covariance matrix of the training data, performing eigenvalue decomposition or singular value decomposition (SVD), and selecting a subset of principal components that capture the majority of the variance in the data. These components define latent feature space 120B.

[0093] Once the latent space is established, CRQA feature vectors computed for the second periods of observation are projected into this space by multiplying them with the matrix of selected principal component vectors. The projected vectors are then reconstructed back into original feature space 120A by reversing the transformation - e.g., multiplying the projected data by the transpose of the principal component matrix and adding the mean vector of the training data (if centering was applied). Reconstruction error 120C may be computed as the Euclidean distance (L2 norm) or squared error between the original CRQA feature vector and its reconstructed counterpart. This reconstruction error serves as a quantitative measure of deviation from the baseline and may be used to detect anomalies or early signs of respiratory deterioration. Accordingly, as an output, module 120 may provide a plurality of reconstruction errors 120C, e.g.,each corresponding to one-minute intervals collected during the last day of observation (the “second” periods of observation).

[0094] It shall be understood that PC A module 120 and the algorithm it implements are provided herein as a non-exclusive example. In some alternative embodiments, other anomaly detection machine learning (ML)-based models may be employed, such as models utilizing autoencoder architectures or other unsupervised learning techniques capable of identifying deviations from baseline patterns.

[0095] Accordingly, in some embodiments, system 100 may include a different anomaly detection ML-based model (not shown). System 100 may be configured to apply, for each of the one or more second periods of observation, the anomaly detection machine-learning (ML)-based model on the respective set of CRQA features HOB, said anomaly detection ML-based model being trained based on sets of CRQA features 110B computed for said one or more first periods of observation, and being configured to determine, as an output, whether the respective second period is anomalous with respect to the one or more first periods. Thereby, system 100 may be configured to evaluate said one or more changes in CRQA features 110B based on the output of the anomaly detection ML-based model.

[0096] In certain embodiments, PCA module 120 may be further configured to pre-process CRQA features HOB prior to projecting them into latent feature space 120B. For example, system 100 may be configured to divide each of said one or more first and one or more second periods of observation into a plurality of sub-periods (e.g., divide a one-minute period into three 20-second sub-periods). For each sub-period, bio-coupling module 110 may compute a set of CRQA features HOB, resulting in three sets of CRQA features per one-minute period.

[0097] PCA module 120 may then be configured to compute, for each one-minute period, a final representation of the CRQA features by calculating the mean and standard deviation of the corresponding features across the three sub-periods. This results in an aggregated feature vector comprising CRQA features (i.e., including both means and standard deviations), which may then be used as input to the PCA projection and reconstruction process, as discussed above. This approach enhances the robustness of the analysis by capturing both the central tendency and variability of physiological coupling within each observation period.

[0098] In some embodiments, system 100 may further include CRQA-based decision aggregation module 130. CRQA-based decision aggregation module 130 may be configured to receive reconstruction errors 120C corresponding to the one or more second periods of observation. CRQA-based decision aggregation module 130 may be configured to evaluate said one or morechanges in CRQA features 110B based on reconstruction error(s) 120C computed for said one or more second periods.

[0099] Specifically, in some embodiments, module 130 may be configured to determine a novelty score (e.g., novelty score 130A) for a target period of observation comprising said one or more second periods of observation (e.g., for the last monitored day). Novelty score 130A may be a statistical metric derived from the distribution of values of reconstruction errors 120C across said one or more second periods (e.g., distribution of errors’ 120C values over the last monitored day) and being indicative of the overall magnitude or variability of said values of reconstruction error 120C.

[0100] As known in the art, a novelty score is a statistical metric used to quantify how much a given observation deviates from a learned baseline or expected pattern. In the context of the present invention, novelty score 130A may be computed based on the distribution of reconstruction errors 120C obtained for a target period of observation (e.g., the most recent monitored day). In some embodiments, the novelty score may be a normalized numerical value, for example, ranging from 0 to 1, where higher values may indicate greater deviation from the baseline and thus a higher likelihood of an abnormal or exacerbation-related pattern.

[0101] To compute novelty score 130A, the reconstruction errors calculated by PCA module 120 for each one-minute segment within the second period of observation may be aggregated using statistical measures such as the mean, standard deviation, or percentile-based thresholds. For example, the novelty score may be defined as the proportion of segments whose reconstruction error exceeds a predefined threshold (e.g., based on the 95th percentile of the baseline error distribution), or as a z- score representing how far the average reconstruction error deviates from the baseline mean in terms of standard deviations. Alternatively, the novelty score may be derived using a kernel density estimation or other probabilistic models to assess the likelihood of the observed error distribution under the baseline model.

[0102] Novelty score 130A may provide a compact and interpretable indicator of whether the subject’s physiological coupling patterns have shifted significantly compared to their recent historical baseline (the data from the first period of observation), thereby enabling early detection of potential respiratory exacerbations.

[0103] In some embodiments, system 100 may further include exacerbation prediction module 170. Module 170 may be configured to issue alert 170A indicative of the potential exacerbation, based on determined novelty score 130A. E.g., module 170 may compare whether the novelty score 130A exceeds a predefined threshold, and issue alert 170A provided that score 130A exceedsthe predefined threshold. The predefined threshold may be predetermined based on normal variation in correlation between ECG and locomotion signals. In some embodiments, system 100 may be further configured to predefine this threshold individually for each subject, e.g., based on the data accumulated during the first period of observation, using known statistical methods.

[0104] System 100 may be further configured to communicate alert 170A to mobile device 103 of the subject or a therapist. Mobile device 103, being a part of system 100 or connectable thereto (e.g., as output devices 8 of Fig. 1), may be configured to provide, via a user interface, a notification alerting about the expected exacerbation of subject’s respiratory condition.

[0105] Fig. 3 illustrates an example of comparative analysis of physiological coupling between a COPD patient (left) and a healthy individual (right), based on cross-recurrence quantification analysis (CRQA). Each panel includes a cross-recurrence plot (CRP) generated from ECG and locomotion signals, along with corresponding time-series and phase-space projections. The CRP visualizes the temporal alignment of dynamic states between the two signals, reconstructed in a common phase space.

[0106] The COPD patient’s CRP (left) shows shorter and less complex diagonal structures, with a determinism score (%DET) of 86.41, mean diagonal line length of 4.65, and entropy of 2.55, indicating reduced coupling stability. In contrast, the healthy individual’s CRP (right) exhibits longer and more diverse diagonal patterns, with a higher %DET of 94.90, mean diagonal line length of 7.61, and entropy of 3.67, reflecting more stable and complex physiological synchronization. These differences underscore the utility of CRQA features in distinguishing between healthy and pathological states.

[0107] It shall be understood that, in some embodiments, system 100 may be configured to operate on pre-recorded data. For example, the subject may accumulate ECG and locomotion data over the course of one or more days using a wearable device (e.g., a smartwatch, mobile phone, or similar). The accumulated data may then be uploaded to a cloud-based service, where the abovedescribed algorithm for predicting respiratory condition exacerbation may be executed upon request.

[0108] Additionally or alternatively, system 100 may be configured for continuous monitoring and real-time exacerbation prediction. E.g., in some embodiments, system 100 may be configured to compute said set of CRQA features 110B in a continuous manner (e.g., using module 110); and evaluate said one or more changes in CRQA features 110B over time across successive periods of observation defined by a sliding time window of predetermined duration. For this purpose, knowntechniques for implementing sliding time window operations may be employed, as will be readily understood by a person skilled in the art.

[0109] Additionally or alternatively to the analysis of ECG signal 102A in combination with locomotion signal 101A, in some embodiments, system 100 may be configured to predict a potential exacerbation of the subject’s respiratory condition based solely on ECG signal 102A, without reference to locomotion signal 101 A.

[0110] Specifically, in some embodiments, system 100 may include ECG analysis module 140. In some embodiments, ECG analysis module 140 may be configured to receive ECG signal 102A. ECG analysis module 140 may be configured to compute one or more ECG-derived metrics 140A indicative of cardiac characteristics of the subject, based on ECG signal 102A.

[0111] The research that underpinned the present invention has shown that specific ECG- derived metrics are highly indicative of COPD condition exacerbation. In particular, features related to ST interval morphology - such as the ST interval gradient, the minimum-to-median amplitude difference, and morphology classification using a trained neural network - have demonstrated strong correlation with disease severity and physiological instability. Additionally, the ECG stability index, computed from the variability of the ECG signal during nocturnal periods, has been shown to reflect the subject’s baseline cardiac -respiratory state. These metrics, when analyzed over time, enable the detection of subtle deviations from a subject’s typical physiological patterns, which may precede clinical symptoms of exacerbation. The identification and validation of these features were based on comparative analysis of ECG data collected from both healthy individuals and COPD patients across varying GOLD severity groups. The results confirmed that these ECG-derived indicators can serve as reliable, non-invasive markers for early detection of respiratory deterioration, and form a critical component of the system’s predictive capabilities.

[0112] To further clarify the aspects of ECG-derived metrics, reference is also made to Fig. 4, showing schematic representation of ECG signal 102A.

[0113] As shown in Fig. 4 and as commonly known in the art, ECG signal 102 may include: P wave 201 in P-wave segment 207, PR segment 208, QRS complex 209 (Q 202 - R 203 - S 204), ST segment 210, T wave 205 in T-wave segment 211, U wave 206 in U-wave segment 212, R-R interval 215 (between R peaks 203 and 203’), PR interval 213, QT interval 214, and ST interval 216. It has been found experimentally that, in COPD patients, the morphology of the ST interval 216, and, in some cases, T-wave 205 specifically may vary significantly depending on the severity of the respiratory condition. In particular, the T-wave may exhibit a greater amplitude range - i.e., a larger difference between its minimum and maximum values - and may present with steepergradients or sharper edges. In some cases, the T-wave may initially deflect downward before rising, as illustrated by line 205'. These morphological variations are indicative of altered cardiac- respiratory interactions and may correlate with physiological stress or deterioration. Accordingly, analysis of ST interval morphology may provide valuable insights into the onset or progression of respiratory condition exacerbations, thereby enhancing the reliability of early detection and prediction.

[0114] Therefore, in some embodiments, ECG analysis module 140 may be configured to calculate the ECG-derived metrics 140A including ST interval morphology metrics. The ST interval morphology metrics may include a ST interval gradient metric, computed from gradient values calculated between the maximum and minimum points within the ST interval 216 of each R-R interval 215 during a predefined period of observation (for the purpose of clarity referred herein as “third” predefined period of observation), using one or more statistical aggregation functions. The ST interval morphology metrics may also include a ST interval min-to-median amplitude difference metric, computed from amplitude differences within the ST interval 216 of each R-R interval 215 during a predefined period of observation (for the purpose of clarity referred herein as “fourth” predefined period of observation). The ST interval morphology metrics may also include a ST interval morphology classification metric, computed by: (a) applying a trained neural network to classify ST interval 216 of each R-R interval 215 during a predefined period of observation (for the purpose of clarity referred herein as “fifth” predefined period of observation) into predefined morphology classes; and (b) determining statistical distribution of the ST interval 216 across the predefined morphology classes.

[0115] In some embodiments, the term statistical aggregation functions may refer to mathematical operations applied to a set of numerical values - such as signal-derived features or segment-level measurements - to summarize or characterize their distribution over a defined period of observation. These functions are used to reduce variability and extract meaningful trends from time-series data, enabling the system to generate stable and interpretable metrics for downstream analysis. Common examples of such aggregation functions include the mean (average value), median (middle value), standard deviation (measure of variability), minimum, maximum, and range (difference between max and min).

[0116] In addition to these conventional functions, the system may also compute higher-order statistical descriptors such as the skewness metric. Skewness quantifies the asymmetry of a distribution around its mean. A positive skew indicates a longer tail on the right side of the distribution, while a negative skew indicates a longer tail on the left. In the context of ECG-derivedmetrics - such as ST interval gradients or amplitude differences - skewness may reveal whether extreme values are more likely to occur in one direction, potentially indicating physiological irregularities or instability. By incorporating skewness and other statistical aggregation functions, the system can capture both central tendencies and distributional nuances of physiological signals, thereby improving the sensitivity and specificity of anomaly detection.

[0117] The abovementioned neural network may be trained using a labeled dataset of ECG segments, where each ST interval segment is annotated according to its morphological similarity to a reference pattern. Training may involve supervised learning techniques, using architectures such as convolutional neural networks (CNNs), long short-term memory (LSTM) networks, or hybrid models combining temporal and spatial feature extraction. The network may be optimized using a loss function such as categorical cross-entropy and trained with an optimizer like Adam.

[0118] Once trained, the neural network may classify each ST interval segment into one of several predefined morphology classes. For example, in some embodiments, the classes may include: class 2, indicating a high similarity to a healthy or expected ST interval morphology; class 1, indicating moderate deviation; and class 0, indicating significant morphological abnormality. These classes may reflect differences in ST interval shape, symmetry, slope, or amplitude distribution. The system may then determine the statistical distribution of ST intervals across these classes over the observation period, providing a quantitative profile of ST interval morphology variability. This distribution may serve as an input to downstream decision-making modules, contributing to the detection of early signs of respiratory condition exacerbation.

[0119] Similar to the approach described with reference to CRQA-based analysis, the neural network model used for ST interval morphology classification may, in some embodiments, be pretrained individually for each subject. This may be achieved by using ECG data accumulated by a wearable device over several days to train or fine-tune the model to the subject’s specific cardiac morphology. Once trained, the personalized model may then be applied to ECG data collected during a subsequent monitoring period (e.g., the most recent day). This individualized training approach allows the system to account for inter-subject variability in ECG morphology, thereby enhancing the sensitivity and specificity of exacerbation detection and enabling a more personalized and adaptive monitoring solution.

[0120] It shall be understood that, in order to calculate ST interval morphology metrics, ECG analysis module 140 may be configured to detect R-R intervals (such as R-R interval 215) in ECG signal 102A, and subsequently identify ST interval (such as ST interval 216) within each of the detected R-R intervals. Detection of R-peaks may be performed using standard peak detectionalgorithms or signal processing techniques known in the art. Once the R-R intervals are established, the ST interval may be identified as the portion of the ECG waveform occurring after the S-wave and before the onset of the next P-wave, typically within a predefined temporal window relative to the R-peak. Following identification, relevant computations may be applied to each ST interval to extract the aforementioned ST interval morphology metrics, including but not limited to gradient-based features, amplitude-based features, and classification-based features as described herein.

[0121] Additionally or alternatively, in some embodiments, ECG analysis module 140 may be configured to calculate the ECG-derived metrics 140A including an ECG stability index, computed by applying one or more statistical aggregation functions - such as the median or mean - to standard deviations of segments of ECG signal 102A (e.g., short, noise-free segments) over a predefined period of observation (referred herein as a “sixth” predefined period of observation). E.g., in some embodiments, ECG signal 102A may be segmented into one-minute intervals during nighttime or resting periods, and segments affected by motion artifacts or signal noise may be excluded using predefined quality criteria. For each valid segment, the standard deviation of the ECG amplitude may be calculated, reflecting the signal’s variability. The ECG stability index may then be derived as the median of these standard deviations across the predefined period (the “sixth” period), with lower values indicating greater signal stability. This metric may serve as an indirect indicator of autonomic balance and physiological restfulness, and may be used to support detection of early signs of respiratory deterioration.

[0122] Additionally or alternatively, in some embodiments, ECG analysis module 140 may be configured to calculate the ECG-derived metrics 140A including a heart rate metric, computed by applying one or more statistical aggregation functions - such as median or mean - to a plurality of heart rate measurements derived from ECG signal 102A over a predefined period of observation (also referred herein as “seventh” predefined period of observation).

[0123] In some embodiments, the system may be configured to apply different periods of observation depending on the nature and sensitivity of the specific metric being calculated. Certain metrics may require high signal fidelity and minimal noise interference to ensure accurate computation. For example, the ST interval gradient metric, which relies on precise detection of the maximum and minimum points within each ST interval, may be calculated during nocturnal periods (e.g., a six-hour nighttime window) when the subject is at rest and motion artifacts are minimal. In contrast, other metrics (such as the ST interval morphology classification metric, which leverages a trained neural network to categorize ST interval shapes) may be computed overa full 24-hour period, as the model is designed to tolerate a broader range of signal variability and still extract meaningful morphological patterns. This flexible assignment of observation windows allows the system to optimize the reliability of each metric while accommodating the practical constraints of real-world, ambulatory monitoring.

[0124] As discussed above, ECG signals may exhibit significant variability between individual subjects. For the purposes of the present invention, such variability may further depend on the severity of the subject’s COPD condition. According to the concept of the present invention, the proposed solution is designed to accommodate this variability by enabling personalized analysis of respiratory condition exacerbation, while maintaining high reliability and robustness despite differences in the input data. This adaptability allows the system to provide clinically meaningful insights across a diverse patient population and varying disease states.

[0125] Therefore, in some embodiments, it is suggested that ECG analysis module 140 may be configured to compute said ECG-derived metrics 140A over one or more first periods of observation and one or more second periods of observation, said one or more first periods preceding said one or more second periods. For example, in some embodiments, module 140 may calculate metrics 140A for a series of one-minute intervals collected over a span of three to five days (the aforementioned “first” periods of observation), and then calculate metrics 140A for one- minute intervals collected during a subsequent day (the “second” period of observation). This temporal structuring enables the system to establish a personalized baseline from the first periods and to evaluate deviations or anomalies in the second period relative to that baseline.

[0126] In some embodiments, system 100 may further include Local Outlier Factor (LOF) calculation module 150. Module 150 may be configured to receive ECG-derived metrics 140A calculated over the first and the second periods of observation. Module 150 may be further configured to evaluate one or more changes in ECG-derived metrics 140A over time by performing the following. For each of the one or more second periods of observation, module 150 may be configured to apply Local Outlier Factor (LOF) algorithm 151 trained on a dataset comprising samples of ECG-derived metrics 140A computed for said one or more first periods of observation to determine whether ECG-derived metrics 140A for the second period represent an outlier relative to the baseline distribution established from the first periods. Accordingly, LOF calculation module 150 may be configured to output determined outliers 150A (e.g., the number of segments of ECG signal 102A marked, based on the respective ECG-derived metrics 140A, as outliers).

[0127] As known in the art, the Local Outlier Factor (LOF) algorithm is an unsupervised anomaly detection technique that identifies data points whose local density significantly deviates from that of their neighbors. In the context of the present invention, LOF algorithm 151 may be applied to ECG-derived metrics 140A computed over a predefined baseline period (i.e., the first periods of observation), which is assumed to represent the subject’s typical or healthy physiological state. For each data point in the second period of observation, the algorithm computes the local reachability density based on its k-nearest neighbors within the baseline dataset. The LOF score is then calculated as the ratio of the average local density of the neighbors to the local density of the data point itself. A score significantly greater than 1 indicates that the point is an outlier.

[0128] The sensitivity of LOF algorithm 151 may be tuned using parameters such as the number of neighbors (k) and contamination factor (c), which influence the granularity and threshold of outlier detection. These parameters may be used to calibrate the sensitivity of the outlier detection process, for example, thereby adjusting the threshold above which a LOF score is considered anomalous. Additionally, module 150 may be configured to adaptively update the baseline dataset used for LOF evaluation. Specifically, if ECG-derived metrics 140A from a second period are classified as non-outliers (i.e., within the expected density range), they may be incorporated into the baseline dataset. This allows the system to evolve over time, refining its understanding of the subject’s physiological norms and improving the accuracy of future outlier detection. To ensure stability, such updates may be conditioned on observing multiple consecutive non-outlier classifications or other statistical criteria.

[0129] In some embodiments, the values of the number of neighbors (k) and the contamination factor (c) used by LOF algorithm 151 may be dynamically adjusted based on the amount of baseline data available. For this purpose, LOF calculation module 150 may utilize additional parameters which allow the LOF algorithm to adapt as the baseline dataset grows during the course of patient monitoring. E.g., as the number of baseline samples increases, the number of neighbors considered by algorithm 151 may also increase proportionally, enabling the algorithm to evaluate each data point within a broader local context. Similarly, the contamination factor, which reflects the expected proportion of outliers in the data, may be gradually increased as more baseline data becomes available, up to a predefined maximum value. This adaptive configuration improves the flexibility and accuracy of the anomaly detection process, particularly in long-term monitoring scenarios where the subject’s physiological baseline may evolve over time.

[0130] In some embodiments, when said one or more second periods of observation represent a comparatively long duration (e.g., a 24-hour period), system 100 may be configured to issue an alert based on a determination by LOF algorithm 151 that the respective long period constitutes an outlier. In alternative embodiments - e.g., when said one or more second periods of observation represent comparatively short durations (e.g., one-minute intervals throughout a day) - the system may be configured to apply additional statistical aggregation to the results, as further described below.

[0131] In some embodiments, system 100 may further include ECG-based decision aggregation module 160. ECG-based decision aggregation module 160 may be configured to receive the data about the determined outliers 150A. Module 160 may be further configured to evaluate one or more changes in ECG-derived metrics 140A by determining anomaly score 160A for a target period of observation comprising said one or more second periods of observation. Anomaly score 160 A may be a statistical metric derived from the proportion of second periods determined as outliers relative to the total number of said one or more second periods.

[0132] In some embodiments, the anomaly score 160A may quantify the degree of deviation of the subject’s current physiological state from a previously established baseline. Specifically, the anomaly may represent a significant shift in the distribution or pattern of ECG-derived metrics - such as ST interval morphology, ECG stability, or heart rate - relative to the subject’s historical norm (e.g., metrics calculated during the second period of observation in relation to the ones calculated during the first period of observation).

[0133] Anomaly score 160A may be computed as a normalized ratio, for example, the ratio between the number of second-period segments classified as outliers by the LOF algorithm, and the total number of segments in the second period.

[0134] In some embodiments, this score may be further refined using statistical smoothing techniques (e.g., moving averages or exponential weighting) or by applying thresholds derived from the distribution of scores observed during the baseline period.

[0135] Optionally, anomaly score 160A may be interpreted probabilistically, such that higher values (e.g., >0.5) indicate a high likelihood of abnormal physiological behavior, potentially triggering an alert. The system may also incorporate temporal trends - such as the persistence of elevated anomaly scores over consecutive days - to increase robustness and reduce false positives.

[0136] In some embodiments, exacerbation prediction module 170 may be configured to issue the alert indicative of the potential exacerbation (e.g., alert 170A), further based on anomaly score 160A.

[0137] In some embodiments, this threshold may be dynamically adjusted based on individual patient data. Specifically, the system may allow the patient to manually annotate or confirm exacerbation events via a user interface on a personal wearable device (e.g., smartwatch or smartphone). These user-reported events may be temporally aligned with previously computed anomaly scores 160A, enabling the system to retrospectively analyze which anomaly score values preceded confirmed exacerbations.

[0138] Based on this historical correlation, exacerbation prediction module 170 may be further configured to learn a personalized threshold, e.g., a value of anomaly score 160A that is statistically associated with an increased likelihood of an impending exacerbation for that specific individual. This adaptive thresholding approach allows the system to account for inter-individual variability in physiological responses and improves the specificity and sensitivity of alert generation. In some embodiments, the threshold may be updated continuously or periodically as more annotated data becomes available, thereby refining the system’s predictive accuracy over time.

[0139] In some embodiments, exacerbation prediction module 170 may be configured to issue an alert only upon cross-check between two independent anomaly detection mechanisms: the CRQA-based novelty score (e.g., novelty score 130A) and the ECG-based anomaly score (e.g., anomaly score 160 A) derived from the Local Outlier Factor (LOF) algorithm. This dual- algorithm approach enhances the robustness of system 100 by requiring mutual confirmation of abnormal physiological patterns before triggering an alert.

[0140] The alert generation process may follow a hierarchical, cross-validation logic that considers both the current and previous day's data. Specifically, an alert is first considered if either novelty score 130A or LOF-based anomaly score 160A exceeds its respective threshold on the current day. If this initial trigger condition is met, the system then performs a cross-check to determine whether the other algorithm also indicated an alert condition - either on the same day or on the immediately preceding day.

[0141] For example, if novelty score 130A exceeds its threshold on the current day, the system checks whether anomaly score 160A also exceeded its threshold either on the current day or the previous day. If so, an alert may be issued. Conversely, if anomaly score 160A is the initial trigger, the system verifies whether novelty score 130A exceeded its threshold on the current or previous day. An alert may be issued only if this mutual confirmation is satisfied within the two-day window.

[0142] This cross-validation mechanism ensures that alerts are not based on transient or isolated anomalies detected by a single algorithm. Instead, it requires temporal and algorithmicconsistency, thereby reducing the likelihood of false positives. The system thus balances sensitivity with specificity, enabling early yet reliable detection of potential exacerbations. In some embodiments, the thresholds for both the novelty and LOF-based scores may be dynamically adjusted based on individual patient data, as described above, further enhancing the personalization and accuracy of the alerting process.

[0143] In some embodiments, exacerbation prediction module 170 may be further configured to assess the long-term stability of the subject’s respiratory condition. Beyond issuing alerts for potential exacerbations, the module may analyze historical anomaly scores (e.g., novelty score 130A and anomaly score 160A) over an extended observation period to determine whether the subject has exhibited a sustained absence of abnormal physiological patterns. If no exacerbation alerts have been triggered over a predefined duration (e.g., several consecutive days or weeks) the system may infer that the subject is in a stable state.

[0144] In such cases, module 170 may be configured to generate a classification or status indicator suggesting that the subject is likely healthy or experiencing a stable respiratory condition. This determination may be based on a configurable threshold, which may be fixed or dynamically adjusted based on the subject’s baseline variability and historical trends. For example, if the system observes that all computed scores remain consistently below alert thresholds for a continuous period exceeding a defined minimum (e.g., 14 days), it may output a message such as “No signs of exacerbation detected - respiratory condition appears stable.” This functionality may be useful for clinicians monitoring recovery, for patients undergoing rehabilitation, or for screening purposes in broader population health applications.

[0145] In some embodiments, system 100 may further include training module 180, configured to receive and process accumulated physiological and analytical data for the purpose of training or updating machine learning models used within system 100. Specifically, training module 180 may be configured to receive one or more of the following: CRQA features 110B, ECG-derived metrics 140 A, and reconstruction errors 120C. These data inputs may be collected over time from the subject or from a broader population of users, and may serve as the basis for model calibration, personalization, or generalization.

[0146] Training module 180 may be operatively connected to other analytical components of the system, including but not limited to the PCA module 120, the LOF calculation module 150, and the neural network used for ST interval morphology classification (ECG analysis module 140). In some embodiments, training module 180 may apply supervised, unsupervised, or semisupervised learning techniques to optimize the performance of these models, as commonly knownin the art. This may include both initial training (e.g., when the system is first deployed or when a new user is enrolled) and re-training or fine-tuning based on newly accumulated data. For example, module 180 may retrain LOF algorithm 151 using updated baseline data, refine PCA latent feature space 120B based on new CRQA feature distributions, or adapt the neural network classifier using newly obtained ST intervals. This architecture enables the system to adapt over time to individual physiological patterns, improve accuracy, and maintain robustness in the face of evolving data characteristics.

[0147] It shall be understood that, in some embodiments, the invention may be implemented not only as a distributed system comprising multiple components, or a solely cloud-based computing solution, but also as a single, integrated wearable device. This device may be configured to perform all necessary signal acquisition, processing, analysis, and alert generation functions locally, without requiring continuous connectivity to external servers or cloud-based infrastructure. Such an embodiment may be particularly advantageous for real-time, ambulatory monitoring in home or outpatient settings, where simplicity, portability, and autonomy are critical.

[0148] In this embodiment, the device may include: at least one non-transitory memory device, wherein modules of instruction code are stored; and at least one processor associated with said memory device and configured to execute the modules of instruction code. Upon execution of the instruction code, the processor may be configured to: receive, from one or more embedded sensors (integrated in the device), physiological signals of the subject, the physiological signals comprising: an electrocardiogram (ECG) signal representing cardiac electrical activity of the subject; and a locomotion signal representing the subject’s physical movement (e.g., from an integrated accelerometer); compute a set of cross-recurrence quantification analysis (CRQA) features based on the ECG signal and the locomotion signal; evaluate one or more changes in the CRQA features over time, optionally using dimensionality reduction and anomaly detection techniques such as PCA and LOF; compute one or more ECG-derived metrics, including ST interval morphology metrics, ECG stability index, and heart rate; determine one or more statistical scores, such as a novelty score and / or anomaly score, based on the evaluated changes; and issue an alert indicative of a potential exacerbation of a respiratory condition based on the evaluated changes.

[0149] The device may further include a user interface, such as a touchscreen or haptic feedback mechanism, to notify the user of alerts, and may optionally support wireless communication (e.g., Bluetooth, Wi-Fi) for data synchronization or remote monitoring. In someembodiments, the device may be implemented as a smartwatch, chest patch, or other wearable form factor, incorporating both sensing and computational capabilities within a single housing.

[0150] This compact embodiment enables fully self-contained respiratory condition monitoring, offering a practical and scalable solution for continuous health tracking in real-world environments.

[0151] Referring now to Fig. 5, a flow diagram is presented, depicting a method for monitoring a respiratory condition of a subject, by at least one processor (e.g., processor 2 of Fig. 1), according to some embodiments.

[0152] As shown in step S1005, the at least one processor (e.g., such as processor 2 of Fig. 1) may acquire, using one or more sensors (e.g., wearable ECG monitor 102 and accelerometer 101 shown in Fig. 2), physiological signals from the subject, the physiological signals comprising: (i) an electrocardiogram (ECG) signal (e.g., ECG signal 102A shown in Fig. 2) representing cardiac electrical activity of the subject; and (ii) a locomotion signal (e.g., locomotion signal 101A shown in Fig. 2) representing the subject’s physical movement. Step S1005 may be carried out by biocoupling module 110 (as described with reference to Fig. 2).

[0153] As shown in step S1010, the at least one processor (e.g., such as processor 2 of Fig. 1) may compute a set of recurrence quantification analysis (RQA) features (e.g., CRQA features HOB shown in Fig. 2) based on the ECG signal (e.g., ECG signal 102A shown in Fig. 2) and the locomotion signal (e.g., locomotion signal 101A shown in Fig. 2). Step S1010 may be carried out by bio-coupling module 110 (as described with reference to Fig. 2).

[0154] As shown in step S1015, the at least one processor (e.g., such as processor 2 of Fig. 1) may evaluate one or more changes in the RQA features (e.g., CRQA features 110B shown in Fig. 2) over time. Step S 1015 may be carried out by PCA module 120 and CRQA-based decision aggregation module 130 (as described with reference to Fig. 2).

[0155] As shown in step S1020, the at least one processor (e.g., such as processor 2 of Fig. 1) may issue an alert (e.g., alert 170A shown in Fig. 2) indicative of a potential exacerbation of the respiratory condition based on the evaluated one or more changes in the RQA features (e.g., CRQA features 110B shown in Fig. 2). Step S 1020 may be carried out by exacerbation prediction module 170 (as described with reference to Fig. 2).

[0156] The present invention is grounded in a series of clinical studies and research efforts conducted by the inventors, which collectively demonstrate the feasibility, accuracy, and clinical relevance of the proposed system for early detection of COPD exacerbations. These studies spanfrom early feasibility trials to a large-scale, multi-center calibration study, and include both retrospective and prospective data analysis.

[0157] Phase I Clinical Study and Feasibility Research. The initial feasibility study involved 10 COPD patients and 10 healthy volunteers. Participants underwent a 6-minute walk test (6MWT) and were monitored using wearable ECG and SpCh sensors. COPD patients were further monitored at home for two weeks. The study introduced the concept of bio-coupling - the synchronization between ECG-derived respiration and locomotor signals - as a novel biomarker for respiratory health.

[0158] Key findings included: a) A ST interval morphology-based classifier and bio-coupling metrics (e.g., %DET, entropy, laminarity) could reliably distinguish COPD patients from healthy individuals. b) A novel Age-Determinism Score (ADS) was developed, combining %DET and patient age to stratify COPD severity (GOLD B vs. GOLD C / D). c) A machine-leaming-based classifier using only 2-3 features achieved full separation between healthy and COPD groups, with sensitivity up to 89% and specificity up to 92%. d) The system demonstrated the ability to differentiate between stable and unstable COPD patients, and to detect early signs of exacerbation.

[0159] These results provided a strong proof-of-concept for the invention’s core algorithms and validated the use of wearable sensors and machine learning for respiratory monitoring.

[0160] Multi-Center Calibration Trial. Building on the Phase I results, inventors launched a multi-center clinical trial, targeting 80 COPD patients across five hospitals in Israel. As of the interim analysis, 25 patients had been enrolled, with nine exacerbation events (four severe, five moderate) already recorded and analyzed.

[0161] Highlights of the trial include : a) Continuous monitoring of ECG, movement, and SpO? data over a 4-6 month period. b) Use of various ML-based models to detect pre-exacerbation patterns. c) In one case, the system identified elevated risk scores up to five days before hospitalization. d) Tailored analytical pipelines were developed for different patient profiles (frequent vs. infrequent exacerbators). e) Algorithms such as TP Min-Median, TP Neural Network, and hierarchical clustering heatmaps were used to refine predictions and classify event severity.

[0162] In summary, together, these studies confirm that the invention:a) Reliably distinguishes between healthy individuals and COPD patients. b) Stratifies COPD severity using physiological coupling and ECG-derived metrics. c) Detects exacerbation risk several days in advance with high sensitivity and specificity. d) Adapts to individual patient profiles through machine learning-based calibration and retraining.

[0163] These findings validate the core concept of the invention and demonstrate its potential to transform COPD management through early, personalized, and non-invasive monitoring.

[0164] As can be seen from the provided description, the claimed invention represents a system and method for respiratory condition monitoring that improve the relevant technological field by enabling continuous, multi-signal analysis to detect early signs of respiratory deterioration. By integrating dynamic physiological coupling metrics through advanced signal processing techniques and machine learning algorithms, the suggested system becomes capable of extracting clinically relevant information from physiological data - thereby increasing the reliability of exacerbation detection and enabling early warnings of potential respiratory decline. These capabilities may significantly improve patient care by supporting timely interventions and reducing the frequency and severity of exacerbations. Furthermore, by leveraging indirect respiratory indicators such as ECG-derived and locomotor data, the system reduces sensor complexity while maintaining diagnostic accuracy.

[0165] Unless explicitly stated, the method embodiments described herein are not constrained to a particular order or sequence. Furthermore, all formulas described herein are intended as examples only and other or different formulas may be used. Additionally, some of the described method embodiments or elements thereof may occur or be performed at the same point in time.

[0166] While certain features of the invention have been illustrated and described herein, many modifications, substitutions, changes, and equivalents may occur to those skilled in the art. It is, therefore, to be understood that the appended claims are intended to cover all such modifications and changes as fall within the true spirit of the invention.

[0167] Various embodiments have been presented. Each of these embodiments may of course include features from other embodiments presented, and embodiments not specifically described may include various features described herein.

Claims

CLAIMS1. A system for monitoring a respiratory condition of a subject, comprising: at least one non-transitory memory device, wherein modules of instruction code are stored; and at least one processor associated with said at least one memory device and configured to execute the modules of instruction code, whereupon execution of said modules of instruction code, the at least one processor is configured to: receive, from one or more sensors, physiological signals of the subject, the physiological signals comprising:(i) an electrocardiogram (ECG) signal representing cardiac electrical activity of the subject; and(ii) a locomotion signal representing the subject’s physical movement; compute a set of recurrence quantification analysis (RQA) features, based on the ECG signal and the locomotion signal; evaluate one or more changes in the RQA features over time; issue an alert indicative of a potential exacerbation of the respiratory condition based on the evaluated one or more changes in the RQA features.

2. The system of claim 1, wherein said at least one processor is configured to compute the set of RQA features as a set of cross -recurrence quantification analysis (CRQA) features by: generating a cross-recurrence plot (CRP) based on the ECG signal and the locomotion signal, each reconstructed in a common phase space; and extracting said CRQA features as numerical features from the CRP that quantify temporal patterns of recurrence and similarity between the ECG signal and the locomotion signal.

3. The system of claim 2, wherein the set of CRQA features comprises features representing diagonal structures of the CRP, said features representing diagonal structures being selected from the list comprising: a determinism percentage score (%DET), representing a proportion of recurrent points that form diagonal lines in the CRP; a mean diagonal line length in the CRP, representing an average duration of coupled behavior between the ECG signal and the locomotion signal; a maximum diagonal line length in the CRP, representing a longest continuous period of coupled behavior; and an entropy of diagonal line length, quantifying diversity of diagonal structures in the CRP.

4. The system according to any one of claims 2-3, wherein the set of CRQA features comprises features representing Vertical or Horizontal (V / H) structures of the CRP, said features representing V / H structures being selected from the list comprising: a laminarity value, representing a proportion of recurrent points that form V / H line structures in the CRP; a mean, non-diagonal line length, representing an average duration in which one of the ECG signal and the locomotion signal persists in a specific state concurrently while the other of the ECG signal and the locomotion signal experiences recurrence; a maximum V / H line length, representing a longest duration in which one of the ECG signal and the locomotion signal persists in a specific state concurrently while the other of the ECG signal and the locomotion signal experiences recurrence; and an entropy of V / H line lengths, quantifying diversity of V / H structures in the CRP.

5. The system according to any one of claims 2-4, wherein the set of CRQA features comprises a weighted columns entropy, quantifying diversity of proximity profiles of the ECG signal and the locomotion signal across temporal points.

6. The system according to any one of claims 2-5, wherein said at least on processor is configured to: compute said set of the CRQA features over one or more first periods of observation and one or more second periods of observation, said one or more first periods preceding said one or more second periods; and evaluate one or more changes in the CRQA features over time by: applying, for each of the one or more second periods of observation, an anomaly detection machine-learning (ML)-based model on the respective set of CRQA features, said anomaly detection ML-based model being trained based on sets of CRQA features computed for said one or more first periods of observation, and being configured to determine, as an output, whether the respective second period is anomalous with respect to the one or more first periods; and evaluating said one or more changes in the CRQA features based on the output of the anomaly detection ML-based model.

7. The system of claim 6, wherein said at least one processor is configured to apply, for each of the one or more second periods of observation, the anomaly detection ML-based model by: for each of the one or more second periods of observation:(a) projecting said computed set of the CRQA features from an original feature space into a latent feature space derived by applying Principal Component Analysis (PCA) to a training dataset comprising training samples each representing the set of CRQA features computed for said one or more first periods of observation;(b) reconstructing the projected set of the CRQA features into the original feature space; and(c) computing a reconstruction error, representing a measure of deviation between the computed set of the CRQA features and the respective reconstructed projected set of the CRQA features; and evaluating said one or more changes in the CRQA features based on the reconstruction error computed for said one or more second periods.

8. The system of claim 7, wherein said evaluating said one or more changes in the CRQA features based on the reconstruction error comprises determining a novelty score for a target period of observation comprising said one or more second periods of observation, the novelty score being a statistical metric derived from the distribution of values of said reconstruction error across said one or more second periods, and being indicative of the overall magnitude or variability of said values of the reconstruction error; and wherein said at least one processor is configured to issue the alert indicative of the potential exacerbation, based on the determined novelty score.

9. The system according to any one of claims 2-8, wherein said at least one processor is configured to compute said set of the CRQA features in a continuous manner; and evaluate said one or more changes in the CRQA features over time across successive periods of observation defined by a sliding time window of predetermined duration.

10. The system according to any one of claims 1-9, wherein said at least one processor is further configured to: compute one or more ECG-derived metrics indicative of cardiac characteristics of the subject, based on the ECG signal; evaluate one or more changes in said one or more ECG-derived metrics over time; and issue said alert, further based on the evaluated one or more changes in said one or more ECG-derived metrics.

11. The system of claim 10, wherein the one or more ECG-derived metrics comprise one or more ST-interval morphology metrics selected from the group consisting of:(i) a ST-interval gradient metric, computed from gradient values calculated between the maximum and minimum points within the ST-interval of each R-R interval during a third predefined period of observation, using one or more statistical aggregation functions;(ii) a ST-interval min-to-median amplitude difference metric, computed from amplitude differences within the ST-interval of each R-R interval during a fourth predefined period of observation;(iii) a ST-interval morphology classification metric, computed by: (a) applying a trained neural network to classify ST-intervals of each R-R interval during a fifth predefined period of observation into predefined morphology classes; and (b) determining statistical distribution of the ST intervals across the predefined morphology classes.

12. The system according to any one of claims 10-11, wherein the one or more ECG-derived metrics comprise an ECG stability index, computed by applying one or more statistical aggregation functions to standard deviations of segments of the ECG signal over a sixth predefined period of observation.

13. The system according to any one of claims 10-12, wherein the one or more ECG-derived metrics comprise a heart rate metric, computed by applying one or more statistical aggregation functions to a plurality of heart rate measurements derived from the ECG signal over a seventh predefined period of observation.

14. The system according to any one of claims 10-13, wherein said at least one processor is configured to: compute the one or more ECG-derived metrics over one or more first periods of observation and one or more second periods of observation, said one or more first periods preceding said one or more second periods; and evaluate said one or more changes in the ECG-derived metrics over time by, for each of the one or more second periods of observation, applying a Local Outlier Factor (LOF) algorithm trained on a dataset comprising samples of the ECG-derived metrics computed for said one or more first periods of observation, to determine whether the ECG-derived metrics for the second period represent an outlier relative to the baseline distribution established from the first periods; andissue the alert indicative of the potential exacerbation, based on determining whether the ECG-derived metrics for the second period represent an outlier.

15. The system of claim 14, wherein said evaluating said one or more changes in the ECG- derived metrics comprises determining an anomaly score for a target period of observation comprising said one or more second periods of observation, the anomaly score being a statistical metric derived from the proportion of second periods determined as outliers relative to the total number of said one or more second periods; and wherein said at least one processor is configured to issue the alert indicative of the potential exacerbation, further based on the determined anomaly score.

16. A system for monitoring a respiratory condition of a subject, comprising: at least one non-transitory memory device, wherein modules of instruction code are stored; and at least one processor associated with said at least one memory device and configured to execute the modules of instruction code, whereupon execution of said modules of instruction code, the at least one processor is configured to: receive, from one or more sensors, physiological signals of the subject, the physiological signals comprising an electrocardiogram (ECG) signal representing cardiac electrical activity of the subject; compute, based on the ECG signal, one or more ST interval morphology metrics selected from the group consisting of:(i) a ST interval gradient metric, computed from gradient values calculated between the maximum and minimum points within the ST interval of each R-R interval during a first predefined period of observation, using one or more statistical aggregation functions;(ii) a ST interval min-to-median amplitude difference metric, computed from amplitude differences within the ST interval of each R-R interval during a second predefined period of observation;(iii) a ST interval morphology classification metric, computed by: (a) applying a trained neural network to classify ST interval of each R-R interval during a third predefined period of observation into predefined morphology classes; and (b) determining statistical distribution of the ST intervals across the predefined morphology classes; evaluate one or more changes in the one or more ST interval morphology metrics over time; andissue an alert indicative of a potential exacerbation of the respiratory condition based on the evaluated one or more changes in the one or more ST interval morphology metrics.

17. A method for monitoring a respiratory condition of a subject, by at least one processor, the method comprising: acquiring, using one or more sensors, physiological signals from the subject, the physiological signals comprising:(i) an electrocardiogram (ECG) signal representing cardiac electrical activity of the subject; and(ii) a locomotion signal representing the subject’s physical movement; computing a set of recurrence quantification analysis (RQA) features based on the ECG signal and the locomotion signal; evaluating one or more changes in the RQA features over time; and issuing an alert indicative of a potential exacerbation of the respiratory condition based on the evaluated one or more changes in the RQA features.

18. The method of claim 17, wherein said computing the set of RQA features comprises: generating a cross-recurrence plot (CRP) based on the ECG signal and the locomotion signal, each reconstructed in a common phase space; and extracting said set of RQA features as a set of cross-recurrence quantification analysis (CRQA) features, representing numerical features from the CRP that quantify temporal patterns of recurrence and similarity between the ECG signal and the locomotion signal.

19. The method of claim 18, wherein the set of CRQA features comprises features representing diagonal structures of the CRP, said features representing diagonal structures being selected from the list comprising: a determinism percentage score (%DET), representing a proportion of recurrent points that form diagonal lines in the CRP; a mean diagonal line length in the CRP, representing an average duration of coupled behavior between the ECG signal and the locomotion signal; a maximum diagonal line length in the CRP, representing a longest continuous period of coupled behavior; and an entropy of diagonal line length, quantifying diversity of diagonal structures in the CRP.

20. The method according to any one of claims 18-19, wherein the set of CRQA features comprises features representing Vertical or Horizontal (V / H) structures of the CRP, said features representing V / H structures being selected from the list comprising: a laminarity value, representing a proportion of recurrent points that form V / H line structures in the CRP; a mean,non-diagonal line length, representing an average duration in which one of the ECG signal and the locomotion signal persists in a specific state concurrently while the other of the ECG signal and the locomotion signal experiences recurrence; a maximum V / H line length, representing a longest duration in which one of the ECG signal and the locomotion signal persists in a specific state concurrently while the other of the ECG signal and the locomotion signal experiences recurrence; and an entropy of V / H line lengths, quantifying diversity of V / H structures in the CRP.

21. The method according to any one of claims 18-20, wherein the set of CRQA features comprises a weighted columns entropy, quantifying diversity of proximity profiles of the ECG signal and the locomotion signal across temporal points.

Images