System for determining cardiac function markers

A myocardial-tissue-contacting sensor array with machine learning predicts cardiac function markers, addressing LVAD monitoring inaccuracies and complications by providing precise, continuous monitoring and enabling automated adjustments.

WO2026238803A1PCT designated stage Publication Date: 2026-11-19MEDIZINISCHE UNIVERSITAET WIEN +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/EP2026/063912
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-13
Filing Date
2026-05-13
Publication Date
2026-11-19

AI Technical Summary

Technical Problem

Conventional LVADs fail to provide accurate, absolute measurements of critical cardiac function markers like left ventricular pressure and contractility, leading to complications such as ventricular wall collapse and suction, due to indirect estimation methods and blood-contacting sensors prone to drift and thrombosis.

Method used

A sensor array positioned directly on myocardial tissue, using force or pressure sensors and a processing unit with machine learning algorithms to predict cardiac function markers like LVSP, LVEDP, dp/dt max, and Emax, eliminating blood-contacting issues and providing precise, continuous monitoring.

Benefits of technology

The system achieves accurate, stable, and real-time monitoring of cardiac function markers with minimal deviation from catheterization values, enabling automated LVAD adjustments to match patient needs, reducing complications and improving patient outcomes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure EP2026063912_19112026_PF_FP_ABST
    Figure EP2026063912_19112026_PF_FP_ABST
Patent Text Reader

Abstract

A system for determining cardiac function markers comprising: a sensor array (8a,8b) configured for positioning in contact with myocardial tissue (5), wherein the sensor array (8a,8b) includes at least one, preferably a plurality of, force or pressure sensor(s) (14) to obtain force or pressure signals representative of a contraction of the myocardial tissue (5); a data acquisition means coupled to the sensor array (8a,8b) and configured to receive the pressure signals from the sensor array (8a,8b); and a processing unit configured to calculate or predict at least one cardiac function marker based on said pressure signals.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] System for determining cardiac function markers

[0002] The invention refers to s system for determining cardiac function markers as well as a left ventricular assist device system comprising such a system.

[0003] Cardiovascular diseases remain the leading cause of mortality worldwide, underscoring the urgent need for advancements in cardiac care and patient monitoring. In patients with endstage heart failure, left ventricular assist devices (LVADs) are increasingly used to support cardiac function by helping to pump blood from the left ventricle to the aorta .

[0004] Conventional LVADs are typically implanted within the left ventricular apex and continuously adjusted to maintain optimal blood flow. However, these adjustments often fail to account for the dynamic nature of a patient ' s condition and hemodynamic status, which continuously change in response to various physiological and environmental factors . This mismatch between the heart ' s actual needs and the LVAD' s operation can lead to serious complications, including ventricular wall collapse and suction within the blood pump . These complications may trigger a cascade of adverse events, such as arrhythmias or thromboembolic complications, which significantly impact patient outcomes .

[0005] Various approaches have been proposed to monitor ventricular hemodynamics in LVAD patients . One common approach involves the indirect estimation of blood flow and derived parameters from motor signals of the LVAD. These methods, however, provide only differential values rather than absolute measurements of critical parameters like left ventricular pressure . Furthermore, other factors influencing motorcurrent, such as thrombosis, can negatively impact the quality of these estimations .

[0006] Alternative approaches include integrating pressure sensors within the inner surface of the inflow cannula, facing the blood, or measuring cardiac impedance to indirectly assess ventricular volume using electrodes placed on the outer surface of the cannula . Some designs employ a tapered inflow cannula combined with pressure measurement on the inner surface and Bernoulli-based flow estimation.

[0007] These prior art solutions suffer from significant limitations . Motor-based estimations cannot provide absolute values for left ventricular pressure and fail to deliver information on left ventricular contractility. Pressure transducers exposed to blood flow are subj ect to drift over time, requiring frequent calibration that is difficult to perform in an implanted device . Impedance-based approaches focus primarily on estimating ventricular volume rather than pressure, providing an incomplete picture of cardiac function .

[0008] The existing approaches to monitoring cardiac function in LVAD fail to provide accurate, absolute measurements of key cardiac function markers, particularly left ventricular pressure parameters . Blood-contacting sensors are prone to drift, thrombus formation, and other complications that reduce long-term reliability. Current methods do not adequately assess left ventricular contractility, which is essential for optimizing LVAD function and preventing complications . Existing monitoring systems do not facilitate continuous, real-time assessment of cardiac function outside clinical settings . The inability to precisely monitor leftventricular function limits the potential for automated, responsive adjustment of LVAD operation to match the patient ' s changing physiological needs .

[0009] The obj ect of the present invention is to overcome these limitations by providing a system that enables continuous, precise monitoring of cardiac function markers, particularly in patients with left ventricular assist devices . The invention aims to provide a more accurate and stable means of monitoring left ventricular function parameters, including left ventricular systolic pressure (LVSP) , left ventricular end-diastolic pressure (LVEDP) , maximum rate of left ventricular pressure increase (dp / dt max) , minimum rate of left ventricular pressure decrease (dp / dt min) , and / or end-systolic elastance (Emax) .

[0010] In order to solve these obj ects, the invention, according to a first aspect thereof, provides a system for determining cardiac function markers comprising: a sensor array configured for positioning in contact with myocardial tissue, wherein the sensor array includes at least one, preferably a plurality of, force or pressure sensor (s) to obtain, by contact with the myocardial tissue (5) , force or pressure signals representative of a contraction of the myocardial tissue; a data acquisition means coupled to the sensor array and configured to receive the pressure signals from the sensor array; and a processing unit configured to calculate or predict at least one cardiac function marker based on said pressure signals . The plurality of force or pressure sensors may be arranged in various geometrical arrangements including but not limited to linear, circular, triangular, or rectangular patterns .By positioning the sensor array in direct contact with myocardial tissue, the system achieves accurate measurements of cardiac function without the complications associated with blood-contacting sensors . This direct measurement yields unexpected precision in predicting ventricular function markers, with minimal deviation from values obtained through conventional catheterization methods . Testing has demonstrated that the system can predict cardiac parameters with mean differences and standard deviations from catheterization-derived values as low as 5.3 mmHg ± 0.06 mmHg for left ventricular systolic pressure (LVSP) and 3. 6 ± 0.15 mmHg for left ventricular end-diastolic pressure (LVEDP) .

[0011] The non-blood-contacting nature of the sensor array eliminates the drift issues commonly experienced with intravascular pressure sensors . Since the sensor (s) is / are positioned against the myocardium rather than exposed to blood flow, they maintain stability and accuracy over extended periods, eliminating the need for frequent recalibration.

[0012] In the context of the present invention, the phrase "in contact with the myocardial tissue" refers to the physical positioning of the sensor array such that it maintains direct mechanical coupling with the muscle tissue of the heart wall, specifically the myocardium. This contact may be established by implanting the sensor array into the myocardial wall itself, affixing it to the epicardial surface (outer layer) of the heart, or integrating it with devices that are positioned against the myocardium, such as on the outer wall of an LVAD inflow cannula that interfaces with heart tissue . The sensor array may for example be embedded between the subendocardial and the subepicardial layers of themyocardium. The sensor array is specifically designed to remain in a fixed position relative to the myocardium, allowing it to detect and measure the biomechanical forces generated during cardiac contraction and relaxation cycles . As the heart muscle contracts and relaxes, these movements and pressure changes are directly transmitted to the sensor array .

[0013] Generally speaking, the sensor (s) of the sensor array may preferably adhere to the myocardial tissue or be embedded into the myocardial tissue .

[0014] The presence of a plurality of force or pressure sensors ensures redundancy and reliability in measurement, regardless of the device ' s precise orientation upon implantation.

[0015] Preferably, the sensor array may comprise two, three or more force or pressure sensors in various geometrical arrangements including but not limited to linear, circular, triangular, or rectangular patterns .

[0016] According to the invention, a processing unit is provided to calculate or predict at least one cardiac function marker based on said pressure signals . It was unexpectedly discovered that properly processed external myocardial pressure signals contain sufficient information to derive accurate cardiac function markers without direct intracardiac measurements .

[0017] In a preferred embodiment of the invention, the processing unit is configured to extract, from the pressure signals, at least one signal feature representing mechanical dynamics of myocardial tissue; and predict, using a machine learning algorithm applied to the at least one extracted signalfeature, the at least one cardiac function marker . By extracting specific signal features that represent the mechanical dynamics of myocardial tissue, such as the minimum rate of change in sensor signal (dS / dt min) , maximum rate of change in sensor signal (dS / dt max) , and sensor peak amplitude (S) , the system captures the essential biomechanical characteristics during the cardiac cycle .

[0018] In a preferred embodiment of the invention, the processing unit is configured to preprocess the pressure signals to segment the signals into individual cardiac cycles and to extract the at least one signal feature from the pressure signals of each cardiac cycle . This enables processing of cardiac function on a beat-by-beat basis . Preferably, the segmentation process identifies the boundaries of each cardiac cycle by detecting characteristic patterns in the pressure signals that correspond to the beginning and end of ventricular contraction. Once the individual cardiac cycles are isolated, the processing unit extracts the specific signal features from each cycle as mentioned above .

[0019] In a preferred embodiment, the preprocessing may also include noise removal operations to enhance signal quality prior to feature extraction. The noise removal may include low-pass filtering to eliminate high-frequency interference .

[0020] The term "using a machine learning algorithm applied to the at least one extracted signal feature" as used in the present invention refers to the application of a predictive model that has been previously trained through machine learning techniques . This trained model, rather than the training algorithm itself, is what operates on the extracted signal features during normal device operation. The training processoccurs separately, typically during device development, where a machine learning algorithm, preferably a random forest regressor, is used to analyze a comprehensive dataset of paired sensor measurements and corresponding invasive catheterization values . During this training phase, the algorithm learns the complex relationships between the extracted signal features (such as dS / dt min, dS / dt max, and S and the target cardiac function markers (LVSP, LVEDP, dp / dt max, dp / dt min, and Emax) . Once trained, the resulting predictive model encapsulates these learned relationships as a fixed set of parameters and decision rules . This trained model is then implemented within the processing unit of the system, where it receives the extracted signal features as inputs and produces predictions of cardiac function markers as outputs .

[0021] The machine learning algorithm employed in the present invention may be developed through a training process using data collected from controlled experiments in animal, such as in swine . Initially, a dataset is compiled by simultaneously recording signals from the implanted sensor array and conventional catheterization measurements, which serve as the ground truth for cardiac function markers . This paired data collection may occur during procedures such as dobutamine stress tests, which induce varying levels of cardiac contractility, pre-load, and after-load, thereby capturing a wide range of cardiac states and loading conditions . The collected dataset is preprocessed to isolate individual heartbeats, remove noise, and extract key features from each cardiac cycle, including the minimum rate of change in sensor signal (dS / dt min) , maximum rate of change in sensor signal (dS / dt max) , and sensor peak amplitude (S) .A certain percentage, such as 80%, of this processed dataset may be used to train a random forest regressor, which may be selected for its robustness against overfitting and its ability to handle non-linear relationships between input features and output parameters . The random forest algorithm operates by constructing multiple decision trees during the training phase, with each tree making independent predictions based on random subsets of the training data . When presented with new sensor data, each decision tree in the forest independently predicts the cardiac function markers based on the extracted features, and the final prediction is determined by averaging the outputs across all trees . This enhances prediction accuracy and stability compared to single-model approaches . The remaining percentage, such as 20%, of the dataset may be used for validation, allowing for obj ective assessment of the algorithm' s performance through metrics such as coefficient of determination (R2score) and Bland-Altman analysis .

[0022] In a preferred embodiment, the sensor array comprises one or more force sensors suitable for measuring mechanical deformation or pressure . Suitable sensor technologies may include, for example, micro-electromechanical systems (MEMS) , piezoresistive, capacitive, or optical sensors, although the invention is not limited to these implementations . The miniaturized nature of MEMS technology enables the creation of very small force or pressure sensors, facilitating minimally invasive implantation with reduced tissue disruption and inflammatory response . MEMS sensors consume minimal power, making them suitable for long-term implantable applications where energy efficiency is important . The sensor array may be encapsulated in biocompatible materials such as medical-grade silicone .According to a second aspect of the invention, a left ventricular assist device system is provided comprising: a suture ring configured to be affixed to the outer surface of the left ventricular wall, providing both access into the ventricle and a mechanically stable anchoring interface for the device; an inflow cannula configured for implantation into a left ventricle of a heart inserted through and anchored to the suturing ring; a pump coupled to the inflow cannula and configured to pump blood from the left ventricle through the inflow cannula; and a system determining cardiac function markers according to the first aspect of the invention. By incorporating the sensor array and processing capabilities of the first aspect directly into an LVAD system, the invention enables continuous monitoring of cardiac function markers in patients receiving mechanical support . The real-time monitoring of cardiac parameters such as LVSP, LVEDP, dp / dt max, dp / dt min, and Emax allows for immediate detection of changes in ventricular function that might necessitate adjustments to LVAD settings or medical therapy. This capability is particularly valuable for preventing complications such as ventricular collapse or suction events that occur when LVAD output does not match the heart ' s actual hemodynamic state .

[0023] In an advantageous embodiment of the invention, the sensor array is positioned on an outer wall of the inflow cannula of the left ventricular assist device or the myocardial-facing side of the suturing ring. The inflow cannula is preferably designed as tubular structure that penetrates the cardiac wall, in order to allow blood flow from the ventricle into the pump housing. The suture ring is preferably designed as an annular structure configured to be affixed to the outersurface of the cardiac wall, serving to secure the inflow cannula at the myocardial interface and provide mechanical stability at the site of insertion. By integrating the sensor (s) directly onto the inflow cannula or the suturing ring, the invention eliminates the need for a separate implantation procedure for the monitoring system, thus reducing surgical complexity and potential complications . The cannula ' s and suture ring' s fixed positioning on or within the left ventricular apex, respectively, provides a stable and consistent platform for the sensor (s) , ensuring reliable contact with the surrounding myocardial tissue . This positioning allows the sensor (s) to detect the biomechanical forces generated by the ventricular wall as it contracts and relaxes around the cannula . Additionally, the inflow cannula, together with the suturing ring securing it, is typically positioned in the apical region of the left ventricle, which undergoes significant deformation during the cardiac cycle, making it a suitable location for detecting contractility changes .

[0024] Generally speaking, the sensor array may preferably be positioned such that when implanted, the sensor (s) of the sensor array face toward myocardial tissue of the left ventricle . This orientation ensures that the sensor (s) is / are directly exposed to the mechanical forces generated by myocardial contraction rather than to blood pressure prevailing within the ventricle . The sensor (s) may therefore be arranged on an outside of the myocardial wall . The orientation of the sensor (s) towards myocardial tissue also facilitates tissue integration over time, as the myocardium naturally grows and adheres to the sensor surfaces, further improving signal transmission and long-term stability.In a particularly advantageous embodiment, the invention further comprises a control module configured to automatically adjust a rotational speed of the pump based on at least one of the predicted cardiac function markers . By dynamically modulating pump speed in response to real-time cardiac function data, the system can maintain optimal hemodynamic support despite fluctuations in the patient ' s physiological state . For instance, when the system detects an elevation in left ventricular end-diastolic pressure (LVEDP) indicating increased preload, the control module can automatically increase pump speed to enhance ventricular unloading and prevent pulmonary congestion. Conversely, if the system observes a decrease in left ventricular systolic pressure (LVSP) or contractility markers that might signal impending suction events, it can reduce pump speed accordingly to prevent ventricular collapse and associated complications . This automated response capability is particularly valuable during periods of changing physiological demands, such as physical activity, sleep, or emotional stress, when manual adjustment would be impractical .

[0025] According to a third aspect, the invention provides a method for processing data from a sensor array to determine cardiac function markers, the method comprising:

[0026] receiving pressure signals from a sensor array that is in contact with myocardial tissue, wherein the sensor array includes at least one, preferably a plurality of, force or pressure sensor (s) , in various geometrical arrangements including but not limited to linear, circular, triangular, or rectangular patterns, to obtain pressure signals representative of a contraction of the myocardial tissue;preprocessing the pressure signals to segment the signals into individual cardiac cycles;

[0027] extracting, for each cardiac cycle, at least one signal feature from the preprocessed signals, wherein the features include at least one from the group consisting of : a minimum rate of change in sensor signal (dS / dt min) , a maximum rate of change in sensor signal (dS / dt max) , and a sensor peak amplitude ( S ) ;

[0028] inputting the extracted features into a trained machine learning algorithm;

[0029] generating, by the machine learning algorithm, predictions for one or more cardiac function markers selected from: left ventricular systolic pressure (LVSP) , left ventricular end-diastolic pressure (LVEDP) , maximum rate of left ventricular pressure increase (dp / dt max) , minimum rate of left ventricular pressure decrease (dp / dt min) , and end-systolic elastance (Emax) .

[0030] In a preferred embodiment of the invention, segmenting the signals into individual cardiac cycles comprises identifying peaks in the pressure signals corresponding to ventricular contractions . By identifying the characteristic peaks in the pressure signals that correspond to ventricular contractions, the system establishes precise temporal boundaries for each cardiac cycle, enabling accurate beat-to-beat analysis . This peak detection method allows the system to automatically adapt to variations in heart rate, maintaining accurate segmentation during both bradycardia and tachycardia, as well as during irregular rhythms where cycle length may vary significantly. The identification of ventricular contraction peaks also serves as a reliable fiducial point for temporal alignment of signals from multiple sensors within the array,facilitating the combination or comparison of data across sensors .

[0031] In the following, the invention will be described in more details with reference to exemplary embodiments shown in the drawings . Therein, Fig. 1 shows a schematic drawing of a LVAD device, Fig. 2 shows a schematic drawing of a sensor array, Fig. 3 is an illustration of the sensor data processing for obtaining cardiac function markers, and Fig. 4 illustrates the cardiac function markers .

[0032] Fig. 1 illustrates an exemplary embodiment of the left ventricular assist device (LVAD) system according to the second aspect of the present invention. The figure depicts a cross-sectional view of the LVAD showing the integration of the sensor array with the inflow cannula and pump components .

[0033] The LVAD system 1 comprises an inflow cannula 2 that is configured for implantation into the left ventricle 3 of a patient ' s heart, and a suture ring 4 configured to secure the inflow cannula to the ventricular wall 5. As shown in the figure, the inflow cannula has a tubular structure with an inner lumen through which blood flows from the left ventricle 3 into the pump housing 6, as depicted by arrow 7.

[0034] The sensor array 8a is positioned on the outer wall of the inflow cannula 2, with the sensing elements facing toward the myocardial tissue . In an alternative embodiment, the sensor array 8b may instead be positioned on the myocardial-facing side of the suture ring 4, allowing direct contact or proximity to the cardiac wall for detecting mechanical deformation. The distal portion of the inflow cannula 2 connects to the main pump housing 6, which contains therotating pump impeller 9. This impeller 9 is designed to create continuous blood flow from the left ventricle 3 through the inflow cannula 2, through the pump mechanism, and ultimately out through the outflow path 10.

[0035] The rotation of the impeller 9 generates the pressure differential necessary to propel blood from the left ventricle 3 into the systemic circulation, typically via an outflow graft connected to the ascending aorta (not shown in the figure) . The continuous flow nature of this pump design allows for consistent unloading of the left ventricle 3, reducing the workload on the failing heart .

[0036] Fig. 2a and b illustrate an exemplary embodiment of the sensor array according to the present invention in which a linear configuration is employed. As seen in the plan view according to Figure 2a, the array includes multiple contact pads 11 positioned along the length of the carrier substrate 12, which serve as interfaces for electrical connections to the data acquisition system. These contact pads 11 are connected to conducting paths 13 that run along the carrier substrate 12, providing the electrical routing necessary to obtain signals from the force or pressure sensors 14.

[0037] As seen in the side view according to Fig. 2b, in one embodiment, each force or pressure sensor 14 is mounted on a support structure 15 that functions as a spacer and maintains the optimal positioning of the sensors 14 relative to the tissue . While a linear arrangement is shown, other geometrical arrangements, such as circular, triangular, or rectangular patterns, may also be used depending on anatomical and functional requirements . In a preferred implantation, the sensors 14 are realized as bare-die siliconpressure sensors, such as model SM5108C-30-AX from Silicon Microstructures Inc . , capable of accurately measuring pressures up to 207 kPa . Each sensor 14 has dimensions of approximately 0. 65 x 0. 65 mm. The sensor array may be encapsulated in e . g. medical-grade silicone 16, to provide biocompatibility for long-term implantation . Alternative sensor technologies and encapsulation materials and methods may be employed depending on the specific application requirements .

[0038] Fig. 3 illustrates an exemplary embodiment of the present invention' s system for acquiring and processing myocardial pressure signals to predict cardiac function markers . The system comprises three primary functional blocks : Signal Acquisition (Fig. 3a) , Preprocessing (Fig. 3b) , and Machine Learning (Fig. 3c) .

[0039] The signal acquisition block, illustrated in Fig, 3a, represents the hardware components responsible for capturing biomechanical signals from the myocardium. At the core of this block is the sensor array, which includes multiple force or pressure sensors . Each sensor in the array detects mechanical forces generated by myocardial contraction and relaxation, converting these biomechanical events into electrical signals .

[0040] The electrical pressure signals from the array are directed to a signal converter, which conditions and amplifies the raw sensor outputs . This signal conditioning stage ensures optimal signal-to-noise ratio and prepares the signals for digitization. The conditioned analog signals are then routed to a multi-channel analog-to-digital converter (ADC) , which transforms the analog sensor outputs into digital datastreams . In a preferred embodiment, a 16-channel multiplexed, 24-bit ADC provides high-resolution digitization of the sensor signals, operating at a sampling rate of approximately 100 Hz to adequately capture the temporal dynamics of cardiac contraction cycles .

[0041] Fig. 3b depicts the preprocessing block, which prepares the digitized sensor data for subsequent analysis . This stage begins with a filtering operation to remove high-frequency noise and other artifacts that might interfere with accurate feature extraction. In the exemplary embodiment, a low-pass filter is applied to eliminate frequencies not associated with physiological cardiac dynamics .

[0042] Following filtration, the segmentation process identifies individual cardiac cycles within the continuous data stream. This process detects characteristic peaks in the pressure signals that correspond to ventricular contractions, establishing temporal boundaries for each heartbeat . This beat-by-beat segmentation is preferably for extracting cyclespecific features that accurately represent the cardiac function .

[0043] In the context of machine learning model development, additional processing steps are applied exclusively during the training phase . These steps are not part of the general signal preprocessing used during routine operation.

[0044] Specifically, the segmented data is divided into training and testing subsets (e . g. , 80% for training and 20% for testing) , allowing the performance of the machine learning algorithm to be validated on previously unseen data . Furthermore, each cardiac cycle in the training dataset is annotated with corresponding reference markers— such as left ventricular end-diastolic pressure (LVEDP) , left ventricular end-systolic pressure (LVESP) , the maximum and minimum rates of pressure change (dP / dt max and dP / dt min) , and the end-systolic elastance (Emax) , obtained from catheter-based measurements . These annotated data pairs serve as ground truth for supervised learning.

[0045] Fig. 3c illustrates the machine learning block, which comprises three key components : feature extraction, random forest, and cardiac function marker prediction.

[0046] The feature extraction component calculates specific characteristics from each cardiac cycle in the preprocessed sensor signals . For each of the sensors (labeled SI, S2, and S3 in the figure) , the system extracts several key features : the maximum rate of change in the sensor signal (dS / dt max) , the minimum rate of change in the sensor signal (dS / dt min) , and the peak signal amplitude (S) .

[0047] The extracted features serve as inputs to the random forest component, which is a machine learning model consisting of multiple decision trees . During the training phase, the model learns the relationships between the sensor-derived features and the corresponding cardiac function markers obtained through conventional catheterization measurements . The random forest architecture enhances prediction robustness by averaging the outputs of numerous individual decision trees, each trained on different subsets of the training data .

[0048] The final component, cardiac function, represents the output of the machine learning pipeline . The trained random forest model generates predictions for several cardiac function markers : left ventricular systolic pressure (LVSP) , leftventricular end-diastolic pressure (LVEDP) , maximum rate of left ventricular pressure increase (dp / dt max) , minimum rate of left ventricular pressure decrease (dp / dt min) , and / or end-systolic elastance (Emax) .

Claims

Claims :1 . A system for determining cardiac function markers comprising :a sensor array ( 8a, 8b) configured for positioning in contact with myocardial tissue (5) , wherein the sensor array (a, 8b) includes at least one, preferably a plurality of, force or pressure sensor (s) ( 14 ) to obtain, by contact with the myocardial tissue (5) , force or pressure signals representative of a contraction of the myocardial tissue (5) ;a data acquisition means coupled to the sensor array and configured to receive the pressure signals from the sensor array; anda processing unit configured to calculate or predict at least one cardiac function marker based on said pressure signals .

2. The system according to claim 1, wherein the processing unit is configured to extract, from the pressure signals, at least one signal feature representing mechanical dynamics of myocardial tissue (5) ; and predict, using a machine learning algorithm applied to the at least one extracted signal feature, the at least one cardiac function marker .

3. The system according to claim 2, wherein the machine learning algorithm comprises a trained random forest regressor .

4. The system according to claim 1 , 2 or 3, wherein the processing unit is configured to preprocess the pressure signals to segment the signals into individual cardiac cycles and to extract the at least one signal feature from the pressure signals of each cardiac cycle .

5. The system according to any one of claims 1 to 4, wherein the at least one signal feature is selected from: a minimum rate of change in sensor signal (dS / dt min) , a maximum rate of change in sensor signal (dS / dt max) , and a sensor peak amplitude (S) .

6. The system according to any one of claims 1 to 5, wherein the at least one cardiac function marker is selected from: left ventricular systolic pressure (LVSP) , left ventricular end-diastolic pressure (LVEDP) , maximum rate of left ventricular pressure increase (dp / dt max) , minimum rate of left ventricular pressure decrease (dp / dt min) , and end-systolic elastance (Emax) .

7. The system according to any one of claims 1 to 6, wherein the at least one force or pressure sensor ( 14 ) is configured as a micro-electromechanical systems (MEMS) sensor, a piezoresistive sensor, a capacitive sensor, or an optical sensor .

8. A left ventricular assist device system comprising:an inflow cannula (2 ) configured for implantation into a left ventricle (3) of a heart;a suture ring (4 ) configured to secure the inflow cannula (2 ) to the ventricular wall (5) ;a pump ( 9) coupled to the inflow cannula (2 ) and configured to pump blood from the left ventricle (3) through the inflow cannula (2 ) ; anda system according to any one of claims 1 to 7 .

9. The device according to claim 8, wherein the sensor array ( 8a, 8b) is positioned on an outer wall of the inflowcannula (2 ) or on a myocardial-facing surface of the suture ring ( 4 ) .

10. The device according to claim 8 or 9, wherein the sensor array ( 8a, 8b) is positioned such that when implanted, the sensor (s) ( 1$ ) of the sensor array ( 8a, 8b) face toward myocardial tissue of the left ventricle (3) .

11. The device according to claim 8, 9 or 10, wherein the sensor array 88a, 8b) is encapsulated in a biocompatible silicone material .

12. The device according to any one of claims 8 to 11, further comprising a control module configured to automatically adjust a rotational speed of the pump based on at least one of the predicted cardiac function markers .

13. A method for processing data from a sensor array to determine cardiac function markers, the method comprising: receiving pressure signals from a sensor array that is in contact with myocardial tissue, wherein the sensor array includes at least one, preferably a plurality of, pressure sensor (s) to obtain pressure signals representative of a contraction of the myocardial tissue;preprocessing the pressure signals to segment the signals into individual cardiac cycles;extracting, for each cardiac cycle, at least one signal feature from the preprocessed signals, wherein the features include at least one from the group consisting of : a minimum rate of change in sensor signal (dS / dt min) , a maximum rate of change in sensor signal (dS / dt max) , and a sensor peak amplitude ( S ) ;inputting the extracted features into a trained machine learning algorithm;generating, by the machine learning algorithm, predictions for one or more cardiac function markers selected from: left ventricular systolic pressure (LVSP) , left ventricular end-diastolic pressure (LVEDP) , maximum rate of left ventricular pressure increase (dp / dt max) , minimum rate of left ventricular pressure decrease (dp / dt min) , and end-systolic elastance (Emax) .

14. The method according to claim 13, wherein segmenting the signals into individual cardiac cycles comprises identifying peaks in the pressure signals corresponding to ventricular contractions .