Apparatus, systems, and methods for cardiac measurements and diagnostics

Abstract

A cardiac diagnostic apparatus 100 includes a phonocardiography (PCG) sensor 108 for sensing heart sounds and a mechanocardiography (MCG) sensor 109 for sensing heart induced motion, for example due t

Description

Field of the invention The present disclosure relates to a sensing apparatus for cardiac measurements and diagnostics, a cardiac diagnostic system, and computer-implemented methods for cardiac diagnostics. In particular, the sensing apparatus comprises a PCG sensor and an MCG sensor arranged in a common housing, configured for synchronous detection of cardiac activity. Background Cardiovascular disease, CVD, is a leading cause of disease burden in the world. Heart Failure, HF, in particular affects a large spectrum of this population, occurring at all ages and equally in both genders. HF increases with age, high blood pressure, obesity, and diabetes. Early detection of Heart Failure is critical to accessing appropriate and timely treatment to achieve the best possible outcomes. Nearly 80% of all HF diagnoses are initially made in emergency care, despite over 60% of these same patients presenting to primary care with signs and symptoms of HF in the 6 months before their diagnosis (van Gils, L. et al. (2017) ‘Prognostic implications of moderate aortic stenosis in patients with left ventricular systolic dysfunction’, Journal of the American College of Cardiology, 69(19), pp. 2383-2392.). Delay and missed diagnoses are common, typically confirmed only after symptoms have become obvious and severe damage has set in. Electrocardiography, ECG, and natriuretic peptide testing play a role in the clinical decision leading to a HF diagnosis, but they have limitations that contribute to high false positive and low detection rates in the early phase. Presently, HF is characterized mainly by echocardiography measurements that can identify functional and structural heart abnormalities, which underpin the HF condition. However, access to echocardiography is typically reserved for secondary care, not primary care, and is all too often delayed due to backlogs and long waiting lists. Due to the lack of a non-invasive medical device for use in primary care which is able to provide echocardiography equivalent findings, the opportunity for early detection of HF is missed. Summary of the invention Aspects of the invention are as set out in the independent claims and optional features are set out in the dependent claims. Aspects of the invention may be provided in conjunction with each other and features of one aspect may be applied to other aspects. An aspect of the invention relates to a sensing apparatus for cardiac diagnostics. The sensing apparatus comprises a phonocardiography, PCG, sensor, configured to sense heart sounds; and a mechanocardiography, MCG, sensor, configured to sense heart-induced motion, wherein the MCG sensor comprising a 3-axis gyroscope and a 3-axis accelerometer. The PCG sensor and the MCG sensor are preferably arranged within a common sensor housing. Heart-induced motion sensed by the MCG sensor may include at least one of heart displacement, acceleration, and velocity. In particular, the 3-axis accelerometer may be configured to capture position information relative to the heart's movement during a cardiac cycle. Systolic time intervals (STIs) estimated from interpreted MCG signal can provide valuable information about cardiac function and contribute to the diagnosis of heart failure. STIs are specific time intervals associated with cardiac movement during the cardiac cycle, particularly the systolic phase, and can reflect various aspects of ventricular function and hemodynamics. The MCG sensor comprising a 3-axis gyroscope and a 3-axis accelerometer (also referred to as a 6-axis IMU sensor) may be configured to stabilize mechanocardiogram (MCG) signals. For example, this may be achieved by leveraging gyroscope inputs to identify the sensor's location relative to the position of the heart based on the sensor's movements, for example induced by breathing. Stabilization may be achieved through correction information derived from the gyroscope signal, facilitating calculation of the relative position of the heart to the sensor to maintain consistent signal amplitudes throughout recording sessions, minimising skewing caused by motion-induced variations. In such examples, the processor may be configured to correct signals obtained from the MCG 3-axis accelerometer based on signals obtained from the MCG 3-axis gyroscope. Thus, providing an MCG sensor comprising a 3-axis gyroscope and a 3-axis accelerometer may advantageously improve the accuracy and reliability of the MCG accelerometer signal, both between heart cycles and recording sessions. Phonocardiography (PCG) may be advantageous for diagnosing valvular heart diseases and heart failure by capturing and analysing the characteristic sounds produced by the heart. PCG helps in detecting abnormal heart sounds, including murmurs, which are indicative of turbulent blood flow caused by valve abnormalities. Different valve diseases produce specific murmurs with distinct characteristics. For example, aortic stenosis typically produces a systolic ejection murmur, while mitral regurgitation produces a holosystolic murmur. PCG can help detect the presence of additional heart sounds, such as the third heart sound (S3) and the fourth heart sound (S4), which are associated with impaired ventricular function and heart failure. An S3 may indicate volume overload and diastolic dysfunction, while an S4 may indicate decreased compliance of the ventricle. Providing MCG and PCG sensors in a common housing may be advantageous to have a single sensor housing for attachment to the patient. It may also be advantageous for signal interpretation and analysis for the PCG and MCG sensors to have a known, fixed positioning relative to one another within the housing. The sensing apparatus may further comprise a processor configured to capture signals from the PCG sensor and the MCG sensor synchronously. This may be advantageous for signal interpretation since the synchronous PCG and MCG signals can be cross referenced during signal analysis and interpretation. For example, a processor may be configured to augment signals obtained from the PCG sensor based on the synchronous signals obtained from the MCG sensor. Due to the known arrangement of the PCG and MCG sensors within the common sensor housing, the gyroscope may also be used to complement the PCG signals in the same manner described above with reference to stabilisation of the MCG accelerometer signal. Thus, the sensing apparatus of the present invention may advantageously also improve the accuracy and reliability of the PCG signal. The sensing apparatus may comprise said processor, or otherwise be configured to communicate with such a processor. The sensing apparatus may further comprise at least one electrocardiography, ECG, electrode, wherein the at least one ECG electrode is arranged within the sensor housing. In some example, the sensing apparatus may comprise one ECG lead (two ECG electrodes). In some examples, the sensing apparatus may comprise three ECG electrodes within the sensor housing. Alternatively, a separate ECG apparatus, such as a 12-Lead ECG, may be used in conjunction with the sensing apparatus. Cardiac problems primarily detected by ECG typically include, but are not limited to, arrhythmia, premature beats, or bundle branch block. Structural, functional or valvular problems may only be anticipated from repolarization anomalies, such as aberrant ST-T in ischemic heart disease, however the PCG / MCG bio-signals advantageously extend the ECG features and provide better prediction of such abnormalities, and more. The processor may be configured to capture signals from the PCG sensor, the MCG sensor, and the at least one ECG electrode synchronously. The sensing apparatus of any preceding claim may further comprise an attachment means configured to attach the sensor housing to a patient. For example, the attachment means may be a strap or harness. This may be advantageous to reduce motion of the sensors relative to the patient during signal acquisition, for example securing the sensor apparatus against the patient’s thoracic wall. In addition, the attachment means may facilitate accurate and reproduceable placement of the sensors relative to the patient. The sensor housing, and / or optionally the attachment means, may comprise an acoustic insulating material. This may be advantageous to reduce external noise and vibrations detected by the sensors. This may be particularly beneficial to reduce noise in the PCG signal. The sensing apparatus may further comprise a grip portion made of malleable material. This may be advantageous for enhancing grip comfort and reducing external vibrations and acoustic noise during handheld positioning of the sensor housing on a patient. The sensing apparatus may further comprise a communication means configured to communicate signals obtained from the PCG sensor and the MCG sensor to a remote device. Preferably, the communication means is a wireless communication means. In some examples, the MCG sensor may be configured to capture mechanical vibrations in the range of 0Hz or 1 Hz to 800Hz; and / or, in some examples, the PCG sensor may be configured to capture acoustic sounds in the frequency range of 20Hz to 1200Hz. In some examples, the sensing apparatus may be configured to be placed on a patient between V4 and V5, wherein V4 represents the 5th intercostal space at midclavicular line, and V5 represents a horizontal position level with V4 at left anterior axillary line. Alternatively, the sensing apparatus may be configured to be placed at any one of the four standard auscultation points on the patient thoracic wall. However, the skilled person will understand that this is not limiting. Another aspect of the invention relates to a cardiac diagnostic system. The system comprises a phonocardiography, PCG, sensor, configured to sense heart sounds, a mechanocardiography, MCG, sensor, configured to sense heart-induced motion, wherein the MCG sensor comprising a 3-axis gyroscope and a 3-axis accelerometer, at least one electrocardiography, ECG, electrode, and a processor. The processor is configured to capture signals from the PCG sensor, the MCG sensor, and the at least one ECG electrode synchronously. In some examples, the PCG sensor and the MCG sensor, and optionally the at least one ECG electrode, are provided by the sensing apparatus for cardiac diagnostics of the preceding aspect, arranged in a common housing. The features of preceding aspect may therefore also apply to such examples. The at least one ECG electrode may comprise ten ECG electrodes configured to provide a 12-Lead ECG. The processor may be configured to capture signals from the PCG sensor, the MCG sensor, and the at least one ECG electrode synchronously for at least a 60 second time period. However the skilled person will understand that this example is not limiting. The cardiac diagnostic system preferably comprises only one PCG sensor and one MCG sensor. This may be advantageous to streamline data acquisition, without relying on multiple PCG and MCG sensors. This may be facilitated by the provision of the MCG sensor comprising the gyroscope which can be used for signal stabilisation to enable robust and reliable data collection from a single PCG and MCG sensor. Providing only one PCG sensor and one MCG sensor may also be advantageous to lightweight the data collection compared to collection of multiple data streams from duplicate sensors, thus reducing required processing power and data storage. Optionally, the PCG sensor and the MCG sensor may be configured to be placed on a patient between ECG electrode placement locations V4 and V5, wherein V4 represents the 5th intercostal space at midclavicular line, and V5 represents a horizontal position level with V4 at left anterior axillary line. Alternatively, the sensing apparatus may be configured to be placed at any one of the four standard auscultation points on the patient thoracic wall. Alternatively, the cardiac diagnostic system may comprise up to four PCG sensors and up to four MCG sensor. For example, this may include up to four sensing apparatuses. In such examples, a sensing apparatus may be positioned at any of the standard auscultation positions, for example including the aortic valve, pulmonary valve, tricuspid valve, and / or mitral valve locations. The cardiac diagnostic system may further comprise a computing device for receiving the captured signals from the processor. Preferably, the computing device may be configured to receive synchronous ECG, PCG, and MCG signals, and determine whether any one of a set of heart abnormalities is likely to be present based on the obtained signals. For example, the computing device may comprise an artificial neural network, the artificial neural network being configured to receive synchronous ECG, PCG, and MCG signals, and determine whether any one of a set of heart abnormalities is likely to be present. Optionally, the computing device may be configured to perform the computer-implemented method for predicting heart failure described in more detail below, based on the obtained signals from the ECG, PCG, and MCG sensors. Another aspect of the invention relates to a computer-implemented method for predicting heart failure. The method comprises: obtaining (i) an ECG signal, (ii) a PCG signal, and (iii) an MCG signal, wherein the ECG signal, the PCG signal, and the MCG signal are synchronised; predicting the presence or absence of each of a set of heart abnormalities, based on the obtained synchronous signals; determining a risk score of heart failure with preserved ejection fraction, HFpEF, based on each heart abnormality which is predicted to be present from the set of heart abnormalities. This method of determining a risk score of heart failure with preserved ejection fraction, HFpEF, may be advantageous as Heart Failure with preserved Ejection Fraction is typically difficult to predict in primary care. This is because, by nature of the preserved ejection fraction, typical indicators used to predict heart failure, such as reduced left ventricle ejection fraction, LVEF, overlook heart failure with preserved ejection fraction. The set of heart abnormalities may include structural, functional, and / or valve diseases. In some examples, determining the risk score of HFpEF may comprise aggregating risk contribution values associated with each heart abnormality which is predicted to be present from the set of heart abnormalities. In some examples, determining the risk score of HFpEF, is based on the predicted presence or absence of each heart abnormality from the set of heart abnormalities and an associated abnormality weighting, wherein each abnormality weighting is associated with one of heart abnormalities from the set of heart abnormalities, and wherein each abnormality weighting represents the relative risk contribution of said heart abnormality to heart failure with preserved ejection fraction. In some examples, predicting the presence or absence of each of a set of heart abnormalities comprises predicting the severity of each of the set of heart abnormalities. This may be, but is not limited to, a severity score or severity classification. Examples severity classifications may include, but are not limited to: Normal, Borderline, Mild, Moderate, Severe, etc. Optionally, determining the risk score of heart failure with preserved ejection fraction, HFpEF, is further based on the severity of each identified heart abnormality. For example, more severe heart abnormalities may be more heavily weighted. In some examples, the method comprises determining a severity classification for each heart abnormality predicted to be present, based on the obtained synchronous signals, and determining the risk score of heart failure with preserved ejection fraction, HFpEF, further based on the severity classification for each identified heart abnormality. Unlike traditional binary interpretations, for example of ECG signals, this approach may advantageously introduce a nuanced dimension by considering the severity of identified conditions. This additional dimension contributes to a comprehensive understanding of the patient's cardiac health, enabling finer analysis of the disease's impact and more accurate predictions. Severity information for identified conditions may also provide clinicians with valuable insights into disease progression and treatment planning. In some examples, predicting the presence or absence of each of a set of heart abnormalities may use an artificial neural network, the artificial neural network being configured to receive synchronous ECG, PCG, and MCG signals, and determine whether a heart abnormality from the set of heart abnormalities is likely to be present. Optionally, the artificial neural network may be configured to receive synchronous ECG, PCG, and MCG signals, and assign a severity classification to each heart abnormality from the set of heart abnormalities. In some examples, the method may further comprise outputting a predicted severity of heart failure, based on the determined risk score. The method may further comprise determining left ventricle ejection fraction, LVEF, based on the obtained synchronous signals, and determining a risk score of borderline heart failure, wherein the risk score for borderline heart failure is based on the first risk score of heart failure with preserved ejection fraction, HFpEF, and the determined LVEF. Borderline heart failure may defined as a condition when a combination of borderline abnormal LVEF and borderline HFpEF score are present. For example, a borderline LVEF range may be approximately 50 to 55%. “Borderline” refers to not yet mild abnormalities, but abnormalities such that the possibility of heart failure should be considered when the entire patient clinical condition is considered. The method may further comprise determining left ventricle ejection fraction, LVEF, based on the obtained synchronous signals, and classifying the severity of left ventricle systolic dysfunction, LVSD, based on the obtained synchronous signals. Optionally, the method may then further comprise outputting a classification of heart failure with reduced ejection fraction, HFrEF, or heart failure with mid-range ejection fraction, HEmrEF, based on the determined LVEF and the classified severity of LVSD. The set of heart abnormalities may comprise one or more structural, functional, valvular, and / or electrical abnormalities. For example, the set of heart abnormalities may comprise one or more of: atrial fibrillation, diastolic dysfunction, wall motion abnormality, left ventricular hypertrophy, left atrial enlargement, right atrial enlargement, ventricular enlargement, aortic stenosis, mitral stenosis, dilated cardiomyopathy, aortic regurgitation, mitral regurgitation, tricuspid regurgitation, and pulmonary hypertension. In a preferred example, the set of heart abnormalities comprises diastolic dysfunction (DDIM), wall motion abnormality (WMA), left ventricular and / or left atrial abnormality (LVH, LAE), aortic stenosis (AS), pulmonary hypertension (PH), and Atrial Fibrillation (AF). However, the skilled person will understand that this is not intended to be limiting. For example, other abnormalities typically detected by ECG may also be used, for example including but not limited to Myocardial infarction, ischeamia, Branch Blocks, ST changes, QT, PR, Rhythm Abnormalities, etc. Determining the risk score of HFpEF may be based on the aggregating the predicted heart abnormalities. The method may further comprise determining whether atrial fibrillation is present based on the ECG signal. In such examples, determining the risk score of heart failure with preserved ejection fraction, HFpEF, may be further based on the determination of atrial fibrillation. The method may further comprise identifying fundamental heart sounds from the PCG signal, based at least in part on the synchronised MCG signal. Optionally, the method may further comprise identifying a location of the identified fundamental heart sounds, based on comparison to the synchronised ECG signal. Predicting the presence or absence of each of a set of heart abnormalities may then be based on at least one of the identified fundamental heart sounds, and optionally the location of the identified fundamental heart sounds, and the ECG signal. In some examples, the method may further comprise obtaining an indication of patient body size; and normalising at least one of the obtained ECG signal, PCG signal, and / or MCG signal based on the indication of patient body size, prior to predicting the presence or absence of each of the set of heart abnormalities. Normalising for body size, such as but not limited to Body Mass Index (BMI), Body Surface Area (BSA), or height, may be advantageous to address uncertainties in predicting the presence or absence of each of the set of heart abnormalities for factors dependent on body size. This may advantageously enable a more accurate prediction compared to traditional comparative measures derived solely from age and gender. Optionally, at least the synchronised PCG signal and MCG signal are obtained from the sensing apparatus of the first aspect of the invention. Alternatively, or on addition, the synchronised ECG signal, PCG signal, and MCG signal may be obtained from the cardiac diagnostic system of the second aspect of the invention. Another aspect of the invention relates to a computer-implemented method for fundamental heart sound detection, the method comprising: obtaining a phonocardiography, PCG, signal from a PCG sensor; obtaining a mechanocardiography, MCG, signal from an MCG sensor, wherein the obtained MCG signal is synchronised relative to the obtained PCG signal; and performing heart sound segmentation on the obtained PCG signal, based on the PCG signal and the MCG signal. Augmenting the PCG signal with the MCG signal may be advantageous to improve the reliability and accuracy of heart sound segmentation and subsequent interpretation and analysis, compared to PCG analysis alone. For example, the method may further comprise determining an indication of patient breathing based on the MCG signal. Performing heart sound segmentation on the obtained PCG signal may then be based on the PCG signal and the indication of patient breathing derived based on the MCG signal. In some examples, performing heart sound segmentation on the obtained PCG signal may comprise segmenting the PCG signal using thresholding to obtain a Shannon Energy envelogram. Another aspect of the invention relates to a computer-implemented method for predicting the presence of a cardiac abnormality based on electrocardiography, ECG, data for a given heart. The method comprises training a neural network with cardiac data sets, each set comprising echocardiography data, identified cardiac abnormality data, electrocardiography (ECG) data, phonocardiography (PCG) data, and mechanocardiography (MCG) data for the same heart, the sets being associated with different hearts. The method may then further comprise obtaining a set of ECG data, PCG data, and MCG data of a given heart, and using the trained neural network to output a prediction of at least one cardiac abnormality for said given heart based on the obtained set of ECG data, PCG data, and MCG data of a given heart. Preferably, the obtained set of ECG data, PCG data, and MCG data for a given heart is synchronised. In some examples, each set of the cardiac data sets further comprises an indication of patient data associated with the same heart, wherein the indication of patient data comprises at least one of age, gender, race, height, weight, blood pressure, a body measurement, such as waist and / or hip size, and / or any other indication of body size. In some examples, each set of the cardiac data sets further comprises wall motion abnormality data for the same heart. This may be advantageous as wall motion abnormality data may enhance the sensitivity in ECG myocardial infarction (Ml) detection, particularly in cases of chest pain episodes the cause of which, in some cases, may not be detected by ECG alone. In some examples, the identified cardiac abnormality data comprises identified heart failure with preserved ejection fraction, HFpEF. Using the trained neural network to output a prediction of at least one cardiac abnormality for said given heart may then further comprise outputting a prediction of heart failure with preserved ejection fraction, HFpEF. The cardiac abnormality data derived based on the echocardiography data may comprise data corresponding to a set of pre-determined heart abnormalities. The set of pre-determined heart abnormalities may comprise, but is not limited to, atrial fibrillation, diastolic dysfunction, wall motion abnormality, left ventricular hypertrophy, left atrial enlargement, aortic stenosis, and pulmonary hypertension. The method may further comprise outputting a severity of the at least one cardiac abnormality for said given heart. The method may further comprise training the neural network with cardiac data sets, wherein each set further comprises associated left ventricular ejection fraction, LVEF, for the same heart. The trained neural network may then be further used to determine a left ventricular ejection fraction, LVEF, for said given heart, and output a prediction of heart failure, based on the LVEF and the at least one predicted cardiac abnormality for said given heart. Another aspect related to a computer-implemented method for predicting the presence of a cardiac abnormality based on electrocardiography, ECG, data for a given heart. The method comprises training a neural network with cardiac diagnostic data sets, each set comprising echocardiography data from a heart diagnosed with a non-ST segment elevation myocardial infarction (NSTEMI), wall motion abnormality data, and electrocardiography, ECG, data for the same heart, the sets being associated with different hearts. The method may then further comprise receiving a set of ECG data and wall motion abnormality data of a given heart, and using the trained neural network to identify the presence of non-ST segment elevation myocardial infarction for said given heart. Another aspect of the invention relates to a method for obtaining cardiac information, the method comprising obtaining a phonocardiography, PCG, signal from a PCG sensor; obtaining a mechanocardiography, MCG, signal from an MCG sensor; and obtaining an electrocardiography, ECG, signal form at least one ECG electrode; wherein the obtained signals are synchronised (e.g., they are obtained synchronously). Preferably, the MCG signal is obtained from an MCG sensor comprising an accelerometer and a gyroscope. In some examples, the PCG signal is obtained from only one PCG sensor, and / or the MCG signal is obtained from only one MCG sensor. Another aspect of the invention relates to a computer program product comprising instructions configured to program a programmable processor to perform the method of any of the preceding aspects of the invention. Drawings Embodiments of the disclosure will now be described, by way of example only, with reference to the accompanying drawings, in which: Fig. 1 shows a box diagram of an example cardiac diagnostic system of the present invention. Fig. 2 shows a schematic illustrating an example sensor configuration for placement on a patient, P. Fig. 3 shows a box diagram of an example of a cardiac diagnostic system, such as the cardiac diagnostic system of Fig. 1, in use. Fig. 4 shows a flow diagram of an example computer-implemented method of the present invention for determining a risk score of heart failure with preserved ejection fraction, HFpEF. Fig. 5A illustrates example findings output by a computer-implemented method, based on ECG signals. Fig. 5B illustrates example findings relating to a set of heart abnormalities output by a computer-implemented method, based on synchronous biosignals, including MCG, PCG, and ECG data. The output data also includes severity indications in relation to a set of heart abnormalities and / or cardiac features. Fig. 5C illustrates example findings relating to myocardial infarction or ischemia output by a computer-implemented method, based on ECG signals. Fig. 6 illustrates an example two-dimensional representation of LVEF plotted against HFpEF score for predicting heart failure, including heart failure severity and type. Fig. 7 illustrates an example training and validation method to train Al models according to the present invention, suitable for predicting the presence (or absence) of a set of heart abnormalities. Specific description Embodiments of the claims relate to a sensing apparatus for cardiac diagnostics, a cardiac diagnostic system, and computer-implemented methods for determining cardiac measurements, predictions, and diagnostics, including predicting heart failure. It will be appreciated from the discussion above that the embodiments shown in the Figures are merely exemplary, and include features which may be generalised, removed or replaced as described herein and as set out in the claims. Fig. 1 shows a box diagram of an example cardiac diagnostic system 100 of the present invention. In particular, the system 100 comprises a sensing apparatus 102 for cardiac diagnostics. The sensing apparatus 102 comprises a phonocardiography, PCG, sensor 108, and a mechanocardiography, MCG, sensor 109, arranged within a common sensor housing. The MCG sensor 109 comprises a 3D gyroscope 110 and a 3D accelerometer 112. The sensing apparatus 102 is coupled to a processor 104. The processor 104 is also coupled to at least one electrocardiography, ECG, electrode 106. Preferably, the at least one ECG electrode 106 comprises a set of ten ECG electrodes configured for a 12-Lead ECG set up. The processor 104 is also coupled to a display 114. Preferably, the processor 104 is coupled to a computing device, including a display 114, such as a computer, laptop, or tablet device. The PCG sensor 108 is configured to sense heart sounds. Preferably, the PCG sensor is configured to capture acoustic sounds in the 20Hz to 1200Hz range. For example, the PCG sensor may comprise a filtrating diaphragm. The MCG sensor 109 is configured to sense heart-induced motion. Preferably, the MCG sensor is configured to sense mechanical vibrations in the 1Hz to 800Hz range, and the MCG sensor is configured to sense position information relative to the heart's movement during a cardiac cycle. The ECG electrode(s) 106 is configured to sense electrical signals in the heart. The processor 104 is configured to capture signals from the PCG sensor 108, the MCG sensor 109, and the at least one ECG electrode 106 synchronously. The processor 104 is further configured to share the synchronised PCG, MCG, and ECG signals with the computing device 114. Fig. 2 shows a schematic illustrating an example placement of a 12-Lead ECG (comprising ten ECG electrodes) on a patient, P. Position C1 is configured to be positioned approximately at the 4th intercostal space to the right of the sternum. Position C2 is configured to be positioned approximately at the 4th intercostal space to the left of the sternum. C3 is configured to be positioned approximately midway between C2 and C4. C4 is configured to be positioned approximately at the 5th intercostal space at the midclavicular line. C5 is configured to be positioned approximately at the anterior axillary line at approximately the same level as C4. C6 is configured to be positioned approximately at the midaxillary line at approximately the same level as C4 and C5, or at approximately the same height as C3. Position R is configured to be positioned anywhere between the right shoulder and the wrist, preferably at the right wrist. Position L is configured to be positioned anywhere between the left shoulder and the wrist, preferably at the left wrist. Position N is configured to be positioned anywhere above the right ankle and below the torso, preferably at the right ankle. Position F is configured to be positioned anywhere above the left ankle and below the torso, preferably at the left ankle. In addition, Fig. 2 indicates a preferred placement position 200 of the sensing apparatus 102, comprising the PCG sensor 108 and MCG sensor 109. In particular, the sensing apparatus 102 is configured to be positioned between C4 and C5, wherein C4 represents the 5th intercostal space at midclavicular line, and C5 represents a horizontal position level with C4 at left anterior axillary line. However, alternatively, the sensing apparatus 102, comprising the PCG 108 and MCG 109 sensors housed in a single common sensor housing, may be placed at any one of the four standard auscultation points on the patient thoracic wall, indicated by 200, 202, 204, and 206 respectively. In some examples, a user, such as a nurse practitioner or other medical professional, manually applies pressure to the top of the sensor apparatus 102 to secure it against the thoracic wall at the placement position 200. The sensor apparatus 102 may further comprise an isolation housing, comprising a malleable material for enhancing grip comfort and reducing external vibrations and acoustic noise during handheld positioning. Alternatively, the sensor apparatus 102 may be secured against the thoracic wall at the placement position 200 using an acoustic and vibration a belt or harness wrapped around the user's body. Preferably, the belt / harness is constructed from vibration isolating and acoustic insulating materials to maintain the sensor apparatus 102 in the optimal position for signal recording, with minimal noise. Fig. 3 shows a box diagram schematic of an example cardiac diagnostic system 100 of the present invention in use. As shown in Fig. 1, the system 100 comprises a sensing apparatus 102 for cardiac diagnostics which comprises one PCG sensor and one MCG sensor (not shown). The PCG sensor is configured for Pulse Density Modulation, PDM, to Pulse Code Modulation, PCM, translation. Preferably the PCG sensor comprises a 4000 Hz sampling rate. The MCG sensor further comprises a 3D (3-axis) gyroscope and a 3D (3-axis) accelerometer. The physical ranges are preferably +-2g, and / or the sampling rate is preferably 4000 Hz. Preferably, the sensing apparatus 102 for cardiac measurements and diagnostics, comprises a digital sensor capable of capturing mechanical vibrations in the 1Hz to 800Hz range, a filtrating diaphragm for capturing acoustic sounds in the 20Hz to 1200Hz range, a second digital sensor for capturing position information relative to the heart's movement during a cardiac cycle; and a digital 3-axis gyroscope sensor for adjusting the recording envelope to maintain consistent angle and direction of vibrations and sounds. Said sensor apparatus is configured to have a higher signal-to-noise ratio compared to analog sensors, thereby enhancing accuracy in signal processing, particularly in detecting low-amplitude heart sounds, and enabling more precise measurements, diagnostic performance, and severity assessment of cardiac conditions. The system 100 further comprises at least one ECG electrode 106. In the example shown, the at least one ECG electrode 106 comprises a set of ten ECG electrodes configured for a 12-Lead ECG set up. Preferably the ECG module 106 has a 4000Hz sampling rate. The ECG electrode set up 106 and the sensing apparatus 102 are coupled to the chest of a patient, P; preferably according to the set up as shown in relation to Fig. 2. The system 100 further comprises a processor 104. The sensing apparatus 102 is coupled to the processor 104. The processor 104 is configured to receive signals obtained from each of the ECG electrode set up 106, the one PCG sensor, the one 3D MCG accelerometer, and the one 3D MCG gyroscope. The processor 104 is configured to ensure that the obtained signals are synchronised via synchronous data acquisition. In particular, the processor 104 is configured to synchronously combine the obtained bio-signals for transmission. The system 100 further comprises a means for data communication configured to communicate the synchronised sensor data to a remote device. In this example, the means for data communication comprises a USB port 302, preferably a high-speed USB port. The skilled person will understand however that other means for data communication may be used, including for example wireless communication modules, configured to wirelessly communicate the synchronised sensor data to a remote device, including but not limited to Wi-Fi (RTM) or Bluetooth (RTM). The USB port 302 is coupled to a corresponding USB port 304 on a remote computing device 306. The remote computing device 306 is preferably a personal computing device, such as a desktop PC or laptop. The remote computing device 306 is configured to perform real-time processing of the obtained signals to ensure signal quality and integrity. In particular, the remote computing device 306 is configured to plot the signals in real-time and perform noise level monitoring. The real time processing comprises applying standard lowpass and highpass filters (signal type specific), preparing signals for real-time monitoring plot, detecting QRS complexes to measure heart rate and check signal quality for every heartbeat. The computing device 306 is configured to identify and generate an alert in the event of electrode fall or slippage, or any quality problems. If the obtained signal recording quality is not sufficiently high, the remote computing device 306 will reject the recorded test and require another, repeat test to be conducted on the patient, for example by a nurse practitioner. The real-time plot, and any alerts, are displayed to a user, such as a nurse practitioner, via a user interface display. The output of the remote computing device 306 is an encrypted file uploaded to a remote server 320. The file comprises patient identification data and the recorded signals obtained from the cardio diagnostic system. The remote server 320 is configured to store the uploaded files. The remote server 320 is also configured to execute a computer-implemented method for predicting heart failure, based on the uploaded file. The method includes processing the ECG, PCG and MCG signals, and using knowledge-based rules and machine learning to output a heart failure prediction. This is discussed in more detail in relation to Fig. 4 below. Optionally, the remote server is further configured to output a predicted classification of heart failure type and severity, based on the heart failure prediction. The remote server 320 may also be configured to output ECG-findings, PCG-findings, MCG-findings, and prediction of heart abnormalities based on the ECG, PCG and MCG signals. The above are output as a report which can be viewed by medical professionals, such as doctors, via a web-based clinical management system. Fig. 4 shows a flow diagram for an example computer-implemented method for predicting heart failure, for example for execution by a remote server or other remote computing device, such as but not limited to remote server 320 or remote computing device 306. For the purposes of this example, reference is made to a remote server, such as remote server 320, however the skilled person will understand that other devices and / or processors may be used. The remote server first obtains (i) an ECG signal, (ii) a PCG signal, and (iii) an MCG signal, wherein the ECG signal, the PCG signal, and the MCG signal are synchronised (402). The ECG signal, PCG signal, and MCG signal may be collectively referred to as bio-signals. As described above, the synchronised biosignals may be received from a cardiac measurement device, such as that described in relation to Fig. 1 or Fig. 3. The remote server may then perform bio-signal pre-processing, including at least one of: cleaning the bio-signals using digital filters, artifact removal, and 3D stabilisation techniques, for example based on the synchronised gyroscope signal which is a component of the MCG signal. In particular, the 6-axis MCG sensor, comprising a 3-axis accelerometer and a 3-axis gyroscope, is utilized to stabilize MCG signals, leveraging the gyroscope data to identify the sensor's location relative to the sensor's movements induced by breathing, relative to the position of the heart. Stabilization can be achieved through correction information derived from the gyroscope inputs, which enables calculation of the relative position of the heart to the sensor to maintain consistent MCG signal amplitudes throughout recording sessions, minimizing skewing caused by motion-induced variations, such as breathing induced motion of the thoracic wall. This advantageously results in robust and reproducible accelerometer signals between heart cycles and recording sessions enhancing signal accuracy and reliability. In a preferred example, the MCG signals are corrected by the gyroscope signals utilizing trained shallow neural networks to mitigate dynamic attitude variability. As one example, a method for signal stabilisation of a MCG signal may comprise utilizing a shallow Cascade Forward Neural Network (CFNN) to process multi-channel time series MCG data by incorporating the integrated 3D accelerometer signal converted to represent velocity (for example using integration with high-pass filter) as the primary input to the CFNN; and employing the gyroscope signal that represents 3D angular velocity as a secondary input to the CFNN. As such, the method generates a corrected velocity signal as the output of the CFNN. The corrected MCG accelerometer signal can then be derived from the first derivative of the CFNN’s output velocity signal. As a results, accelerometer signals become more robust and reproducible between heart cycles and between recording sessions. The remote server may also perform signal segmentation to segment the synchronised bio-signals into heart cycles. Signal segmentation may be determined collectively for all bio-signals, based on their synchronicity. For example, the remote server may perform fundamental heart sound detection on phonocardiogram (PCG) and mechanocardiogram (MCG) signals. This may comprise detecting fundamental heart sounds (first and second sounds) on PCG signals by employing adaptive thresholding on Shannon sound envelope method. When uncertainty arises in the PCG pattern during sound detection, for example when a confidence threshold is not met for a given sound, the remote server may utilise the MCG signal by performing fusion of synchronised PCG and MCG signal envelopes for fundamental sound detection. In addition, the remote server may optionally utilise synchronised electrocardiogram (ECG) information to localize potential locations of the first and second heart sounds. For example, QRS annotations are labelled on the ECG signal as an array of time moments identifying the QRS peak (R wave) placements. These annotations may be used to separate heart beats for beat averaging and further processing of ECG, PCG and MCG signals. As such, the remote server may then classify the detected sounds, for example as first or second heart sounds, based on incorporation of knowledge-based rules with the synchronised signals. The remote server may also employ adaptive threshold-based recognition to identify sounds within the most likely and expected time region. The remote server also performs heart sound sub-segmentation or sub-segment recognition on the PCG signal to find the characteristic points of the fundamental heart sounds: onset, first peak (M1 or A2), second peak (T1 or P2) and offset. The remote server takes into account breathing expiration and inspiration information derived from MCG signal during sound identification, because in most cases the second sound pattern depends on breathing phase. The aim of breathing segmentation and measurement derived from MCG signals is threefold: 1) Detecting respiration extreme points, such as start of inhalation and start of exhalation. These segment annotations are used in the PCG sub-segmentation, e.g., the wide second sound depends on the breathing phase. 2) The breathing phase is analysed and used in typical beat sequence detection. The breathing phase represents the actual state of inhalation or exhalation for each heartbeat and therefore introduces measurement consistency through the breathing phase. 3) The breathing measurements are included in a combined feature vector for the bio-signals. MCG segmentation in this context also refers to the MCG heart event detection on a seismocardiogram (SCG)-like signal. Importantly, the aortic and mitral valve opening and closure events are identified from the MCG signal. The PCG sub-segments are used to help the identification of MCG heart events. As they are in sync, in all the cases the PCG provided heart sounds coincide with the aortic and mitral opening / closure events. The events or sub-segments for each heartbeat identified by the remote server include the following: Mitral Closure (MC), Aortic Opening (AO), Isovolumic Movement (IM), Isovolumic Contraction (IC), Rapid Ejection (RE), Aortic Closure (AC), Mitral Opening (MO), Rapid Filling (RF). The synchronized PCG and MCG signals may therefore enable a unique method for detection of heart events and sounds due to their synchronicity, improving the reliability and accuracy of sound detection and heart event based on PCG and MCG signals. This detection may then be utilised for signal segmentation into heart signals, for example based on knowledge-based rules. Based on the segmented signals, the remote server may then extract measurements from the segmented bio-signals, for example including amplitudes, intervals, time-frequency representation, cross-power spectrum density and statistical features. In particular, the systolic time intervals can be calculated based on to the synchronized and segmented bio-signals. Simple ECG analysis is also performed using rules, such findings include but are not limited to: Axis related findings (Leftward Axis I Rightward Axis / Extreme Axis), Interval related findings (Bord. Long PR I Long PR / Bord. Short PR / Short PR I Uninterpretable PR, Bord. Long QT I Prolonged QT I Bord. Short QT / Short QT I Uninterpretable QT), R wave related findings (PRWP, Broad R in V1 V2, Low QRS voltage, High QRS voltage), premature beat findings (PVC I Frequent PVC, PAC I Frequent PAC), and quality findings (Good Quality / Passable Quality I Bad Quality). Neural networks are trained and utilised for complex ECG findings, which are used in Knowledge Enhanced Neural Network (KENN) models. These findings include: Rhythms (RHY), Bundle Brunch Block (BBB), Myocardial Infarction (Ml), ST-T deviation (STT dev), LV Hypertrophy (LVH), RV Hypertrophy (RVH), LA and RA abnormality (LAE, RAE). A feature-compression model is typically used. These complex findings can be derived from the following inputs: average beat signal, beat sequence signal, Poincare plot histogram. A KENN model is also used to output myocardial infarction (Ml) findings. It interprets Ml in three dimensions: 1. Type or age of Ml: Ischemic ST-T, Non-STEMI, STEMI and Old Ml, 5 2. Location: Anterior, Inferior, Lateral, Posterior and Septal, and 3. Severity: Absent, Borderline, Moderate (Non-extensive) and Extensive. Example ECG findings output by the remote server are illustrated in the table below. Finding group Finding Rhythm Sinus / S. Tachycardia / S. Bradycardia / A. Fibrillation / A. Flutter / Ectopic Atrial R. / Supraventricular / Other Rhy. I Pacemaker PVC Absent / PVC / Frequent PVC PAC Absent / PAC / Frequent PAC PR interval Normal / Bord. Long PR / Long PR 1 Bord. Short PR / Short PR / Uninterpretable PR QT interval Normal / Bord. Long QT / Prolonged QT / Bord. Short QT / Short QT 1 Uninterpretable QT Axis Normal 1 Leftward Axis I Rightward Axis / Extreme Axis R-wave Absent 1 PRWP 1 Broad R in V1 V2 QRS voltage Normal / Low QRS volt 1 High QRS volt BBB Absence / LBBB / RBBB 11 VCD / ILBBB 1IRBBB / Vent. Pre-exc. Fasc. Block Absent / LAFB 1 LPFB ST-T dev Absence 1 non-specific 1 ST deviation / ST-T deviation / Twave Abnormality LVH Absence / Bord. LVH / LVH / LV strain RVH Absence / Bord. RVH / RVH LAE Absence / Bord. LAAbn. / Possible LAAbn. RAE Absence / Bord. RA Abn. I Possible RAAbn. Ischemic STT Absence / Borderline / Moderate / Extensive Ant / Inf / Lat / Pos / Sep Non-STEMI Absence 1 Borderline 1 Moderate 1 Extensive Ant / Inf / Lat / Pos / Sep STEMI Absence 1 Borderline 1 Moderate 1 Extensive Ant / Inf / Lat / Pos / Sep Old Ml Absence / Borderline / Moderate / Extensive Ant / Inf / Lat / Pos / Sep Quality Good Quality / Passable Quality / Bad Quality Summary Normal ECG / Borderline ECG / Abnormal ECG / Uninterpretable 10 The Ml model is trained utilizing both ECG-based and Echo-based ground truth data, specifically incorporating wall motion abnormality data. The inclusion of wall motion abnormality data during training enhances the sensitivity of ECG Ml detection, proving valuable in cases where the ECG is nondiagnostic during chest pain episodes. The severity of Ml is calculated based on the probabilities derived from applied rules and the neural network outputs. A higher probability result is indicative of a larger ST change or changes in multiple leads, providing a nuanced assessment of Ml severity. Accurately identifying the type and severity of myocardial infarction (Ml) may be advantageous for users, enabling a precise assessment of the actual state of coronary artery disease (CAD). This recognition, whether a patient has had an acute or old heart attack or is displaying ischemic changes, may advantageously facilitate informed clinical decisionmaking and prognosis evaluation. Example Ml Findings output by the remote server are illustrated in Fig. 5C. In more detail, an ECG interpretation algorithm places a primary focus on identifying the most prevalent ECG abnormalities and assessing their severity, with particular attention given to myocardial ischemia and infarction. The outputs, referred to as ECG Findings, cover the most common ECG abnormalities for the adult outpatients. The ECG Findings are output to a report, including severity classifications for each identified abnormality. Example ECG Findings are illustrated in Fig. 5A. A first colour, in this case green, represents normal or absent classification; a second colour, for example yellow, represents borderline or mild classification; and a third colour, for example orange, represent abnormal classification. The remote server also predicts the presence or absence of each of a set of heart abnormalities, based on the obtained synchronous signals (404). The predictions may be based at least in part based on medical rules. Preferably, dynamic thresholds are applied to the medical rules as a function of patient body size and / or heart rate. In addition, the predictions are based at least in part on machine learning models which output echoequivalence findings based on the segmented bio-signals. An example machine learning model is discussed in more detail in relation to Fig. 7. For example, feature vectors comprise ECG, PCG and MCG global and local measurements, extended with body size parameters. Signal extraction includes the beat representative and the time-frequency representation features are extracted from the typical beats to provide the information to the machine learning or neural networks. The aim of feature extraction is to produce a lesser number of more valuable features for the model input. Applying dynamic thresholds as a function of body size, such as but not limited to Body Mass Index (BMI), Body Surface Area (BSA), or height, may be advantageous to address uncertainties in threshold values dependent on body size, ensuring more accurate output compared to traditional medical rules derived solely from age and gender. The applicant has found that this may advantageously reduce false negative classifications by considering individual body size alongside gender in threshold value determination. For example, this has been shown to be particularly effective mitigating biases in small, thin men whose heart characteristics have been shown to closely resemble those of average women. In addition, the applicant has found considering individual body size alongside gender in threshold value determination reduces false positive classification, mitigating biases particularly in larger woman whose heart characteristics have been shown to closely resemble those of average male. By way of comparison, in the case of a larger woman with a body size significantly above the average, traditional ECG threshold values derived solely from gender might lead to false positive classifications. So-called “HART-findings” are disease findings output by the remote server, often equivalent to echocardiography-derived findings (or “Echo-Findings”), but are derived from the synchronous bio-signals, whereas Echo-Findings are derived from echocardiography images. While HART-findings are derived from bio-signals and Echofindings are derived from images, both modalities can indicate the presence of the same disease — they are considered disease equivalent. Thus, the HART Findings are advantageous to make advanced cardiac insights which were previously only available in secondary care, using echocardiography, available in primary care, based on the novel synchronous bio-signals. For example, the HART left ventricular hypertrophy (LVH) finding is disease equivalent to the Echo-derived LVH Finding. The following Echo-equivalent measurements may also be estimated by learning models, based on the synchronised bio-signals: Aortic Valve Peak Velocity (AVpV), Mitral Inflow E Wave Velocity (E), EJA Wave Velocity (E / A), Left ventricular ejection fraction (LVEF), Interventricular Septum Thickness in Diastole (IVSd), Left Atrial Volume Index {LAVI), Left Ventricular Internal Diameter in Diastole (LVIDd), Left Ventricular Internal Diameter in Systole (LVIDs), Left Ventricular Mass (LVmass), Left Ventricular Mass Index (LVMI), Mitral Regurgitation Jet Ratio in Left Atrium (MRjpLA), Right Atrial Volume Index (RAVI), Right Ventricular Systolic Pressure (RVSP), Tricuspid Regurgitation Jet Ratio in Right Atrium (TrjpRA), Rate of Aortic Regurgitation (AOR), Rate of Mitral regurgitation (MR), Transversal Diameter of the Right Atrium (RAD), End-diastolic Right Ventricular Diameter (RVDD), Right Ventricular Outflow Tract Proximal (RVOT), Rate of Tricuspid Regurgitation (TR), Wall Motion Score (WMSC). Machine learning techniques are used for the classification of a set of heart abnormalities, based at least in part on the above Echo-equivalent measurements. HART findings are classified as “Normal / Mild / Abnormal” for a set of heart abnormalities, including, but not limited to: • structural problems (such as: LVH, dilated cardiomyopathy (DCM), right ventricular enlargement (RVE), left atrial enlargement (LAE), right atrial enlargement (RAE)); • functional problems (such as: wall motion abnormalities (WMA), left ventricular systolic dysfunction (LVSD), diastolic dysfunction with impaired relaxation (DDIM)); and • valve problems (such as: aortic stenosis (AS), mitral stenosis (MS), aortic regurgitation (AR), mitral regurgitation (MR), tricuspid valve regurgitation (TR), pulmonary hypertension (PH)). The HART Findings are output to a report, including severity classifications for each identified abnormality. Example HART Findings are illustrated in Fig. 5B. A first colour, in this case green, represents normal or absent classification; a second colour, for example yellow, represents borderline or mild classification; and a third colour, for example orange, represent abnormal classification. Preferably the set of heart abnormalities comprises all of the following: LVH, DCM, RVE, LAE, RAE, WMA, LVSD, DDIM, AS, MS, AR, MR, TR, PH. The insights derived from ECG interpretations are therefore enhanced by the "HART" findings, which encompass prevalent structural, functional, and valve diseases. ECG findings of LVH and LAE (and RVH and RAE), carry specific meaning within the ECG knowledge base. Despite sharing common labels such as “LVH” with Echocardiography, the ECG interpretation of LVH differs significantly from Echocardiography interpretation of LVH, ECG LVH relying on distinct medical information. To illustrate, an ECG LVH finding does not equate to an Echocardiography LVH finding, as they are derived from different medical data Consequently, the present method enables corroboration of ECG LVH findings with HART-Findings (equivalent to Echo-findings in terms of identifying the same disease). This unique capability enhances diagnostic accuracy by corroborating ECG findings through a different modality, reliant on the synchronized bio-signals, providing a more comprehensive and reliable assessment and prediction of cardiac conditions. This may be particularly helpful in primary care settings or other point-of-care environments. In addition, it is noted that when LVSD (Left Ventricular Systolic Dysfunction) is present, the NICE (National Institute for Health and Care Excellence, UK) guidelines currently require a referral to cardiology. However, the current diagnostic devices for use in primary care - the ECG - typically cannot indicate whether this LVSD condition is present, much less mild, moderate or severe. Echocardiography is presently only available in secondary care, and thus many referrals to cardiology are missed. For example, the presence of LVSD is typically only detectable with ECG when LVEF is lower than 35%. By contrast, the present method, based on the analysis of synchronised bio-signals, is more sensitive to mildly reduced LVEF (<50%) and can therefore predict mild LVSD, as well as moderate LVSD (LVEF <40%). Based on the heart abnormalities which are predicted to be present by the above HART-findings, the remote server then also determines and outputs a risk score of heart failure, including phenotype classification into heart failure with preserved ejection fraction (HFpEF), heart failure with mildly reduced ejection fraction (HFmrEF), and heart failure with reduced ejection fraction (HFrEF). ln particular, the remote server determines and outputs a risk score of heart failure with preserved ejection fraction (HFpEF) (406). This is determined by aggregating a score based on the presence and severity of multiple structural and functional heart 5 abnormalities identified during step 404. In particular, structural and functional abnormalities associated with HFpEF include at least one of: diastolic dysfunction (DDIM), wall motion abnormality (WMA), left ventricular or left atrial abnormality (LVH, LAE), aortic stenosis (AS), pulmonary hypertension (PH). In addition, the risk score of heart failure with preserved ejection fraction (HFpEF) is preferably based on the 10 presence and severity of atrial fibrillation (AF or AFib) which may be determined based on the ECG signal. An example table for determining a risk score for HFpEF is included below, however the skilled person will understand that this is merely for illustration and that other weightings 15 may be encompassed. HART finding Value Score points Comment LVH Mild 0.37 Typical structural abnormality of HFpEF - low weight Abnormal 0.93 LAE Mild 1.48 Typical structural abnormality of HFpEF - high weight Abnormal 3.70 WMA Mild 1.30 Typical functional abnormality of HFpEF - moderate weight Abnormal 3.24 DDIM Mild 0.75 Typical functional abnormality of HFpEF - low weight Abnormal 1.85 AS Mild 1.30 Typical functional abnormality of HFpEF - moderate weight Abnormal 3.24 PH Mild 1.85 Typical functional abnormality of HFpEF - high weight Abnormal 4.63 AF Presence 4.63 Typical ECG functional abnormality of HFpEF - high weight HF score 0-10 domain sum of above The HFpEF score advantageously addresses the challenge of using LVEF alone in determination and prediction of heart failure, as LVEF is blind to HFpEF pathological abnormalities since HFpEF patient has normal LVEF (> 50%) similarly to normal or mildly abnormal patients. However, there is no single echocardiographic measurement that can represent all HFpEF types. Thus, outputting a HF score, where the score aggregates multiple structural and functional diseases, may advantageously address this. For example, in this case, the Atrial Fibrillation from ECG findings and HFpEF related HART-findings are aggregated for the HFpEF score. Thus, determining a risk score for HFpEF based on the presence and severity of multiple structural and functional heart abnormalities ensures more accurate prediction of HFpEF severity by considering a range of relevant clinical (and typically echocardiographic-derived) parameters. In addition, the remote server may optionally output predictions of other types of heart failure (HF). For example, the present method may predict heart failure based on the following rules: 1. HFrEF |S likely if LVSD is abnormal and estimated LVEF <40% 2. HFmrEF is likely if LVSD is mild and estimated LVEF is between 41-50% 3. Borderline HF prediction derived from estimated LVEF and HFpEF score 5. Otherwise, HF is Unlikely. An example two-dimensional presentation of LVEF plotted against HFpEF score for predicting heart failure is illustrated in Fig. 6. In the example illustrated, an LVEF estimate of 49 % and a HFpEF score of 4 predicts mild HFmrEF. The patient can then be referred on to cardiology as appropriate. Training Fig. 7 illustrates an example training and validation method to train Al models to predict the presence (or absence) of a set of heart abnormalities, and the severity thereof, based on obtained synchronous bio-signals (based on synchronous ECG, MCG, and PCG data). Firstly, clinical studies are performed to collect training and validation data (41). In this example, the obtained data records include 12-lead ECG and four standard auscultation points for PCG and MCG signals, collectively referred to as “bio-signal data” (42). The typical records have three times 60 second length records with a short break in between, altogether 180 second records. The patient data includes: Age, Gender, Race, Height, Weight, Waist and Hip size, Blood Pressure. In parallel with collecting the above data records, standard Transthoracic Echocardiogram (TTE) measurements and findings are obtained along with patient history (anamnesis) (43). These elements collectively establish the ground truth for diagnosing common heart diseases. A cardiologist consensus study (44) is conducted to establish the ground truth of echofindings which was used to validate the various findings, in particular the HART-findings output by the trained model. The ground truth is based on audited, verified and validated echo measurements, preferably with an average of five cardiologists per test. Databases containing parallel bio-signal data with ground truth data can therefore be used for training and validation. A first subset (45a) is used as the training set of data used for training of machine learning classification models and neural networks regression models (46). A second subset (45b) is used as the internal test and validation data set used to check and stop the learning of the models (47). An external validation data set (45c) may be used to estimate the real performance of models on the independent data (48). These records and patients are not used for any training purpose. The resulting trained classification model (referred to as the HART model) (49) may then be used to determine and output ECG-findings and HART-findings. Regression neural networks are used for compressed features, that precede the classification models. Standard binary performance metrics are calculated for each classification model (50): Sensitivity (SE), specificity (SP), positive predictive value (PPV) and F1-score (F1) as the best scalar performance for imbalanced classification (when prevalence is <25%). In use, a cardiac diagnostic system, such as that disclosed above, for collecting synchronous PCG, MCG, and ECG data is used for capturing bio-signals in a clinical environment. This data can be input into a trained model in order to determine and output ECG-findings and HART-f indings. 5 Performance The table below compares the HART findings to the traditional ECG-findings for common heart diseases. HART-findings refer to outputs from the HART model, the trained classification model described herein. 10 In the ECG and HART performance evaluation, sensitivity (SE), specificity (SP), positive predictive value (PPV), negative predictive value (NNPV), and F1 -score (F1) are indicated. The F1-score, an average of sensitivity and positive predictive value (PPV), is the predominant global performance metric. 15 The standard reading of ECG shows low sensitivity (20-50%) with moderate-high specificity (85-90%) for ventricular hypertrophy and atrial enlargement, where these structural abnormalities are defined by Standard based LVMI, IVSd, RWT, LAVI, and RAVI echocardiographic parameters. Disease Model SE% SP% PPV% NPV% F1% LVH ECG Finding: LVHe 30.5 94.5 64.4 80.7 41.4 HART-f inding 80.5 84.7 63.1 93.0 70.8 DCM ECG Finding: LVHe 26.4 93.2 45.9 85.3 33.5 HART-f inding 78.1 91.8 67.7 95.0 72.5 RVE ECG finding: RVHe 1.3 99.8 26.3 94.9 2.5 HART-finding 79.1 95.5 48.5 98.8 60.1 LAE ECG finding: LAEe 6.3 97.9 41.2 81.5 11.0 HART-finding 81.7 91.8 70.2 95.5 75.5 RAE ECG finding: RAEe 4.0 99.8 66.7 90.6 7.6 HART-finding 69.2 96.0 65.4 96.6 67.3 WMA ECG finding: STTd 53.4 94.6 83.1 80.2 65.0 HART-finding 81.8 91.9 83.4 91.0 82.6 LVSD ECG finding: STTd 43.0 86.6 48.5 83.7 45.6 HART-f inding 82.4 88.7 68.2 94.5 74.6 DDIM ECG finding: LAX 26.5 89.6 53.4 73.0 35.4 HART-f inding 77.3 85.4 70.4 89.3 73.7 AS ECG finding: LVHe 32.5 91.9 25.7 94.0 28.7 HART-finding 73.8 98.7 83.1 97.8 78.2 MS ECG finding: LVHe 43.6 90.8 13.2 98.0 20.3 HART-finding 79.2 94.2 30.7 99.3 44.2 AR ECG finding: LVHe 21.7 92.4 41.6 82.6 28.5 HART-finding 73.5 85.3 55.4 92.8 63.2 MR ECG finding: Isch 44.9 84.5 37.0 88.3 40.6 HART-finding 74.8 89.3 58.7 94.6 65.8 TR ECG finding: AFib 41.8 94.8 54.2 91.8 47.2 HART-finding 61.4 96.1 69.6 94.5 65.2 PH ECG finding: AFib 53.5 93.0 36.7 96.4 43.5 HART-finding 72.6 96.1 58.4 97.9 64.7 In the table above, the HART models are compared with the best ECG findings. The best individual ECG findings typically produce more false negatives and more false positives. 5 The HART models reach the highest performance in all cases: typical sensitivity of 70- 80% with specificity of 85-95%. Compared to the best ECG-findings, the sensitivity and F1 scores of the HART models are double those of ECG findings. The HART model drastically reduces false negatives and is able to provide more clinically meaningful results in the prediction and determination of structural, functional and valvular heart diseases. 15 It will be appreciated from the discussion above that the embodiments shown in the Figures are merely exemplary, and include features which may be generalised, removed or replaced as described herein and as set out in the claims. In the context of the present disclosure other examples and variations of the apparatus and methods described herein will be apparent to a person of skill in the art. It will be appreciated in the context of the foregoing disclosure that the described embodiments are not to be construed as limiting. For example, with reference to the drawings in general, it will be appreciated that schematic functional block diagrams are used to indicate functionality of systems and apparatus described herein. It will be appreciated however that the functionality need not be divided in this way and should not be taken to imply any particular structure of hardware other than that described and claimed below. The function of one or more of the elements shown in the drawings may be further subdivided, and / or distributed throughout apparatus of the disclosure. In some embodiments the function of one or more elements shown in the drawings may be integrated into a single functional unit. The processor or controller 110, remote computing device 306, or remote server 320, may be provided by one or more distributed services, configured to provide processing logic which performs the function of the said devices as described and claimed herein. In some examples the functionality of the controllers and processing means described herein (such as controller 110, etc.) may be provided by mixed analogue and / or digital processing and / or control functionality. It may comprise any general-purpose processor, which may be configured to perform a method according to any one of those described herein. In some examples the controller may comprise digital logic, such as field programmable gate arrays, FPGA, application specific integrated circuits, ASIC, a digital signal processor, DSP, or by any other appropriate hardware. In some examples, one or more memory elements can store data and / or program instructions used to implement the operations described herein. Embodiments of the disclosure provide computer program products such as tangible, nontransitory storage media comprising program instructions operable to program a processor to perform any one or more of the methods described and / or claimed herein and / or to provide data processing apparatus as described and / or claimed herein. Such a controller may comprise an analogue control circuit which provides at least a part of this control functionality. An embodiment provides an analogue control circuit configured to perform any one or more of the methods described herein. Another embodiment provides a digital control circuit configured to perform any one or more of the methods described herein. The above embodiments are to be understood as illustrative examples. Further 5 embodiments are envisaged. It is to be understood that any feature described in relation to any one embodiment may be used alone, or in combination with other features described, and may also be used in combination with one or more features of any other of the embodiments, or any combination of any other of the embodiments. Furthermore, equivalents and modifications not described above may also be employed without 10 departing from the scope of the invention, which is defined in the accompanying claims. These claims are to be interpreted with due regard for equivalents.

Claims

1. A sensing apparatus for cardiac diagnostics, the sensing apparatus comprising:a phonocardiography, PCG, sensor, configured to sense heart sounds; anda mechanocardiography, MCG, sensor, configured to sense heart-induced motion, wherein the MCG sensor comprising a 3-axis gyroscope and a 3-axis accelerometer;wherein the PCG sensor and the MCG sensor are arranged within a common sensor housing.

2. The sensing apparatus of any preceding claim further comprising a processor configured to capture signals from the PCG sensor and the MCG sensor synchronously.

3. The sensing apparatus of claim 2 wherein the processor is configured to augment signals obtained from the PCG sensor based on the synchronous signals obtained from the MCG sensor.

4. The sensing apparatus of any preceding claim, further comprising at least one electrocardiography, ECG, electrode, wherein the at least one ECG electrode is arranged within the sensor housing.

5. The sensing apparatus of claim 4, dependent on claim 2 or 3, wherein the processor is configured to capture signals from the PCG sensor, the MCG sensor, and the at least one ECG electrode synchronously.

6. The sensing apparatus of any preceding claim wherein the processor is configured to correct signals obtained from the MCG 3-axis accelerometer based on signals obtained from the MCG 3-axis gyroscope.

7. The sensing apparatus of any preceding claim, further comprising an attachment means configured to attach the sensor housing to a patient.

8. The sensing apparatus of any preceding claim wherein the sensor housing comprises an acoustic insulating material.

9. The sensing apparatus of any preceding claim further comprises a grip portion made of malleable material.

10. The sensing apparatus of any preceding claim further comprising a wireless communication means configured to wirelessly communicate signals obtained from the PCG sensor and the MCG sensor to a remote device.

11. A cardiac diagnostic system comprising:a phonocardiography, PCG, sensor, configured to sense heart sounds;a mechanocardiography, MCG, sensor, configured to sense heart-induced motion, wherein the MCG sensor comprising a 3-axis gyroscope and a 3-axis accelerometer;at least one electrocardiography, ECG, electrode; anda processor, wherein the processor is configured to capture signals from the PCG sensor, the MCG sensor, and the at least one ECG electrode synchronously.

12. The cardiac diagnostic system of claim 11 wherein the PCG sensor and the MCG sensor are provided by the sensing apparatus for cardiac diagnostics of any of claims 1 to 10, optionally wherein at least one of the at least one ECG electrode is provided by the sensing apparatus for cardiac diagnostics of claim 5.

13. The cardiac diagnostic system of any of claims 11 to 12 wherein the at least one ECG electrode comprises ten ECG electrodes configured to provide a 12-Lead ECG.

14. The cardiac diagnostic system of any of claims 11 to 13 wherein the processor is configured to capture signals from the PCG sensor, the MCG sensor, and the at least one ECG electrode synchronously for at least a 60 second time period.

15. The cardiac diagnostic system of any of claims 11 to 14 wherein the system comprises one PCG sensor and one MCG sensor.

16. The cardiac diagnostic system of any of claims 11 to 15 wherein the PCG sensor andthe MCG sensor are configured to be placed on a patient between ECG electrode placement locations V4 and V5, wherein V4 represents the 5th intercostal space at midclavicular line, and V5 represents the 5th intercostal space at left anterior axillary line.

17. The cardiac diagnostic system of any of claims 11 to 14 wherein the system comprises up to four PCG sensors and up to four MCG sensors, optionally wherein each PCG sensor and each MCG sensor is configured to be positioned at a standard auscultation position.

18. The cardiac diagnostic system of any of claims 11 to 17 further comprising a computing device for receiving the captured signals from the processor, wherein the computing device is configured to perform the method of any claims 20 to 36.

19. The cardiac diagnostic system of claim 18 wherein the computing device comprise an artificial neural network, the artificial neural network being configured to receive synchronous ECG, PCG, and MCG signals, and determine whether any one of a set of heart abnormalities is likely to be present.

20. A computer-implemented method for predicting heart failure, the method comprising: obtaining (i) an ECG signal, (ii) a PCG signal, and (iii) an MCG signal, wherein the ECG signal, the PCG signal, and the MCG signal are synchronised;predicting the presence or absence of each of a set of heart abnormalities, based on the obtained synchronous signals;determining a risk score of heart failure with preserved ejection fraction, HFpEF, based on each heart abnormality which is predicted to be present from the set of heart abnormalities.

21. The computer-implemented of claim 20 wherein determining the risk score of HFpEF comprises aggregating risk contribution values associated with each heart abnormality which is predicted to be present from the set of heart abnormalities22. The computer-implemented of claim 20 wherein determining the risk score of HFpEF, is based on the predicted presence or absence of each heart abnormality from the set ofheart abnormalities and an associated abnormality weighting, wherein each abnormality weighting is associated with one of heart abnormalities from the set of heart abnormalities, and wherein each abnormality weighting represents the relative risk contribution of said heart abnormality to heart failure with preserved ejection fraction.

23. The method of any of claims 20 to 22 wherein predicting the presence or absence of each of a set of heart abnormalities comprises predicting the severity of each of the set of heart abnormalities.

24. The method of any of claims 20 to 23 wherein predicting the presence or absence of each of a set of heart abnormalities utilises an artificial neural network, the artificial neural network being configured to receive synchronous ECG, PCG, and MCG signals, and determine whether a heart abnormality from the set of heart abnormalities is likely to be present.

25. The method of any of claims 20 to 24, further comprising:determining a severity classification for each heart abnormality predicted to be present, based on the obtained synchronous signals; andwherein determining the risk score of heart failure with preserved ejection fraction, HFpEF, is further based on the severity classification for each identified heart abnormality.

26. The method of any of claims 20 to 25 further comprising:determining left ventricle ejection fraction, LVEF, based on the obtained synchronous signals; anddetermining a risk score of borderline heart failure, wherein the risk score for borderline heart failure is based on the first risk score of heart failure with preserved ejection fraction, HFpEF, and the determined LVEF.

27. The method of claims 20 to 26 further comprising:determining left ventricle ejection fraction, LVEF, based on the obtained synchronous signals; andclassifying the severity of left ventricle systolic dysfunction, LVSD, based on theobtained synchronous signals; andoutputting a classification of heart failure with reduced ejection fraction, HFrEF, or heart failure with mid-range ejection fraction, HEmrEF, based on the determined LVEF and the classified severity of LVSD.

28. The method of any of claims 20 to 27 wherein the set of heart abnormalities comprises one or more structural, functional, valvular, and / or electrical abnormalities, optionally comprising one or more of: atrial fibrillation, diastolic dysfunction, wall motion abnormality, left ventricular hypertrophy, left atrial enlargement, right atrial enlargement, ventricular enlargement, aortic stenosis, mitral stenosis, dilated cardiomyopathy, aortic or mitral, or tricuspid regurgitation, and pulmonary hypertension.

29. The method of any of claims 20 to 28 wherein determining the risk score of HFpEF is based on the aggregating the predicted heart abnormalities.

30. The method of any of claims 20 to 29, further comprising:determining whether atrial fibrillation is present based on the ECG signal;wherein determining the risk score of heart failure with preserved ejection fraction, HFpEF, is further based on the determination of atrial fibrillation is present.

31. The method of any of claims 20 to 30, further comprisingidentifying fundamental heart sounds from the PCG signal, based at least in part on the synchronised MCG signal;identifying a location of the identified fundamental heart sounds, based on comparison to the synchronised ECG signal; andpredicting the presence or absence of each of a set of heart abnormalities based on at least one of the identified fundamental heart sounds, the location of the identified fundamental heart sounds, and the ECG signal.

32. The method of any of claims 20 to 31, further comprising:obtaining an indication of patient body size; andnormalising at least one of the obtained ECG signal, PCG signal, and / or MCG signal based on the indication of patient body size, prior to predicting the presence orabsence of each of the set of heart abnormalities.

33. The method of any of claims 20 to 32 wherein at least the synchronised PCG signal and MCG signal are obtained from the sensing apparatus of claims 1 to 9.

34. The method of any of claims 20 to 33 wherein the synchronised ECG signal, PCG signal, and MCG signal are obtained from the cardiac diagnostic system of claims 11 to 19.

35. A computer-implemented method for fundamental heart sound detection, the method comprising:obtaining a phonocardiography, PCG, signal from a PCG sensor;obtaining a mechanocardiography, MCG, signal from a MCG sensor, wherein the obtained MCG signal is synchronised relative to the obtained PCG signal; andperforming heart sound segmentation on the obtained PCG signal, based on the PCG signal and the MCG signal.

36. The method of claim 35 further comprising:determining an indication of patient breathing based on the MCG signal;wherein performing heart sound segmentation on the obtained PCG signal is based on the PCG signal and the indication of patient breathing derived based on the MCG signal.

37. A computer program product comprising instructions configured to program a programmable processor to perform the method of any of claims 20 to 36.Amendments have been added to the claims as follows :12 08 2541CLAIMS:

1. A sensing apparatus for cardiac diagnostics, the sensing apparatus comprising: a phonocardiography, PCG, sensor, configured to sense heart sounds; and5 a mechanocardiography, MCG, sensor, configured to sense heart-induced motion, wherein the MCG sensor comprising a 3 axis gyroscope and a 3 axis accelerometer;wherein the PCG sensor and the MCG sensor are arranged within a common sensor housing.

102. The sensing apparatus of any preceding claim further comprising a processor configured to capture signals from the PCG sensor and the MCG sensor synchronously.

3. The sensing apparatus of claim 2 wherein the processor is configured to augment15 signals obtained from the PCG sensor based on the synchronous signals obtained from the MCG sensor.

4. The sensing apparatus of any preceding claim, further comprising at least one electrocardiography, ECG, electrode, wherein the at least one ECG electrode is20 arranged within the sensor housing.

5. The sensing apparatus of claim 4, dependent on claim 2 or 3, wherein the processor is configured to capture signals from the PCG sensor, the MCG sensor, and the at least one ECG electrode synchronously.

256. The sensing apparatus of any preceding claim wherein the processor is configured to correct signals obtained from the MCG 3 axis accelerometer based on signals obtained from the MCG 3 axis gyroscope.30 7. The sensing apparatus of any preceding claim, further comprising an attachment means configured to attach the sensor housing to a patient.

8. The sensing apparatus of any preceding claim wherein the sensor housing comprises an acoustic insulating material.12 08 259. The sensing apparatus of any preceding claim further comprises a grip portion made of malleable material.5 10. The sensing apparatus of any preceding claim further comprising a wireless communication means configured to wirelessly communicate signals obtained from the PCG sensor and the MCG sensor to a remote device.

11. A cardiac diagnostic system comprising:10 a phonocardiography, PCG, sensor, configured to sense heart sounds;a mechanocardiography, MCG, sensor, configured to sense heart-induced motion, wherein the MCG sensor comprising a 3 axis gyroscope and a 3 axis accelerometer;at least one electrocardiography, ECG, electrode; and15 a processor, wherein the processor is configured to capture signals from the PCG sensor, the MCG sensor, and the at least one ECG electrode synchronously.

12. The cardiac diagnostic system of claim 11 wherein the PCG sensor and the MCG sensor are provided by the sensing apparatus for cardiac diagnostics of any of claims 120 to 10, optionally wherein at least one of the at least one ECG electrode is provided by the sensing apparatus for cardiac diagnostics of claim 5.

13. The cardiac diagnostic system of any of claims 11 to 12 wherein the at least one ECG electrode comprises ten ECG electrodes configured to provide a 12-Lead ECG.2514. The cardiac diagnostic system of any of claims 11 to 13 wherein the processor is configured to capture signals from the PCG sensor, the MCG sensor, and the at least one ECG electrode synchronously for at least a 60 second time period.30 15. The cardiac diagnostic system of any of claims 11 to 14 wherein the system comprises one PCG sensor and one MCG sensor.

16. The cardiac diagnostic system of any of claims 11 to 15 wherein the PCG sensor and the MCG sensor are configured to be placed on a patient between ECG electrode12 08 25placement locations V4 and V5, wherein V4 represents the 5th intercostal space at midclavicular line, and V5 represents the 5th intercostal space at left anterior axillary line.

17. The cardiac diagnostic system of any of claims 11 to 14 wherein the system 5 comprises up to four PCG sensors and up to four MCG sensors, optionally wherein eachPCG sensor and each MCG sensor is configured to be positioned at a standard auscultation position.

18. The cardiac diagnostic system of any of claims 11 to 17 further comprising a 10 computing device for receiving the captured signals from the processor, wherein the computing device is configured to perform the method of any claims 20 to 36.

19. The cardiac diagnostic system of claim 18 wherein the computing device comprise an artificial neural network, the artificial neural network being configured to receive 15 synchronous ECG, PCG, and MCG signals, and determine whether any one of a set of heart abnormalities is likely to be present.

20. A computer program product comprising instructions configured to program a programmable processor to:20 obtain (i) an ECG signal, (ii) a PCG signal, and (iii) an MCG signal, wherein the ECG signal, the PCG signal, and the MCG signal are synchronised;predict the presence or absence of each of a set of heart abnormalities, based on the obtained synchronous signals;determine a risk score of heart failure with preserved ejection fraction, HFpEF, 25 based on each heart abnormality which is predicted to be present from the set of heart abnormalities.

21. The computer program product of claim 20 wherein determining the risk score of HFpEF comprises aggregating risk contribution values associated with each heart 30 abnormality which is predicted to be present from the set of heart abnormalities22. The computer program product of claim 20 wherein determining the risk score of HFpEF, is based on the predicted presence or absence of each heart abnormality from the set of heart abnormalities and an associated abnormality weighting, wherein each12 08 25abnormality weighting is associated with one of heart abnormalities from the set of heart abnormalities, and wherein each abnormality weighting represents the relative risk contribution of said heart abnormality to heart failure with preserved ejection fraction.5 23. The computer program product of any of claims 20 to 22 wherein predicting the presence or absence of each of a set of heart abnormalities comprises predicting the severity of each of the set of heart abnormalities.

24. The computer program product of any of claims 20 to 23 wherein predicting the 10 presence or absence of each of a set of heart abnormalities utilises an artificial neural network, the artificial neural network being configured to receive synchronous ECG, PCG, and MCG signals, and determine whether a heart abnormality from the set of heart abnormalities is likely to be present.15 25. The computer program product of any of claims 20 to 24, further comprising instructions configured to program a programmable processor to:determine a severity classification for each heart abnormality predicted to be present, based on the obtained synchronous signals; andwherein determining the risk score of heart failure with preserved ejection fraction, 20 HFpEF, is further based on the severity classification for each identified heart abnormality.

26. The computer program product of any of claims 20 to 25 further comprising instructions configured to program a programmable processor to:25 determine left ventricle ejection fraction, LVEF, based on the obtained synchronous signals; anddetermine a risk score of borderline heart failure, wherein the risk score for borderline heart failure is based on the first risk score of heart failure with preserved ejection fraction, HFpEF, and the determined LVEF.3027. The computer program product of claims 20 to 26 further comprising instructions configured to program a programmable processor to:determine left ventricle ejection fraction, LVEF, based on the obtained synchronous signals; and12 08 25classify the severity of left ventricle systolic dysfunction, LVSD, based on the obtained synchronous signals; andoutput a classification of heart failure with reduced ejection fraction, HFrEF, or heart failure with mid-range ejection fraction, HEmrEF, based on the determined LVEF 5 and the classified severity of LVSD.

28. The computer program product of any of claims 20 to 27 wherein the set of heart abnormalities comprises one or more structural, functional, valvular, and / or electrical abnormalities, optionally comprising one or more of: atrial fibrillation, diastolic 10 dysfunction, wall motion abnormality, left ventricular hypertrophy, left atrial enlargement, right atrial enlargement, ventricular enlargement, aortic stenosis, mitral stenosis, dilated cardiomyopathy, aortic or mitral, or tricuspid regurgitation, and pulmonary hypertension.

29. The computer program product of any of claims 20 to 28 wherein determining the risk 15 score of HFpEF is based on the aggregating the predicted heart abnormalities.

30. The computer program product of any of claims 20 to 29, further comprising instructions configured to program a programmable processor to:determine whether atrial fibrillation is present based on the ECG signal;20 wherein determining the risk score of heart failure with preserved ejection fraction, HFpEF, is further based on the determination of atrial fibrillation is present.

31. The computer program product of any of claims 20 to 30, further comprising instructions configured to program a programmable processor to:25 identify fundamental heart sounds from the PCG signal, based at least in part on the synchronised MCG signal;identify a location of the identified fundamental heart sounds, based on comparison to the synchronised ECG signal; andpredict the presence or absence of each of a set of heart abnormalities based on 30 at least one of the identified fundamental heart sounds, the location of the identified fundamental heart sounds, and the ECG signal.

32. The computer program product of any of claims 20 to 31, further comprising instructions configured to program a programmable processor to::obtain an indication of patient body size; andnormalise at least one of the obtained ECG signal, PCG signal, and / or MCG signal based on the indication of patient body size, prior to predicting the presence or absence of each of the set of heart abnormalities.

33. The computer program product of any of claims 20 to 32 wherein at least the synchronised PCG signal and MCG signal are obtained from the sensing apparatus of claims 1 to 9.10 34. The computer program product of any of claims 20 to 33 wherein the synchronised ECG signal, PCG signal, and MCG signal are obtained from the cardiac diagnostic system of claims 11 to 19.LD

Images