Non-contact quantification of aortic stenosis and mitral regurgitation using carotid waveforms from skin displacements

Objective. Early diagnosis of heart problems is essential for improving patient prognosis. Approach. We created a non-contact imaging system that calculates the vessel-induced deformation of the skin to estimate the carotid artery pressure displacement waveforms. We present a clinical study of the system in patients (n = 27) with no underlying condition, aortic stenosis (AS), or mitral regurgitation (MR). Main results. Displacement waveforms were compared to aortic catheter pressures in the same patients. The morphologies of the pressure and displacement waveforms were found to be similar, and pulse wave analysis metrics, such as our modified reflection indices (RI) and waveform duration proportions, showed no significant differences. Compared with the control group, AS patients displayed a greater proportion of time to peak (p = 0.026 and p = 0.047 for catheter and displacement, respectively), whereas augmentation index (AIx) was greater for the displacement waveform only (p = 0.030). The modified RI for MR (p = 0.047 and p = 0.004 for catheter and displacement, respectively) was lower than in the controls. AS and MR were also significantly different for the proportion of time to peak (p = 0.018 for the catheter measurements), RI (p = 0.045 and p = 0.002 for the catheter and displacement, respectively), and AIx (p = 0.005 for the displacement waveform). Significance. These findings demonstrate the ability of our system to provide insights into cardiac conditions and support further development as a diagnostic/telehealth-based screening tool.


Introduction
Cardiovascular disease (CVD) affects millions worldwide and is the principal cause of mortality (World Health Organisation 2017). Devices that can efficiently and non-invasively provide early and clinically useful diagnostic information may help reduce CVD morbidity and mortality. CVD severity is often assessed by investigating the flow and/or pressure waveforms within the heart (DeSilva 2013, Sam et al 2021). The current assessment of cardiac health includes measures based on pressure cuff, ultrasound, echocardiography, and catheterisation techniques. However, many of these methods require expensive equipment, are invasive, and need trained medical professionals to record and interpret the results. While blood pressure cuff measurement of the brachial artery is widely used and relatively easy to conduct, this measurement has limitations (Sam et al 2021). Cuff measurements cannot easily differentiate cardiac diseases, do not account for the effects of atherosclerosis, and vary from the blood pressure that directly loads the heart (McEniery et al 2014).
The measurement of skin deformations near peripheral vessels shows promise to overcome the above issues. Changes in fluid pressure within blood vessels cause neighbouring tissues and skin deformations, such as in the neck where pulsations of the carotid artery (CA) and jugular vein can be seen (figure 1). By assessing the morphology of the CA neck pulsations (via auscultation, touch examination, and visualisation), trained clinicians can get an indication of heart and valvular diseases. Clinicians identify the shape of the carotid blood pressure waveform, which in normal subjects consists of a quick upstroke and a slow downstroke (Douglas 1990). Variations of the carotid arterial waveforms can be a sign of heart diseases such as aortic stenosis (AS) (Pujari and Agasthi 2023) (figure 1), aortic regurgitation, or mitral regurgitation (MR) (Maganti et al 2010, Pickett et al 2011. Examination of the CA requires clinical expertise and is difficult to master, and is thus often neglected, or performed inconsistently (Pickett et al 2011, Sam et al 2021. Due to difficulties in directly measuring/visualising pressure waveforms, alternative non-contact methods have been developed. These methods rely on the use of sophisticated tools such as photoplethysmography (PPG) ( (Moco et al 2017), for acquisition and illumination to identify the pulsatile waveforms. To address these limitations, we have developed a non-contact system employing a robust image registration algorithm that can be used with an ordinary video camera.
The algorithm has been previously used to measure displacements of jugular venous pulsations with subpixel accuracy in healthy subjects (Lam Po Tang et al 2018). The present study builds on this earlier work by clinically testing the system in CVD patients who typically exhibit an abnormal pressure waveform. Here we present an improved imaging system and compare the skin displacement deformation against gold standard catheter measurements. Pulse wave analysis techniques are applied to quantify differences between modalities and conditions. These techniques are increasingly being recognised as non-invasive tools to assess cardiovascular state/timings within each pulse and can provide an indication of arterial stiffness (Gurovich andBraith 2011, Charlton et al 2019). The benefits of using PWA to assess cardiovascular risk have been shown in clinical trials (Roman et al 2007, Poleszczuk et al 2018. Over the last few decades, these techniques have been utilised as non-invasive tools to assess cardiovascular state (Pereira et al 2019), however further ongoing trials are needed to guide clinicians. To the best of our knowledge, this is the first study to compare diagnostic information from invasive catheter pressure waveforms to those derived from displacement waveforms obtained with camera-based image registration techniques. The outcomes of this study provide justification for the future use of the system in the clinical setting and show the potential to develop a novel non-contact, cost-effective system to provide easily accessible diagnostic information for CVDs. The technique also shows promise for CVD telehealth applications.

Methods
This study was approved by the Auckland Health Research Ethics Committee (AHREC) University of Auckland (reference number AH1082). All experiments were performed under the guidelines and regulations of AHREC. The study was also conducted in line with the Declaration of Helsinki.

Patient information
Patients undergoing routine clinical investigations (cardiac ultrasound and catheterisation) for their conditions were recruited for this study. Written informed consent was obtained from 27 (6 female, 21 male) patients aged 18 years or older, who were already undergoing clinical investigations for their conditions which necessitated invasive arterial catheterisation, at Auckland City Hospital. Exclusion criteria were diagnosed arrhythmia, and inability to give informed consent.
Patients with a range of cardiac conditions that are expected to affect the carotid pulse were recruited to compare neck displacement measurements to cardiac catheter pressure waveforms. Three groups of patients were studied: 1. Control group with normal LV function (ejection fraction > 55%) having coronary angiography for suspected coronary disease. (n = 10) 2. Aortic stenosis (n = 10) 3. Mitral regurgitation (n = 7) The majority of patients were classified as having severe disease. In AS, 8/10 patients were considered severe, and 2/10 moderate. Two cases of severe AS were described to have calcified leaflets. MR was mild/moderate in 2/7 cases, and severe in 5/7 cases. None of the patients reported overlapping AS or MR aetiologies. Table 1 show the demographics of the participants. All groups had similar average ages (68.2 ± 7.6, 75.7 ± 7.5, 73.1 ± 5.5) and BMIs (29.1 ± 4.6, 28.9 ± 4.5, 26.7 ± 6.0) for control, AS, MR respectively. System design Our custom imaging system is comprised of two cameras (FLIR, GS3-U3-32S4M-C) angled at 53°with respect to each other. The cameras were fitted with 16 mm lenses (Tamron, M118FM16) with a working distance of approximately 110 mm from the subject's neck (figure 2). The cameras were mounted in an acrylic housing attached to a tripod with a ball head to enable adjustment of the stereoscope to account for variations in participant neck sizes, vessel positions, and target locations. For this study, video recordings from Camera 1 (figure 2) were used. In the future, 3D reconstructions using the stereoscope (both cameras) will be explored.
An illumination system was positioned between the cameras. This consisted of two crosshair laser diodes, with beams that intersected at the working distance along both camera's optical axes, which ensured that the imaging system was positioned at the correct working distance and target location on each subject's neck. Three 470 nm blue LEDs provided illumination during the recordings. The blue light, although not necessary, provided a consistent light source for this initial study. The blue light enhances the contrast of intrinsic features of the skin, due to its higher absorption by melanin, compared to longer wavelengths. The cameras and light sources were hardware-triggered at 90 frames per second, and a 3-lead chest electrocardiogram (ECG) was simultaneously recorded by a data acquisition unit (National Instruments, USB-6002). The video and ECG recordings were acquired using custom software written in LabVIEW (National Instruments, 2019). Data were recorded on a laptop (Dell, 17 R4, 32 GB RAM, Intel(R) Core (TM) i7-7700HQ CPU @ 2.80 GHz), which enabled continuous recordings of duration up to approximately 70 seconds.

Experimental methods
Imaging data were collected from participants in a cardiology day-stay ward before their catherisation procedures. Each subject's neck and carotid artery were imaged (figure 2) while the subject's upper trunk and head were positioned at an incline of 45°and their legs flat on the bed. This is a common position for manual clinical examinations. The patient was asked to remain still during the recordings. The angle between the camera and the skin surface on the subject's neck was approximated at 45°. Video recordings of the neck were captured at 90 frames per second with a resolution of 2048 pixels by 1536 pixels and an A/D converter resolution/bit depth of 8-bit. Each recording lasted 5s and was saved in a lossless format. Clinically-indicated cardiac catheterisation was undertaken within ∼1-3 h of the video acquisition. In most cases, the recordings were taken within an hour. A 1.7 mm or 2 mm angiography catheter (Cordis, USA) 1 m in length, internal diameter of 1.2-1.4 mm was inserted via a vascular sheath, under local anaesthesia, predominantly via the right radial artery. Most patients received intravenous analgesia with fentanyl (25-50 μg) and midazolam (0.5-1.0 mg). The transducer connected to the intra-arterial catheter was calibrated to the level of the mid-thorax, with the patient supine. Continuous invasive blood pressures were recorded with Mac-Lab software (version 6.9.6., GE Medical Systems). Pressure traces were taken from the aortic root, close to the carotid artery in the neck, providing comparable pressure waveforms to that of the displacement waveforms.

Data analysis
The ECG data were filtered using a first-order Butterworth lowpass filter with a cut-off frequency of 100 Hz. The displacement waveforms were calculated as described previously (Lam Po Tang et al 2018). A summary of the methods used is as follows: a polygonal region of interest was manually drawn over the field of view to exclude control points that were not located on the neck. A 64 pixel × 64 pixel sub-image with a 49 pixel overlap was selected. An accurate subpixel image registration algorithm (HajiRassouliha et al 2018) (which utilises a phasebased approach and Savitzky-Golay differentiator in gradient correlation) was then used to calculate the displacement field between consecutive image pairs. These displacements were then presented as a function of time to produce the displacement waveforms. The highest power in the frequency range of 0.7-2 Hz was extracted to identify a component of the signal related to the heartbeat.
Outliers were also removed via two methods. Firstly, temporal segments from videos that showed large global movements were not selected for quantitative analysis. Additionally, following quantitative analysis the derivatives greater than 0.25 pixel (0.015 mm) per frame were removed as this motion was unlikely to be related to pulsation deformation.
Using the ECG recordings, both the catheter pressure and displacement waveform signals were segmented into separate waveforms corresponding to each heart cycle (2-4 cycles). These single cycle waveforms were normalised in time using the duration of one complete cardiac cycle (RR interval), and in amplitude using the trough to the peak of each pulse.

Extracting pulse wave indices
To quantitatively compare the displacement and catheter pressure waveforms, pulse wave analysis was conducted for both measurements. Three waveform characteristics were extracted from each of the cycles of normalised displacement and pressure waveforms (figure 3), including the time-to-peak or proportion of time to peak relative to the whole cardiac cycle, the augmentation index (AIx), and the reflection index (RI). These data were then averaged over the number of cycles. While computing an average over multiple cycles would help to generate a more consistent waveform, averaging waveforms seems inappropriate as some pathologic features may present in only some of the cardiac cycles. All analysis was completed using MATLAB (version R2020a). The normalised time to peak describes the proportion of time to peak pressure/displacement to be achieved. This metric was calculated by determining the time from pulse initiation to the time of peak displacement/pressure (P1 in figure 3) and dividing it by the cardiac cycle duration (equation (1)). Augmentation pressure (AP) is a measure of the contribution that the reflected wave makes to the systolic arterial pressure. It was calculated by determining the local minimum of the gradient (first derivative) during the initial rise of the waveform to the peak P1 (see figure 3). AIx was then determined using the following equation (2). . Pulse wave analysis. Definition of arterial pulse waveform characteristics derived from the normalised displacement or pressure waveforms. An example of one cardiac cycle displacement waveform from the control group is shown. The first derivative (lower trace) is used to calculate the timing of the measurements required to calculate the indices. AP = augmentation pressure, AI = augmentation index, P1 = first peak, P2 = second peak, Dnotch = dicrotic notch, modified RI = reflection index.
The RI has been used as an index of arterial stiffness in PPG and radial pressure signals (Li et al 2018 (3)) was determined as the ratio of the peak (P2 in figure 3) following the dicrotic notch and max peak (P1 in figure 3). We have termed this the 'modified (camera and catheter based) RI'.

Statistical analysis
The values of all derived arterial pulse waveform characteristics (proportion of time to peak, AIx, RI) were calculated for all participants. Analysis of variance was then performed (Prism statistical software, version 9.2.0) to investigate the effect of cardiac condition and modality on all the displacement and pressure waveform characteristics after normalisation. Tukey post hoc comparisons were used to determine significant differences between pairs (p < 0.05; n = 10, 10, 7 for control, AS, and MR groups, respectively).

Morphological waveform comparisons
Visual comparison of the displacement and pressure waveforms revealed similar morphologies. Figure 4 shows representative displacement and pressure waveforms for a control subject (Patient 5), demonstrating that the initial notch in the rise time and the dicrotic notch could be observed in both recordings. Catheter and displacement waveforms showed similar augmentation pressure (anacrotic) and dicrotic notches, (rectangles below) and timings for the dicrotic notch (figure 4). Figure 5 shows the displacement waveforms for all patient groups. The displacement waveforms were more complex, with additional higher frequency components compared to the arterial pressure waveforms, and are discussed in the following sections. Characteristic morphological differences were observed for the specific cardiac conditions. Firstly, the dicrotic notch was difficult to identify in the aortic stenosis traces, while 5 (patients 21, 22, 23, 26, 27) of the 7 mitral regurgitation subjects showed a pronounced deflection or secondary peak following the dicrotic notch ( figure 5).
Pulse wave analysis-cardiac disease waveform metrics Figure 6 shows the proportions of time to peak for both the normalised skin displacement and catheter pressure waveforms for the control, AS, and MR groups. Comparisons of the pressure and displacement waveform indices showed no significant differences between the proportions of time to peak for each group (p > 0.05).
Post hoc comparisons revealed that the proportions of time to peak for both the pressure and displacement waveforms were significantly greater for the AS group compared to the control group. The proportion of time to . Control cohort carotid artery, patient 5 A: pressure and B: displacement waveforms, normalised by pressure/displacement trough to peak amplitude and RR interval. Green and red boxes represent the anacrotic and dicrotic notches respectively. peak for the AS group was also significantly greater than that of the MR group for the pressure waveforms, but not for the displacement waveforms (figure 6 and table 2), perhaps due to the low numbers. Figure 7 shows the derived RI values for the normalised pressure and displacement waveforms for the control, AS, and MR groups. There were no significant differences between the pressure and displacement waveform indices for each group.
The dicrotic notch was not identifiable using our gradient-based analysis methods in 2 of the 10 AS patients (No. 18, 19) pressure measurements, and hence these cases were excluded from the modified RI analysis. Comparisons of the modified RIs between the control and AS groups showed no significant differences. On the other hand, the modified RI for MR was significantly lower compared to the control and AS groups for both the pressure and displacement waveforms ( figure 7 and table 2). Figure 8 displays the AIx values for the control and AS groups. AIx values were significantly greater for the AS group compared to control and MR groups for the displacement waveforms only ( figure 6 and table 2). Some AIx metrics from the MR group (Patients 21, 26) were not available, because in some cases it was difficult to identify the local minimum of the first derivatives (AP, figure 3) from the pressure and displacement waveforms.

Discussion
This work highlights the clinically based proof-of-concept study of a non-contact imaging system for quantifying skin displacements due to carotid artery pulsation. We evaluated the carotid artery displacements and catheter pressure waveforms in patients with two common cardiac conditions, AS and MR. Features such as Figure 5. Carotid pressure (blue) and displacement (black) waveforms over two to three cardiac cycles for each of the 27 patients. Detected dicrotic notches identified by purple arrows, and red crosses indicate traces where notches were not able to be identified. systolic rise, diastolic decline, and dicrotic notches were observed in both modalities, and the quantitative pulse wave metrics were able to provide initial insights into the cardiac cohorts/conditions. Visualisation of the carotid artery and the jugular vein pulsations enables clinicians to detect heart problems (Bickley et al 2009). Carotid arterial pressure is often used as a substitute for aortic pressure due to its close proximity (McEniery et al 2014). In addition, carotid arterial waveforms have a characteristic shape that is like that of the aortic pressure waveforms (Millasseau et al 2003). Our study shows that the displacement waveforms presented the characteristic systolic rise and diastolic decline characteristic of catheter pressure waveforms. Features such as the augmentation pressure (AP) notch (figure 8) (indicative of the reflection of pulse pressure from peripheral locations (Trudeau 2014)), and dicrotic notch (due to closure of the aortic valve (Trudeau 2014)) were evident in both modalities in most of the patients. However, the displacement waveforms were more complex. Other studies also reported this complexity and beat-to-beat variability in jugular venous pulse neck waveforms (García-López and Rodriguez-Villegas 2020) and carotid skin displacements (Li et al 2020). The complexity is believed to be due to several factors contributing to the displacement waveform, such as the mechanical properties of the surrounding tissues and dermis that lie between the epidermis and the arterial wall (McEniery et al 2014). The variabilities in the displacement waveforms are also likely to be influenced by other factors, such as the thickness of intermediate tissues of the subject in question (Garbey et al 2014) and vibrations from the muscular and subcutaneous tissues interposed between the artery and the skin (Li et al 2020). Further investigation using stereoscopic 3D reconstruction of the carotid artery displacement may uncover some of the variability.  While surrounding tissues may contribute to additional overlaying signals, it is unlikely that these factors alter the timing of the cardiac cycle events (systole, diastole, and aortic valve closure) within pressure-induced skin deformation waveforms. The quantitative pulse wave analysis technique showed no significant differences between the displacement and pressure metrics. The proportions of time to peak also showed no significant differences across the cohorts. The relatively close proximities of the aortic root and carotid artery lead to pulse timings being similar (McEniery et al 2014). Conversely, in AS these proportions were significantly extended in comparison to the controls, which was expected and is attributed to the narrowing of the aortic valve slowing the  exit of blood from the heart (Carità et al 2016). This, in turn, causes a shallower pressure gradient within the aorta, and a longer proportion of time for systolic ejection (Eleid and Nishimura 2020). These gradients and timings have been used to assess the severity of AS (Saikrishnan et al 2014).
For the AS/MR cohorts, the AIx was also extracted as an exploratory metric. AIx is not commonly used for AS/Mr However, it does provide an indication of stiffness. Echocardiography has been used to demonstrate the relationship between greater stiffness and AS severity (Bruschi et al 2017), where increased aortic and carotid arterial AIx is associated with increased risk of cardiovascular disease, mortality, and morbidity (Chirinos et al 2005). Our results showed that AS AIx was significantly different from the control and MR group for the displacement waveforms only. Conversely, in agreement with our findings, studies also show AS AIx for aortic (i.e. catheter) pressures do not significantly differ to control subjects (Jones et al 2021). The reason why AIx was significantly different for only the displacement waveforms may be that, with greater aortic stiffness, less blood is pooled within the stiff carotid arteries and more blood flows out toward the more compliant head/neck arteries, and thus the late systolic flow is further augmented (Hashimoto et al 2018). Furthermore, the proximity of the carotid artery to cerebral arterial beds likely results in an earlier pressure wave reflection which leads to a greater overlap of forward and reflected/backward pressure waves. The earlier arrival of reflected waves and greater overlap in stiff arteries has been shown to increase augmented pressures (Karpetas et al 2015, Hungerford et al 2021. However, further investigations are required to confirm this. In addition, AIx increases with age and is higher in females (Fantin et al 2006). The cohorts in this study were comprised of similar age ranges and, while the MR cohort contained mostly female participants, significantly greater AIx values were not observed for this group. Irrespective of age and gender, the significance of AIx for displacement waveforms highlights the potential importance of monitoring carotid stiffness, in line with clinical mortality correlations (van Sloten and Stehouwer 2016). Further studies comparing ultrasound of the carotid artery and the skin displacement waveforms will provide more insights into the mechanisms at play.
The MR cohort and traces displayed features characteristic of an incompetent valve. The majority of both waveform modalities exhibited more pronounced secondary peaks following the dicrotic notch. While the RI is not normally used to quantify MR waveforms, the large secondary peak enables our modified RI to be used for quantifying changes. Accordingly, we see that, compared with MR cases, the controls/AS display a larger modified RI. This secondary notch in MR is believed to be attributable to the contiguous wall of the aortic root and the anterior wall of the left atrium, where sudden left atrial pressure changes can reverberate directly to the aortic root. This reverberation is then transmitted to the aorta/carotid artery, potentially leading to a subsequent rise in pressure (J and M 1985). Furthermore, the secondary peak was more pronounced in the displacement waveform. Carotid sounds and auscultation have been shown to display this larger secondary peak, where an expansion of the large left atrium, which is commonly dilated in MR (Rusinaru et al 2011), during systole may cause a late systolic thrust in the parasternal region, which may mimic right ventricular enlargement. With severe MR there is a sharp carotid upstroke, and a downwardly and outwardly displaced brisk hyperdynamic apical impulse (Chatterjee 2018). Niki et al (1999) also showed that secondary wave intensity peaks associated with the dicrotic notch occur later in the cardiac cycle in MR compared to normal patients. This is in line with the reduction in the amplitudes following the dicrotic notch for MR patients observed in our study. These factors likely all contribute to the large secondary pulse observed. These results promote further investigation of pressure waveform morphology in MR, an area that is not yet well investigated.
The results of this study also indicate that the system and PWA metrics with further development can potentially be used to screen for and identify various CVD conditions. Patients are typically referred for specialist assessment and cardiac imaging, usually by echocardiography, when general practitioners suspect valvular heart disease. This is resource-intensive and expensive. Our device may provide additional clinically useful information, with which to better triage those patients potentially needing a specialist referral. The combination of PWA metrics presented in this study highlights the potential ability to distinguish between AS and MR, two common causes of systolic heart murmurs (Chorba et al 2021). In comparison to control and MR, the AS cohort had a greater mean, AIx, and a greater mean proportion of time to peak (see table 2). Conversely, in comparison to control and AS, the MR cohort had lower RI values. In AS, the dicrotic notch was difficult to identify due to the valvular insufficiency (Petzoldt et al 2013). Taken together, these metrics could be used to differentiate these cardiac conditions. For example, if the proportion of time to peak and AIx are higher than control values, then AS is the likely diagnosis. Alternatively, if RI is lower and AIx is similar to control values, then the likely diagnosis is Mr This preliminary classification scheme however requires further testing with a greater number of patients and a range of diseases/severities (i.e. other conditions associated with a systolic murmur). With further development, there is potential to improve cardiac condition differentiation without the need for relatively expensive diagnostic modalities such as echocardiography, catheterisation, or recent techniques that make use of digital stethoscopes (Chorba et al 2021).
Overall, aortic pressure measurements provide a useful predictor of cardiovascular mortality and morbidity. Carotid waveform measurements provide similar non-invasive insights. Current clinical methods of measuring the carotid waveforms either require invasive catheterisation, expensive cardiac imaging, specialised equipment, physical contact with the patient, or visualization methods that are highly subjective. In contrast, our device is non-invasive, non-contact, and can work with a standard video camera to measure the skin deformations caused directly by the carotid pressure pulses. Since the device is non-contact, there is no force exerted on the vessels, and hence the measurements are not influenced by compression as they are, for example, with ultrasound imaging. This study highlights the potential of our novel system to identify and provide insights into cardiac conditions, such as AS and Mr To build on this work, we aim to validate this system across a wider range of conditions, such as right heart valvular diseases, and explore smartphone implementation. This will enable the creation of an accessible tool that can provide clinicians with supplementary information to aid in the diagnosis and remote monitoring of CVDs. It has the potential to reduce the need for invasive and expensive techniques and to support telemedicine options for remote communities where long-distance travel can be a barrier to specialised cardiac healthcare.

Future considerations and limitations
Several considerations/limitations need to be taken for the future development of the system. Firstly, blue light was added to this study to provide a consistent light source. In future experiments, environmental light will be used. Ideally, the video acquisition and catheterisation would be performed concurrently. However, due to restrictions in catheter laboratory access and patient well-being, this was not possible in this initial study. It has been shown that large hemodynamic (blood volume, ventricular contractility, or systemic vascular resistance changes) changes of 15%-60% (Mulder et al 2022) and body position alterations in the lower trunk (Kubota et al 2015) are required to elicit significant changes in waveform morphology and hemodynamic features respectively. These changes are not expected within our experimental setting.
Large frame-to-frame motion was filtered out and patients were asked to remain still during the recording, they were however breathing as normal (to reduce patient discomfort) which may have induced some variability. In the future, we will test the implementation of breath-hold and further motion artifact correction methods if necessary. The video data was captured from two cameras and will also be used in the future to investigate whether stereoscopic imaging provides more information about the pulse wave and observed beat-to-beat variability.
It should also be noted that this study was exploratory in nature and patient recruitment in the clinical setting was difficult for some of the cohorts. This in turn has resulted in a relatively small cohort and has not allowed for more in-depth statistical assessment, such as sensitivity, specificity, and receiver operating characteristic calculations. Nevertheless, we were able to observe significant condition-based differences in our acquired measures.
The patients within each group were of similar ages and BMIs, thus reducing the possible confounding effects of these factors. However, there were a low number of females in the current cohort hence it was not possible to analyse and report on sex and gender differences. The information to participate in the present study was distributed equally among all the patients that were in the triage for cardiac catheterisation over the recruitment period of approximately 1 year. Most eligible participants in triage were male and opportunities to recruit were challenging in the clinical setting. This prevented us from balancing the male/female ratio, which reflects the individual rates of patient presentation. Similarly, while the cohort was diverse in terms of skin tone (Māori, Pacific People, Fijian Indian, NZ European), the numbers within each demographic sub-group were too small to analyse and report on by these groups. Building on the outcomes of the present research, we aim to investigate and overcome any potential sex, gender and ethnicity biases and limitations.
We also aim to compare our system against results derived from non-invasive imaging methods such as echocardiography. This will allow us to investigate the influence of the factors noted above and determine the utility of our method in identifying AS and MR prospectively in a larger, unseen cohort, and allow us to identify potential mimics that may lead to false positives. We also only used our system for binary diagnosis-related analyses (i.e. control versus CVD). With more data, we plan to subclassify these conditions (mild, moderate, severe) so that our tool can also potentially monitor the progression of the disease.
Lastly, while we have proposed possible reasons behind the observed relationships between RI in MR, the exact mechanisms remain unclear. MR aortic pressure waveforms have not been well established, additional studies with cardiovascular imaging are needed to identify the sequence of physiological steps that lead to the RI changes observed in MR. We believe that reporting on the MR waveforms is an important finding that requires further investigation.

Data availability statement
The data cannot be made publicly available upon publication because they are not available in a format that is sufficiently accessible or reusable by other researchers. The data that support the findings of this study are available upon reasonable request from the authors.

Sources of funding
This work was funded by the MedTech Centre of Research Excellence funded by the New Zealand Government. AJT is supported by a James Cook Fellowship provided by the Royal Society of New Zealand.

Disclosures
Written informed consent was also obtained for any identifying images to be published in an online open-access publication.

Contributions
Conception and design of the work (PK, AD, MW, AT, PN, MN, YC); acquisition and analysis (PK, AD, EL, AH, MW, PN, AT, MN, YC), interpretation of data (PK, MN, PN, AT, YC, MW); and drafting the work (PK, MN, PN, AT, YC). All authors approved the final version of the manuscript and agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved, and all persons designated as authors qualify for authorship, while all those who qualify for authorship are listed.