Publications
487 results found
Mashar M, Chawla S, Chen F, et al., 2023, Artificial Intelligence Algorithms in Health Care: Is the Current Food and Drug Administration Regulation Sufficient?, JMIR AI, Vol: 2, Pages: e42940-e42940
<jats:p>Given the growing use of machine learning (ML) technologies in health care, regulatory bodies face unique challenges in governing their clinical use. Under the regulatory framework of the Food and Drug Administration, approved ML algorithms are practically locked, preventing their adaptation in the ever-changing clinical environment, defeating the unique adaptive trait of ML technology in learning from real-world feedback. At the same time, regulations must enforce a strict level of patient safety to mitigate risk at a systemic level. Given that ML algorithms often support, or at times replace, the role of medical professionals, we have proposed a novel regulatory pathway analogous to the regulation of medical professionals, encompassing the life cycle of an algorithm from inception, development to clinical implementation, and continual clinical adaptation. We then discuss in-depth technical and nontechnical challenges to its implementation and offer potential solutions to unleash the full potential of ML technology in health care while ensuring quality, equity, and safety. References for this article were identified through searches of PubMed with the search terms “Artificial intelligence,” “Machine learning,” and “regulation” from June 25, 2017, until June 25, 2022. Articles were also identified through searches of the reference list of the articles. Only papers published in English were reviewed. The final reference list was generated based on originality and relevance to the broad scope of this paper.</jats:p>
On Y, Vimalesvaran K, Galazis C, et al., 2023, Automatic Aortic Valve Pathology Detection from 3-Chamber Cine MRI with Spatio-Temporal Attention Maps, Pages: 648-657, ISSN: 0302-9743
The assessment of aortic valve pathology using magnetic resonance imaging (MRI) typically relies on blood velocity estimates acquired using phase contrast (PC) MRI. However, abnormalities in blood flow through the aortic valve often manifest by the dephasing of blood signal in gated balanced steady-state free precession (bSSFP) scans (Cine MRI). We propose a 3D classification neural network (NN) to automatically identify aortic valve pathology (aortic regurgitation, aortic stenosis, mixed valve disease) from Cine MR images. We train and test our approach on a retrospective clinical dataset from three UK hospitals, using single-slice 3-chamber cine MRI from N = 576 patients. Our classification model accurately predicts the presence of aortic valve pathology (AVD) with an accuracy of 0.85 ± 0.03 and can also correctly discriminate the type of AVD pathology (accuracy: 0.75 ± 0.03 ). Gradient-weighted class activation mapping (Grad-CAM) confirms that the blood pool voxels close to the aortic root contribute the most to the classification. Our approach can be used to improve the diagnosis of AVD and optimise clinical CMR protocols for accurate and efficient AVD detection.
Davies HJ, Williams I, Peters NS, et al., 2023, In-Ear Blood Oxygen Saturation: A Tool for Wearable, Unobtrusive Monitoring of Core Blood Oxygen Saturation, Advances in Medical Imaging, Detection, and Diagnosis, Pages: 667-682, ISBN: 9789814877466
One of the major roles of blood is to supply oxygen to tissues throughout the body. This is achieved through the protein haemoglobin within red blood cells, which has a high affinity to oxygen. The term blood oxygen saturation specifically refers to the proportion of haemoglobin in the blood that is carrying oxygen and is given by Oxygen Saturation==HbO2HbO2+Hb, where Hb refers to haemoglobin not bound with oxygen and HbO2 refers to haemoglobin bound to oxygen. The blood oxygen estimation delay was calculated using the button release point as the marker for minimal blood oxygen, corresponding to the point at which the breath hold ends. The time between this point of minimal blood oxygen and the first trough of the SpO2 waveform for the ear and the finger was then used to calculate the SpO2 delay for the ear, the finger and then the relative delay between the ear and the finger.
Wu H, Patel KHK, Li X, et al., 2022, A fully-automated paper ECG digitisation algorithm using deep learning, Scientific Reports, Vol: 12, ISSN: 2045-2322
There is increasing focus on applying deep learning methods to electrocardiograms (ECGs), with recent studies showing that neural networks (NNs) can predict future heart failure or atrial fibrillation from the ECG alone. However, large numbers of ECGs are needed to train NNs, and many ECGs are currently only in paper format, which are not suitable for NN training. We developed a fully-automated online ECG digitisation tool to convert scanned paper ECGs into digital signals. Using automated horizontal and vertical anchor point detection, the algorithm automatically segments the ECG image into separate images for the 12 leads and a dynamical morphological algorithm is then applied to extract the signal of interest. We then validated the performance of the algorithm on 515 digital ECGs, of which 45 were printed, scanned and redigitised. The automated digitisation tool achieved 99.0% correlation between the digitised signals and the ground truth ECG (n = 515 standard 3-by-4 ECGs) after excluding ECGs with overlap of lead signals. Without exclusion, the performance of average correlation was from 90 to 97% across the leads on all 3-by-4 ECGs. There was a 97% correlation for 12-by-1 and 3-by-1 ECG formats after excluding ECGs with overlap of lead signals. Without exclusion, the average correlation of some leads in 12-by-1 ECGs was 60–70% and the average correlation of 3-by-1 ECGs achieved 80–90%. ECGs that were printed, scanned, and redigitised, our tool achieved 96% correlation with the original signals. We have developed and validated a fully-automated, user-friendly, online ECG digitisation tool. Unlike other available tools, this does not require any manual segmentation of ECG signals. Our tool can facilitate the rapid and automated digitisation of large repositories of paper ECGs to allow them to be used for deep learning projects.
Patel K, Bajaj N, Statton B, et al., 2022, Bariatric surgery reverses ventricular repolarisation heterogeneity in obesity: mechanistic insights into fat-related arrhythmic risk, ESC Congress 2022, Publisher: Oxford University Press, Pages: 658-658, ISSN: 1554-2815
Auton A, Padayachee Y, Samways J, et al., 2022, Smartphone-based remote monitoring in chronic heart failure: patient & clinician user experience, impact on patient engagement and quality of life, European Heart Journal, Vol: 43, Pages: 2808-2808, ISSN: 0195-668X
BackgroundHeart failure with reduced ejection fraction (HFrEF) lowers patients' quality of life (QoL) [1]. Digital interventions such as ESC's “Heart Failure Matters” website aim to encourage patient-engagement & self-management [2], which remain major challenges in HFrEF care. Although remote monitoring (RM) has been tested in HFrEF with inconclusive impact on prognosis [3], its impact on patients' experience and engagement is unclear [4]. Furthermore, the perspective of clinicians using RM technologies remains unknown. We present users' experience of Luscii, a novel smartphone-based RM platform enabling HFrEF patients to submit clinical measurements, symptoms, complete educational modules, & communicate with HF specialist nurses (HFSNs).Purpose(I) To evaluate the usage-type & user experience of patients and HFSNs.(II) To assess the impact of using the RM platform on self-reported QoLMethodsA two-part retrospective analysis of HFrEF patients from our regional service using the RM platform: Part A: Thematic analysis of patient feedback provided via the platform and a focus group of six HFSNs. Part B: Scores for a locally-devised HF questionnaire (HFQ), depression (PHQ-9) & anxiety (GAD-7) questionnaires were extracted from the RM platform at two timepoints: at on-boarding and 3 months after. Paired non-parametric tests were used to evaluate difference between median scores across the two time points.Results83 patients (mean age 62 years; 27% female) used the RM platform between April and November 2021. 2 dropped out & 2 died before 3 months. Part A: Patients and HFSNs exchanged information on many topics via the platform, including patient educational modules (Figure 1). Thematic analysis revealed positive and negative impacts with many overlapping subthemes between the two user groups (Figure 2). Part B: At 3 months there was no difference in HFQ score (19 vs. 18, p=0.57, maximum possible score = 50). PHQ-9 (3 vs. 3, p=0.48, maximum
Padayachee Y, Shah M, Auton A, et al., 2022, Smartphone-based remote monitoring (RM) in chronic heart failure reduces emergency hospital attendances, unplanned admissions and secondary care costs: a retrospective cohort study, ESC Congress, Publisher: Oxford University Press, Pages: 2816-2816, ISSN: 0195-668X
Sau A, Ibrahim S, Ahmed A, et al., 2022, Artificial intelligence-enabled electrocardiogram to distinguish cavotricuspid isthmus dependence from other atrial tachycardia mechanisms, European Heart Journal – Digital Health, Vol: 3, Pages: 405-414, ISSN: 2634-3916
Aims:Accurately determining atrial arrhythmia mechanisms from a 12-lead electrocardiogram (ECG) can be challenging. Given the high success rate of cavotricuspid isthmus (CTI) ablation, identification of CTI-dependent typical atrial flutter (AFL) is important for treatment decisions and procedure planning. We sought to train a convolutional neural network (CNN) to classify CTI-dependent AFL vs. non-CTI dependent atrial tachycardia (AT), using data from the invasive electrophysiology (EP) study as the gold standard.Methods and results:We trained a CNN on data from 231 patients undergoing EP studies for atrial tachyarrhythmia. A total of 13 500 five-second 12-lead ECG segments were used for training. Each case was labelled CTI-dependent AFL or non-CTI-dependent AT based on the findings of the EP study. The model performance was evaluated against a test set of 57 patients. A survey of electrophysiologists in Europe was undertaken on the same 57 ECGs. The model had an accuracy of 86% (95% CI 0.77–0.95) compared to median expert electrophysiologist accuracy of 79% (range 70–84%). In the two thirds of test set cases (38/57) where both the model and electrophysiologist consensus were in agreement, the prediction accuracy was 100%. Saliency mapping demonstrated atrial activation was the most important segment of the ECG for determining model output.Conclusion:We describe the first CNN trained to differentiate CTI-dependent AFL from other AT using the ECG. Our model matched and complemented expert electrophysiologist performance. Automated artificial intelligence-enhanced ECG analysis could help guide treatment decisions and plan ablation procedures for patients with organized atrial arrhythmias.
Patel K, li X, xu X, et al., 2022, Increasing adiposity is associated with QTc interval prolongation and increased ventricular arrhythmic risk in the context of metabolic dysfunction: results from the UK Biobank, Frontiers in Cardiovascular Medicine, Vol: 9, Pages: 1-11, ISSN: 2297-055X
Background: Small-scale studies have linked obesity (Ob) and metabolic ill-health with proarrhythmic repolarisation abnormalities. Whether these are observed at a population-scale, modulated by individuals’ genetics and confer higher risks of ventricular arrhythmias (VA) are not known. Methods and Results: Firstly, using the UK Biobank, the association between adiposity and QTc interval was assessed in participants with resting 12-lead ECG (n=23,683), and a polygenic risk score was developed to investigate any modulatory effect of genetics. Participants were also categorised into four phenotypes according to presence (+) or absence (-) of Ob, and if they were metabolically unhealthy (MU+) or not (MU-). QTc was positively associated with body mass index, body fat, waist:hip ratio, and hip and waist girths. Individuals’ genetics had no significant modulatory effect on QTc-prolonging effects of increasing adiposity. QTc was comparably longer in those with metabolic perturbationwithout obesity (Ob-MU+) and obesity alone (Ob+MU-) compared to individuals with neither (Ob-MU-), and their co-existence (Ob+MU+) had an additive effect on QTc interval. Secondly, for 502,536 participants in the UK Biobank, odds ratios (OR) for ventricular arrhythmias (VA) were computed for the four clinical phenotypes above using their past medical records. Referenced to Ob-MU-, ORs for VA in Ob-MU+ males and females were 5.96 (95%CI: 4.70-7.55) and 5.10 (95%CI: 3.34-7.80), respectively. OR for Ob+MU+ were 6.99 (95%CI: 5.72-8.54) and 3.56 (95%CI: 2.66-4.77) in males and females, respectively. Conclusion: Adiposity and metabolic perturbation increase QTc to a similar degree, and their co-existence exerts an additive effect. These effects are not modulated by individuals’ genetics. Metabolic ill-health is associated with higher OR for VA than obesity.
Falkenberg McGillivray M, Coleman JA, Dobson S, et al., 2022, Identifying locations susceptible to micro-anatomical reentry using a spatial network representation of atrial fibre maps, PLoS One, Vol: 17, Pages: 1-24, ISSN: 1932-6203
Micro-anatomical reentry has been identified as a potential driver of atrial fibrillation (AF). In this paper, we introduce a novel computational method which aims to identify which atrial regions are most susceptible to micro-reentry. The approach, which considers the structural basis for micro-reentry only, is based on the premise that the accumulation of electrically insulating interstitial fibrosis can be modelled by simulating percolation-like phenomena on spatial networks. Our results suggest that at high coupling, where micro-reentry is rare, the micro-reentrant substrate is highly clusteredin areas where the atrial walls are thin and have convex wall morphology, likely facilitating localised treatment via ablation. However, as transverse connections between fibres are removed, mimicking the accumulation of interstitial fibrosis, the substrate becomes less spatially clustered, and the bias to forming in thin, convex regions of the atria is reduced, possibly restricting the efficacy of localised ablation. Comparing our algorithm on image-based models with and without atrial fibre structure, we find thatstrong longitudinal fibre coupling can suppress the micro-reentrant substrate, whereas regions with disordered fibre orientations have an enhanced risk of micro-reentry. With further development, these methods may be useful for modelling the temporal development of the fibrotic substrate on an individualised basis.
Patel K, Bajaj N, Statton B, et al., 2022, Bariatric surgery reverses ventricular repolarisation heterogeneity – mechanistic insights into fat-related arrhythmic risk, British Cardiovascular Society Annual Conference, ‘100 years of Cardiology’, 6–8 June 2022, Publisher: BMJ Publishing Group, Pages: A60-A61, ISSN: 1355-6037
Hesketh LM, Sikkel MB, Mahoney-Sanchez L, et al., 2022, OCT2013, an ischaemia-activated antiarrhythmic prodrug, devoid of the systemic side effects of lidocaine, British Journal of Pharmacology, Vol: 179, Pages: 2037-2053, ISSN: 0007-1188
BACKGROUND AND PURPOSE: Sudden cardiac death (SCD) caused by acute myocardial ischaemia and ventricular fibrillation (VF) is an unmet therapeutic need. Lidocaine suppresses ischaemia-induced VF, but utility is limited by side effects and a narrow therapeutic index. Here we characterise OCT2013, a putative ischaemia-activated prodrug of lidocaine. EXPERIMENTAL APPROACH: The rat Langendorff-perfused isolated heart, anaesthetised rat and rat ventricular myocyte preparations were utilised in a series of blinded and randomised studies to investigate the antiarrhythmic effectiveness, adverse effects and mechanism of action of OCT2013, compared with lidocaine. KEY RESULTS: In isolated hearts, OCT2013 and lidocaine prevented ischaemia-induced VF equi-effectively, but OCT2013 did not share lidocaine's adverse effects (PR widening, bradycardia and negative inotropy). In anesthetised rats, i.v. OCT2013 and lidocaine suppressed VF and increased survival equi-effectively; OCT2013 had no effect on cardiac output even at 64 mg.kg-1 i.v., whereas lidocaine reduced it even at 1 mg.kg-1 . In adult rat ventricular myocytes, OCT2013 had no effect on Ca2+ handling whereas lidocaine impaired it. In paced isolated hearts, lidocaine caused rate-dependent conduction slowing and block, whereas OCT2013 was inactive. However, during regional ischaemia, OCT2013 and lidocaine equi-effectively hastened conduction block. Chromatography and mass spectrometry analysis revealed that OCT2013, detectable in normoxic OCT2013-perfused hearts, became undetectable during global ischaemia, with lidocaine becoming detectable. CONCLUSIONS AND IMPLICATIONS: OCT2013 is inactive but is bioreduced locally in ischaemic myocardium to lidocaine, acting as an ischaemia-activated and ischaemia-selective antiarrhythmic prodrug with a large therapeutic index, mimicking lidocaine's benefit without adversity.
Nagy SZ, Kasi P, Afonso VX, et al., 2022, Cycle length evaluation in persistent atrial fibrillation using kernel density estimation to identify transient and stable rapid atrial activity, Cardiovascular Engineering and Technology, Vol: 13, Pages: 219-233, ISSN: 1869-408X
PurposeLeft atrial (LA) rapid AF activity has been shown to co-localise with areas of successful atrial fibrillation termination by catheter ablation. We describe a technique that identifies rapid and regular activity.MethodsEight-second AF electrograms were recorded from LA regions during ablation for psAF. Local activation was annotated manually on bipolar signals and where these were of poor quality, we inspected unipolar signals. Dominant cycle length (DCL) was calculated from annotation pairs representing a single activation interval, using a probability density function (PDF) with kernel density estimation. Cumulative annotation duration compared to total segment length defined electrogram quality. DCL results were compared to dominant frequency (DF) and averaging.ResultsIn total 507 8 s AF segments were analysed from 7 patients. Spearman’s correlation coefficient was 0.758 between independent annotators (P < 0.001), 0.837–0.94 between 8 s and ≥ 4 s segments (P < 0.001), 0.541 between DCL and DF (P < 0.001), and 0.79 between DCL and averaging (P < 0.001). Poorer segment organization gave greater errors between DCL and DF.ConclusionDCL identifies rapid atrial activity that may represent psAF drivers. This study uses DCL as a tool to evaluate the dynamic, patient specific properties of psAF by identifying rapid and regular activity. If automated, this technique could rapidly identify areas for ablation in psAF.
Kim M-Y, Coyle C, Tomlinson DR, et al., 2022, Ectopy-triggering ganglionated plexus ablation to prevent atrial fibrillation: GANGLIA-AF study., Heart Rhythm, Vol: 19, Pages: 516-524, ISSN: 1547-5271
BACKGROUND: The ganglionated plexuses (GP) of the intrinsic cardiac autonomic system may play a role in atrial fibrillation (AF). OBJECTIVES: We hypothesized that ablating the ectopy-triggering GPs (ET-GP) prevents AF. METHODS: GANGLIA-AF (NCT02487654) was a prospective, randomized, controlled, 3-centre trial. ET-GP were mapped using high frequency stimulation (HFS), delivered within the atrial refractory period and ablated until non-functional. If triggered AF became incessant, atrioventricular dissociating GPs (AVD-GP) were ablated. We compared GP ablation (GPA) without pulmonary vein isolation (PVI) against PVI, in patients with paroxysmal AF. Follow-up was for 12 months including 3-monthly 48hr Holter monitors. The primary endpoint was documented ≥30s atrial arrhythmia after a 3-month blanking period. RESULTS: 102 randomized patients were analysed on a per-protocol basis after GPA (n=52) or PVI (n=50). GPA patients had 89±26 HFS sites tested, identifying median 18.5 (IQR 16; 21%) GPs. RF ablation time in GPA was 22.9±9.8mins and 38±14.4mins in PVI (p<0.0001). The freedom from ≥30s atrial arrhythmia at 12-month follow-up with GPA was 50% (26/52) vs 64% (32/50) with PVI (log rank p=0.09). ET-GP ablation without AVD-GP ablation achieved 58% (22/38) freedom from the primary endpoint. There was a significantly higher reduction in AAD usage post-ablation after GPA vs PVI (55.5% vs 36%; p=0.05). Patients were referred for redo ablations in 31% (16/52) after GPA and 24% (12/50) after PVI (p=0.53). CONCLUSIONS: GPA did not prevent atrial arrhythmias more than PVI. However, less RF ablation was delivered to achieve a higher reduction in AAD usage with GPA than PVI.
Sau A, Kaura A, Ahmed A, et al., 2022, Prognostic significance of ventricular arrhythmias in 13444 patients with acute coronary syndrome: a retrospective cohort study based on routine clinical data (NIHR Health Informatics Collaborative VA-ACS Study), Journal of the American Heart Association, Vol: 11, Pages: 1-19, ISSN: 2047-9980
Background: A minority of acute coronary syndrome (ACS) cases are associated with ventricular arrhythmias (VA) and/or cardiac arrest (CA). We investigated the effect of VA/CA at time of ACS on long-term outcomes.Methods and Results: We analysed routine clinical data from 5 NHS Trusts in the United Kingdom, collected between 2010 and 2017, by the National Institute for Health Research Health Informatics Collaborative (NIHR HIC).13,444 patients with ACS, of which 376 (2.8%) had concurrent VA, survived to hospital discharge and were followed up for a median of 3.42 years. Patients with VA or CA at index presentation had significantly increased risks of subsequent VA during follow-up (VA group: adjusted HR 4.15, 95% CI 2.42-7.09, CA group: adjusted HR 2.60 95% CI 1.23-5.48). Patients who suffered a CA in the context of ACS and survived to discharge also had a 36% increase in long-term mortality (adjusted hazard ratio 1.36 (95% 1.04-1.78)), though the concurrent diagnosis of VA alone during ACS did not affect all-cause mortality (adjusted HR 1.03, 95% CI 0.80-1.33). Conclusions: Patients who develop VA or CA during ACS, who survive to discharge, have increased risks of subsequent VA, while those who have CA during ACS also have an increase in long-term mortality. These individuals may represent a subgroup at greater risk of subsequent arrhythmic events due to intrinsically lower thresholds for developing VA.
Davies HJ, Bachtiger P, Williams I, et al., 2022, Wearable in-ear PPG: detailed respiratory variations enable classification of COPD, IEEE Transactions on Biomedical Engineering, Vol: 69, ISSN: 0018-9294
An ability to extract detailed spirometry-like breath-ing waveforms from wearable sensors promises to greatly improve respiratory health monitoring. Photoplethysmography (PPG) has been researched in depth for estimation of respiration rate, given that it varies with respiration through overall intensity, pulse amplitude and pulse interval. We compare and contrast the extraction of these three respiratory modes from both the ear canal and finger and show a marked improvement in the respiratory power for respiration induced intensity variations and pulse amplitude variations when recording from the ear canal. We next employ a data driven multi-scale method, noise assisted multivariate empirical mode decomposition (NA-MEMD), which allows for simultaneous analysis of all three respiratory modes to extract detailed respiratory waveforms from in-ear PPG. For rigour, we considered in-ear PPG recordings from healthy subjects, both older and young, patients with chronic obstructive pulmonary disease (COPD) and idiopathic pulmonary fibrosis (IPF) and healthy subjects with artificially obstructed breathing. Specific in-ear PPG waveform changes are observed for COPD, such as a decreased inspiratory duty cycle and an increased inspiratory magnitude, when compared with expiratory magnitude. These differences are used to classify COPD from healthy and IPF waveforms with a sensitivity of 87% and an overall accuracy of 92%. Our findings indicate the promise of in-ear PPG for COPD screening and unobtrusive respiratory monitoring in ambulatory scenarios and in consumer wearables.
Bachtiger P, Adamson A, Maclean W, et al., 2022, Increasing but inadequate intention to receive Covid-19 vaccination over the first 50 days of impact of the more infectious variant and roll-out of vaccination in UK: indicators for public health messaging, Publisher: MedRxiv
Objectives To inform critical public health messaging by determining how changes in Covid-19 vaccine hesitancy, attitudes to the priorities for administration, the emergence of new variants and availability of vaccines may affect the trajectory and achievement of herd immunity.Methods >9,000 respondents in an ongoing cross-sectional participatory longitudinal epidemiology study (LoC-19, n=18,581) completed a questionnaire within their personal electronic health record in the week reporting first effective Covid-19 vaccines, and then again after widespread publicity of the increased transmissibility of a new variant (November 13th and December 31st 2020 respectively). Questions covered willingness to receive Covid-19 vaccination and attitudes to prioritisation. Descriptive statistics, unadjusted and adjusted odds ratios (ORs) and natural language processing of free-text responses are reported, and how changes over the first 50 days of both vaccination roll-out and new-variant impact modelling of anticipated transmission rates and the likelihood and time to herd immunity.Findings Compared with the week reporting the first efficacious vaccine there was a 15% increase in acceptance of Covid-19 vaccination, attributable in one third to the impact of the new variant, with 75% of respondents “shielding” – staying at home and not leaving unless essential – regardless of health status or tier rules. 12.5% of respondents plan to change their behaviour two weeks after completing vaccination compared with 45% intending to do so only when cases have reduced to a low level. Despite the increase from 71% to 86% over this critical 50-day period, modelling of planned uptake of vaccination remains below that required for rapid effective herd immunity – now estimated to be 90 percent in the presence of a new variant escalating R0 to levels requiring further lockdowns. To inform the public messaging essential therefore to improve uptake, age and female
Bachtiger P, Petri CF, Scott FE, et al., 2022, Point-of-care screening for heart failure with reduced ejection fraction using artificial intelligence during ECG-enabled stethoscope examination in London, UK: a prospective, observational, multicentre study, The Lancet Digital Health, Vol: 4, ISSN: 2589-7500
BACKGROUND: Most patients who have heart failure with a reduced ejection fraction, when left ventricular ejection fraction (LVEF) is 40% or lower, are diagnosed in hospital. This is despite previous presentations to primary care with symptoms. We aimed to test an artificial intelligence (AI) algorithm applied to a single-lead ECG, recorded during ECG-enabled stethoscope examination, to validate a potential point-of-care screening tool for LVEF of 40% or lower. METHODS: We conducted an observational, prospective, multicentre study of a convolutional neural network (known as AI-ECG) that was previously validated for the detection of reduced LVEF using 12-lead ECG as input. We used AI-ECG retrained to interpret single-lead ECG input alone. Patients (aged ≥18 years) attending for transthoracic echocardiogram in London (UK) were recruited. All participants had 15 s of supine, single-lead ECG recorded at the four standard anatomical positions for cardiac auscultation, plus one handheld position, using an ECG-enabled stethoscope. Transthoracic echocardiogram-derived percentage LVEF was used as ground truth. The primary outcome was performance of AI-ECG at classifying reduced LVEF (LVEF ≤40%), measured using metrics including the area under the receiver operating characteristic curve (AUROC), sensitivity, and specificity, with two-sided 95% CIs. The primary outcome was reported for each position individually and with an optimal combination of AI-ECG outputs (interval range 0-1) from two positions using a rule-based approach and several classification models. This study is registered with ClinicalTrials.gov, NCT04601415. FINDINGS: Between Feb 6 and May 27, 2021, we recruited 1050 patients (mean age 62 years [SD 17·4], 535 [51%] male, 432 [41%] non-White). 945 (90%) had an ejection fraction of at least 40%, and 105 (10%) had an ejection fraction of 40% or lower. Across all positions, ECGs were most frequently of adequate quality for AI-ECG interpretation at the p
Letchumy MJ, Brook J, Ntagiantas K, et al., 2022, The Effects of Electrode Configuration on Omnipolar Electrograms: An In-Silico Approach, ISSN: 2325-8861
Atrial Fibrillation (AF) is the most common cardiac ar-rhythmia, involving pathological triggers and substrate in the atria. In the clinical catheter laboratory, contact electrograms are an essential tool to characterise AF. Omnipolar electrograms (OE), derived from three or more neighbouring electrodes, are thought to be superior compared to traditional unipolar and bipolar electrograms by eliminating far-field effects and correcting for wavefront incidence angle. We sought to understand the changes in OE morphology under different electrode configurations using 2D simulations of healthy tissue and scarred tissue. Virtual unipolar electrograms (UE) were generated from single electrodes which were used to predict the local electric field and subsequently calculate OEs in cliques of 3, 4, and 6 electrodes at different inter-electrode spacings. Five features were identified on each OE to measure changes in OE morphology under different clique configurations. Additionally, the morphology of the OE signals in the presence of fibrosis was examined. OE signals obtained from scarred tissue are more fractionated compared to healthy tissue. The most appropriate inter-electrode distance for interpreting the OE signals was found to be 2-3mm, using either three or four electrodes.
Zaman S, Petri C, Vimalesvaran K, et al., 2022, Automatic diagnosis labeling of cardiovascular MRI by using semisupervised natural language processing of text reports, Radiology: Artificial Intelligence, Vol: 4, ISSN: 2638-6100
A semisupervised natural language processing (NLP) algorithm, based on bidirectional transformers, accurately categorized diagnoses from cardiac MRI text of radiology reports for the labeling of MR images; the model had a higher accuracy than traditional NLP models and performed faster labeling than clinicians.
Ntagiantas K, Panagopoulos D, Poon WM, et al., 2022, Electrogram-based Estimation of Myocardial Conduction Using Deep Neural Networks, ISSN: 2325-8861
Contact electrograms (EGMs) can be used to guide catheter ablation in the treatment of atrial fibrillation. However, our understanding of the link between electrophysiology (EP) and the underlying myocardial substrate is limited. We use neural networks and EGMs to estimate the amount of collagen within the field of view of the recording electrodes. EGMs were recorded from rat ventricular slices (n=15), Samples were imaged using second harmonic generation (SHG) microscopy, allowing for quantification of collagen. A convolutional neural network (1D-ResNet) was trained to estimate collagen distribution from the recorded EGMs. Each electrogram, recorded for one cycle length, was paired with a collagen index for the corresponding electrode. The total number of samples was 91,239. We successfully estimated collagen index in the testing set, with an absolute error of 0.022± 0.024, and a correlation coefficient of R=0.81 between the predicted and true collagen amount. The network identifies main morphological features of the EGMs as useful features for quantifying collagen underneath the electrode. This work provides a framework and proof of concept that location of scar can be predicted from EGMS using neural networks.
Ciaccio EJ, Anter E, Coromilas J, et al., 2022, Structure and function of the ventricular tachycardia isthmus, Heart Rhythm, Vol: 19, Pages: 137-153, ISSN: 1547-5271
Catheter ablation of postinfarction reentrant ventricular tachycardia (VT) has received renewed interest owing to the increased availability of high-resolution electroanatomic mapping systems that can describe the VT circuits in greater detail and the emergence and need to target noninvasive external beam radioablation. These recent advancements provide optimism for improving the clinical outcome of VT ablation in patients with postinfarction and potentially other scar-related VTs. The combination of analyses gleaned from studies in swine and canine models of postinfarction reentrant VT, and in human studies, suggests the existence of common electroanatomic properties for reentrant VT circuits. Characterizing these properties may be useful for increasing the specificity of substrate mapping techniques and for noninvasive identification to guide ablation. Herein, we describe properties of reentrant VT circuits that may assist in elucidating the mechanisms of onset and maintenance, as well as a means to localize and delineate optimal catheter ablation targets.
Herrero Martin C, Oved A, Chowdhury R, et al., 2021, EP-PINNs: cardiac electrophysiology characterisation using physics-informed neural networks, Frontiers in Cardiovascular Medicine, ISSN: 2297-055X
Accurately inferring underlying electrophysiological (EP) tissue properties from action potential recordings is expected to be clinically useful in the diagnosis and treatment of arrhythmias such as atrial fibrillation, but it is notoriously difficult to perform. We present EP-PINNs (Physics-Informed Neural Networks), a novel tool for accurate action potential simulation and EP parameter estimation, from sparse amounts of EP data. We demonstrate, using 1D and 2D in silico data, how EP-PINNs are able to reconstruct the spatio-temporal evolution of action potentials, whilst predicting parameters related to action potential duration (APD), excitability and diffusion coefficients. EP-PINNs are additionally able to identify heterogeneities in EP properties, making them potentially useful for the detection of fibrosis and other localised pathology linked to arrhythmias. Finally, we show EP-PINNs effectiveness on biological in vitro preparations, by characterising the effect of anti-arrhythmic drugs on APD using optical mapping data. EP-PINNs are a promising clinical tool for the characterisation and potential treatment guidance of arrhythmias.
Bachtiger P, Petri CF, Scott FE, et al., 2021, Artificial Intelligence For Point-of-Care Heart Failure Screening During ECG-Enabled Stethoscope Examination: Independent Real-World Prospective Multicenter External Validation Study, Scientific Sessions of the American-Heart-Association / Resuscitation Science Symposium, Publisher: LIPPINCOTT WILLIAMS & WILKINS, Pages: E586-E587, ISSN: 0009-7322
Bachtiger P, Petri CF, Scott FE, et al., 2021, Artificial Intelligence For Point-of-Care Heart Failure Screening During ECG-Enabled Stethoscope Examination: Independent Real-World Prospective Multicenter External Validation Study, Scientific Sessions of the American-Heart-Association / Resuscitation Science Symposium, Publisher: LIPPINCOTT WILLIAMS & WILKINS, Pages: E586-E587, ISSN: 0009-7322
Martin CH, Oved A, Chowdhury RA, et al., 2021, EP-PINNs: cardiac electrophysiology characterisation using physics-informed neural networks, Publisher: arXiv
Accurately inferring underlying electrophysiological (EP) tissue propertiesfrom action potential recordings is expected to be clinically useful in thediagnosis and treatment of arrhythmias such as atrial fibrillation, but it isnotoriously difficult to perform. We present EP-PINNs (Physics-Informed NeuralNetworks), a novel tool for accurate action potential simulation and EPparameter estimation, from sparse amounts of EP data. We demonstrate, using 1Dand 2D in silico data, how EP-PINNs are able to reconstruct the spatio-temporalevolution of action potentials, whilst predicting parameters related to actionpotential duration (APD), excitability and diffusion coefficients. EP-PINNs areadditionally able to identify heterogeneities in EP properties, making thempotentially useful for the detection of fibrosis and other localised pathologylinked to arrhythmias. Finally, we show EP-PINNs effectiveness on biological invitro preparations, by characterising the effect of anti-arrhythmic drugs onAPD using optical mapping data. EP-PINNs are a promising clinical tool for thecharacterisation and potential treatment guidance of arrhythmias.
Chow J-J, Leong KMW, Yazdani M, et al., 2021, A Multicenter External Validation of a Score Model to Predict Risk of Events in Patients With Brugada Syndrome, AMERICAN JOURNAL OF CARDIOLOGY, Vol: 160, Pages: 53-59, ISSN: 0002-9149
- Author Web Link
- Cite
- Citations: 4
Patel K, Li X, Sun L, et al., 2021, Neural networks applied to 12-lead electrocardiograms predict body mass index, visceral adiposity and concurrent cardiometabolic ill-health, Cardiovascular Digital Health Journal, Vol: 2, Pages: S1-S10, ISSN: 2666-6936
BackgroundObesity is associated with electrophysiological remodeling, which manifests as detectable changes on the surface electrocardiogram (ECG).ObjectiveTo develop neural networks (NN) to predict body mass index (BMI) from ECGs and test the hypothesis that discrepancies between NN-predicted BMI and measured BMI are indicative of underlying adiposity and/or concurrent cardiometabolic ill-health.MethodsNN models were developed using 36,856 12-lead resting ECGs from the UK Biobank. Two architectures were developed for continuous and categorical BMI estimation (normal weight [BMI <25 kg/m2] vs overweight/obese [BMI ≥25 kg/m2]). Models for male and female participants were trained and tested separately. For each sex, data were randomly divided into 4 folds, and models were evaluated in a leave-1-fold-out manner.ResultsECGs were available for 17,807 male and 19,049 female participants (mean ages: 61 ± 7 and 63 ± 8 years; mean BMI 26 ± 5 kg/m2 and 27 ± 4 kg/m2, respectively). NN models detected overweight/obese individuals with average accuracies of 75% and 73% for male and female subjects, respectively. The magnitudes of difference between NN-predicted BMI and actual BMI were significantly correlated with visceral adipose tissue volumes. Concurrent hypertension, diabetes, dyslipidemia, and/or coronary heart disease explained false-positive classifications (ie, calculated BMI <25 kg/m2 misclassified as ≥25 kg/m2 by NN model, P < .001).ConclusionNN models applied to 12-lead ECGs predict BMI with a reasonable degree of accuracy. Discrepancies between NN-predicted and calculated BMI may be indicative of underlying visceral adiposity and concomitant cardiometabolic perturbation, which could be used to identify individuals at risk of cardiometabolic disease.
Li X, Shi X, Handa BS, et al., 2021, Classification of fibrillation organisation using electrocardiograms to guide mechanism-directed treatments, Frontiers in Physiology, Vol: 12, Pages: 1-14, ISSN: 1664-042X
Background: Atrial fibrillation (AF) and ventricular fibrillation (VF) are complex heart rhythm disorders and may be sustained by distinct electrophysiological mechanisms. Disorganised self-perpetuating multiple-wavelets and organised rotational drivers (RDs) localising to specific areas are both possible mechanisms by which fibrillation is sustained. Determining the underlying mechanisms of fibrillation may be helpful in tailoring treatment strategies. We investigated whether global fibrillation organisation, a surrogate for fibrillation mechanism, can be determined from electrocardiograms (ECGs) using band-power (BP) feature analysis and machine learning.Methods: In this study, we proposed a novel ECG classification framework to differentiate fibrillation organisation levels. BP features were derived from surface ECGs and fed to a linear discriminant analysis classifier to predict fibrillation organisation level. Two datasets, single-channel ECGs of rat VF (n = 9) and 12-lead ECGs of human AF (n = 17), were used for model evaluation in a leave-one-out (LOO) manner.Results: The proposed method correctly predicted the organisation level from rat VF ECG with the sensitivity of 75%, specificity of 80%, and accuracy of 78%, and from clinical AF ECG with the sensitivity of 80%, specificity of 92%, and accuracy of 88%.Conclusion: Our proposed method can distinguish between AF/VF of different global organisation levels non-invasively from the ECG alone. This may aid in patient selection and guiding mechanism-directed tailored treatment strategies.
Bachtiger P, Scott F, Park S, et al., 2021, Multicentre validation of point-of-care screening tool for heart failure: single-lead ECG recorded by smart stethoscope predicts low ejection fraction using artificial intelligence, ESC Congress 2021, Publisher: European Society of Cardiology, Pages: 3071-3071, ISSN: 0195-668X
Background/IntroductionArtificial intelligence (AI) applied to 12-lead ECG can identify left ventricular ejection fraction (EF) ≤35% with a sensitivity and specificity of 86.3% and 85.7%, respectively. Whether AI algorithms trained on 12-lead can accurately predict EF from single-lead ECGs (recorded by a smart stethoscope) remains unknown. This could facilitate point-of-care screening for low EF during routine clinical examination.PurposeFirst independent multicentre real-world UK National Health Service (NHS) prospective validation of 12-lead-ECG-trained AI algorithm applied to single-lead ECG recorded by a smart stethoscope, with AI algorithm tuned to detect EF ≤40%.MethodsProspective recruitment of unselected patients attending for echocardiography across six urban NHS hospital sites (UK). In addition to transthoracic echocardiogram (routine care), all participants had 15 seconds of supine, single-lead ECG recorded at six different positions (figure), encompassing standard anatomical positions for cardiac auscultation. A convolutional neural network (CNN) previously trained on 35,970 independent pairings of 12-lead-ECG and echocardiograms was retrained to use the single-lead ECG as input. Accuracy of CNN detection of low EF (binary ≤40%) is reported at a threshold of 0.5 against gold-standard; echo-determined percentage EF.ResultsAmong 353 patients recruited (mean age 63±17; 58% male, 43.1% non-white), 309 (87.5%) had an EF >40%, and 44 (12.5%) had EF ≤40%. The best single recording position in isolation was position 3 (sensitivity 57.9% [42.2–73.6], specificity 86.3% [82.2–90.3]). Taking any of the six positions performed during the examination as predicting EF ≤40%, this achieved a sensitivity of 81.2% and specificity of 61.5%.Conclusion(s)In this first prospective multicentre validation study the retrained AI algorithm reliably detected low EF from single-lead ECGs acquired using a novel ECG-enabled stethoscope in standard
This data is extracted from the Web of Science and reproduced under a licence from Thomson Reuters. You may not copy or re-distribute this data in whole or in part without the written consent of the Science business of Thomson Reuters.