tinyHLS: a novel open source high level synthesis tool targeting hardware accelerators for artificial neural network inference

Objective.In recent years, wearable devices such as smartwatches and smart patches have revolutionized biosignal acquisition and analysis, particularly for monitoring electrocardiography (ECG). However, the limited power supply of these devices often precludes real-time data analysis on the patch itself.Approach.This paper introduces a novel Python package, tinyHLS (High Level Synthesis), designed to address these challenges by converting Python-based AI models into platform-independent hardware description language code accelerators. Specifically designed for convolutional neural networks, tinyHLS seamlessly integrates into the AI developer's workflow in Python TensorFlow Keras. Our methodology leverages a template-based hardware compiler that ensures flexibility, efficiency, and ease of use. In this work, tinyHLS is first-published featuring templates for several layers of neural networks, such as dense, convolution, max and global average pooling. In the first version, rectified linear unit is supported as activation. It targets one-dimensional data, with a particular focus on time series data.Main results.The generated accelerators are validated in detecting atrial fibrillation on ECG data, demonstrating significant improvements in processing speed (62-fold) and energy efficiency (4.5-fold). Quality of code and synthesizability are ensured by validating the outputs with commercial ASIC design tools.Significance.Importantly, tinyHLS is open-source and does not rely on commercial tools, making it a versatile solution for both academic and commercial applications. The paper also discusses the integration with an open-source RISC-V and potential for future enhancements of tinyHLS, including its application in edge servers and cloud computing. The source code is available on GitHub:https://github.com/Fraunhofer-IMS/tinyHLS.

Atrial Fibrillation Burden on a 14-Day ECG Monitor: Findings From the GUARD-AF Trial Screening Arm

BACKGROUND

The "burden" of atrial fibrillation (AF) detected by screening likely influences stroke risk, but the distribution of burden is not well described.

OBJECTIVES

This study aims to determine the frequency of AF and the distribution of AF burden found when screening individuals ≥70 years of age with a 14-day electrocardiograph monitor.

METHODS

This is a cohort study of the screening arm of a randomized AF screening trial among those ≥70 years of age without a prior AF diagnosis (between 2019 and 2021). Screening was performed with a 14-day continuous electrocardiogram patch monitor.

RESULTS

Analyzable patches were returned by 5,684 (95%) of screening arm participants; the median age was 75 years (Q1-Q3: 72-78 years), 57% were female, and the median CHA2DS2-VASc score was 3 (Q1-Q3: 2-4). AF was detected in 252 participants (4.4%); 29 (0.5%) patients had continuous AF and 223 (3.9%) had paroxysmal AF. Among those with paroxysmal AF, the average indices of AF burden were of low magnitude with right-skewed distributions. The median percent time in AF was 0.46% (Q1-Q3: 0.02%-2.48%), or 75 (Q1-Q3: 3-454) minutes, and the median longest episode was 38 (Q1-Q3: 2-245) minutes. The upper quartile threshold of 2.48% time in AF corresponded to 7.6 hours. Age >80 years was associated with screen-detected AF in our multivariable model (OR: 1.46; 95% CI: 1.06-2.02).

CONCLUSIONS

Most AF detected in these older patients was very low burden. However, one-quarter of those with AF had multiple hours of AF, raising concern about stroke risk. These findings have implications for targeting populations for AF screening trials and for responding to heart rhythm alerts from mobile devices (GUARD-AF [A Study to Determine if Identification of Undiagnosed Atrial Fibrillation in People at least 70 Years of Age Reduces the Risk of Stroke]; NCT04126486).

Early detection of atrial fibrillation in the digital era, risk factors, treatment options, and the need for new definitions

Graphical abstract.

Comparing the outcomes and costs of cardiac monitoring with implantable loop recorders and mobile cardiac outpatient telemetry following stroke using real-world evidence

Aim: Patients with ischemic stroke (IS) commonly undergo monitoring to identify atrial fibrillation with mobile cardiac outpatient telemetry (MCOT) or implantable loop recorders (ILRs). The authors compared readmission, healthcare cost and survival in patients monitored post-stroke with either MCOT or ILR. Materials & methods: The authors used claims data from Optum's de-identified Clinformatics® Data Mart Database to identify patients with IS hospitalized from January 2017 to December 2020 who were prescribed ambulatory cardiac monitoring via MCOT or ILR. They compared the costs associated with the initial inpatient visit as well as the rate and causes of readmission, survival and healthcare costs over the following 18 months. Datasets were balanced using patient baseline and hospitalization characteristics. Multivariable generalized linear gamma regression was used for cost comparisons. Cox proportional hazard regression was used for survival and readmission analysis. Sub-cohorts were analyzed based on the severity of the index IS. Results: In 2244 patients, readmissions were significantly lower in the MCOT monitored group (30.2%) compared with the ILR group (35.4%) (hazard ratio [HR] 1.23; 95% CI: 1.04-1.46). Average cost over 18 months starting with the index IS was $27,429 (USD) lower in the MCOT group (95% CI: $22,353-$32,633). Survival difference bordered on statistical significance and trended to lower mortality in MCOT (8.9%) versus ILR (11.3%) (HR 1.30; 95% CI: 1:00-1.69), led by significance in patients with complications or comorbidities with the index event (MCOT 7.5%, ILR 11.5%; HR 1.62; 95% CI: 1.11-2.36). Conclusion: The use of MCOT versus ILR as the primary monitor following IS was associated with significant decreases in readmission, lower costs for the initial IS and total care over the next 18 months, significantly lower mortality for patients with complications and comorbidities at the index stroke, and a trend toward improved survival across all patients.

Length of hospital stay for elective electrophysiological procedures: a survey from the European Heart Rhythm Association

AIMS

Electrophysiological (EP) operations that have traditionally involved long hospital lengths of stay (LOS) are now being undertaken as day case procedures. The coronavirus disease-19 pandemic served as an impetus for many centres to shorten LOS for EP procedures. This survey explores LOS for elective EP procedures in the modern era.

METHODS AND RESULTS

An online survey consisting of 27 multiple-choice questions was completed by 245 respondents from 35 countries. With respect to de novo cardiac implantable electronic device (CIED) implantations, day case procedures were reported for 79.5% of implantable loop recorders, 13.3% of pacemakers (PMs), 10.4% of implantable cardioverter defibrillators (ICDs), and 10.2% of cardiac resynchronization therapy (CRT) devices. With respect to CIED generator replacements, day case procedures were reported for 61.7% of PMs, 49.2% of ICDs, and 48.2% of CRT devices. With regard to ablations, day case procedures were reported for 5.7% of atrial fibrillation (AF) ablations, 10.7% of left-sided ablations, and 17.5% of right-sided ablations. A LOS ≥ 2 days for CIED implantation was reported for 47.7% of PM, 54.5% of ICDs, and 56.9% of CRT devices and for 54.5% of AF ablations, 42.2% of right-sided ablations, and 46.1% of left-sided ablations. Reimbursement (43-56%) and bed availability (20-47%) were reported to have no consistent impact on the organization of elective procedures.

CONCLUSION

There is a wide variation in the LOS for elective EP procedures. The LOS for some procedures appears disproportionate to their complexity. Neither reimbursement nor bed availability consistently influenced LOS.

Atrial Fibrillation Recurrence in Patients With Transient New-Onset Atrial Fibrillation Detected During Hospitalization for Noncardiac Surgery or Medical Illness : A Matched Cohort Study

BACKGROUND

Atrial fibrillation (AF) is often detected for the first time in patients who are hospitalized for another reason. Long-term risks for AF recurrence in these patients are unclear.

OBJECTIVE

To estimate risk for AF recurrence in patients with new-onset AF during a hospitalization for noncardiac surgery or medical illness compared with a matched population without AF.

DESIGN

Matched cohort study. (ClinicalTrials.gov: NCT03221777).

SETTING

Three academic hospitals in Hamilton, Ontario, Canada.

PARTICIPANTS

The study enrolled patients hospitalized for noncardiac surgery or medical illness who had transient new-onset AF. For each participant, an age- and sex-matched control participant with no history of AF from the same hospital ward was recruited. All participants left the hospital in sinus rhythm.

MEASUREMENTS

14-day electrocardiographic (ECG) monitor at 1 and 6 months and telephone assessment at 1, 6, and 12 months. The primary outcome was AF lasting at least 30 seconds on the monitor or captured by ECG 12-lead during routine care at 12 months.

RESULTS

Among 139 participants with transient new-onset AF (70 patients with medical illness and 69 surgical patients) and 139 matched control participants, the mean age was 71 years (SD, 10), the mean CHA2DS2-VASc score was 3.0 (SD, 1.5), and 59% were male. The median duration of AF during the index hospitalization was 15.8 hours (IQR, 6.4 to 49.6 hours). After 1 year, recurrent AF was detected in 33.1% (95% CI, 25.3% to 40.9%) of participants in the transient new-onset AF group and 5.0% (CI, 1.4% to 8.7%) of matched control participants; after adjustment for the number of ECG monitors worn and for baseline clinical differences, the adjusted relative risk was 6.6 (CI, 3.2 to 13.7). After exclusion of participants who had electrical or pharmacologic cardioversion during the index hospitalization (n = 40) and their matched control participants and limiting to AF events detected by the patch ECG monitor, recurrent AF was detected in 32.3% (CI, 23.1% to 41.5%) of participants with transient new-onset AF and 3.0% (CI, 0% to 6.4%) of matched control participants.

LIMITATIONS

Generalizability is limited, and the study was underpowered to evaluate subgroups and clinical predictors.

CONCLUSION

Among patients who have transient new-onset AF during a hospitalization for noncardiac surgery or medical illness, approximately 1 in 3 will have recurrent AF within 1 year.

PRIMARY FUNDING SOURCE

Peer-reviewed grants.

Opioid Impacts on Cardiovascular Health

The prevalence of opioid use in the current opioid epidemic era has led to a public health emergency due to the ties to mortality and morbidity. Studies have investigated opioids' impacts on different aspects of cardiovascular health, although there seems to be a lack of a current concise review. Therefore, the aim of this literature review is to provide a summary of the most recent studies from the past decade that postulate a connection between opioids and their impact on cardiovascular health while highlighting conflicting areas among published research. For this literature review, three databases, PubMed (NLM), EMBASE, and Web of Science (Core Collection), were searched for full peer-reviewed articles written in English about human subjects and published between 2013 and 2023 inclusive. The following initial approach was to search for terms in the title of articles: "opioid AND ("vascular" OR "artery" OR "vein" OR "heart rate" OR "infarct" OR "stroke" OR "aortic" OR "cardiovascular disease"). After assessing for duplicate articles from the three databases, the remaining articles were assessed for inclusion eligibility. In the present review, a brief description of the overall role of opioid receptors is provided followed by the literature findings. These findings indicate potentially important negative impacts of opioid use on cardiovascular health in a number of areas. These include opioid-associated increases in the following: (1) vascular aging based on demonstrated increases in arterial stiffness, (2) opioid-related reductions in heart rate variability (HRV) and its implications on morbidity and mortality, (3) opioid's impacts on coronary artery and coronary heart disease (CHD), (4) opioids as a risk factor for atrial fibrillation (AF) and (5) opioid use as a risk factor for vascular occlusion processes. In addition to these broad cardiovascular effects, other aspects of concern are related to the potential impacts of withdrawal from opioid use, which, when done rapidly, are associated with increases in blood pressure and a decrease in HRV.

Insertable cardiac monitoring results in higher rates of atrial fibrillation diagnosis and oral anticoagulation prescription after ischaemic stroke

AIMS

After an ischaemic stroke, atrial fibrillation (AF) detection allows for improved secondary prevention strategies. This study aimed to compare AF detection and oral anticoagulant (OAC) initiation in patients with an insertable cardiac monitor (ICM) vs. external cardiac monitor (ECM) after ischaemic stroke.

METHODS AND RESULTS

Medicare Fee-for-Service (FFS) insurance claims and Abbott Labs device registration data were used to identify patients hospitalized with an ischaemic stroke in 2017-2019 who received an ICM or ECM within 3 months. Patients with continuous Medicare FFS insurance and prescription drug enrolment in the prior year were included. Patients with prior AF, atrial flutter, cardiac devices, or OAC were excluded. Insertable cardiac monitor and ECM patients were propensity score matched 1:4 on demographics, comorbidities, and stroke hospitalization characteristics. The outcomes of interest were AF detection and OAC initiation evaluated with Kaplan-Meier and Cox proportional hazard regression analyses. A total of 5702 Medicare beneficiaries (ICM, n = 444; ECM, n = 5258) met inclusion criteria. The matched cohort consisted of 2210 Medicare beneficiaries (ICM, n = 442; ECM, n = 1768) with 53% female, mean age 75 years, and mean CHA₂DS₂-VASc score 4.6 (1.6). Insertable cardiac monitor use was associated with a higher probability of AF detection [(hazard ratio (HR) 2.88, 95% confidence interval (CI) (2.31, 3.59)] and OAC initiation [HR 2.91, CI (2.28, 3.72)] compared to patients monitored only with ECM.

CONCLUSION

Patients with an ischaemic stroke monitored with an ICM were almost three times more likely to be diagnosed with AF and to be prescribed OAC compared to patients who received ECM only.

The influence of atrial high-rate episodes on stroke and cardiovascular death: an update

Atrial high-rate episodes (AHRE) are atrial tachyarrhythmias detected by continuous rhythm monitoring by pacemakers, defibrillators, or implantable cardiac monitors. Atrial high-rate episodes occur in 10-30% of elderly patients without atrial fibrillation. However, it remains unclear whether the presence of these arrhythmias has therapeutic consequences. The presence of AHRE increases the risk of stroke compared with patients without AHRE. Oral anticoagulation would have the potential to reduce the risk of stroke in patients with AHRE but is also associated with a rate of major bleeding of ∼2%/year. The stroke rate in patients with AHRE appears to be lower than the stroke rate in patients with atrial fibrillation. Wearables like smart-watches will increase the absolute number of patients in whom atrial arrhythmias are detected. It remains unclear whether anticoagulation is effective and, equally important, safe in patients with AHRE. Two randomized clinical trials, NOAH-AFNET6 and ARTESiA, are expected to report soon. They will provide much-needed information on the efficacy and safety of oral anticoagulation in patients with AHRE.

Incidence of clinical atrial fibrillation and related complications using a screening algorithm at a nationwide level

AIMS

In a recent position paper, the European Heart Rhythm Association (EHRA) proposed an algorithm for the screening and management of arrhythmias using digital devices. In patients with prior stroke, a systematic screening approach for atrial fibrillation (AF) should always be implemented, preferably immediately after the event. Patients with increasing age and with specific cardiovascular or non-cardiovascular comorbidities are also deemed to be at higher risk. From a large nationwide database, the aim was to analyse AF incidence rates derived from this new EHRA algorithm.

METHODS AND RESULTS

Using the French administrative hospital discharge database, all patients hospitalized in 2012 without a history of AF, and with at least a 5-year follow-up (FU) (or if they died earlier), were included. The yearly incidence of AF was calculated in each subgroup defined by the algorithm proposed by EHRA based on a history of previous stroke, increasing age, and eight comorbidities identified via International Classification of Diseases 10th Revision codes. Out of the 4526 104 patients included (mean age 58.9 ± 18.9 years, 64.5% women), 1% had a history of stroke. Among those with no history of stroke, 18% were aged 65-74 years and 21% were ≥75 years. During FU, 327 012 patients had an incidence of AF (yearly incidence 1.86% in the overall population). Implementation of the EHRA algorithm divided the population into six risk groups: patients with a history of stroke (group 1); patients > 75 years (group 2); patients aged 65-74 years with or without comorbidity (groups 3a and 3b); and patients < 65 years with or without comorbidity (groups 4a and 4b). The yearly incidences of AF were 4.58% per year (group 2), 6.21% per year (group 2), 3.50% per year (group 3a), 2.01% per year (group 3b), 1.23% per year (group 4a), and 0.35% per year (group 4b). In patients aged < 65 years, the annual incidence of AF increased progressively according to the number of comorbidities from 0.35% (no comorbidities) to 9.08% (eight comorbidities). For those aged 65-75 years, the same trend was observed, i.e. increasing from 2.01% (no comorbidities) to 11.47% (eight comorbidities).

CONCLUSION

These findings at a nationwide scale confirm the relevance of the subgroups in the EHRA algorithm for identifying a higher risk of AF incidence, showing that older patients (>75 years, regardless of comorbidities) have a higher incidence of AF than those with prior ischaemic stroke. Further studies are needed to evaluate the usefulness of algorithm-based risk stratification strategies for AF screening and the impact of screening on major cardiovascular event rates.

Design of Hardware Accelerators for Optimized and Quantized Neural Networks to Detect Atrial Fibrillation in Patch ECG Device with RISC-V

Atrial Fibrillation (AF) is one of the most common heart arrhythmias. It is known to cause up to 15% of all strokes. In current times, modern detection systems for arrhythmias, such as single-use patch electrocardiogram (ECG) devices, have to be energy efficient, small, and affordable. In this work, specialized hardware accelerators were developed. First, an artificial neural network (NN) for the detection of AF was optimized. Special attention was paid to the minimum requirements for the inference on a RISC-V-based microcontroller. Hence, a 32-bit floating-point-based NN was analyzed. To reduce the silicon area needed, the NN was quantized to an 8-bit fixed-point datatype (Q7). Based on this datatype, specialized accelerators were developed. Those accelerators included single-instruction multiple-data (SIMD) hardware as well as accelerators for activation functions such as sigmoid and hyperbolic tangents. To accelerate activation functions that require the e-function as part of their computation (e.g., softmax), an e-function accelerator was implemented in the hardware. To compensate for the losses of quantization, the network was expanded and optimized for run-time and memory requirements. The resulting NN has a 7.5% lower run-time in clock cycles (cc) without the accelerators and 2.2 percentage points (pp) lower accuracy compared to a floating-point-based net, while requiring 65% less memory. With the specialized accelerators, the inference run-time was lowered by 87.2% while the F1-Score decreased by 6.1 pp. Implementing the Q7 accelerators instead of the floating-point unit (FPU), the silicon area needed for the microcontroller in 180 nm-technology is below 1 mm2.

Embolic Stroke of Undetermined Source: Current Perspectives on Diagnosis, Investigations, and Management

In 2014, Hart et al. introduced the concept of "embolic stroke of undetermined source" (ESUS) to the clinical-research stroke community. The hypothesis underlying the development of the ESUS construct was that this potentially heterogenous group of stroke mechanisms were largely thromboembolic, and would thus benefit from anticoagulation over antiplatelet for secondary prevention. Since then, 2 large clinical trials have shown that, to date, there is not a clear uniform antithrombotic strategy for secondary prevention after ESUS as it was originally broadly defined. However, this work has yielded valuable information about the patient phenotypes that experience ESUS strokes, as well as hypothesis-generating substudies that have given rise to the next generation of secondary prevention trials aimed at more personalized approaches for different suspected mechanisms of embolic stroke. In parallel with the evolution of ESUS, several studies aimed at screening for atrial fibrillation in the secondary stroke prevention population have generated additional questions about the mechanistic relevance of atrial fibrillation detected after stroke, and how this should inform poststroke workup, and secondary prevention strategies. Herein, we provide a synthesis of the current understanding surrounding the patient phenotypes that experience ESUS strokes, and previous, ongoing, and anticipated clinical trials that will guide earlier and later secondary prevention strategies and poststroke cardiac investigations.