Arrhythmia Database Research Articles

Towards Reliable ECG Analysis: Addressing Validation Gaps in the Electrocardiographic R-Peak Detection

Electrocardiographic (ECG) R-peak detection is essential for every sensor-based cardiovascular health monitoring system. To validate R-peak detectors, comparing the predicted results with reference annotations is crucial. This comparison is typically performed using tools provided by the waveform database (WFDB) or custom methods. However, many studies fail to provide detailed information on the validation process. The literature also highlights inconsistencies in reporting window size, a crucial parameter used to compare predictions with expert annotations to distinguish false peaks from the true R-peak. Additionally, there is also a need for uniformity in reporting the total number of beats for individual or collective records of the widely used MIT-BIH arrhythmia database. Thus, we aim to review validation methods of various R-peak detection methodologies before their implementation in real time. This review discusses the impact of non-beat annotations when using a custom validation method, allowable window tolerance, the effects of window size deviations, and implications of varying numbers of beats and skipping segments on ECG testing, providing a comprehensive guide for researchers. Addressing these validation gaps is critical as they can significantly affect validatory outcomes. Finally, the conclusion section proposes a structured concept as a future approach, a guide to integrate WFDB R-peak validation tools for testing any QRS annotated ECG database. Overall, this review underscores the importance of complete transparency in reporting testing procedures, which prevents misleading assessments of R-peak detection algorithms and enables fair methodological comparison.

A multi-adaptive neuro-fuzzy inference system with variable thresholds for heartbeat classification

Various heart disorders are non-invasively diagnosed using electrocardiograms (ECGs). An ECG records a variety of waveforms, including P, QRS, and T waves, which represent the electrical activity of the human heart. Cardiovascular diseases are diagnosed by examining the length, form, and spacing of these waveforms. This research develops a multi-adaptive, neuro-fuzzy inference system (MANFIS), enhanced by a variable threshold approach, to enhance heartbeat classification accuracy. The MIT-BIH arrhythmia database was utilized, and seven features were extracted from each record. A subtractive clustering method was employed to prepare the inputs for the MANFIS, enabling heartbeat classification. By applying a variable threshold to the MANFIS outputs, classification accuracy was further enhanced. The proposed method, termed variable-threshold MANFIS, can separately detect normal sinus rhythm, left bundle branch block, right bundle branch block, premature ventricular contraction, atrial premature condition, and paced beat. This is achieved using six different ANFIS classifiers, each with its own threshold. The system was evaluated, achieving an accuracy of 98.33%, a sensitivity of 93.12%, a specificity of 99.66%, a precision of 98.33, and an F1-score of 95.44. A distinct feature of this machine-learning-based model is its controllable threshold, which delivers promising results across all training, testing, and validation datasets. The proposed diagnostic system is applicable in new automated medical instrumentation and serves as a valuable tool in cardiology.

A hybrid deep learning network for automatic diagnosis of cardiac arrhythmia based on 12-lead ECG

Cardiac arrhythmias are the leading cause of death and pose a huge health and economic burden globally. Electrocardiography (ECG) is an effective technique for the diagnosis of cardiovascular diseases because of its noninvasive and cost-effective advantages. However, traditional ECG analysis relies heavily on the clinical experience of physicians, which can be challenging and time-consuming to produce valid diagnostic results. This work proposes a new hybrid deep learning model that combines convolutional neural network (CNN) and bidirectional gated recurrent unit (BiGRU) with multi-head attention (CBGM model). Specifically, the model consists of seven convolutional layers with varying filter sizes (4, 16, 32, and 64) and three pooling layers, respectively, while the BiGRU module includes two layers with 64 units each followed by multi-head attention (8-heads). The combination of CNN and BiGRU effectively captures spatio-temporal features of ECG signals, with multi-head attention comprehensively extracted global correlations among multiple segments of ECG signals. The validation in the MIT-BIH arrhythmia database achieved an accuracy of 99.41%, a precision of 99.15%, a specificity of 99.68%, and an F1-Score of 99.21%, indicating its robust performance across different evaluation metrics. Additionally, the model’s performance was evaluated on the PTB Diagnostic ECG Database, where it achieved an accuracy of 98.82%, demonstrating its generalization capability. Comparative analysis against previous methods revealed that our proposed CBGM model exhibits more higher performance in automatic classification of arrhythmia and can be helpful for assisting clinicians by enabling real-time detection of cardiac arrhythmias during routine ECG screenings.

ElectrApp: Deep learning framework to enhance cardiac arrhythmia detection

Abstract Medical specialists who diagnose patients with cardiovascular arrhythmias are not promptly available in rural areas. The proposed information system provides access to upload the data to a cloud-based processing module in which different arrhythmias are classified, allowing the classification of various arrhythmias to solve the scarcity of physicians in remote areas through an automatic system that helps with the diagnosis. This information system has a controller module for the operation that manages data collection, processing, beautification and visualization tasks. We used the MIT-BIH Arrhythmia Database during the training and validation of the classification module. The module uses deep learning to identify five different types of heartbeats according to standards, reporting 98.9% accuracy. The information system will improve health coverage in the prediagnosis of cardiovascular diseases.

Signal Quality Index Estimator for Complex-Lead type ECG beat detectors

Abstract The Electro Cardio Gram is a very valuable clinical tool to access the electric function of the heart. It provides insight into the different phases of the heart beat and various kinds of disorders which may affect them. In the past decades various algorithms to detect and delineate ECG beats have been developed. The complex-lead based algorithms compute a virtual-lead by averaging the samples of an arbitrary number of channels. In this study exponential weighted moving powers based estimates for the skewness, kurtosis and signal to noise ratio are used to select the channels which are included in the resulting complex lead signal. According to these parameters the channels are labelled bad, poor, scaled and good. A weighted average is used to to compute the next sample of the complex lead from the channels labelled good and scaled. The 75 30minutes records of the ”St.-Petersburg Institute of Cardiological Technics 12-lead Arrhythmia Database” are used to evaluate the proposed approach. The average sensitivity increases to 99.61% and to 99.63% in total compared to 99.57% and 99.59% for the standard approach. The average and overall specificity increases to 99.65% and the F1 and F-M scores to 0.9964. This shows that the estimates provide sufficient information about the quality, to successfully reduce the distortion of the complex-lead signal by artefacts and strong muscle signals while avoiding the additional time delay imposed by standard window based methods.

An Arrhythmia Classification Model Based on a CNN-LSTM-SE Algorithm

Arrhythmia is the main cause of sudden cardiac death, and ECG signal analysis is a common method for the noninvasive diagnosis of arrhythmia. In this paper, we propose an arrhythmia classification model based on the combination of a channel attention mechanism (SE module), convolutional neural network (CNN), and long short-term memory neural network (LSTM). The data of this model use the MIT-BIH arrhythmia database, and after noise reduction of raw ECG data by the EEMD denoising algorithm, a CNN-LSTM is used to learn features from the data, and the fusion channel attention mechanism is used to adjust the weight of the feature map. The CNN-LSTM-SE model is compared with the LSTM, CNN-LSTM, and LSTM-attention models, and the models are evaluated using Precision, Recall, and F1-Score. The classification performance of the tested CNN-LSTM-SE classification prediction model is better, with a classification accuracy of 98.5%, a classification precision rate of more than 97% for each label, a recall rate of more than 98%, and an F1-score of more than 0.98. It meets the requirements of arrhythmia classification prediction and has a certain practical value.

AOCBLS: A novel active and online learning system for ECG arrhythmia classification with less labeled samples

Electrocardiogram (ECG) is a pivotal determinant of cardiac arrhythmia. In practice, ECG data are often acquired as continuous unlabeled chunks with high labeling costs and severe class imbalances. This poses formidable challenges to existing offline learning methods. To address these issues, we propose a novel model called active online class-specific broad learning system (AOCBLS) to learn streaming imbalanced ECG and classify arrhythmias. First, a two-stage active learning strategy combining diversity and uncertainty queries was designed to select the most valuable beats from the current chunk for further labeling and learning, thereby reducing labeling costs. Subsequently, a stable training weight update rule was developed to address the dynamic imbalances in the streaming data. Finally, an incremental learning algorithm was proposed to update network weights online using selected beats, eliminating the need for retraining from scratch. Thus, the AOCBLS can effectively learn online from imbalanced data streams with less label information. Intra-patient experiments on the benchmark MIT-BIH arrhythmia database showed our model achieved an overall accuracy of 99.10% and a G-mean of 96.35% in identifying the four AAMI-recommended beat types, with only 10% additional annotation, outperforming advanced methods at lower labeling cost. Additionally, patient-specific experimental results on the benchmark database and two other databases, INCARTDB and SVDB, suggest the potential of AOCBLS for personalized ECG monitoring on portable devices.

Ecg signal watermarking using QR decomposition.

This study introduces a novel watermarking technique for electrocardiogram (ECG) signals. Watermarking embeds critical information within the ECG signal, enabling data origin authentication, ownership verification, and ensuring the integrity of research data in domains like telemedicine, medical databases, insurance, and legal proceedings. Drawing inspiration from image watermarking, the proposed method transforms the ECG signal into a two-dimensional format for QR decomposition. The watermark is then embedded within the first row of the resulting R matrix. Three implementation scenarios are proposed: one in the spatial domain and two in the transform domain utilizing discrete wavelet transform (DWT) for improved watermark imperceptibility. Evaluation on real ECG signals from MIT-BIH Arrhythmia database and comparison to existing methods demonstrate that the proposed method achieves: (1) higher Peak Signal-to-Noise Ratio (PSNR) indicating minimal alterations to the watermarked signal, (2) lower bit error rates (BER) in robustness tests against external modifications such as AWGN noise (additive white Gaussian noise), line noise and down-sampling, and (3) lower computational complexity. These findings emphasize the effectiveness of the proposed QR decomposition-based watermarking method, achieving a balance between robustness and imperceptibility. The proposed approach has the potential to improve the security and authenticity of ECG data in healthcare and legal contexts, while its lower computational complexity enhances its practical applicability.

Fractional calculus integration for improved ECG modeling: A McSharry model expansion

This study introduces a new method for modeling electrocardiogram (ECG)11- CR: Compression Ratio; FDE: Fractional Differential Equation; GA: Genetic Algorithm; IDE: Integer Differential Equation; MAE: Mean Absolute Error; MSE: Mean Square Error; NRMSE: Normalized Root Mean Square Error; PC: Predictor Corrector; PRD: Percentage Root Mean Square Difference; QS: Quality Score; RMSE: Root Mean Square Error waveforms using Fractional Differential Equations (FDEs). By incorporating fractional calculus into the well-established McSharry model, the proposed approach achieves improved representation and high precision for a wide range of ECG waveforms. The research focuses on the impact of integrating fractional derivatives into Integer Differential Equation (IDE) models, enhancing the fidelity of ECG signal modeling.To optimize the model's unknown parameters, a combination of the Predictor-Corrector method for solving FDEs and genetic algorithms for optimization is utilized. The effectiveness of the fractional-order model is assessed through distortion metrics, providing a comprehensive evaluation of the modeling quality.Comparisons show that the fractional-order model outperforms the traditional McSharry IDE model in modeling quality and compression efficiency. It improves modeling quality by 48.40 % in MSE and compression efficiency by 23.18 % when applied on five beat types of MIT/BIH arrhythmia database. The fractional-order model demonstrates enhanced flexibility while preserving essential McSharry model characteristics, with fractional orders (α) ranging from 0.96 to 0.99 across five beat types.

LDCNN: A new arrhythmia detection technique with ECG signals using a linear deep convolutional neural network.

The electrocardiogram (ECG) is a fundamental and widely used tool for diagnosing cardiovascular diseases. It involves recording cardiac electrical signals using electrodes, which illustrate the functioning of cardiac muscles during contraction and relaxation phases. ECG is instrumental in identifying abnormal cardiac activity, heart attacks, and various cardiac conditions. Arrhythmia detection, a critical aspect of ECG analysis, entails accurately classifying heartbeats. However, ECG signal analysis demands a high level of expertise, introducing the possibility of human errors in interpretation. Hence, there is a clear need for robust automated detection techniques. Recently, numerous methods have emerged for arrhythmia detection from ECG signals. In our research, we developed a novel one-dimensional deep neural network technique called linear deep convolutional neural network (LDCNN) to identify arrhythmias from ECG signals. We compare our suggested method with several state-of-the-art algorithms for arrhythmia detection. We evaluate our methodology using benchmark datasets, including the PTB Diagnostic ECG and MIT-BIH Arrhythmia databases. Our proposed method achieves high accuracy rates of 99.24% on the PTB Diagnostic ECG dataset and 99.38% on the MIT-BIH Arrhythmia dataset.

ECG classification via integration of adaptive beat segmentation and relative heart rate with deep learning networks

We propose a state-of-the-art deep learning approach for accurate electrocardiogram (ECG) signal analysis, addressing both waveform delineation and beat type classification tasks. For beat type classification, we integrated two novel schemes into the deep learning model, significantly enhancing its performance. The first scheme is an adaptive beat segmentation method that determines the optimal duration for each heartbeat based on RR-intervals, mitigating segmenting errors from conventional fixed-period segmentation. The second scheme incorporates relative heart rate information of the target beat compared to neighboring beats, improving the model’s ability to accurately detect premature atrial contractions (PACs) that are easily confused with normal beats due to similar morphology. Extensive evaluations on the PhysioNet QT Database, MIT-BIH Arrhythmia Database, and real-world wearable device data demonstrated the proposed approach’s superior capabilities over existing methods in both tasks. The proposed approach achieved sensitivities of 99.81% for normal beats, 99.08% for premature ventricular contractions, and 97.83% for PACs in beat type classification. For waveform delineation, we achieved F1-scores of 0.9842 for non-waveform, 0.9798 for P-waves, 0.9749 for QRS complexes, and 0.9848 for T-waves. It significantly outperforms existing methods in PAC detection while maintaining high performance across both tasks. The integration of aforementioned two schemes into the deep learning model improved the accuracy of normal sinus rhythms and arrhythmia detection.

A rate-responsive duty-cycling protocol for leadless pacemaker synchronization

Dual-chamber leadless pacemakers (LLPMs) consist of two implants, one in the right atrium and one in the right ventricle. Inter-device communication, required for atrioventricular (AV) synchrony, however, reduces the projected longevity of commercial dual-chamber LLPMs by 35–45%. This work analyzes the power-saving potential and the resulting impact on AV-synchrony for a novel LLPM synchronization protocol. Relevant parameters of the proposed window scheduling algorithm were optimized with system-level simulations investigating the resulting trade-off between transceiver current consumption and AV-synchrony. The parameter set included the algorithm’s setpoint for the target number of windows per cardiac cycle and the number of averaging cycles used in the window update calculation. The sensing inputs for the LLPM model were derived from human electrocardiogram recordings in the MIT-BIH Arrhythmia Database. Transceiver current consumption was estimated by combining the simulation results on the required communication resources with electrical measurements of a receiver microchip developed for LLPM synchronization in previous work. The performance ratio given by AV-synchrony divided by current consumption was maximized for a target of one window per cardiac cycle and three averaging cycles. Median transceiver current of both LLPMs combined was 166 nA (interquartile range: 152–183 nA) and median AV-synchrony was 92.5%. This corresponded to median reduction of 18.3% and 3.2% in current consumption and AV-synchrony, respectively, compared to a non-rate-responsive implementation of the same protocol, which prioritized maximum AV-synchrony. In conclusion, adopting a rate-responsive communication protocol may significantly increase device longevity of dual-chamber LLPMs without compromising AV-synchrony, potentially reducing the frequency of device replacements.

Enhancing Inter-Patient Performance for Arrhythmia Classification with Adversarial Learning Using Beat-Score Maps

Research on computer-aided arrhythmia classification is actively conducted, but the limited generalization capacity constrains its applicability in practical clinical settings. One of the primary challenges in deploying such techniques in real-world scenarios is the inter-patient variability and the consequent performance degradation. In this study, we leverage our previous innovation, the n-beat-score map (n-BSM), to introduce an adversarial framework to mitigate the issue of poor performance in arrhythmia classification within the inter-patient paradigm. The n-BSM is a 2D representation of the ECG signal, capturing its constituent beat characteristics through beat-score vectors derived from a pre-trained beat classifier. We employ adversarial learning to eliminate patient-dependent features during the training of the beat classifier, thereby generating the patient-independent n-BSM (PI-BSM). This approach enables us to concentrate primarily on the learning characteristics associated with beat type rather than patient-specific features. Through a beat classifier pre-trained with adversarial learning, a series of beat-score vectors are generated for the beat segments that make up a given ECG signal. These vectors are then concatenated chronologically to form a PI-BSM. Utilizing PI-BSMs as the input, an arrhythmia classifier is trained to differentiate between distinct types of rhythms. This approach yields a 14.27% enhancement in the F1-score in the MIT-BIH arrhythmia database and a 4.97% improvement in cross-database evaluation using the Chapman–Shaoxing 12-lead ECG database.

Development and Validation of a Real-Time Service Model for Noise Removal and Arrhythmia Classification Using Electrocardiogram Signals.

Arrhythmias range from mild nuisances to potentially fatal conditions, detectable through electrocardiograms (ECGs). With advancements in wearable technology, ECGs can now be monitored on-the-go, although these devices often capture noisy data, complicating accurate arrhythmia detection. This study aims to create a new deep learning model that utilizes generative adversarial networks (GANs) for effective noise removal and ResNet for precise arrhythmia classification from wearable ECG data. We developed a deep learning model that cleans ECG measurements from wearable devices and detects arrhythmias using refined data. We pretrained our model using the MIT-BIH Arrhythmia and Noise databases. Least squares GANs were used for noise reduction, maintaining the integrity of the original ECG signal, while a residual network classified the type of arrhythmia. After initial training, we applied transfer learning with actual ECG data. Our noise removal model significantly enhanced data clarity, achieving over 30 dB in a signal-to-noise ratio. The arrhythmia detection model was highly accurate, with an F1-score of 99.10% for noise-free data. The developed model is capable of real-time, accurate arrhythmia detection using wearable ECG devices, allowing for immediate patient notification and facilitating timely medical response.

Improved bidirectional long short term memory-based QRS complex detection using autoencoder

In this paper, we propose a new technique to improve QRS complex detection. This technique consists of incorporating an autoencoder and bidirectional long short term memory (BiLSTM). The autoencoder used is a stacked autoencoder and functions as signal filtering. Meanwhile, BiLSTM is used as a detector. Exploration of the effect of hyperparameter in the autoencoder was also carried out to determine the effect on QRS complex detection. Furthermore, the dataset used in this study is the MIT-BIH arrhythmia database. Based on the experimental results, the hyperparameter in the autoencoder that gives a better effect on QRS complex detection is 16-8. Finally, the proposed method out-of-perform state of the art algorithm with accuracy 99.94%.

Detection of atrial fibrillation using a nonlinear Lorenz Scattergram and deep learning in primary care

BackgroundAtrial fibrillation (AF) is highly correlated with heart failure, stroke and death. Screening increases AF detection and facilitates the early adoption of comprehensive intervention. Long-term wearable devices have become increasingly popular for AF screening in primary care. However, interpreting data obtained by long-term wearable ECG devices is a problem in primary care. To diagnose the disease quickly and accurately, we aimed to build AF episode detection model based on a nonlinear Lorenz scattergram (LS) and deep learning.MethodsThe MIT-BIH Normal Sinus Rhythm Database, MIT-BIH Arrhythmia Database and the Long-Term AF Database were extracted to construct the MIT-BIH Ambulatory Electrocardiograph (MIT-BIH AE) dataset. We converted the long-term ECG into a two-dimensional LSs. The LSs from MIT-BIH AE dataset was randomly divided into training and internal validation sets in a 9:1 ratio, which was used to develop and internally validated model. We built a MOBILE-SCREEN-AF (MS-AF) dataset from a single-lead wearable ECG device in primary care for external validation. Performance was quantified using a confusion matrix and standard classification metrics.ResultsDuring the evaluation of model performance based on the LS, the sensitivity, specificity and accuracy of the model in diagnosing AF were 0.992, 0.973, and 0.983 in the internal validation set respectively. In the external validation set, these metrics were 0.989, 0.956, and 0.967, respectively. Furthermore, when evaluating the model’s performance based on ECG records in the MS-AF dataset, the sensitivity, specificity and accuracy of model diagnosis paroxysmal AF were 1.000, 0.870 and 0.876 respectively, and 0.927, 1.000 and 0.973 for the persistent AF.ConclusionsThe model based on the nonlinear LS and deep learning has high accuracy, making it promising for AF screening in primary care. It has potential for generalization and practical application.

A High-Performance Anti-Noise Algorithm for Arrhythmia Recognition.

In recent years, the incidence of cardiac arrhythmias has been on the rise because of changes in lifestyle and the aging population. Electrocardiograms (ECGs) are widely used for the automated diagnosis of cardiac arrhythmias. However, existing models possess poor noise robustness and complex structures, limiting their effectiveness. To solve these problems, this paper proposes an arrhythmia recognition system with excellent anti-noise performance: a convolutionally optimized broad learning system (COBLS). In the proposed COBLS method, the signal is convolved with blind source separation using a signal analysis method based on high-order-statistic independent component analysis (ICA). The constructed feature matrix is further feature-extracted and dimensionally reduced using principal component analysis (PCA), which reveals the essence of the signal. The linear feature correlation between the data can be effectively reduced, and redundant attributes can be eliminated to obtain a low-dimensional feature matrix that retains the essential features of the classification model. Then, arrhythmia recognition is realized by combining this matrix with the broad learning system (BLS). Subsequently, the model was evaluated using the MIT-BIH arrhythmia database and the MIT-BIH noise stress test database. The outcomes of the experiments demonstrate exceptional performance, with impressive achievements in terms of the overall accuracy, overall precision, overall sensitivity, and overall F1-score. Specifically, the results indicate outstanding performance, with figures reaching 99.11% for the overall accuracy, 96.95% for the overall precision, 89.71% for the overall sensitivity, and 93.01% for the overall F1-score across all four classification experiments. The model proposed in this paper shows excellent performance, with 24 dB, 18 dB, and 12 dB signal-to-noise ratios.

Weighted Dual-Channel Model for Arrhythmia Detection using Single Electrocardiogram Signal

The single-lead method plays a central role in arrhythmia warning, and its anomaly detection is of practical significance for the prevention of heart diseases. However, the detection performance is often limited due to the limited information content of single-lead electrocardiogram (ECG) signals. To address this issue, this paper proposes a weighted dualchannel arrhythmia classification model aimed at improving the detection performance of single-lead ECG signals. The model employs a parallel structure of CNN-LSTM and Transformer to capture both local and global features of the ECG signals. By applying weighted learning to the features from the two channels, the model achieves precise arrhythmia classification. Additionally, to alleviate the class imbalance problem, a weighted cross-entropy loss function is introduced, further enhancing the classification performance. Experimental results based on the MIT-BIH arrhythmia database demonstrate that the proposed model exhibits excellent performance in terms of overall accuracy, F1 score, and sensitivity, achieving 99.33%, 94.71%, and 99.33%, respectively. Compared with existing classification models, the proposed model shows superior classification performance and significant practical value, providing solid technical support for the auxiliary diagnosis of arrhythmias and bringing new possibilities for actual clinical applications.

FPGA implementation of IIR elliptic filters for de-noising ECG signal

De-noising of ECG signal is very much necessary to monitor the heart health and to diagnose disease. This paper presents a real time de-noising system that efficiently wiped out the noises from corrupted ECG signal and improves its SNR which enhances the interpretation of features present in ECG signal. This proposed system is designed with three Infinite Impulse Response (IIR) elliptic filters in cascaded direct form II structure and is implemented on the FPGA platform. Total 376 ECG samples are taken from the open access public database physiobank. All the architectural level FPGA designs have been developed in Xilinx System Generator (XSG) and implemented on Xilinx Zynq 7000 board. Our design utilized only 3.87 % of the total available resource and consumed 0.321 watt on-chip power. Worst Negative Slack (WNS) for the critical path is 2.101 ns. The positive value of WNS signifies our hardware design implementation run successfully by meeting the required time constraint. To judge the performance of the proposed design, the measurement matrices SNR Improvement, RMSE, PRD, SSNR and SINAD are calculated and achieved 81.11% , 88.89% and 85.41% accuracy for ECG recordings of ECG-ID, MIT-BIH Normal Sinus Rhythm and MIT-BIH Arrhythmia database respectively. The novelty of this proposed denoising system is its robustness, which work satisfactorily for Baseline Wander (BW), Electrode Motion (EM), Power Line Interference (PLI), Muscle or Electromyography (EMG) and as well as Additive White Gaussian Noise (AWGN).

A 36-nW Electrocardiogram Anomaly Detector Based on a 1.5-bit Non-Feedback Delta Quantizer for Always-on Cardiac Monitoring.

An always-on electrocardiogram (ECG) anomaly detector (EAD) with ultra-low power (ULP) consumption is proposed for continuous cardiac monitoring applications. The detector is featured with a 1.5-bit non-feedback delta quantizer (DQ) based feature extractor, followed by a multiplier-less convolutional neural network (CNN) engine, which eliminates the traditional high-resolution analog-to-digital converter (ADC) in conventional signal processing systems. The DQ uses a computing-in-capacitor (CIC) subtractor to quantize the sample-to-sample difference of ECG signal into 1.5-bit ternary codes, which is insensitive to low-frequency baseline wandering. The subsequent event-driven classifier is composed of a low-complexity coarse detector and a systolic-array-based CNN engine for ECG anomaly detection. The DQ and the digital CNN are fabricated in 65-nm and 180-nm CMOS technology, respectively, and the two chips are integrated on board through wire bonding. The measured detection accuracy is 90.6% ∼ 91.3% when tested on the MIT-BIH arrhythmia database, identifying three different ECG anomalies. Operating at 1 V and 1.4 V power supplies for the DQ and the digital CNN, respectively, the measured long-term average power consumption of the core circuits is 36 nW, which makes the detector among those state-of-the-art always-on cardiac anomaly detection devices with the lowest power consumption.