ConceFT: concentration of frequency and time via a multitapered synchrosqueezed transform.

A new method is proposed to determine the time-frequency content of time-dependent signals consisting of multiple oscillatory components, with time-varying amplitudes and instantaneous frequencies. Numerical experiments as well as a theoretical analysis are presented to assess its effectiveness.

Automated J wave detection from digital 12-lead electrocardiogram.

In this report we provide a method for automated detection of J wave, defined as a notch or slur in the descending slope of the terminal positive wave of the QRS complex, using signal processing and functional data analysis techniques. Two different sets of ECG tracings were selected from the EPICARE ECG core laboratory, Wake Forest School of Medicine, Winston Salem, NC. The first set was a training set comprised of 100 ECGs of which 50 ECGs had J-wave and the other 50 did not. The second set was a test set (n=116 ECGs) in which the J-wave status (present/absent) was only known by the ECG Center staff. All ECGs were recorded using GE MAC 1200 (GE Marquette, Milwaukee, Wisconsin) at 10mm/mV calibration, speed of 25mm/s and 500HZ sampling rate. All ECGs were initially inspected visually for technical errors and inadequate quality, and then automatically processed with the GE Marquette 12-SL program 2001 version (GE Marquette, Milwaukee, WI). We excluded ECG tracings with major abnormalities or rhythm disorder. Confirmation of the presence or absence of a J wave was done visually by the ECG Center staff and verified once again by three of the coauthors. There was no disagreement in the identification of the J wave state. The signal processing and functional data analysis techniques applied to the ECGs were conducted at Duke University and the University of Toronto. In the training set, the automated detection had sensitivity of 100% and specificity of 94%. For the test set, sensitivity was 89% and specificity was 86%. In conclusion, test results of the automated method we developed show a good J wave detection accuracy, suggesting possible utility of this approach for defining and detection of other complex ECG waveforms.

Deep learning from atrioventricular plane displacement in patients with Takotsubo syndrome: lighting up the black-box.

Aims: The spatiotemporal deep convolutional neural network (DCNN) helps reduce echocardiographic readers' erroneous 'judgement calls' on Takotsubo syndrome (TTS). The aim of this study was to improve the interpretability of the spatiotemporal DCNN to discover latent imaging features associated with causative TTS pathophysiology. Methods and results: We applied gradient-weighted class activation mapping analysis to visualize an established spatiotemporal DCNN based on the echocardiographic videos to differentiate TTS (150 patients) from anterior wall ST-segment elevation myocardial infarction (STEMI, 150 patients). Forty-eight human expert readers interpreted the same echocardiographic videos and prioritized the regions of interest on myocardium for the differentiation. Based on visualization results, we completed optical flow measurement, myocardial strain, and Doppler/tissue Doppler echocardiography studies to investigate regional myocardial temporal dynamics and diastology. While human readers' visualization predominantly focused on the apex of the heart in TTS patients, the DCNN temporal arm's saliency visualization was attentive on the base of the heart, particularly at the atrioventricular (AV) plane. Compared with STEMI patients, TTS patients consistently showed weaker peak longitudinal displacement (in pixels) in the basal inferoseptal (systolic: 2.15 ± 1.41 vs. 3.10 ± 1.66, P < 0.001; diastolic: 2.36 ± 1.71 vs. 2.97 ± 1.69, P = 0.004) and basal anterolateral (systolic: 2.70 ± 1.96 vs. 3.44 ± 2.13, P = 0.003; diastolic: 2.73 ± 1.70 vs. 3.45 ± 2.20, P = 0.002) segments, and worse longitudinal myocardial strain in the basal inferoseptal (-8.5 ± 3.8% vs. -9.9 ± 4.1%, P = 0.013) and basal anterolateral (-8.6 ± 4.2% vs. -10.4 ± 4.1%, P = 0.006) segments. Meanwhile, TTS patients showed worse diastolic mechanics than STEMI patients (E'/septal: 5.1 ± 1.2 cm/s vs. 6.3 ± 1.5 cm/s, P < 0.001; S'/septal: 5.8 ± 1.3 cm/s vs. 6.8 ± 1.4 cm/s, P < 0.001; E'/lateral: 6.0 ± 1.4 cm/s vs. 7.9 ± 1.6 cm/s, P < 0.001; S'/lateral: 6.3 ± 1.4 cm/s vs. 7.3 ± 1.5 cm/s, P < 0.001; E/E': 15.5 ± 5.6 vs. 12.5 ± 3.5, P < 0.001). Conclusion: The spatiotemporal DCNN saliency visualization helps identify the pattern of myocardial temporal dynamics and navigates the quantification of regional myocardial mechanics. Reduced AV plane displacement in TTS patients likely correlates with impaired diastolic mechanics.

Predicting adverse cardiac events in sarcoidosis: deep learning from automated characterization of regional myocardial remodeling.

Recognizing early cardiac sarcoidosis (CS) imaging phenotypes can help identify opportunities for effective treatment before irreversible myocardial pathology occurs. We aimed to characterize regional CS myocardial remodeling features correlating with future adverse cardiac events by coupling automated image processing and data analysis on cardiac magnetic resonance (CMR) imaging datasets. A deep convolutional neural network (DCNN) was used to process a CMR database of a 10-year cohort of 117 consecutive biopsy-proven sarcoidosis patients. The maximum relevance - minimum redundancy method was used to select the best subset of all the features-24 (from manual processing) and 232 (from automated processing) left ventricular (LV) structural/functional features. Three machine learning (ML) algorithms, logistic regression (LogR), support vector machine (SVM) and multi-layer neural networks (MLP), were used to build classifiers to categorize endpoints. Over a median follow-up of 41.8 (inter-quartile range 20.4-60.5) months, 35 sarcoidosis patients experienced a total of 43 cardiac events. After manual processing, LV ejection fraction (LVEF), late gadolinium enhancement, abnormal segmental wall motion, LV mass (LVM), LVMI index (LVMI), septal wall thickness, lateral wall thickness, relative wall thickness, and wall thickness of 9 (out of 17) individual LV segments were significantly different between patients with and without endpoints. After automated processing, LVEF, end-diastolic volume, end-systolic volume, LV mass and wall thickness of 92 (out of 216) individual LV segments were significantly different between patients with and without endpoints. To achieve the best predictive performance, ML algorithms selected lateral wall thickness, abnormal segmental wall motion, septal wall thickness, and increased wall thickness of 3 individual segments after manual image processing, and selected end-diastolic volume and 7 individual segments after automated image processing. LogR, SVM and MLP based on automated image processing consistently showed better predictive accuracies than those based on manual image processing. Automated image processing with a DCNN improves data resolution and regional CS myocardial remodeling pattern recognition, suggesting that a framework coupling automated image processing with data analysis can help clinical risk stratification.

Mortality Correlates in Patients With Takotsubo Syndrome During the COVID-19 Pandemic.

We completed a systematic review of Takotsubo syndrome (TTS) cases reported during the coronavirus disease 2019 (COVID-19) pandemic and performed clustering and feature importance analysis and statistical testing for independence on the demographic, clinical, and imaging parameters. Compared with the data before the COVID-19 pandemic, TTS was increasingly diagnosed in physical stress (mostly COVID-19 pneumonia)-triggered male patients without psychiatric/neurologic disorders, warranting further investigation to establish new reference criteria to improve diagnostic specificity. In clustering analysis, sex and inpatient mortality primarily contributed to the automated classification of the TTS. Both sex and inpatient mortality had essential correlations with COVID-19 infection/pneumonia. There is effect modification of sex on outcomes in patients with COVID-19 infection and TTS, with male patients having significantly worse inpatient mortality. Meanwhile, significantly more male patients with TTS were classified as high risk according to International Takotsubo Registry prognostic scores, suggesting that male COVID-19/TTS survivors will likely have worse long-term outcome.

Spatio-temporal hybrid neural networks reduce erroneous human "judgement calls" in the diagnosis of Takotsubo syndrome.

BACKGROUND: We investigate whether deep learning (DL) neural networks can reduce erroneous human "judgment calls" on bedside echocardiograms and help distinguish Takotsubo syndrome (TTS) from anterior wall ST segment elevation myocardial infarction (STEMI). METHODS: We developed a single-channel (DCNN[2D SCI]), a multi-channel (DCNN[2D MCI]), and a 3-dimensional (DCNN[2D+t]) deep convolution neural network, and a recurrent neural network (RNN) based on 17,280 still-frame images and 540 videos from 2-dimensional echocardiograms in 10 years (1 January 2008 to 1 January 2018) retrospective cohort in University of Iowa (UI) and eight other medical centers. Echocardiograms from 450 UI patients were randomly divided into training and testing sets for internal training, testing, and model construction. Echocardiograms of 90 patients from the other medical centers were used for external validation to evaluate the model generalizability. A total of 49 board-certified human readers performed human-side classification on the same echocardiography dataset to compare the diagnostic performance and help data visualization. FINDINGS: The DCNN (2D SCI), DCNN (2D MCI), DCNN(2D+t), and RNN models established based on UI dataset for TTS versus STEMI prediction showed mean diagnostic accuracy 73%, 75%, 80%, and 75% respectively, and mean diagnostic accuracy of 74%, 74%, 77%, and 73%, respectively, on the external validation. DCNN(2D+t) (area under the curve [AUC] 0·787 vs. 0·699, P = 0·015) and RNN models (AUC 0·774 vs. 0·699, P = 0·033) outperformed human readers in differentiating TTS and STEMI by reducing human erroneous judgement calls on TTS. INTERPRETATION: Spatio-temporal hybrid DL neural networks reduce erroneous human "judgement calls" in distinguishing TTS from anterior wall STEMI based on bedside echocardiographic videos. FUNDING: University of Iowa Obermann Center for Advanced Studies Interdisciplinary Research Grant, and Institute for Clinical and Translational Science Grant. National Institutes of Health Award (1R01EB025018-01).

Ventricular geometry-regularized QRSd predicts cardiac resynchronization therapy response: machine learning from crosstalk between electrocardiography and echocardiography.

Up to one-third of patients selected by current guidelines do not respond to cardiac resynchronization therapy (CRT), the aim of this study was to find out novel analytical approaches to improve pre-implantation CRT response prediction. Among 31 pre-implantation features of clinical, laboratory, electrocardiography (ECG), and echocardiography variables in a consecutive cohort of patients receiving a first-time CRT device (CRT-pacemaker or CRT-defibrillator), we developed a machine learning (ML) model with three classification algorithms (support vector machines (SVM), K nearest neighbors, and random subspaces) with the best features combination to predict CRT response. Three categorical variables, left bundle branch block (LBBB), nonischemic cardiomyopathy, and female gender, were independently associated with CRT responses. Among continuous variables, including septal wall thickness, posterior wall thickness, and relative wall thickness (RWT), could regularize ECG QRS duration (QRSd) and significantly enhance the correlation between QRSd and CRT response. The 3 ML algorithms in a total of 38 features combinations constantly recognized that the features combined with QRSd/RWT outperformed the combinations without it. For each of three algorithms, the triplet feature combination of QRSd/RWT, LBBB, and nonischemic cardiomyopathy repeatedly increased the classification rate more than 8%. The best performance for CRT response prediction occurred with SVM model, which proposed actual QRSd/RWT values that favored CRT responses in patients both with and without LBBB. Lower QRSd/RWT values were required for CRT responses in patients with ischemic cardiomyopathy compared to those with non-ischemic cardiomyopathy. ML from ventricular remodeling characteristics-regularized QRSd improves CRT response prediction.

Electrocardiographic J Wave and Cardiovascular Outcomes in the General Population (from the Atherosclerosis Risk In Communities Study).

The association between the J wave, a key component of the early repolarization pattern, and adverse cardiovascular outcomes remains unclear. Inconsistencies have stemmed from the different methods used to measure the J wave. We examined the association between the J wave, detected by an automated method, and adverse cardiovascular outcomes in 14,592 (mean age = 54 ± 5.8 years; 56% women; 26% black) participants from the Atherosclerosis Risk In Communities (ARIC) study. The J wave was detected at baseline (1987 to 1989) and during follow-up study visits (1990 to 1992, 1993 to 1995, and 1996 to 1998) using a fully automated method. Sudden cardiac death, coronary heart disease death, and cardiovascular mortality were ascertained from hospital discharge records, death certificates, and autopsy data through December 31, 2010. A total of 278 participants (1.9%) had evidence of a J wave. Over a median follow-up of 22 years, 4,376 of the participants (30%) died. In a multivariable Cox regression analysis adjusted for demographics, cardiovascular risk factors, and potential confounders, the J wave was not associated with an increased risk of sudden cardiac death (hazard ratio [HR] 0.74, 95% CI 0.36 to 1.50), coronary heart disease death (HR 0.72, 95% CI 0.40 to 1.32), or cardiovascular mortality (HR 1.16, 95% CI 0.87 to 1.56). An interaction was detected for cardiovascular mortality by gender with men (HR 1.54, 95% CI 1.09 to 2.19) having a stronger association than women (HR 0.74, 95% CI 0.43 to 1.25; P-interaction = 0.030). In conclusion, our findings suggest that the J wave is a benign entity that is not associated with an increased risk for sudden cardiac arrest in middle-aged adults in the United States.