Improving generalization performance of electrocardiogram classification models.

Objective.Recently, many electrocardiogram (ECG) classification algorithms using deep learning have been proposed. Because the ECG characteristics vary across datasets owing to variations in factors such as recorded hospitals and the race of participants, the model needs to have a consistently high generalization performance across datasets. In this study, as part of the PhysioNet/Computing in Cardiology Challenge (PhysioNet Challenge) 2021, we present a model to classify cardiac abnormalities from the 12- and the reduced-lead ECGs.Approach.To improve the generalization performance of our earlier proposed model, we adopted a practical suite of techniques, i.e. constant-weighted cross-entropy loss, additional features, mixup augmentation, squeeze/excitation block, and OneCycle learning rate scheduler. We evaluated its generalization performance using the leave-one-dataset-out cross-validation setting. Furthermore, we demonstrate that the knowledge distillation from the 12-lead and large-teacher models improved the performance of the reduced-lead and small-student models.Main results.With the proposed model, our DSAIL SNU team has received Challenge scores of 0.55, 0.58, 0.58, 0.57, and 0.57 (ranked 2nd, 1st, 1st, 2nd, and 2nd of 39 teams) for the 12-, 6-, 4-, 3-, and 2-lead versions of the hidden test set, respectively.Significance.The proposed model achieved a higher generalization performance over six different hidden test datasets than the one we submitted to the PhysioNet Challenge 2020.

Feeding strategy for single-stage deammonification to treat moderate-strength ammonium under low free ammonia conditions.

Single-stage deammonification (SSD) processes have been successfully operated using the step-feeding strategy to treat high-strength NH4+ (>300 mg/L), but often failed to treat moderate-strength NH4+ (100-300 mg/L). Because it is hard to maintain the free ammonia (FA) above 1 mg/L, which is a concentration in which the activity of NO2- oxidizing bacteria (NOB) can be selectively suppressed. In this study, to evaluate the effectiveness of the step-feeding strategy on the long-term stability of treating moderate-strength NH4+, two SSD sequential-batch reactors (SBRs) were operated under one-step feeding and multi-step feeding strategies. The one-step feeding SBR achieved a higher nitrogen removal efficiency (86 %), nitrogen removal rate (0.61 kg/m3/d), and COD removal efficiency (95 %) than the multi-step feeding SBR (73 %, 0.39 kg/m3/d, and 95 %, respectively). This means the appropriate FA to selectively suppress NOB activity was successfully maintained in the one-step feeding SBR (FA > 1 mg/L). Therefore, it the necessary to apply a step feed strategy that can be maintained above FA (1 mg/L) from the start-up of operation to treat moderate-strength NH4+.

CNN-based Two Step R Peak Detection Method: Combining Segmentation and Regression.

For semantic segmentation, U-Net provides an end-to-end trainable framework to detect multiple class objects from background. Due to its great achievements in computer vision tasks, U-Net has broadened its application to biomedical signal processing, especially, segmentation of waveforms in ECG signal. Despite its superior performance for QRS complex detection to other traditional signal processing methods, direct application of the U-Net to R peak detection has limitation since the U-Net structures tend to predict high probability around true peak. Such multiple detection results require additional process to determine a unique peak location in each QRS complex. In this study, we use a regression process to detect R peak instead of pixel-wise classification. Such regression process guarantees a unique peak location prediction. We collect data from resting ECG systems and wearable ECG devices as well as public ECG databases and the proposed model is trained on various combinations of the data sources. Especially, we investigate the robustness of the model for input data from the wearable devices when the model is trained by data from heterogeneous devices.

Origin of macroscopic adhesion in organic light-emitting diodes analyzed at different length scales.

Organic light-emitting diodes (OLEDs) have been widely studied because of their various advantages. OLEDs are multi-layered structures consisting of organic and inorganic materials arranged in a heterojunction; the nature of adhesion at their heterogeneous interfaces has a significant effect on their properties. In this study, the origin of macroscopic adhesion was explored in OLEDs using a combination of microscopy techniques applied at different length scales. The different techniques allowed the identification of layers exposed by a peel test, which aided direct characterization of their macroscopic adhesion. Further, the contribution of each exposed layer to macroscopic adhesion could be determined through an analysis of photographic images. Finally, analysis of the local roughness and adhesion confirmed that the interface between an anode and emission layer could play a predominant role in determining the nature of macroscopic adhesion in OLEDs. These results may provide guidelines for exploring the origin of macroscopic adhesion properties through a combination of various microscopy techniques.

Evenly transferred single-layered graphene membrane assisted by strong substrate adhesion.

We explored the transfer of a single-layered graphene membrane assisted by substrate adhesion. A relatively larger adhesion force was measured on the SiO2 substrate compared with its van der Waals contribution, which is expected to result from the additional contribution of the chemical bonding force. Density functional theory calculations verified that the strong adhesion force was indeed accompanied by chemical bonding. The transfer of single-layered graphene and subsequent deposition of the dielectric layer were best performed on the SiO2 substrate exhibiting a larger adhesion force. This study suggests the selection and/or modification of the underlying substrate for proper transfer of graphene as well as other 2D materials similar to graphene.

Determination of ferroelectric contributions to electromechanical response by frequency dependent piezoresponse force microscopy.

Hysteresis loop analysis via piezoresponse force microscopy (PFM) is typically performed to probe the existence of ferroelectricity at the nanoscale. However, such an approach is rather complex in accurately determining the pure contribution of ferroelectricity to the PFM. Here, we suggest a facile method to discriminate the ferroelectric effect from the electromechanical (EM) response through the use of frequency dependent ac amplitude sweep with combination of hysteresis loops in PFM. Our combined study through experimental and theoretical approaches verifies that this method can be used as a new tool to differentiate the ferroelectric effect from the other factors that contribute to the EM response.

Probing of multiple magnetic responses in magnetic inductors using atomic force microscopy.

Even though nanoscale analysis of magnetic properties is of significant interest, probing methods are relatively less developed compared to the significance of the technique, which has multiple potential applications. Here, we demonstrate an approach for probing various magnetic properties associated with eddy current, coil current and magnetic domains in magnetic inductors using multidimensional magnetic force microscopy (MMFM). The MMFM images provide combined magnetic responses from the three different origins, however, each contribution to the MMFM response can be differentiated through analysis based on the bias dependence of the response. In particular, the bias dependent MMFM images show locally different eddy current behavior with values dependent on the type of materials that comprise the MI. This approach for probing magnetic responses can be further extended to the analysis of local physical features.

Comparison of gadoxetic acid-enhanced MR imaging versus four-phase multi-detector row computed tomography in assessing tumor regression after radiofrequency ablation in subjects with hepatocellular carcinomas.

PURPOSE: To assess the diagnostic value of gadoxetic acid-enhanced magnetic resonance (MR) imaging in follow-up of patients with hepatocellular carcinomas (HCCs) who were treated with radiofrequency (RF) ablation and to compare it with that of four-phase multi-detector row computed tomography (CT). MATERIALS AND METHODS: From July 2007 to May 2008, 36 patients (43 HCCs) were enrolled who were treated with RF ablation (tumor size, 20-47 mm; mean, 24.5 mm) and underwent gadoxetic acid-enhanced MR imaging and four-phase (precontrast, arterial, portal venous, and equilibrium) multidetector CT for follow-up. Two radiologists independently reviewed these images, and conspicuity of tumor margins and detection of residual or recurrent tumor were assessed on a five-point scale with receiver operating characteristic (ROC) curve analysis. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were evaluated. RESULTS: The mean conspicuity value of tumor margins was significantly higher on MR imaging than on multidetector CT (P < .001). The degree of differentiation between residual/recurrent tumor and hyperemia was significantly greater on MR imaging (P < .001). The mean area under the ROC curve was significantly higher with MR imaging (P = .015), as were sensitivity, specificity, PPV, NPV, and accuracy of detection rate (mean, 100%, 96.2%, 82.4%, 100%, and 96.7%, respectively, vs 41.7%, 56.8%, 13.5%, 85.7%, and 54.7% for multidetector CT). The interobserver agreement rate for MR imaging was higher (0.919) than for multidetector CT (0.672; P < .05). CONCLUSIONS: Diagnostic accuracy, conspicuity of tumor margins, and detection rate of residual or recurrent tumor were found to be better with gadoxetic acid-enhanced MR imaging than with four-phase multidetector CT.