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.

Lightweight Convolutional Neural Network for Real-Time Arrhythmia Classification on Low-Power Wearable Electrocardiograph.

In this study, a lightweight CNN-based electrocardiogram (ECG) classification model is implemented to operate it on a wearable device for real-time arrhythmia detection by efficiently reducing the number of parameters of the model. Ten second-windowed ECGs from three different public ECG databases were used to learn and classify them into four classes: normal sinus rhythm, atrial fibrillation, atrial premature contraction, and ventricular premature contraction. The model implemented in the workstation environment was converted using the TensorFlow Lite framework and then imported into an ARM Cortex-M4 architecture-based nRF52840 microprocessor. The proposed model shows high performance (97.7% accuracy and 97.4% F1 score) with reasonable execution time: 298ms and current consumption: 3.55mA at optimized for speed and execution time: 480ms and current consumption: 3.82mA at optimized for size, respectively.

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.

Compressed Deep Learning to Classify Arrhythmia in an Embedded Wearable Device.

The importance of an embedded wearable device with automatic detection and alarming cannot be overstated, given that 15-30% of patients with atrial fibrillation are reported to be asymptomatic. These asymptomatic patients do not seek medical care, hence traditional diagnostic tools including Holter are not effective for the further prevention of associated stroke or heart failure. This is likely to be more so in the era of COVID-19, in which patients become more reluctant on hospitalization and checkups. However, little literature is available on this important topic. For this reason, this study developed efficient deep learning with model compression, which is designed to use ECG data and classify arrhythmia in an embedded wearable device. ECG-signal data came from Korea University Anam Hospital in Seoul, Korea, with 28,308 unique patients (15,412 normal and 12,896 arrhythmia). Resnets and Mobilenets with model compression (TensorFlow Lite) were applied and compared for the diagnosis of arrhythmia in an embedded wearable device. The weight size of the compressed model registered a remarkable decrease from 743 MB to 76 KB (1/10000), whereas its performance was almost the same as its original counterpart. Resnet and Mobilenet were similar in terms of accuracy, i.e., Resnet-50 Hz (97.3) vs. Mo-bilenet-50 Hz (97.2), Resnet-100 Hz (98.2) vs. Mobilenet-100 Hz (97.9). Here, 50 Hz/100 Hz denotes the down-sampling rate. However, Resnets took more flash memory and longer inference time than did Mobilenets. In conclusion, Mobilenet would be a more efficient model than Resnet to classify arrhythmia in an embedded wearable device.

Study on the use of standard 12-lead ECG data for rhythm-type ECG classification problems.

BACKGROUND AND OBJECTIVES: Most deep-learning-related methodologies for electrocardiogram (ECG) classification are focused on finding an optimal deep-learning architecture to improve classification performance. However, in this study, we proposed a methodology for fusion of various single-lead ECG data as training data in the single-lead ECG classification problem. METHODS: We used a squeeze-and-excitation residual network (SE-ResNet) with 152 layers as the baseline model. We compared the performance of a 152-layer SE-ResNet trained on ECG signals from various leads of a standard 12-lead ECG system to that of a 152-layer SE-ResNet trained on only single-lead ECG data with the same lead information as the test set. The experiments were performed using five different types of rhythm-type single-lead ECG data obtained from Konkuk University Hospital in South Korea. RESULTS: Experiment results based on the combination from the relationship experiments of the leads showed that lead -aVR or II revealed the best classification performance. In case of -aVR, this model achieved a high F1 score for normal (98.7%), AF (98.2%), APC (95.1%), and VPC (97.4%), indicating its potential for practical use in the medical field. CONCLUSION: We concluded that the 152-layer SE-ResNet trained by fusion of single-lead ECGs had better classification performance than the 152-layer SE-ResNet trained on only single-lead ECG data, regardless of the single-lead ECG signal type. We also found that the best performance directions for single-lead ECG classification are Lead -aVR and II.

Gender Differences in Patients with Metastatic Pancreatic Cancer Who Received FOLFIRINOX.

BACKGROUND: The combination of 5-fluorouracil, leucovorin, irinotecan, and oxaliplatin (FOLFIRINOX) is a very effective chemotherapeutic regimen for unresectable pancreatic cancer. Previous studies have reported that female gender may be a predictor of a better response to FOLFIRINOX. This study was aimed at investigating the clinical outcomes and dose modification patterns of FOLFIRINOX by gender. METHODS: Patients with metastatic pancreatic cancer (MPC) who began FOLFIRINOX as the first-line therapy at Seoul National University Bundang Hospital between 2013 and 2018 were enrolled. The patients received at least four chemotherapy cycles. Local regression and a linear mixed model were used to analyze dose modification patterns by gender. RESULTS: Ninety-seven patients with MPC (54 men; 43 women) were enrolled. In the first FOLFIRINOX cycle, there were significant differences in age and body surface area between the genders (58.8 (men) and 64.9 years (women), p = 0.005; 1.7 (men) and 1.6 m2 (women), p < 0.001, respectively). The median progression-free survival (PFS) and overall survival (OS) were 10.8 and 18.0 months, respectively. There was a trend of longer PFS (10.3 (men) and 11.9 months (women), p = 0.153) and a significantly longer OS (17.9 (men) and 25.9 months (women), p = 0.019) in female patients. During the first year of FOLFIRINOX treatment, there was a significant difference of the age-corrected dose reduction pattern by gender (a mean of 95.6% dose at the initial cycle and -0.35% of dose reduction per week in men versus a mean of 90.7% dose at the initial cycle and -0.53% of dose reduction per week in women, p-value of the slope: <0.001). There was no difference in the adverse event rates between the genders. CONCLUSIONS: Female patients showed longer OS despite a more rapid dose reduction during each cycle. Gender differences should be considered during FOLFIRINOX treatment.

Comparison of FOLFIRINOX and Gemcitabine Plus Nab-paclitaxel for Treatment of Metastatic Pancreatic Cancer: Using Korean Pancreatic Cancer (K-PaC) Registry.

OBJECTIVE: By using the Korean Pancreatic Cancer (K-PaC) registry, we compared the clinical outcomes of FOLFIRINOX (FFX) and gemcitabine plus nab-paclitaxel (GNP) in patients with metastatic pancreatic cancer (MPC). METHODS: We constructed a web-based database of 3748 anonymized patients diagnosed with pancreatic ductal adenocarcinoma. MPC patients who received first-line FFX or GNP were enrolled. Overall survival (OS), progression-free survival, grade III to IV toxicity, and cross-over treatment were analyzed. RESULTS: A total of 413 patients (232 vs. 181, FFX vs. GNP; all data are presented in this sequence) were eligible. Median age was 63 years (60 vs. 69 y) with 43% (39% vs. 47%) comprising female individuals. The major metastatic sites were the liver (64%), peritoneum (25%), and distant lymph nodes (18%). The median OS was 11.5 versus 12.7 months (hazard ratio [HR]=0.87, 95% confidence interval [CI]: 0.68-1.12, P=0.286), and median progression-free survival was 7.5 versus 8.1 months (HR=0.92, 95% CI: 0.70-1.20, P=0.517), respectively. The frequency of grade III to IV febrile neutropenia was higher in the FFX group (18% vs. 11%, P=0.040), and that of peripheral neuropathy was higher in the GNP group (8% vs. 14%, P=0.046). The chance to receive second-line chemotherapy was higher in the GNP group (45% vs. 56%, P=0.036). In the cross-over treatment, the median OS of the FFX-GNP group (n=43) and the GNP-FFX group (n=47) was 16.8 versus 17.7 months (HR=0.79, 95% CI: 0.44-1.41, P=0.425). CONCLUSIONS: FFX and GNP showed similar efficacy and comparable toxicity in MPC patients. Although the GNP group had a higher chance to receive second-line chemotherapy, they did not have improved overall survival.

Clinicopathologic characteristics of paranasal sinus fungus ball: retrospective, multicenter study in Korea.

PURPOSE: Fungus ball (FB) is the most common type of fungal rhinosinusitis and the prevalence of FB has increased over the past 10 years. The aim of this study was to compare the clinical characteristics of Korean adult patients with FB and chronic rhinosinusitis (CRS) without FB. METHODS: We retrospectively analyzed data on 1362 patients (147 FB and 1215 CRS) who underwent endoscopic sinus surgery at nine Korean medical centers in 2005, 2010, and 2016. We evaluated the prevalence of FB and compared the clinical characteristics of FB and CRS. Medical records, computed tomography (CT) findings, atopic status, concomitant diseases, tissue, and blood eosinophil count were assessed. RESULTS: The prevalence of FB was significantly higher in 2016 (15.9%) than in the other years (7.8% in 2005 and 7.5% in 2010). The FB patients were more likely to be female, older, have unilateral disease and less likely to have allergy compared to the CRS patients. The most common main complaint related to CRS and FB was nasal obstruction. CT determined that unilateral disease and maxillary sinus dominancy were common in patients with FB. The incidence of concomitant diseases was much higher in FB, with lower tissue and blood eosinophilia. CONCLUSION: FB is commonly encountered in older women with the increased prevalence. FB had a different clinical presentation, radiological findings, and prognosis than CRS. Further studies are needed to understand the pathophysiologic mechanisms underlying the development of FB.

Comparison of laser in situ keratomileusis flaps created by 3 femtosecond lasers and a microkeratome.

PURPOSE: To evaluate the thickness and side-cut angle of laser in situ keratomileusis (LASIK) flaps created by 1 of 3 femtosecond lasers or a microkeratome using Fourier-domain optical coherence tomography (OCT). SETTING: Department of Ophthalmology, Institute of Vision Research, Yonsei University College of Medicine, Seoul, Korea. DESIGN: Comparative case series. METHODS: Flap creation for bilateral LASIK was performed using an IntraLase (femtosecond group 1), VisuMax (femtosecond group 2), or Femto LDV (femtosecond group 3) femtosecond laser or an M2 microkeratome. Flap thickness was determined at 14 points. The side-cut angle was measured in 4 directions at the margin interface. Measurements were taken 2 months postoperatively using an RTVue Fourier-domain OCT device and integrated software. RESULTS: Femtosecond group 1 comprised 50 eyes; femtosecond group 2, 40 eyes; femtosecond group 3, 64 eyes; and the microkeratome group, 52 eyes. Eyes in femtosecond groups 1 and 2 had relatively even flap configuration. Flaps in femtosecond group 3 and the microkeratome group had a meniscus shape. Flaps in femtosecond group 1 had the least difference between the mean peripheral and the central flap thickness (P<.001). The greatest flap thickness predictability (measured versus intended thickness) was in femtosecond group 3 (P<.001). Flaps in femtosecond group 1 had a side-cut angle closest to 90 degrees (P<.001). CONCLUSIONS: Flap morphology differed according to the system used. The 3 femtosecond laser systems appeared to be superior to the microkeratome system generally. The 3 femtosecond laser systems also produced different flap configurations depending on their individual mechanisms.

Clinical impact of phosphorylated signal transducer and activator of transcription 3, epidermal growth factor receptor, p53, and vascular endothelial growth factor receptor 1 expression in resected adenocarcinoma of lung by using tissue microarray.

BACKGROUND: The signal transducer and activator of transcription 3 (STAT3) play a key role in the downstream pathway of the epidermal growth factor receptor (EGFR) in nonsmall cell lung cancer and promote cell proliferation, invasion, and angiogenesis. The clinical significance of phosphorylated STAT3 (pSTAT3), EGFR, p53, and vascular endothelial growth factor receptor 1 (VEGFR-1) expression in patients with completely resected lung adenocarcinoma was evaluated to determine the effects of pSTAT3 in tumor angiogenesis and proliferation. METHODS: The expressions of pSTAT3, EGFR, p53, and VEGFR-1 were evaluated by immunohistochemical staining of tissue microarrays from 162 samples of resected lung adenocarcinoma. RESULTS: The median age of the 162 patients was 62 years, the median disease-free survival was 41.7 months, and the median OS (OS) was 80.2 months. Expression of pSTAT3, EGFR, p53, and VEGFR-1 was detected in 51.2%, 71%, 35.2%, and 35.2% of the samples, respectively. pSTAT3 expression was correlated significantly with VEGFR-1 expression (P = .025). The coexpression of pSTAT3 and VEGFR-1 was correlated with increased lymph node involvement (P = .021) and a trend toward a short OS (P = .085). In multivariate analysis, the expression levels of p53 and VEGFR-1 were identified as independent prognostic factors that affected OS. CONCLUSIONS: The results of this study suggested that pSTAT3 and VEGFR-1 expression may play roles in the tumor progression and angiogenesis of lung adenocarcinoma. Further studies are needed, however, to uncover the detailed mechanisms that underlie the roles of these proteins in lung adenocarcinoma.

Effect of peroxisome proliferator-activated receptor gamma agonists on myofibroblast differentiation and collagen production in nasal polyp-derived fibroblasts.

OBJECTIVES: We investigated the effect of transforming growth factor-beta1 (TGF-beta1) on the differentiation of nasal polyp-derived fibroblasts (NPDFs) and on their production of collagen, and investigated the effect of peroxisome proliferator-activated receptor gamma (PPARgamma) agonists on the effects of TGF-beta1. METHODS: Primary fibroblast lines were established from 5 human nasal polyp tissues. We activated the NPDFs with TGF-beta1 and exposed them to the PPARy agonists ciglitazone and 15-deoxy-Delta-prostaglandin J2. The expression of alpha-smooth muscle actin (alpha-SMA) and collagen was measured by reverse transcriptase-polymerase chain reactions, immunocytochemical staining, and collagen assays. RESULTS: The treatment with TGF-beta1 increased the expression of alpha-SMA and collagen, and expression was markedly attenuated by the PPARy agonists. CONCLUSIONS: PPARy agonists inhibit the differentiation of TGF-beta1-activated NPDFs and their production of collagen.

Discordance of molecular biomarkers associated with epidermal growth factor receptor pathway between primary tumors and lymph node metastasis in non-small cell lung cancer.

INTRODUCTION: For the identification of the patients who most likely benefit from epidermal growth factor receptor (EGFR) tyrosine kinase inhibitors in non-small cell lung cancer (NSCLC), molecular assays are considered to be of paramount importance. Given the heterogeneity of NSCLC at the molecular level, this study was conducted to determine the discrepancy in EGFR mutations between primary tumors and the corresponding lymph node metastasis. PATIENTS AND METHODS: Surgically resected 101 paired primary NSCLC and metastatic lymph nodes were evaluated for the EGFR mutations by direct DNA sequencing and heteroduplex analysis. RESULTS: EGFR mutation was detected in 29.7% (30 of 101) of the primary tumors and in 27.7% of lymph node metastases (28 of 101) by either direct sequencing or heteroduplex analysis, respectively. By direct sequencing, 12 cases (11.9%) showed discordance in EGFR mutations between primary tumors and metastasis. In 11 cases, EGFR mutations were detected only in the primary tumor, whereas 1 case only in lymph node metastases. By heteroduplex analysis, 17 cases (16.8%) were discordant. Ten cases were primary tumor positive and lymph node negative, whereas seven cases were lymph node positive and primary tumor negative. CONCLUSIONS: A considerable proportion of NSCLC showed discrepancy in EGFR mutations between primary tumors and metastatic lymph nodes, suggesting tumor heterogeneity at the molecular level during the process of metastasis.

Endoscopic surgery for inverted papilloma originating from the sphenoid sinus and related clinical characteristics.

CONCLUSIONS: Bony changes on CT imaging of inverted papilloma (IP) are useful for predicting tumor origin and recurrence sites. Because the lateral wall and floor of the sphenoid sinus are the most common origin and recurrence sites, the anterior wall of the sphenoid sinus should be opened as wide as the lateral wall and inferiorly to the level of the floor, especially in deeply pneumatized sphenoid sinuses. OBJECTIVES: The incidence of isolated sphenoid IP is exceedingly low. So far, there have not been studies on the usual origin and recurrence sites of the sphenoid sinus. We sought to identify the sites of origin and recurrence and describe clinical characteristics, radiological features, and proper endoscopic management. PATIENTS AND METHODS: We retrospectively reviewed medical records of patients with IP of the sphenoid sinus. Data collection included clinical presentations, origin and recurrent sites, radiological features, and surgical methods. RESULTS: Seven patients were identified. The most common symptoms were headache and nasal obstruction. Radiological findings showed that most tumors extended into the nasal cavity or ethmoid sinus combined frequently with erosion of the lateral wall or intersinus septum. We noted simultaneous attachment to multiple walls in five subjects, including both lateral wall and floor attachment in three subjects.

Biological profiles of Korean atomic bomb survivors in residence at Daegu and Kyungbuk, Republic of Korea.

In 1945, many Koreans, in addition to Japanese, were killed or injured by the atomic bombs dropped on Hiroshima and Nagasaki, Japan. This study compared the biological profiles of Korean atomic bomb survivors in residence at Daegu and Kyungbuk, Republic of Korea with those of a representative sample of Koreans obtained during a similar period. We evaluated anthropometric measurements, blood pressure, blood cell counts, blood chemistry, and urinalysis of survivors (n=414) and age- and sex-matched controls (n=414) recruited from the third Korea National Health and Nutrition Examination Survey conducted in 2005. Univariate analyses revealed significantly higher systolic blood pressure, white blood cell count, and serum total cholesterol, triglycerides, high-density lipoprotein-cholesterol, and aspartate aminotransferase levels (p<0.01) in the survivors. Conversely, hemoglobin concentration, hematocrit, red blood cell count, and the proportion of positive urine occult blood (p<0.01) were lower in the survivors. Our findings suggest that biological profiles of Korean atomic bomb survivors were adversely affected by radiation exposure.

Korean atomic bomb survivors' report on the present status and health of their children: a mail questionnaire survey.

A mail questionnaire survey on the present status and health of children of Korean survivors of the 1945 atomic bombing of Japan was conducted in 2004. In total, 1,115 Korean A-bomb survivors' reports on demographic factors and the health of their children (n = 4,079) were analysed. In 2004, 293 of their children were reported to have died and 3,786 children still living. Among the deceased children, the cause of death was unidentified in 64 per cent of cases. The most frequently reported cause of death was infectious disease (9.6%). Most of the deaths (51%) were reported to have occurred in children less than ten years old. The most frequently reported congenital disorders among living children were mental retardation (0.18%) and congenital bone disorders (0.18%). Further study and administrative support are needed to more accurately determine the status of Korean A-bomb survivors and their children.

Non-cancer diseases of Korean atomic bomb survivors in residence at Hapcheon, Republic of Korea.

Many Koreans, in addition to Japanese, were killed or injured by the atomic bombs detonated over Hiroshima and Nagasaki, Japan, in 1945. Our study examined noncancer diseases of Korean A-bomb survivors in residence at Hapcheon, Republic of Korea and evaluated whether they had significantly higher prevalence of noncancer diseases than non-exposed people. We evaluated a number of tests, including anthropometric measurements, blood pressure, blood chemistry, hepatitis B surface antigen, and urinalysis, of survivors (n=223) and controls (n=372). Univariate analysis revealed significantly lower fasting glucose and creatinine, and higher diastolic blood pressure, aspartate aminotransferase, alanine aminotransferase, and blood urea nitrogen levels in the survivors than in the controls. The calculation of crude prevalence ratios (PRs) revealed that A-bomb survivors had a significantly higher prevalence of hypertension (PR, 1.16; 95% CI, 1.00-1.35) and chronic liver disease (2.20; 1.59-3.06) than controls. After adjusting for covariates (age, sex, body mass index, marital status, education, alcohol consumption, and smoking), A-bomb survivors had a significantly higher prevalence of hypertension (1.24; 1.06-1.44), chronic liver disease (2.07; 1.51-2.84), and hypercholesterolemia (1.79; 1.11-2.90) than controls. This study suggests that A-bomb exposure is associated with a higher prevalence of non-cancer diseases in Korean survivors.

Present status and self-reported diseases of the Korean atomic bomb survivors: a mail questionnaire survey.

Many Koreans were forced to move to Japan while Korea was occupied by Japan. Consequently, when the atomic bombs were dropped on Hiroshima and Nagasaki an estimated 40,000 Koreans died and 30,000 survived. In 2004, 2,235 Koreans were registered as A-bomb survivors in South Korea. A mail questionnaire survey to evaluate the present status and self-reported diseases of the Korean survivors was conducted. In total, 1,256 questionnaires were returned and analysed. The most frequent chronic diseases reported by Korean survivors were hypertension (40.1 per cent), peptic ulcer disease (25.7 per cent), anaemia (23.3 per cent) and cataracts (23.1 per cent). The most frequent malignant diseases were stomach cancer (1.9 per cent), colon cancer (0.5 per cent) and leukaemia/multiple myeloma (0.4 per cent). This study suggests that further investigations are needed into the health concerns of the survivors and into health protection measures.