Continual learning framework for a multicenter study with an application to electrocardiogram.

Deep learning has been increasingly utilized in the medical field and achieved many goals. Since the size of data dominates the performance of deep learning, several medical institutions are conducting joint research to obtain as much data as possible. However, sharing data is usually prohibited owing to the risk of privacy invasion. Federated learning is a reasonable idea to train distributed multicenter data without direct access; however, a central server to merge and distribute models is needed, which is expensive and hardly approved due to various legal regulations. This paper proposes a continual learning framework for a multicenter study, which does not require a central server and can prevent catastrophic forgetting of previously trained knowledge. The proposed framework contains the continual learning method selection process, assuming that a single method is not omnipotent for all involved datasets in a real-world setting and that there could be a proper method to be selected for specific data. We utilized the fake data based on a generative adversarial network to evaluate methods prospectively, not ex post facto. We used four independent electrocardiogram datasets for a multicenter study and trained the arrhythmia detection model. Our proposed framework was evaluated against supervised and federated learning methods, as well as finetuning approaches that do not include any regulation to preserve previous knowledge. Even without a central server and access to the past data, our framework achieved stable performance (AUROC 0.897) across all involved datasets, achieving comparable performance to federated learning (AUROC 0.901).

A development of a graph-based ensemble machine learning model for skin sensitization hazard and potency assessment.

Many defined approaches (DAs) for skin sensitization assessment based on the adverse outcome pathway (AOP) have been developed to replace animal testing because the European Union has banned animal testing for cosmetic ingredients. Several DAs have demonstrated that machine learning models are beneficial. In this study, we have developed an ensemble prediction model utilizing the graph convolutional network (GCN) and machine learning approach to assess skin sensitization. The model integrates in silico parameters and data from alternatives to animal testing of well-defined AOP to improve DA predictivity. Multiple ensemble models were created using the probability produced by the GCN with six physicochemical properties, direct peptide reactivity assay, KeratinoSens™, and human cell line activation test (h-CLAT), using a multilayer perceptron approach. Models were evaluated by predicting the testing set's human hazard class and three potency classes (strong, weak, and non-sensitizer). When the GCN feature was used, 11 models out of 16 candidates showed the same or improved accuracy in the testing set. The ensemble model with the feature set of GCN, KeratinoSens™, and h-CLAT produced the best results with an accuracy of 88% for assessing human hazards. The best three-class potency model was created with the feature set of GCN and all three assays, resulting in 64% accuracy. These results from the ensemble approach indicate that the addition of the GCN feature could provide an improved predictivity of skin sensitization hazard and potency assessment.

Development of a Novel Prognostic Model to Predict 6-Month Swallowing Recovery After Ischemic Stroke.

Background and Purpose- The aim of this study was to explore clinical and radiological prognostic factors for long-term swallowing recovery in patients with poststroke dysphagia and to develop and validate a prognostic model using a machine learning algorithm. Methods- Consecutive patients (N=137) with acute ischemic stroke referred for swallowing examinations were retrospectively reviewed. Dysphagia was monitored in the 6 months poststroke period and then analyzed using the Kaplan-Meier method and Cox regression model for clinical and radiological factors. Bayesian network models were developed using potential prognostic factors to classify patients into those with good (no need for tube feeding or diet modification for 6 months) and poor (tube feeding or diet modification for 6 months) recovery of swallowing function. Results- Twenty-four (17.5%) patients showed persistent dysphagia for the first 6 months with a mean duration of 65.6 days. The time duration of poststroke dysphagia significantly differed by tube feeding status, clinical dysphagia scale, sex, severe white matter hyperintensities, and bilateral lesions at the corona radiata, basal ganglia, or internal capsule (CR/BG/IC). Among these factors, tube feeding status (P<0.001), bilateral lesions at CR/BG/IC (P=0.001), and clinical dysphagia scale (P=0.042) were significant prognostic factors in a multivariate analysis using Cox regression models. The tree-augmented network classifier, based on 10 factors (sex, lesions at CR, BG/IC, and insula, laterality, anterolateral territory of the brain stem, bilateral lesions at CR/BG/IC, severe white matter hyperintensities, clinical dysphagia scale, and tube feeding status), performed better than other benchmarking classifiers developed in this study. Conclusions- Initial dysphagia severity and bilateral lesions at CR/BG/IC are revealed to be significant prognostic factors for 6-month swallowing recovery. The prediction of 6-month swallowing recovery was feasible based on clinical and radiological factors using the Bayesian network model. We emphasize the importance of bilateral subcortical lesions as prognostic factors that can be utilized to develop prediction models for long-term swallowing recovery.

Fimasartan increases glucose-stimulated insulin secretion in patients with type 2 diabetes and hypertension compared with amlodipine.

AIM: To study the effects of angiotensin receptor blockers (ARBs) on insulin secretion in hypertensive patients with type 2 diabetes. MATERIALS AND METHODS: A total of 41 patients were enrolled in this open-label, active comparator-controlled, crossover study. After a 2-week run-in period with amlodipine, the participants were assigned to receive either fimasartan (60-120 mg daily) or amlodipine (5-10 mg daily) for 16 weeks. Thereafter, they were treated with the other drug for another 16 weeks. Physical examinations and laboratory tests were performed before and after each treatment. RESULTS: Blood pressure, glycated haemoglobin and oral glucose tolerance test (OGTT) values were similar with each treatment. Fimasartan treatment significantly increased median (range) homeostatic assessment of ß-cell function values (49.9 [22.5-174.4] vs 46.9 [15.6-148.0]), area under the curve of insulin during OGTT (27 284 [9501-94 525] vs 26 818 [8112-76 704] pmol/L × min), insulinogenic index at 60 minutes (19.7 [3.0-131.2] vs 15.0 [2.4-103.8] pmol/mmol) and at 120 minutes (19.1 [1.9-85.5] vs 12.6 [-4.3-178.8] pmol/mmol) compared with those with amlodipine (all P < .05); however, acute insulin response and insulin resistance indices were similar for both agents. CONCLUSIONS: Compared with amlodipine, fimasartan increased late-phase glucose-stimulated insulin secretion in patients with type 2 diabetes and hypertension. This finding suggests that ARBs would be more beneficial in such patients compared with other classes of anti-hypertensives.

Computational Discrimination of Breast Cancer for Korean Women Based on Epidemiologic Data Only.

Breast cancer is the second leading cancer for Korean women and its incidence rate has been increasing annually. If early diagnosis were implemented with epidemiologic data, the women could easily assess breast cancer risk using internet. National Cancer Institute in the United States has released a Web-based Breast Cancer Risk Assessment Tool based on Gail model. However, it is inapplicable directly to Korean women since breast cancer risk is dependent on race. Also, it shows low accuracy (58%-59%). In this study, breast cancer discrimination models for Korean women are developed using only epidemiological case-control data (n = 4,574). The models are configured by different classification techniques: support vector machine, artificial neural network, and Bayesian network. A 1,000-time repeated random sub-sampling validation is performed for diverse parameter conditions, respectively. The performance is evaluated and compared as an area under the receiver operating characteristic curve (AUC). According to age group and classification techniques, AUC, accuracy, sensitivity, specificity, and calculation time of all models were calculated and compared. Although the support vector machine took the longest calculation time, the highest classification performance has been achieved in the case of women older than 50 yr (AUC = 64%). The proposed model is dependent on demographic characteristics, reproductive factors, and lifestyle habits without using any clinical or genetic test. It is expected that the model could be implemented as a web-based discrimination tool for breast cancer. This tool can encourage potential breast cancer prone women to go the hospital for diagnostic tests.

Towards optimal design of patient isolation units in emergency rooms to prevent airborne virus transmission: From computational fluid dynamics to data-driven modeling.

BACKGROUND: Patient isolation units (PIUs) can be an effective method for effective infection control. Computational fluid dynamics (CFD) is commonly used for PIU design; however, optimizing this design requires extensive computational resources. Our study aims to provide data-driven models to determine the PIU settings, thereby promoting a more rapid design process. METHOD: Using CFD simulations, we evaluated various PIU parameters and room conditions to assess the impact of PIU installation on ventilation and isolation. We investigated particle dispersion from coughing subjects and airflow patterns. Machine-learning models were trained using CFD simulation data to estimate the performance and identify significant parameters. RESULTS: Physical isolation alone was insufficient to prevent the dispersion of smaller particles. However, a properly installed fan filter unit (FFU) generally enhanced the effectiveness of physical isolation. Ventilation and isolation performance under various conditions were predicted with a mean absolute percentage error of within 13%. The position of the FFU was found to be the most important factor affecting the PIU performance. CONCLUSION: Data-driven modeling based on CFD simulations can expedite the PIU design process by offering predictive capabilities and clarifying important performance factors. Reducing the time required to design a PIU is critical when a rapid response is required.

A practical approach based on learning-based model predictive control with minimal prior knowledge of patients for artificial pancreas.

BACKGROUND AND OBJECTIVES: Complete identification of the glucose dynamics for a patient generally requires prior clinical procedures and several measurements for the patient. However, these steps may not be always feasible. To address this limitation, we propose a practical approach integrating learning-based model predictive control (MPC), adaptive basal and bolus injections, and suspension with minimal requirements of prior knowledge of the patient. METHODS: The glucose dynamic system matrices were periodically updated using only input values, without any pretrained models. The optimal insulin dose was calculated based on a learning-based MPC algorithm. Meal detection and estimation modules were also introduced. The basal and bolus insulin injections were fine-tuned using the performance of glucose control from the previous day. To validate the proposed method, evaluations with 20 virtual patients from a type 1 diabetes metabolic simulator were employed. RESULTS: Time-in-range (TIR) and time-below-range (TBR) were 90.8% (84.1% - 95.6%) and 0.3% (0% - 0.8%), as represented by the median, first (Q1), and third quartiles (Q3), respectively, when meal intakes were fully announced. When one out of three meal intake announcements was missing, TIR and TBR were 85.2% (75.0% - 88.9%) and 0.9% (0.4% - 1.1%), respectively. CONCLUSIONS: The proposed approach obviates the need for prior tests from patients and shows effective regulation of blood glucose levels. From the perspective of practical implementation in clinical environments, to deal with minimal prior information of the patient, our study demonstrates how essential clinical knowledge and learning-based modules can be integrated into a control framework for an artificial pancreas.

Development of an artificial intelligence bacteremia prediction model and evaluation of its impact on physician predictions focusing on uncertainty.

Prediction of bacteremia is a clinically important but challenging task. An artificial intelligence (AI) model has the potential to facilitate early bacteremia prediction, aiding emergency department (ED) physicians in making timely decisions and reducing unnecessary medical costs. In this study, we developed and externally validated a Bayesian neural network-based AI bacteremia prediction model (AI-BPM). We also evaluated its impact on physician predictive performance considering both AI and physician uncertainties using historical patient data. A retrospective cohort of 15,362 adult patients with blood cultures performed in the ED was used to develop the AI-BPM. The AI-BPM used structured and unstructured text data acquired during the early stage of ED visit, and provided both the point estimate and 95% confidence interval (CI) of its predictions. High AI-BPM uncertainty was defined as when the predetermined bacteremia risk threshold (5%) was included in the 95% CI of the AI-BPM prediction, and low AI-BPM uncertainty was when it was not included. In the temporal validation dataset (N = 8,188), the AI-BPM achieved area under the receiver operating characteristic curve (AUC) of 0.754 (95% CI 0.737-0.771), sensitivity of 0.917 (95% CI 0.897-0.934), and specificity of 0.340 (95% CI 0.330-0.351). In the external validation dataset (N = 7,029), the AI-BPM's AUC was 0.738 (95% CI 0.722-0.755), sensitivity was 0.927 (95% CI 0.909-0.942), and specificity was 0.319 (95% CI 0.307-0.330). The AUC of the post-AI physicians predictions (0.703, 95% CI 0.654-0.753) was significantly improved compared with that of the pre-AI predictions (0.639, 95% CI 0.585-0.693; p-value < 0.001) in the sampled dataset (N = 1,000). The AI-BPM especially improved the predictive performance of physicians in cases with high physician uncertainty (low subjective confidence) and low AI-BPM uncertainty. Our results suggest that the uncertainty of both the AI model and physicians should be considered for successful AI model implementation.

Multi-Task Disentangled Autoencoder for Time-Series Data in Glucose Dynamics.

The objective of this study is to propose MD-VAE: a multi-task disentangled variational autoencoders (VAE) for exploring characteristics of latent representations (LR) and exploiting LR for diverse tasks including glucose forecasting, event detection, and temporal clustering. We applied MD-VAE to one virtual continuous glucose monitoring (CGM) data from an FDA-approved Type 1 Diabetes Mellitus simulator (T1DMS) and one publicly available CGM data of real patients for glucose dynamics of Type 1 Diabetes Mellitus. LR captured meaningful information to be exploited for diverse tasks, and was able to differentiate characteristics of sequences with clinical parameters. LR and generative models have received relatively little attention for analyzing CGM data so far. However, as proposed in our study, VAE has the potential to integrate not only current but also future information on glucose dynamics and unexpected events including interactions of devices in the data-driven manner. We expect that our model can provide complementary views on the analysis of CGM data.

Colonoscopic image synthesis with generative adversarial network for enhanced detection of sessile serrated lesions using convolutional neural network.

Computer-aided detection (CADe) systems have been actively researched for polyp detection in colonoscopy. To be an effective system, it is important to detect additional polyps that may be easily missed by endoscopists. Sessile serrated lesions (SSLs) are a precursor to colorectal cancer with a relatively higher miss rate, owing to their flat and subtle morphology. Colonoscopy CADe systems could help endoscopists; however, the current systems exhibit a very low performance for detecting SSLs. We propose a polyp detection system that reflects the morphological characteristics of SSLs to detect unrecognized or easily missed polyps. To develop a well-trained system with imbalanced polyp data, a generative adversarial network (GAN) was used to synthesize high-resolution whole endoscopic images, including SSL. Quantitative and qualitative evaluations on GAN-synthesized images ensure that synthetic images are realistic and include SSL endoscopic features. Moreover, traditional augmentation methods were used to compare the efficacy of the GAN augmentation method. The CADe system augmented with GAN synthesized images showed a 17.5% improvement in sensitivity on SSLs. Consequently, we verified the potential of the GAN to synthesize high-resolution images with endoscopic features and the proposed system was found to be effective in detecting easily missed polyps during a colonoscopy.

Hyoid kinematic features for poor swallowing prognosis in patients with post-stroke dysphagia.

Identification of prognostic factors for swallowing recovery in patients with post-stroke dysphagia is crucial for determining therapeutic strategies. We aimed at exploring hyoid kinematic features of poor swallowing prognosis in patients with post-stroke dysphagia. Of 122 patients who experienced dysphagia following ischemic stroke, 18 with poor prognosis, and 18 age- and sex-matched patients with good prognosis were selected and retrospectively reviewed. Positional data of the hyoid bone during swallowing were obtained from the initial videofluoroscopic swallowing study after stroke onset. Normalized hyoid profiles of displacement/velocity and direction angle were analyzed using functional regression analysis, and maximal or mean values were compared between the good and poor prognosis patient groups. Kinematic analysis showed that maximal horizontal displacement (P = 0.031) and velocity (P = 0.034) in forward hyoid motions were significantly reduced in patients with poor prognosis compared to those with good prognosis. Mean direction angle for the initial swallowing phase was significantly lower in patients with poor prognosis than in those with good prognosis (P = 0.0498). Our study revealed that reduced horizontal forward and altered initial backward motions of the hyoid bone during swallowing can be novel kinematic features indicating poor swallowing prognosis in patients with post-stroke dysphagia.

3D cell culture using a clinostat reproduces microgravity-induced skin changes.

Exposure to microgravity affects human physiology in various ways, and astronauts frequently report skin-related problems. Skin rash and irritation are frequent complaints during space missions, and skin thinning has also been reported after returning to Earth. However, spaceflight missions for studying the physiological changes in microgravity are impractical. Thus, we used a previously developed 3D clinostat to simulate a microgravity environment and investigate whether physiological changes of the skin can be reproduced in a 3D in vitro setting. Our results showed that under time-averaged simulated microgravity (taSMG), the thickness of the endothelial cell arrangement increased by up to 59.75%, indicating skin irritation due to vasodilation, and that the diameter of keratinocytes and fibroblast co-cultured spheroids decreased by 6.66%, representing skin thinning. The α1 chain of type I collagen was upregulated, while the connective tissue growth factor was downregulated under taSMG. Cytokeratin-10 expression was significantly increased in the taSMG environment. The clinostat-based 3D culture system can reproduce physiological changes in the skin similar to those under microgravity, providing insight for understanding the effects of microgravity on human health before space exploration.

Crack Width Estimation of Mortar Specimen Using Gas Diffusion Experiment.

Maintenance of structures using self-healing concrete technologies has recently been actively studied. However, unlike the technological development of self-healing concrete, research focused on evaluating the self-healing performance is insufficient. Although water permeability experiments are widely used, the reliability of the test results may be reduced due to the viscosity of water and the possibility of elution of material inside the specimen. In this study, we propose a gas diffusion test for estimating the crack width and eventually for application to evaluation of the self-healing performance. The results verified that the proposed method can be effectively applied to the estimation of crack width.

Differential kinematic features of the hyoid bone during swallowing in patients with Parkinson's disease.

This study aimed to investigate spatiotemporal characteristics of the hyoid bone during swallowing in patients with Parkinson's disease (PD) and dysphagia. Spatiotemporal data of the hyoid bone was obtained from videofluoroscopic images of 69 subjects (23 patients with PD, 23 age- and sex-matched healthy elderly controls, and 23 healthy young controls). Normalized profiles of displacement/velocity were analyzed during different periods (percentile) of swallowing using functional regression analysis, and the maximal values were compared between the groups. Maximal horizontal displacement and velocity were significantly decreased during the initial backward (P = 0.006 and P < 0.001, respectively) and forward (P = 0.008 and P < 0.001, respectively) motions in PD patients compared to elderly controls. Maximal vertical velocity was significantly lower in PD patients than in elderly controls (P = 0.001). No significant difference was observed in maximal displacement and velocity in both horizontal and vertical planes between the healthy elderly and young controls, although horizontal displacement was significantly decreased during the forward motion (51st-57th percentiles) in the elderly controls. In conclusion, reduced horizontal displacement and velocity of the hyoid bone during the forward motion would be due to combined effects of disease and aging, whereas those over the initial backward motion may be considered specific to patients with PD.

Synthesis and characterization of heteroleptic titanium MOCVD precursors for TiO2 thin films.

Heteroleptic titanium alkoxides with three different ligands, i.e., [Ti(OiPr)(X)(Y)] (X = tridentate, Y = bidentate ligands), were synthesized to find efficient metal organic chemical vapor deposition (MOCVD) precursors for TiO2 thin films. Acetylacetone (acacH) or 2,2,6,6-tetramethyl-3,5-heptanedione (thdH) was employed as a bidentate ligand, while N-methyldiethanolamine (MDEA) was employed as a tridentate ligand. It was expected that the oxygen and moisture susceptibility of titanium alkoxides, as well as their tendency to form oligomers, would be greatly reduced by placing multidentate and bulky ligands around the center Ti atom. The synthesized heteroleptic titanium alkoxides were characterized both physicochemically and crystallographically, and their thermal behaviors were also investigated. [Ti(OiPr)(MDEA)(thd)] was found to be monomeric and stable against moisture; it also showed good volatility in the temperature window between volatilization and decomposition. This material was used as a single-source precursor during MOCVD to generate TiO2 thin films on silicon wafers. The high thermal stability of [Ti(OiPr)(MDEA)(thd)] enabled the fabrication of TiO2 films over a wide temperature range, with steady growth rates between 500 and 800 °C.

Microgravity induces autophagy via mitochondrial dysfunction in human Hodgkin's lymphoma cells.

Gravitational forces can impose physical stresses on the human body as it functions to maintain homeostasis. It has been reported that astronauts exposed to microgravity experience altered biological functions and many subsequent studies on the effects of microgravity have therefore been conducted. However, the anticancer mechanisms of simulated microgravity remain unclear. We previously showed that the proliferation of human Hodgkin's lymphoma (HL) cells was inhibited when these cells were cultured in time-averaged simulated microgravity (taSMG). In the present study, we investigated whether taSMG produced an anticancer effect. Exposure of human HL cells to taSMG for 2 days increased their reactive oxygen species (ROS) production and NADPH oxidase family gene expression, while mitochondrial mass, ATPase, ATP synthase, and intracellular ATP levels were decreased. Furthermore, human HL cells exposed to taSMG underwent autophagy via AMPK/Akt/mTOR and MAPK pathway modulation; such autophagy was inhibited by the ROS scavenger N-acetylcysteine (NAC). These results suggest an innovative therapeutic approach to HL that is markedly different from conventional chemotherapy and radiotherapy.

Application of the Oral Minimal Model to Korean Subjects with Normal Glucose Tolerance and Type 2 Diabetes Mellitus.

BACKGROUND: The oral minimal model is a simple, useful tool for the assessment of ß-cell function and insulin sensitivity across the spectrum of glucose tolerance, including normal glucose tolerance (NGT), prediabetes, and type 2 diabetes mellitus (T2DM) in humans. METHODS: Plasma glucose, insulin, and C-peptide levels were measured during a 180-minute, 75-g oral glucose tolerance test in 24 Korean subjects with NGT (n=10) and T2DM (n=14). The parameters in the computational model were estimated, and the indexes for insulin sensitivity and ß-cell function were compared between the NGT and T2DM groups. RESULTS: The insulin sensitivity index was lower in the T2DM group than the NGT group. The basal index of ß-cell responsivity, basal hepatic insulin extraction ratio, and post-glucose challenge hepatic insulin extraction ratio were not different between the NGT and T2DM groups. The dynamic, static, and total ß-cell responsivity indexes were significantly lower in the T2DM group than the NGT group. The dynamic, static, and total disposition indexes were also significantly lower in the T2DM group than the NGT group. CONCLUSION: The oral minimal model can be reproducibly applied to evaluate ß-cell function and insulin sensitivity in Koreans.