Chun-Hung Richard Lin's research while affiliated with Sun Yat-Sen University and other places

Background The global aging population presents a significant challenge, with older adults experiencing declining physical and cognitive abilities and increased vulnerability to chronic diseases and adverse health outcomes. This study aims to develop an interpretable deep learning (DL) model to predict adverse events in geriatric patients within 72 hours of hospitalization. Methods The study used retrospective data (2017–2020) from a major medical center in Taiwan. It included non-trauma geriatric patients who visited the emergency department and were admitted to the general ward. Data preprocessing involved collecting prognostic factors like vital signs, lab results, medical history, and clinical management. A deep feedforward neural network was developed, and performance was evaluated using accuracy, sensitivity, specificity, positive predictive value (PPV), and area under the receiver operating characteristic curve (AUC). Model interpretation utilized the Shapley Additive Explanation (SHAP) technique. Results The analysis included 127,268 patients, with 2.6% experiencing imminent intensive care unit transfer, respiratory failure, or death during hospitalization. The DL model achieved AUCs of 0.86 and 0.84 in the validation and test sets, respectively, outperforming the Sequential Organ Failure Assessment (SOFA) score. Sensitivity and specificity values ranged from 0.79 to 0.81. The SHAP technique provided insights into feature importance and interactions. Conclusion The developed DL model demonstrated high accuracy in predicting serious adverse events in geriatric patients within 72 hours of hospitalization. It outperformed the SOFA score and provided valuable insights into the model’s decision-making process.

Background Hyperkalemia poses a significant risk of sudden cardiac death, especially for those with end-stage renal diseases (ESRD). Smartwatches with ECG capabilities offer a promising solution for continuous, non-invasive monitoring using AI. Objectives To develop an AI-ECG algorithm to predict serum potassium level in ESRD patient with smartwatch generated ECG waveforms. Methods A cohort of 152,508 patients with 293,557 ECGs paired serum potassium levels obtained within one hour at Cedars Sinai Medical Center (CSMC) was used to train an AI-ECG model (Kardio-Net) to predict serum potassium level. The model was further fine-tuned on 4,337 ECGs from 1,463 patients with ESRD using inputs from 12-lead and single-lead ECGs. Kardio-Net was evaluated in held-out test cohorts from CSMC and Stanford Healthcare (SHC) as well as a prospective international cohort of 40 ESRD patients with smartwatch ECGs at Chang Gung Memorial Hospital (CGMH). Results The Kardio-Net, when applied to 12-lead ECGs, identified severe hyperkalemia with an AUC of 0.852 and a mean absolute error (MAE) of 0.527 mEq/L. In external validation at SHC, the model achieved an AUC of 0.849 and an MAE of 0.599 mEq/L. For single-lead ECGs, Kardio-Net detected hyperkalemia with an AUC of 0.876 and had an MAE of 0.575 mEq/L in the CSMC test cohort. Using prospectively obtained smartwatch data, the AUC was 0.831, with an MAE of 0.580 mEq/L. Conclusions We validate a deep learning model to predict serum potassium levels from both 12-lead ECGs and single-lead smartwatch data, demonstrating its utility for remote monitoring of hyperkalemia.

Simultaneous wireless information and power transfer (SWIPT) has been advocated as a highly promising technology for enhancing the capabilities of 5G and 6G devices. However, the challenge of dealing with large propagation path loss poses a significant hurdle. To address this issue, massive multiple-input multiple-output (MIMO) is employed to enhance the efficiency of SWIPT in cellular-based networks with multiple small cells, and especially increase the energy for cell-edge users. In addition, by leveraging a large set of spatially distributed base stations to collaboratively serve SWIPT-enabled user equipment, the cell-free massive MIMO has the potential to provide even better performance than the conventional small-cell systems. In this work, we extend the investigation to include the application of SWIPT technology with alternating current (AC) logic in the cell-free networks and the small-cell networks and propose joint beamforming and power splitting optimization frameworks to maximize the system sum-rate, subject to the constraints on harvested energy, AC logic energy supply, and total transmit power. The optimization problem is shown to be non-convex, posing a significant challenge. To address this challenge, we resort to a two-stage decomposition approach. Specifically, we first introduce quadratic transform-based fractional programming (FP) algorithms to iteratively solve the non-convex optimization problems in the first stage, achieving near-optimal solutions with low time complexities. To further reduce the complexities, we also incorporate conventional schemes such as zero forcing, maximum ratio transmission, and signal-to-leakage-and-noise ratio for the design of beamforming vectors. Second, to determine the optimal power splitting ratio within the framework, we develop a one-dimensional (1-D) search algorithm to tackle the single variable optimization problem reduced in the second stage. These algorithms are then evaluated in the context of cell-free MIMO and small-cell networks with numerical experiments. The results show that the FP-based algorithms can consistently outperform those utilizing the conventional beamforming schemes, and the solutions of this work can achieve up to fivefold improvement in the system sum-rate than the small-cell counterpart while providing different but comparable performance trends in energy harvesting (EH).

Imbalance classification is common in scenarios like fault diagnosis, intrusion detection, and medical diagnosis, where obtaining abnormal data is difficult. This article addresses a one-class problem, implementing and refining the One-Class Nearest-Neighbor (OCNN) algorithm. The original inter-quartile range mechanism is replaced with the K-means with outlier removal (KMOR) algorithm for efficient outlier identification in the target class. Parameters are optimized by treating these outliers as non-target-class samples. A new algorithm, the Location-based Nearest-Neighbor (LBNN) algorithm, clusters one-class training data using KMOR and calculates the farthest distance and percentile for each test data point to determine if it belongs to the target class. Experiments cover parameter studies, validation on eight standard imbalanced datasets from KEEL, and three applications on real medical imbalanced datasets. Results show superior performance in precision, recall, and G-means compared to traditional classification models, making it effective for handling imbalanced data challenges.

This paper proposes a method to record automobiles driving data by using Internet of Vehicles (IoV) and blockchain technologies. The automobiles physical driving data are collected from On-Board Diagnostics (OBD) or J1939 through IoV, whereas the blockchain used in paper is built by the use of Hyperledger Fabric and RSA digital signature. The proposed method could solve the major issues of verification whether the forged data enters the blockchain. The proposed Internet of Vehicles data transmission is to record driving operation data every second, and upload driving data through smart contract (Chaincode) automation to ensure the correctness, integrity and immediacy of the blockchain content of driving data. The blockchain platform uses practical Byzantine Fault Tolerance (PBFT) consensus mechanism, PBFT has fault tolerance, and effectively improves the blockchain transaction volume per second (Transactions Per Second, TPS). Finally, through Hyperledger Explorer monitoring the blockchain status and Hyperledger Caliper performance test, the experimental results show that the proposed method could effectively solve the false data entering the blockchain. The proposed method can be applied to practical application services such as highway passenger transport management, financial technology and Usage-Based Insurance (UBI) in the future.

Objectives The aim of this study was to develop a deep-learning pipeline for the measurement of pericardial effusion (PE) based on raw echocardiography clips, as current methods for PE measurement can be operator-dependent and present challenges in certain situations. Methods The proposed pipeline consisted of three distinct steps: moving window view selection (MWVS), automated segmentation, and width calculation from a segmented mask. The MWVS model utilized the ResNet architecture to classify each frame of the extracted raw echocardiography files into selected view types. The automated segmentation step then generated a mask for the PE area from the extracted echocardiography clip, and a computer vision technique was used to calculate the largest width of the PE from the segmented mask. The pipeline was applied to a total of 995 echocardiographic examinations. Results The proposed deep-learning pipeline exhibited high performance, as evidenced by intraclass correlation coefficient (ICC) values of 0.867 for internal validation and 0.801 for external validation. The pipeline demonstrated a high level of accuracy in detecting PE, with an area under the receiving operating characteristic curve (AUC) of 0.926 (95% CI: 0.902–0.951) for internal validation and 0.842 (95% CI: 0.794–0.889) for external validation. Conclusion The machine-learning pipeline developed in this study can automatically calculate the width of PE from raw ultrasound clips. The novel concepts of moving window view selection for image quality control and computer vision techniques for maximal PE width calculation seem useful in the field of ultrasound. This pipeline could potentially provide a standardized and objective approach to the measurement of PE, reducing operator-dependency and improving accuracy.

Several risk factors are related to glycemic control in patients with type 2 diabetes mellitus (T2DM), including demographics, medical conditions, negative emotions, lipid profiles, and heart rate variability (HRV; to present cardiac autonomic activity). The interactions between these risk factors remain unclear. This study aimed to use machine learning methods of artificial intelligence to explore the relationships between various risk factors and glycemic control in T2DM patients. The study utilized a database from Lin et al. (2022) that included 647 T2DM patients. Regression tree analysis was conducted to identify the interactions among risk factors that contribute to glycated hemoglobin (HbA1c) values, and various machine learning methods were compared for their accuracy in classifying T2DM patients. The results of the regression tree analysis revealed that high depression scores may be a risk factor in one subgroup but not in others. When comparing different machine learning classification methods, the random forest algorithm emerged as the best-performing method with a small set of features. Specifically, the random forest algorithm achieved 84% accuracy, 95% area under the curve (AUC), 77% sensitivity, and 91% specificity. Using machine learning methods can provide significant value in accurately classifying patients with T2DM when considering depression as a risk factor.

Importance: Early awareness of Kawasaki disease (KD) helps physicians administer appropriate therapy to prevent acquired heart disease in children. However, diagnosing KD is challenging and relies largely on subjective diagnosis criteria. Objective: To develop a prediction model using machine learning with objective parameters to differentiate children with KD from other febrile children. Design, setting, and participants: This diagnostic study included 74 641 febrile children younger than 5 years who were recruited from 4 hospitals, including 2 medical centers and 2 regional hospitals, between January 1, 2010, and December 31, 2019. Statistical analysis was performed from October 2021 to February 2023. Main outcomes and measures: Demographic data and laboratory values from electronic medical records, including complete blood cell count with differential, urinalysis, and biochemistry, were collected as possible parameters. The primary outcome was whether the febrile children fulfilled the diagnostic criteria of KD. The supervised eXtreme Gradient Boosting (XGBoost) machine learning method was applied to establish a prediction model. The confusion matrix and likelihood ratio were used to evaluate the performance of the prediction model. Results: This study included a total of 1142 patients with KD (mean [SD] age, 1.1 [0.8] years; 687 male patients [60.2%]) and 73 499 febrile children (mean [SD] age, 1.6 [1.4] years; 41 465 male patients [56.4%]) comprising the control group. The KD group was predominantly male (odds ratio, 1.79; 95% CI, 1.55-2.06) with younger age (mean difference, -0.6 years [95% CI, -0.6 to -0.5 years]) compared with the control group. The prediction model's best performance in the testing set was able to achieve 92.5% sensitivity, 97.3% specificity, 34.5% positive predictive value, 99.9% negative predictive value, and a positive likelihood ratio of 34.0, which indicates outstanding performance. The area under the receiver operating characteristic curve of the prediction model was 0.980 (95% CI, 0.974-0.987). Conclusions and relevance: This diagnostic study suggests that results of objective laboratory tests had the potential to be predictors of KD. Furthermore, these findings suggested that machine learning with XGBoost can help physicians differentiate children with KD from other febrile children in pediatric emergency departments with excellent sensitivity, specificity, and accuracy.

Background: Machine learning models have demonstrated superior performance in predicting invasive bacterial infection (IBI) in febrile infants compared to commonly used risk stratification criteria in recent studies. However, the black-box nature of these models can make them difficult to apply in clinical practice. In this study, we developed and validated an explainable deep learning model that can predict IBI in febrile infants ≤ 60 days of age visiting the emergency department. Methods: We conducted a retrospective study of febrile infants aged ≤ 60 days who presented to the pediatric emergency department of a medical center in Taiwan between January 1, 2011 and December 31, 2019. Patients with uncertain test results and complex chronic health conditions were excluded. IBI was defined as the growth of a pathogen in the blood or cerebrospinal fluid. We used a deep neural network to develop a predictive model for IBI and compared its performance to the IBI score and step-by-step approach. The SHapley Additive Explanations (SHAP) technique was used to explain the model's predictions at different levels. Results: Our study included 1847 patients, 53 (2.7%) of whom had IBI. The deep learning model performed similarly to the IBI score and step-by-step approach in terms of sensitivity and negative predictive value, but provided better specificity (54%), positive predictive value (5%), and area under the receiver-operating characteristic curve (0.87). SHapley Additive exPlanations identified five influential predictive variables (absolute neutrophil count, body temperature, heart rate, age, and C-reactive protein). Conclusion: We have developed an explainable deep learning model that can predict IBI in febrile infants aged 0-60 days. The model not only performs better than previous scoring systems, but also provides insight into how it arrives at its predictions through individual features and cases.