A machine-learning approach to predict postprandial hypoglycemia.

BACKGROUND: For an effective artificial pancreas (AP) system and an improved therapeutic intervention with continuous glucose monitoring (CGM), predicting the occurrence of hypoglycemia accurately is very important. While there have been many studies reporting successful algorithms for predicting nocturnal hypoglycemia, predicting postprandial hypoglycemia still remains a challenge due to extreme glucose fluctuations that occur around mealtimes. The goal of this study is to evaluate the feasibility of easy-to-use, computationally efficient machine-learning algorithm to predict postprandial hypoglycemia with a unique feature set. METHODS: We use retrospective CGM datasets of 104 people who had experienced at least one hypoglycemia alert value during a three-day CGM session. The algorithms were developed based on four machine learning models with a unique data-driven feature set: a random forest (RF), a support vector machine using a linear function or a radial basis function, a K-nearest neighbor, and a logistic regression. With 5-fold cross-subject validation, the average performance of each model was calculated to compare and contrast their individual performance. The area under a receiver operating characteristic curve (AUC) and the F1 score were used as the main criterion for evaluating the performance. RESULTS: In predicting a hypoglycemia alert value with a 30-min prediction horizon, the RF model showed the best performance with the average AUC of 0.966, the average sensitivity of 89.6%, the average specificity of 91.3%, and the average F1 score of 0.543. In addition, the RF showed the better predictive performance for postprandial hypoglycemic events than other models. CONCLUSION: In conclusion, we showed that machine-learning algorithms have potential in predicting postprandial hypoglycemia, and the RF model could be a better candidate for the further development of postprandial hypoglycemia prediction algorithm to advance the CGM technology and the AP technology further.

Deep ECG-Respiration Network (DeepER Net) for Recognizing Mental Stress.

Unmanaged long-term mental stress in the workplace can lead to serious health problems and reduced productivity. To prevent this, it is important to recognize and relieve mental stress in a timely manner. Here, we propose a novel stress detection algorithm based on end-to-end deep learning using multiple physiological signals, such as electrocardiogram (ECG) and respiration (RESP) signal. To mimic workplace stress in our experiments, we used Stroop and math tasks as stressors, with each stressor being followed by a relaxation task. Herein, we recruited 18 subjects and measured both ECG and RESP signals using Zephyr BioHarness 3.0. After five-fold cross validation, the proposed network performed well, with an average accuracy of 83.9%, an average F1 score of 0.81, and an average area under the receiver operating characteristic (ROC) curve (AUC) of 0.92, demonstrating its superiority over conventional machine learning models. Furthermore, by visualizing the activation of the trained network's neurons, we found that they were activated by specific ECG and RESP patterns. In conclusion, we successfully validated the feasibility of end-to-end deep learning using multiple physiological signals for recognition of mental stress in the workplace. We believe that this is a promising approach that will help to improve the quality of life of people suffering from long-term work-related mental stress.

WavelNet: A novel convolutional neural network architecture for arrhythmia classification from electrocardiograms.

BACKGROUND AND OBJECTIVE: Automated detection of arrhythmias from electrocardiograms (ECGs) can be of considerable assistance to medical professionals in providing efficient treatment for patients with cardiovascular diseases. In recent times, convolutional neural network (CNN)-based arrhythmia classification models have been introduced, but their decision-making processes remain unclear and their performances are not reproducible. This paper proposes an accurate, interpretable, and reproducible end-to-end arrhythmia classification model based on a novel CNN architecture named WavelNet, which is interpretable and optimal for dealing with ECGs. METHODS: Inspired by SincNet, which is capable of band-pass filtering-based spectral analysis, WavelNet was devised to achieve wavelet transform-based spectral analysis. WavelNet was trained using a subject-oriented five-class ECG arrhythmia dataset generated from the MIT-BIH Arrhythmia Database while following a benchmark scheme. By adopting various mother wavelets, multiple WavelNet-based arrhythmia classification models were implemented. To investigate whether our wavelet transform-based approach outperforms original end-to-end and band-pass filtering-based approaches, our proposed models were compared with vanilla CNN- and SincNet-based models. Model implementation and evaluation processes were repeated ten times in a Google Colab Pro+ environment. Furthermore, our most successful model was compared with state-of-the-art arrhythmia classification models for performance evaluation. RESULTS: The proposed WavelNet-based models showed excellent performance on classifying non-ectopic, supraventricular ectopic, and ventricular ectopic beats because of their ability to perform adaptive spectral analysis while preserving temporal ECG information compared with vanilla CNN- and SincNet-based models. In particular, a Symlet 4 wavelet-adopting WavelNet-based model achieved the best performance with nearly 90% overall accuracy as well as the highest levels of sensitivity in classifying each arrhythmia class: 91.4%, 49.3%, and 91.4% for non-ectopic, supraventricular ectopic, and ventricular ectopic beat classifications, respectively. These results were comparable to those of state-of-the-art models. In addition, the results are reproducible, which differentiates our study from previous studies. CONCLUSIONS: Our proposed WavelNet-based arrhythmia classification model achieved remarkable performance based on a reasonable decision-making process, in comparison with other models. As its noteworthy performance is clinically reasonable and reproducible, our proposed model can contribute toward implementing a real-world precision healthcare system for patients with cardiovascular diseases.

A personalized blood glucose level prediction model with a fine-tuning strategy: A proof-of-concept study.

BACKGROUND: The accurate prediction of blood glucose (BG) level is still a challenge for diabetes management. This is due to various factors such as diet, personal physiological characteristics, stress, and activities influence changes in BG level. To develop an accurate BG level predictive model, we propose a personalized model based on a convolutional neural network (CNN) with a fine-tuning strategy. METHODS: We utilized continuous glucose monitoring (CGM) datasets from 1052 professional CGM sessions and split them into three groups according to type 1, type 2, and gestational diabetes mellitus (T1DM, T2DM, and GDM, respectively). During the preprocessing, only CGM data points were utilized, and future BG levels of four different prediction horizons (PHs, 15, 30, 45, and 60 min) were used as output. In training, we trained a general CNN and a multi-output random forest regressor using a hold-out method for each group. Next, we developed two personalized models: (1) by fine-tuning the general CNN on partial sample points of each CGM dataset, and (2) by learning a CNN from scratch on the points. RESULTS: For all groups, the fine-tuned CNN showed the lowest average root mean squared error, average mean absolute percentage error, highest average time gain (PH = 15 and 60 min in T1DM) and highest percentage in region A of Clarke error grid analysis at all PHs. In the performance comparison between the fine-tuned CNN and other models, we found that the fine-tuned CNN improved the performance of the general CNN in most cases and outperformed the scratch CNN at all PHs in all groups, making the fine-tuning strategy was useful for accurate BG level prediction. We analyzed all cases of four predictive patterns in each group, and found that the input BG level trend and the BG level at the time of prediction were related to the future BG level trend. CONCLUSIONS: We demonstrated the efficacy of the fine-tuning method in a large number of CGM datasets and analyzed the four predictive patterns. Therefore, we believe that the proposed method will significantly contribute to the development of an accurate personalized model and the analysis for its predictions.

Machine learning-based analysis of adolescent gambling factors.

BACKGROUND AND AIMS: Problem gambling among adolescents has recently attracted attention because of easy access to gambling in online environments and its serious effects on adolescent lives. We proposed a machine learning-based analysis method for predicting the degree of problem gambling. METHODS: Of the 17,520 respondents in the 2018 National Survey on Youth Gambling Problems dataset (collected by the Korea Center on Gambling Problems), 5,045 students who had gambled in the past 3 months were included in this study. The Gambling Problem Severity Scale was used to provide the binary label information. After the random forest-based feature selection method, we trained four models: random forest (RF), support vector machine (SVM), extra trees (ETs), and ridge regression. RESULTS: The online gambling behavior in the past 3 months, experience of winning money or goods, and gambling of personal relationship were three factors exhibiting the high feature importance. All four models demonstrated an area under the curve (AUC) of >0.7; ET showed the highest AUC (0.755), RF demonstrated the highest accuracy (71.8%), and SVM showed the highest F1 score (0.507) on a testing set. DISCUSSION: The results indicate that machine learning models can convey meaningful information to support predictions regarding the degree of problem gambling. CONCLUSION: Machine learning models trained using important features showed moderate accuracy in a large-scale Korean adolescent dataset. These findings suggest that the method will help screen adolescents at risk of problem gambling. We believe that expandable machine learning-based approaches will become more powerful as more datasets are collected.