Search

Background Fatty liver disease (FLD) arises from the accumulation of fat in the liver and may cause liver inflammation, which, if not well controlled, may develop into liver fibrosis, cirrhosis, or even hepatocellular carcinoma. Objective We describe the construction of machine-learning models for current-visit prediction (CVP), which can help physicians obtain more information for accurate diagnosis, and next-visit prediction (NVP), which can help physicians provide potential high-risk patients with advice to effectively prevent FLD. Methods The large-scale and high-dimensional dataset used in this study comes from Taipei MJ Health Research Foundation in Taiwan. We used one-pass ranking and sequential forward selection (SFS) for feature selection in FLD prediction. For CVP, we explored multiple models, including k-nearest-neighbor classifier (KNNC), Adaboost, support vector machine (SVM), logistic regression (LR), random forest (RF), Gaussian naïve Bayes (GNB), decision trees C4.5 (C4.5), and classification and regression trees (CART). For NVP, we used long short-term memory (LSTM) and several of its variants as sequence classifiers that use various input sets for prediction. Model performance was evaluated based on two criteria: the accuracy of the test set and the intersection over union/coverage between the features selected by one-pass ranking/SFS and by domain experts. The accuracy, precision, recall, F-measure, and area under the receiver operating characteristic curve were calculated for both CVP and NVP for males and females, respectively. Results After data cleaning, the dataset included 34,856 and 31,394 unique visits respectively for males and females for the period 2009-2016. The test accuracy of CVP using KNNC, Adaboost, SVM, LR, RF, GNB, C4.5, and CART was respectively 84.28%, 83.84%, 82.22%, 82.21%, 76.03%, 75.78%, and 75.53%. The test accuracy of NVP using LSTM, bidirectional LSTM (biLSTM), Stack-LSTM, Stack-biLSTM, and Attention-LSTM was respectively 76.54%, 76.66%, 77.23%, 76.84%, and 77.31% for fixed-interval features, and was 79.29%, 79.12%, 79.32%, 79.29%, and 78.36%, respectively, for variable-interval features. Conclusions This study explored a large-scale FLD dataset with high dimensionality. We developed FLD prediction models for CVP and NVP. We also implemented efficient feature selection schemes for current- and next-visit prediction to compare the automatically selected features with expert-selected features. In particular, NVP emerged as more valuable from the viewpoint of preventive medicine. For NVP, we propose use of feature set 2 (with variable intervals), which is more compact and flexible. We have also tested several variants of LSTM in combination with two feature sets to identify the best match for male and female FLD prediction. More specifically, the best model for males was Stack-LSTM using feature set 2 (with 79.32% accuracy), whereas the best model for females was LSTM using feature set 1 (with 81.90% accuracy).

Diversified research focusing on early prediction of diseases based on small datasets and/or reducing the execution time on computer was published. However, no convincing evidence was showed that the conventional wrapper based feature selection was applicable for processing large-scale high-dimension dataset in an efficient way. In this study, our wrapper based feature selection method is designed to apply to a large-scale dataset with high dimension and long longitudinal breadth for predicting two types of metabolic syndrome diseases. Specifically, according to the components, the method adopts dimensionality reduction and/or feature selection to optimize subsets of features. Subsequently, the selected features will be further identified with statistics provided by clinic experts to exam the significance and accuracy. Accordingly, the method provides a sufficient mean for early diagnosis of metabolic diseases and achieves the efficiency by reducing notable computational time.

In-hospital cardiac arrest (IHCA) diminish the survival rate of patients, despite most of the IHCA cases are preventable. More than 54% IHCA patient had abnormal clinical manifestation before they suffered a cardiac arrest. If appropriate steps were taken, patients’ survival rate would be higher and medical expense would be decreased. This paper proposes a novel approach to detect IHCA before the event occurred. We construct two types of shifting windows (corresponding to two tasks) that allow machine learning to be applied for our dataset which is severely imbalanced. The results show that our approach can effectively handle the imbalanced dataset for detecting cardiac arrest. As the selection of performance index, we used the area under the receiver operating characteristic curve (AUROC) and the area under the precision–recall curve (AUPRC). In our experiments, the best classifier is random forest for task 1, with AUROC of 0.88. LSTM is the best for task 2, with AUPRC of 0.71 for the second task.

The structured electronic medical record is the basis for computers to process and achieve the target of precise diagnosis and treatment automatically using the knowledge and features of the techniques such as machine learning and artificial intelligence (AI). Because of the increasing demands on improving the efficiency and the flexibility during the step or phase of classification and extraction, providing the expansion mechanism for the automatic adaption of new NER (Named Entity Recognition, NER) model training during the NER model training stage anytime when the new entities/tags shall be learned and classified and hence the related knowledge database (DB) shall be expanded automatically. The proposed method includes a training stage involving the step of adaptive improved NER model training for the chest x-ray medical reports/files and a test stage involving the step of the dependency parsing and the relation extracting to be perform sequentially, and thus the goals of automatic information extraction and structured medical report generation using the machine learning technique, and the optimization and accuracy improvement of the doctor's work and performance through referring to the structured medical report for diagnosis and treatment can be achieved.

AdLocus is an APP developed by HyXen Company for mobile advertisements. This advertising software can push the message to the target users within specified locations. Based on the real big data provided by AdLocus, we design a dynamic advertisements recommendation system to increase the advertising efficiency. The proposed method uses the regression models and the click probability to recommend the amount of mobile advertisements for every 30 minutes. The results show that our recommendation can efficiently raise the successful clicking rate to satisfy the required clicks number. Moreover, the proposed method avoids the redundant advertisements to reduce the system cost.

This study evaluated the impact of multi-sensory information cues from on-bicycle rider information assistance devices (OBRAD) on hazard perception performance. Experiments tested the impact of distraction from different combinations of visual, auditory and tactile sensory aids on the subject’s ability to maintain peddling frequency while conducting eight different tasks. The results indicate that the integrated use of different sensory cues (e.g., text, audible alerts and vibration) can decrease cognitive loading, with each sensory combination, particularly those involving tactile stimulation, having different levels of effect. Tactile sensory aids helped reduce the degree of rider distraction, thus helping maintain a high sensitivity to danger (hit rate mean: 0.34). Cycling performance was further improved through combining tactile stimuli with auditory cues for assistance in the secondary task. The implications of these findings and the need to integrate and manage complex OBRAD information systems are discussed.

Technologies have been developed to assist the cyclist for precise training exercise such as monitoring their body and bicycle state. Due to the limited position placing on-bicycle rider assistant device and riding posture, it enlarges the essential view range of the chance the cyclist distracted from the road. The risks using the system have been evaluated through an interactive video-based hazard perception/reaction test in which response times were measured. 5 participants, 3 male and 2 female participated a 2-phase section including 7 traffic scenes in a 5 minutes cycling exercise with and without using on-bicycle rider assistant device. As the results, average reaction times tended to increase 0.33 seconds and average mis-maintained pushing cadence time was 31.4 seconds when cycling with the device assistance. It is concluded that cadence control assisted by the device can interfere the efficiency of hazard perception, an important safety factor of cycling.