Long short-term memory and Logistic regression for mortality risk prediction of intensive care unit patients with stroke / 北京大学学报(医学版)

OBJECTIVE@#To select variables related to mortality risk of stroke patients in intensive care unit (ICU) through long short-term memory (LSTM) with attention mechanisms and Logistic regression with L1 norm, and to construct mortality risk prediction model based on conventional Logistic regression with important variables selected from the two models and to evaluate the model performance.@*METHODS@#Medical Information Mart for Intensive Care (MIMIC)-Ⅳ database was retrospectively analyzed and the patients who were primarily diagnosed with stroke were selected as study population. The outcome was defined as whether the patient died in hospital after admission. Candidate predictors included demogra-phic information, complications, laboratory tests and vital signs in the initial 48 h after ICU admission. The data were randomly divided into a training set and a test set for ten times at a ratio of 8 ∶2. In training sets, LSTM with attention mechanisms and Logistic regression with L1 norm were constructed to select important variables. In the test sets, the mean importance of variables of ten times was used as a reference to pick out the top 10 variables in each of the two models, and then these variables were included in conventional Logistic regression to build the final prediction model. Model evaluation was based on the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy. And the model performance was compared with the forward Logistic regression model which hadn't conducted variable selection previously.@*RESULTS@#A total of 2 755 patients with 2 979 ICU admission records were included in the analysis, of which 526 recorded deaths. The AUC of Logistic regression model with L1 norm was statistically better than that of LSTM with attention mechanisms (0.819±0.031 vs. 0.760±0.018, P < 0.001). Age, blood glucose, and blood urea nitrogen were at the top ten important variables in both of the two models. AUC, sensitivity, specificity, and accuracy of Logistic regression models were 0.85, 85.98%, 71.74% and 74.26%, respectively. And the final prediction model was superior to forward Logistic regression model.@*CONCLUSION@#The variables selected by Logistic regression with L1 norm and LSTM with attention mechanisms had good prediction performance, which showed important implications on the mortality prediction of stroke patients in ICU.

Automatic labeling and extraction of terms in natural language processing in acupuncture clinical literature / 中国针灸

The paper analyzes the specificity of term recognition in acupuncture clinical literature and compares the advantages and disadvantages of three named entity recognition (NER) methods adopted in the field of traditional Chinese medicine. It is believed that the bi-directional long short-term memory networks-conditional random fields (Bi LSTM-CRF) may communicate the context information and complete NER by using less feature rules. This model is suitable for term recognition in acupuncture clinical literature. Based on this model, it is proposed that the process of term recognition in acupuncture clinical literature should include 4 aspects, i.e. literature pretreatment, sequence labeling, model training and effect evaluation, which provides an approach to the terminological structurization in acupuncture clinical literature.

Prediction of malaria using deep learning models: A case study on city clusters in the state of Amazonas, Brazil, from 2003 to 2018

ABSTRACT Background: Malaria is curable. Nonetheless, over 229 million cases of malaria were recorded in 2019, along with 409,000 deaths. Although over 42 million Brazilians are at risk of contracting malaria, 99% percent of all malaria cases in Brazil are located in or around the Amazon rainforest. Despite declining cases and deaths, malaria remains a major public health issue in Brazil. Accurate spatiotemporal prediction of malaria propagation may enable improved resource allocation to support efforts to eradicate the disease. Methods: In response to calls for novel research on malaria elimination strategies that suit local conditions, in this study, we propose machine learning (ML) and deep learning (DL) models to predict the probability of malaria cases in the state of Amazonas. Using a dataset of approximately 6 million records (January 2003 to December 2018), we applied k-means clustering to group cities based on their similarity of malaria incidence. We evaluated random forest, long-short term memory (LSTM) and dated recurrent unit (GRU) models and compared their performance. Results: The LSTM architecture achieved better performance in clusters with less variability in the number of cases, whereas the GRU presents better results in clusters with high variability. Although Diebold-Mariano testing suggested that both the LSTM and GRU performed comparably, GRU can be trained significantly faster, which could prove advantageous in practice. Conclusions: All models showed satisfactory accuracy and strong performance in predicting new cases of malaria, and each could serve as a supplemental tool to support regional policies and strategies.

A comparative study of ARIMA and LSTM models in predicting hospital discharge number / 公共卫生与预防医学

Objective To fit and predict the monthly discharge number of a specialist hospital using Autoregressive Integrated Moving Average model (ARIMA) and Long Short-Term Memory Neural Network model (LSTM), and compare the prediction effects of the two models. Methods ARIMA and LSTM models were constructed based on the monthly discharge number of a specialist hospital from 2013 to 2018. The resulting models were then used to predict the monthly discharge numbers in 2019, which were compared with actual data. The mean absolute percentage error (MAPE) was used to evaluate the prediction effect of these two models. Results The MAPE values of ARIMA and LSTM compared to actual data in 2019 were 7.90% and 14.26%, respectively. Conclusion The prediction effect of ARIMA was better than that of LSTM. The prediction results of ARIMA showed that the number of patients discharged from the specialist hospital in 2019 was increasing, which fit well with the actual data.

Comparison of ARIMAX and multivariate LSTM model in predicting daily death toll in Yancheng City / 公共卫生与预防医学

Objective To compare the effects of Autoregressive Integrated Moving Average model-X (ARIMAX) and multivariate Long Short Term Memory Network (multivariate LSTM) in the prediction of daily total death toll in Yancheng City. Methods Based on total death toll data, meteorological data and air quality data from January 1st, 2014 to June 30th,2017 in Yancheng City, Jiangsu province, ARIMAX model and multivariate LSTM model were established to predict the daily total death toll from July 1st,2017 to July 14th,2017. RMSE, MAE and MAPE were used as evaluation indexes to compare the prediction effects of these two models. Results RMSE, MAE and MAPE of ARIMAX model and multivariate LSTM model were 20.742、15.094、9.921 and 47.182、35.863、19.633, respectively. Conclusion ARIMAX model is better than multivariate LSTM model to predict the daily death toll in Yancheng city.

Classification of Fetal Heart Rate Based on Poincare Plot and LSTM / 中国医疗器械杂志

Fetal heart rate plays an essential role in maternal and fetal monitoring and fetal health detection. In this study, a method based on Poincare Plot and LSTM is proposed to realize the high performance classification of abnormal fetal heart rate. Firstly, the original fetal heart rate signal of CTU-UHB database is preprocessed via interpolation, then the sequential fetal heart rate signal is converted into Poincare Plot to obtain nonlinear characteristics of the signals, and then SquenzeNet is used to extract the features of Poincare Plot. Finally, the features extracted by SqueezeNet are classified by LSTM. And the accuracy, the true positive rate and the false positive rate are 98.00%, 100.00%, 92.30% respectively on 2 000 test set data. Compared with the traditional fetal heart rate classification method, all respects are improved. The method proposed in this study has good performance in CTU-UHB fetal monitoring database and has certain practical value in the clinical diagnosis of auxiliary fetal heart rate detection.

An Enhanced Focused Web Crawler for Biomedical Topics Using Attention Enhanced Siamese Long Short Term Memory Networks

Abstract The Internet is chosen to be one among the primary source of biomedical information. To retrieve necessary biomedical information, the search engine needs an efficient, focused crawler mechanism. But the area of research concerned with the focused crawler for biomedical topics is notably scanty. However, the quantity, momentum, diversity, and quality of the available online biomedical information, challenges and calls for enhanced aid to crawl. This paper surmounts the challenges and proposes a new learning approach for focused web crawling adopting Attention Enhanced Siamese Long Short Term Memory (AE-SLSTM) Networks with peephole connections which predicts topical relevance of the web page. The proposed AE-SLSTM model accurately computes the semantic similarity between the topic and the web pages. The performance of the newly designed crawler is assessed using two well known metrics namely harvest rate ( h r a t e ) and irrelevance ratio ( p r a t e ). The presented crawler surpass the existing focused crawlers with an average h r a t e of 0.39 and an average p r a t e of 0.61 after crawling 5,000 web pages relating to biomedical topics. The results clearly depicts that the proposed methodology aids to download more relevant biomedical web pages related to the particular topic from the internet.

Estimación en la Intención de Agarres: Cilíndrico, Esférico y Gancho Utilizando Redes Neuronales Profundas / Estimation of Hand-Grip Intention: Cylindrical, Spherical and Hook Using Deep Neural Networks

Resumen Las amputaciones de extremidades superiores pueden producir diversos grados de incapacidad en la persona afectada, esto es exacerbado aún más, si se presenta durante un periodo de su vida laboral activa, por esta razón es de importancia social el estudio de las prótesis y algoritmos que ayuden a un mejor control de estas por parte del usuario. En esta investigación, se propone una arquitectura basada en redes neuronales recurrentes del tipo Long Short-Term Memory y redes convolucionales para la clasificación de señales electromiográficas, con aplicaciones para control de prótesis de mano. La red propuesta clasifica tres tipos de agarres realizados con la mano: cilíndrico, esférico y de gancho. El modelo propuesto al ser evaluado mostró una eficiencia (accuracy) del 89 %, en contraste con una red neuronal artificial basada en capas completamente conectadas que solo obtuvo una eficiencia del 80% en la predicción de los agarres. El presente trabajo se limita solamente a evaluar la red ante una entrada de electromiograma y no se implementó un sistema de control para la prótesis de la mano. Así, una arquitectura de redes convolucionales para el control de prótesis de mano que pueden ser entrenadas con las señales del sujeto.

Abstract Upper extremities amputations can produce different disability degrees in the amputated person, this is acerbated even more, when it happens during active working life. So, for this reason, it is of social importance the study of prostheses and algorithms that help a better control of these by the user. In this research, we propose an architecture based on recurrent neural networks, called Long Short-Term Memory, and convolutional neural networks for classification of electromyographic signals, with applications for hand prosthesis control. The proposed network classifies three types of movements made by the hand: cylindrical, spherical and hook grips. The proposed model showed an efficiency (accuracy) of 89%, in contrast to an artificial neural network based on completely connected layers that only obtained an efficiency of 80% in the prediction of the hand movements. The present work is limited to evaluate the network with an electromyogram input, the control system for hand prosthesis was not implemented. Thus, an architecture of convolutional networks for the control of hand prostheses that can be trained with the signals of the subject.

Comparison on predictive capacity of ARIMA model and LSTM model for incidence of hand, foot and mouth disease in Shijiazhuang / 中华疾病控制杂志

Objective To predict the incidence of hand, foot and mouth disease (HFMD) in Shijiazhuang using the multiple seasonal autoregressive integrated moving average model (ARIMA) and long short term memory (LSTM) model, lay theoretical foundation for the prevention and control of HFMD. Methods Multiple seasonal ARIMA model and LSTM model were established separately by using Eviews 8.0 and python 3.7.1 according to the data of monthly incidence of HFMD from January 2013 to May 2018 in Shijiazhuang, and the data from June 2018 to May 2019 were used to verify the prediction precision of model. Finally, the monthly incidence from June to August 2019 was predicted. Results Based on the monthly incidence from January 2013 to May 2018, the optimal models, ARIMA(1,0,0)×(1,1,2)12 and LSTM model were established. Mean absolute percentage of error (MAPE) of ARIMA and LSTM model were 22.14 and 10.03 respectively based on the monthly incidence from June to December 2018, while MAPE of ARIMA and LSTM model were 43.84 and 25.26 respectively based on the monthly incidence from June 2018 to May 2019. These results indicated that LSTM model was superior to ARIMA model in model fitting degree and predicting accuracy, which was relatively consistent with the actual situation. Conclusions LSTM model is able to fit and predict the incidence trend of HFMD well in Shijiazhuang. It can provide guidance to HFMD epidemic prediction and alerting.

Study on the relationship between meteorological factors and the number of hypertension outpatients based on CNN-LSTM / 中华疾病控制杂志

Objective To study the effect of meteorological factors on the number of hypertension outpatients in four areas of Gansu Province, then predict and analyze the trend of the number of hypertension outpatients, so as to provide reference for the prevention and control of hypertension diseases. Methods On the basis of controlling the confounding factors such as long-term trends, date effects, meteorological information and contaminant influence, a mixed model of convolutional neural network (CNN) and long-short term memory (LSTM) was constructed for the number of hypertension outpatients in the four regions of Baiyin, Chengxian, Qingcheng and Liangzhou by Python programming language. Results The root mean square errors of the CNN-LSTM model for the number of hypertensive outpatients in the four regions was 6.330 9, 6.814 2, 6.393 6 and 6.867 6. The mean absolute percentage error was 74.082 2, 78.508 2, 56.618 3 and 50.235 4. And the average absolute errors was 4.875 7, 5.431 1, 4.542 0 and 6.460 8. All the results was superior to those of support vector machine (SVM), autoregressive integrated moving average model (ARIMA), random forest (RF), CNN and LSTM. Conclusion The CNN-LSTM model can accurately predict the number of hypertension outpatients in Gansu. The hospital can rationally allocate medical resources according to the needs of hypertension for medical treatment at different times.

Integration of A Deep Learning Classifier with A Random Forest Approach for Predicting Malonylation Sites / 基因组蛋白质组与生物信息学报·英文版

As a newly-identified protein post-translational modification, malonylation is involved in a variety of biological functions. Recognizing malonylation sites in substrates represents an initial but crucial step in elucidating the molecular mechanisms underlying protein malonylation. In this study, we constructed a deep learning (DL) network classifier based on long short-term memory (LSTM) with word embedding (LSTM) for the prediction of mammalian malonylation sites. LSTM performs better than traditional classifiers developed with common pre-defined feature encodings or a DL classifier based on LSTM with a one-hot vector. The performance of LSTM is sensitive to the size of the training set, but this limitation can be overcome by integration with a traditional machine learning (ML) classifier. Accordingly, an integrated approach called LEMP was developed, which includes LSTM and the random forest classifier with a novel encoding of enhanced amino acid content. LEMP performs not only better than the individual classifiers but also superior to the currently-available malonylation predictors. Additionally, it demonstrates a promising performance with a low false positive rate, which is highly useful in the prediction application. Overall, LEMP is a useful tool for easily identifying malonylation sites with high confidence. LEMP is available at http://www.bioinfogo.org/lemp.