A prediction model of blood pressure for telemedicine
In this article, a prediction model for High Blood Pressure was generated for telemedicine purposes, and people that do not take regular BP measurements. The model generates an alarm based on the Systolic Blood Pressure predicted value. The authors proposed two models one based in a multilayer perceptron neural network, and another one base on a radial basis neural network. The data was collected from the health and body conditions of 498 people and included features such as systolic blood pressure, gender, age, body mass index, smoking status, exercise level, alcohol consumption level, stress level, and salt intake. According to the results of the article, the accuracy of the multilayer perceptron was 94.28% with 4 hidden nodes while the redial basis function had an accuracy of 91.06% with 5 hidden nodes.
Noninvasive classification of blood pressure based on photoplethysmography signals Using Bidirectional Long Short-Term Memory and Time-Frequency Analysis.
The authors proposed the use of time-frequency analysis and bidirectional long short-term memory neural network to classify blood pressure. The dataset was collected from 219 adult subjects aged from 21 to 86 years. Also, 900 recorded data values from the PPG-BP figshare database were added. By using a time-frequency analysis for feature extraction, the author claim that the training time of the bidirectional LSTM neural network is reduced instead of inputting the whole-time domain sequence of the PPG signal, also the accuracy is improved. The classification process as divided in three categories normotension, prehypertension, and hypertension. Accuracy, sensitivity, and specificity values were of 97.33%, 100%, and 94.87%, respectively. The F1 scores of the three classes were of 97.29%, 97.39%, and 93.93%, respectively
Prediction of Incident Hypertension Within the Next Year: Prospective Study Using Statewide Electronic Health Records and Machine Learning
Given the increase of available EHR data, this paper aims to construct and validate a new hypertension risk prediction model taking one year of the patient clinical information. The analyzed dataset was taken from the EHTs of the state of Maine, gathering data from 35 hospitals, 34 health centers, and more than 400 ambulatory practices. This dataset covered almost 95% of the population of Maine and is a subset of the health information exchange (HIE) network provided by the HealthInfoNet organization. To generate the model, first, the k-nearest neighbors’ algorithm was used to fill the missing data from the database. To remove features that were not related to the outcome variable, a univariate correlation filtering was applied to the features of the dataset. The chosen machine learning algorithm for the prediction of hypertension was an XGBoost algorithm. The 1-year incident hypertension risk model attained areas under the curve (AUCs) of 0.917 and 0.870 in the retrospective and prospective cohorts, respectively.
A hybrid neural network model for predicting kidney disease in hypertension patients based on electronic health records
DOI: 10.1186/s12911-019-0765-4
This work proposes a hybrid deep learning model for the prediction of kidney disease on patients suffering from hypertension. The proposed model was a bidirectional long short-term memory and an Autoencoder. The bidirectional long short-term memory neural network was used to classify text sequence from HER, while the Autoencoder was used to classify numerical data from the EHR. The output of both classifiers is merged with a fully connected layer. This model achieved an accuracy of 89.70%.
Predicting increased blood pressure using machine learning.
doi:10.1155/2014/637635
This work is interested in developing a model based on a decision tree to demonstrate the availability of classifying pre hypertense or hypertense patient based on features such as body mass index (BMI), waist (WC) and hip circumference (HC), and waist-hip ratio (WHR), The data was collected and published by the same authors in the figshare repository, and according to the authors, the decision three models achieved a sensibility: 72%, specificity: 86.25%, and AUC: 0.688.
Development of disease prediction model based on ensemble learning approach for diabetes and hypertension.
In this article, type 2 diabetes and hypertension data sets were used to generate an ensemble machine learning model to prevent or predict the risk factor of both diseases. The data set attributes for the hypertension model are age, obesity, body mass index, waist, and hip circumference. The data come from the Men’s and Women’s Dataset From the “Predicting Increased Blood Pressure Using Machine Learning” paper available in the figshare repository. The proposed model uses a combination of a multilayer perceptron, a support vector machine, a decision tree, and logistic regression achieving an accuracy of the ensemble model of75.78% and are under the curve 0.76.
The Prediction of Hypertension Based on Convolution Neural Network
This proposal implements a Convolutional Artificial Neural Network for the prediction of hypertension. The data was taken from the MIMIC II Waveform Database Matched Subset of Pyshionet which contains data from arterial blood pressure, heart rate, respiration, and oxygen levels. Furthermore, the work presents a comparison of other machine learning techniques such as k-Nearest Neighbor, random forest, and logistic regression been the CNN the one achieving the best performance in terms of accuracy. The accuracy of the CNN was tested with different activation functions, with the ReLu achieving the highest accuracy of 91.7%.
Using machine learning to predict hypertension from a clinical dataset (LaFreniere, 2016)
In this article, a feedforward neural network was proposed to predict the presence of hypertension in an individual. The model was generated using the Matlab Neural Network Toolkit. The training and test data were taken from the Canadian Primary Care Sentinel Surveillance Network (CPCSSN) data set. According to the authors, the model achieved an accuracy of 81%.