Prediction of H-type hypertension based on pulse-taking and inquiry diagnosis

Abstract

H-type hypertension increases the morbidity of stroke and cardiovascular diseases, posing a great threat to human health. The existing diagnosis of H-type hypertension detects patients’ plasma homocysteine content, which is inefficient and leaves wounds. The pulse-taking diagnosis of traditional Chinese medicine (TCM) can implement the non-invasive auxiliary diagnosis of H-type hypertension by the physiological activities of patient’s pulse wave combined with the clinical inquiry information. Therefore, we propose a combined model of pulse-taking and inquiry diagnosis, which includes a pulse-taking model based on CNN-BiLSTM and an inquiry diagnosis model based on the integrated Cluster-RFs. We search the optimal weights for the heterogeneous integrated model dynamically by using Grid Search to classify H-type hypertension. The datasets are the collection of 364 cases H-type hypertension from Longhua Hospital affiliated to Shanghai University of Chinese Medicine and Hospital of Integrated Traditional Chinese and Western Medicine concerning pulse-taking and inquiry diagnosis. The evaluation indicator of classification of H-type hypertension, i.e. Sensitivity, Specificity, Accuracy, F1-score, AUC, is 82.31%, 73.75%, 79.34%, 85.92%, 87.78% respectively, is higher than that of the other typical machine learning models on pulse-taking or inquiry diagnosis. In addition, this article also studies the correlation between pulse-taking or inquiry diagnosis and its attributes, and analyzes the feature importance ranking on pulse-taking and inquiry diagnosis, which aids clinicians seek the occurrence mechanisms of H-type hypertension, and find the reasonable measurements for timely prevention and treatment.