Prediction of Ankle Brachial Index with Photoplethysmography Using Convolutional Long Short Term Memory

Abstract

Early detection is critical for effective prevention of cardiovascular disease. One of the representative indicators of cardiovascular disease is the ankle-brachial index (ABI). It is mainly used to measure artery disease in primary care. However, the ABI measurement is difficult because the patient must wear cuffs on four limbs. The purpose of this study is to predict ABI using photoplethysmography (PPG), to overcome this difficulty. PPG is known to be closely correlated with cardiovascular conditions. An ABI prediction model based on deep learning is proposed, as it does not require feature extraction from the PPG signals. The ABI values are classified into six classes depending on the cardiovascular disease severity, and the ABI class is predicted by the designed deep learning model. In this study, a convolutional long short term memory (C-LSTM) model consisting of five convolutional layers, five pooling layers, and one LSTM layer was designed. As a result of evaluating the performance of the C-LSTM model, the accuracy was 98.3429% and the F1 score was 97.4293%. Therefore, this model achieves high performance. The method proposed in this study is a novel method for predicting the ABI class using PPG signals that can be easily measured. The proposed model can classify ABI class automatically without feature extraction. The proposed model enables fast and simple evaluation of the cardiovascular disease in primary care without requiring an ABI measuring instrument.