Rede neural convolucional u-net para inferência do sinal eletrocardiograma a partir do sinal fotopletismograma

Abstract

To measure the cardiac cycle and obtain heart rate measurements, there are two widely used methods: the electrocardiogram (ECG) and the photoplethysmogram (PPG). The sensors used in these methods have gained great popularity in wearable devices, which has extended cardiac monitoring beyond the hospital environment. In this sense, wearable devices have become a strong ally in the continuous monitoring of cardiac vital signs, being considered important tools to assist in the early identification of heart diseases. However, monitoring the ECG signal continuously via mobile device is still a problem, as it requires the user to keep their fingers pressed on the device to form closed circuits during data collection, which makes monitoring the ECG signals unfeasible in the long term. On the other hand, the PPG does not contain this limitation, but the medical knowledge to diagnose these anomalies from this sign is limited by the lack of familiarity, since the ECG is studied and used in the literature as the gold standard. To minimize this problem, this work proposes a method that uses the correlation between domains of PPG and ECG signals to infer from the PPG signal the waveform of the ECG signal. The proposed method, called PPG2ECG, consists of mapping between domains through the application of a set of convolution filters, learning to transform a PPG input signal into an ECG output signal. To perform this transformation, the PPG2ECG method uses a U-net Inception neural network architecture that performs convolutions on different filter sizes in parallel. For the evaluation of the proposed method, two evaluation strategies based on the personalized and generalized models were used. The results show the mean error value (MSE) of 0.015 and 0.026 for the custom and generalized models, respectively. The results prove the feasibility of the method of mapping the PPG signal to the ECG, due to the short distances between the inferred ECG and the original ECG.