Non-contact PPG measurement system incorporating image analyzed photoplethysmogram signals and deep learning algorithms

We designed a new non-contact photoplethysmographic (PPG) measurement system by adopting an imaging-PPG (iPPG) method which used specific wavelengths detection and signal processing algorithms. PPG signals can be used to obtain information about blood oxygen saturation and provide diagnostic data on cardiovascular condition and atrial fibrillation (AFib). Most traditional PPG detection methods requires contacting sensors to the measuring surface. In our study, we developed a non-contact iPPG system with its high-potential performance to reduce the processing time. The selective narrow-band filter and incorporated an active illumination lighting array are used to designed according to the blood absorption spectrum. After collecting a series of iPPG images of a preset duration, the iPPG signals were remotely analyzed using traditional methods and using our deep learning (DL) algorithms. The DL algorithm based on a long short-term memory (LSTM) model was developed to fulfill waveform improvement. An InGaAs camera and a monochrome Si camera, both set at a 50fps frame rate, were used for field image detection. The measurement data collected at 550nm wavelength are shown in the discussion. Several indicators such as heart rhythm, peak-to-peak interval error, similarity, and root mean squared errors (RMSE) were adopted to compare the DL detection iPPG signal with the ECG and PPG signals. Our newly developed iPPG system has a high potential application for personnel health monitoring.