Abstract
<title>Abstract</title> <p>Assessment of arterial stiffness is important in the diagnosis and management of cardiovascular diseases. A strong biomarker for arterial stiffness is pulse-wave velocity (PWV) – the speed at which a heartbeat-induced pulse wave travels through the large arteries. By measuring the pulse-transit time (PTT) – the time delay between the arrival of the pulse wave – at two points on the arterial pathway, PWV can be estimated. PTT can be calculated from skin-displacement signals, measured via laser Doppler vibrometry (LDV). This process requires an accurate indication of heartbeat onsets, to be able to utilize the correct signal features or fiducial points that lead to PTT. Typically, electrocardiography (ECG) is a reliable, parallel measurement to segment LDV signals. However, common practice for LDV does not include ECG due to increased measurement time and complexity, necessitating alternative means for identifying heartbeat cycles from LDV signals. A proposed solution is a deep neural network (DNN) which returns an ECG-proxy for each LDV signal, containing information on the timing of the heartbeats. Heart-carotid PTT can then be calculated after segmenting the signals based on the predicted heartbeat timing. This PTT was additionally calculated via two other techniques i.e. a state-of-the-art template matching algorithm and a reference technique in which PTT was calculated via manual inspection of the data. The DNN was able to correctly predict the onset of 68% and 77% of heartbeats present in the test and validation sets respectively. When applied to unseen data of 100 subjects (45 female), it provided PTT values for 64 subjects that correlated well with the reference (ρ = 0.71). In comparison, a state-of-the-art knowledge-driven approach yielded a similar correlation (ρ = 0.69) but was only able to return PTT values for 46 subjects. An ECGbased ground truth method reached a correlation of ρ = 0.69 for 90 subjects. Finally, the technique and rationale used for the DNN training demonstrates potential for broader biomedical LDV applications beyond heart-carotid PTT.</p>