A Deep Learning Model Based on the Bi-LSTM Neural Networks with Self-attention Mechanism to Enhance the Blood Flow Sensor's Ability to Track Fluctuating Flow Rate.

Blood flow sensors based on thermal sensing principles are widely used for health monitoring, but they produce large errors when tracking fluctuating blood flow rate (BFR). In order to enhance the ability of these sensors to track BFR, this study establishes a deep learning model based on the bidirectional long short-term memory network with self-attention mechanism (SA-BiLSTM-BFR). The model effect is quantified and evaluated using the middle cerebral artery BFR dataset. Compared with traditional models, the SA-BiLSTM-BFR model reduces the flow prediction root mean square error (RMSEBFR) by 36.8%, while almost completely eliminating the sensing delay. The results shows that the SA-BiLSTM-BFR model is a highly significant factor in promoting the enhancement of blood flow sensing ability, expanding the applications of thermal-based blood flow sensors.Clinical Relevance- This reduces the measurement error of thermal-based wearable blood flow sensors by close to 40%, thereby improving the reliability of haemodynamic monitoring.