Bi-LSTM for Non-Invasive Forecasting of Hypotension Using Photoplethysmography Signals

Intraoperative hypotension is a major risk factor for postoperative complications, yet current hemodynamic monitoring remains invasive, costly, and limited in scope. Digital plethysmography offers a non-invasive alternative, but accurately predicting hypotensive episodes in real time remains challenging due to its complex and dynamic nature. In this work,
we propose a BiLSTM-based approach to effectively capture both short- and long-term temporal dependencies in plethysmography signals, ensuring robust prediction of hypotension. Its bidirectional structure enables comprehensive feature extraction, allowing anticipation of hypotensive episodes up to five minutes before onset. We evaluated our approach on the VitalDB dataset, where it demonstrated superior performance compared to state-of-the-art methods. This approach enhances predictive hemodynamic monitoring, enabling proactive intervention
and improving surgical outcomes.