Lowering blood pressure using guided breathing with reinforcement learning

Hypertension is a major health risk, often managed through pharmacological means, which can have side effects. Guided breathing offers a promising non-pharmacological alternative, yet existing methods typically rely on fixed breathing rates with limited success. In our work we propose a machine-learning approach to lowering blood pressure through personalized guided breathing, using reinforcement learning to adapt breathing rates in real time. Addressing the limitations of fixed-rate methods, a closed-loop system was developed that provides breathing cues based on physiological feedback constructed from EEG, ECG, and PPG signals. The system was tested across three experimental stages and provided superior outcomes in systolic blood pressure reduction than common fixed-rate approaches. This research demonstrates the feasibility of machine learning based guided breathing as a non-pharmacological option for hypertension management, contributing to advances in digital healthcare.

M.Sc. student under the supervision of Prof. Danny Lange.