Construction and hardware validation of a transition-aware quantum kernel for pre-hypotensive PPG and ECG

Abstract

Full quantum-kernel estimation requires a quadratic number of pairwise fidelity evaluations, which limits hardware studies on public cloud QPUs. We constructed paired stable and pre-hypotensive observations from VitalDB and represented each observation by its current physiological state and the change across two adjacent 60-s windows. Development-only analysis selected a 30-s prediction gap and four PPG/ECG-derived features: PPG amplitude, PPG peak width at half prominence, pulse interval, and pulse-arrival time. The Hemodynamic Transition Quantum Kernel (HTQK) encodes four state values with Ry[jls-end-space/]rotations and their four corresponding changes with Rz[jls-end-space/]rotations on the same four qubits, separated by an internal nearest-neighbour CX bridge. This placement avoids the cancellation that affected a fixed terminal entangler in the original compute–uncompute circuit. The locked kernel was then validated through exact, repeated finite-shot, backend-derived-noise, and physical-QPU stages. Physical validation was restricted to a deterministically selected subset of 16 development and eight holdout observations. The entangled kernel achieved holdout balanced accuracy of 0.623 (95% bootstrap CI 0.576–0.674). On ibm_marrakesh, 252 circuits used 148 QPU s. Relative Frobenius distortion from the exact training kernel was 0.0420 and centred kernel alignment was 0.9992. The method evaluates resource-efficient kernel preservation rather than quantum computational advantage. This is an open access article under the CC BY license. http://creativecommons.org/licenses/by/4.0/