Derived physiological features and IBM QPU measurements for a transition-aware quantum kernel workflow for pre-hypotension analysis
Description
This dataset contains derived physiological features and experimental results supporting a transition-aware quantum kernel workflow for analysis of pre-hypotensive physiological changes. Source recordings were obtained from the publicly available VitalDB version 1.0.0 dataset. The released material includes cohort-selection and signal-quality-control records, features extracted from paired stable and pre-hypotensive observations, development and non-overlapping holdout assignments, multi-gap feature representations, locked four-feature quantum inputs, classical model-validation results, exact and finite-shot quantum-kernel results, backend-derived noise simulation, IBM Quantum backend and physical-qubit metadata, raw QPU measurement counts, hardware kernel matrices, and final kernel-preservation statistics. The final HTQK representation uses PPG amplitude median, PPG width at half height, pulse-interval median and pulse-arrival-time median with a 30-s pre-event gap. State and temporal-transition components are encoded on four qubits. Physical hardware validation was performed on IBM Marrakesh using physical qubits 53, 54, 55 and 59 at 2048 shots per circuit. Original VitalDB waveform data are not redistributed. Public source data are available from PhysioNet under DOI 10.13026/czw8-9p62.
Files
Steps to reproduce
1. Obtain VitalDB version 1.0.0 from PhysioNet (DOI 10.13026/czw8-9p62) if waveform-level regeneration is required. 2. Use `01_cohort_and_qc/` to inspect cohort selection, 180-s context availability and multi-gap QC. 3. Use `02_derived_features/features_multigap.csv` to inspect the 0/30/60-s development feature representations. 4. Use `03_method_lock/` and `02_derived_features/features_locked.csv` for the frozen 30-s four-feature representation and quantum angles. 5. Use `04_development_validation/` to reproduce reported development summaries and finite-shot aggregation. 6. Use `05_holdout_validation/holdout_validation_final.csv` plus the prediction files to verify holdout metrics. 7. Use `06_backend_noise/` and `07_ibm_qpu/` to verify backend-noise and physical-QPU kernel preservation. 8. `07_ibm_qpu/qpu_measurements_v2.csv` contains the raw bitstring counts from all 252 physical circuits.
Institutions
- Petro Mohyla Black Sea National UniversityMykolaiv, Mykolaiv