Fidelity-Derived Quantum Dissimilarity-Enhanced k-Nearest Neighbor Algorithm for Arterial Hypertension Detection

We present a quantum-enhanced version of the classic k-Nearest Neighbors (kNN) classification algorithm, applied to the detection of arterial hypertension. The traditional Euclidean distance metric of the kNN algorithm is replaced with a fidelity-derived quantum dissimilarity measure to evaluate the similarity between data samples. We map classical real-world clinical and ECG-derived data features into quantum states via the dense-angle encoding, which efficiently packs multiple features into minimal qubits while maintaining pure states. We evaluate the performance of the dissimilarity measure using both the noiseless state-vector Simulator and the IBM Qiskit Estimator primitives on 25 test sets. We found that the quantum approaches present a lower mean number of neighbors (6.8 ± 4.1), while the classical approach has a mean of (7.9 ± 4.4) neighbors. The classical kNN algorithm achieves a balanced accuracy of (70.9 ± 2.9)% and an F1-macro of (71.5 ± 3.0)%, while the quantum algorithm scores (67.9 ± 2.4)% and (68.3 ± 2.4)%, respectively, indicating a marginally non-statistically significant difference with a p-value of 6.6%. In addition, the classical kNN algorithm with cosine similarity achieves performance equivalent to the quantum dissimilarity metric, with a p-value of 75.5% for the F1-macro metric. While the quantum algorithm does not claim computational supremacy over the classical baseline, the framework provides a promising simulation-based proof of concept that fidelity-based similarity is a physically meaningful approach to hybrid quantum-classical classification.