QUANTUM-INSPIRED MACHINE LEARNING FRAMEWORK FOR HEALTHCARE DECISION SUPPORT SYSTEMS

Abstract

Healthcare decision support demands models that are accurate on heterogeneous clinical data, fast enough for the point of care, and transparent enough to earn clinician trust. Classical learners meet parts of this brief but struggle to capture the high-order correlations that bind physiological variables together, and they typically expose little of their internal reasoning. This paper introduces QIDSS, a quantum-inspired machine-learning framework that runs entirely on commodity classical hardware yet borrows the mathematical machinery of quantum computation—amplitude encoding, parameterised unitary evolution, and entanglement-style feature interaction—to enrich the representation of clinical evidence. A patient record is mapped into a normalised state vector, a shallow variational circuit emulated through tensor operations transforms that state, and a Born-rule measurement yields a calibrated risk estimate together with a per-feature attribution. Training proceeds by a parameter-shift gradient over the emulated circuit, combined with cross-entropy and an entanglement-regularisation term that discourages spurious correlations. We evaluate the framework on a consolidated clinical dataset and benchmark it against logistic regression, gradient-boosted trees, a deep neural network, and a simulator-based variational quantum classifier. QIDSS reaches 94.5% accuracy with matching gains in precision and recall while holding average decision latency to 9.2 ms, comfortably below the quantum baseline. Beyond the headline numbers, the framework surfaces the features that most influenced each verdict and routes low-confidence cases to clinician review. Implementation overhead, qubit-count scaling, and deployment constraints are examined candidly.