Quantum Reasoner Agents for Adaptive Medical Imaging with Hybrid Agentic Quantum Machine Learning Framework

Quantum computing, agentic artificial intelligence (AI), and machine learning (ML) are converging to reshape adaptive medical imaging. This paper presents a hybrid Agentic Quantum Machine Learning (AQML) framework featuring a Quantum Reasoner Agent (QRA), a specialized sub-agent that performs high-dimensional optimization, kernel reasoning, and uncertainty modelling using variational quantum circuits within an LLM-orchestrated agentic pipeline. To address the documented limitations of outdated baselines and insufficient benchmarking in prior quantum medical imaging work, the AQML framework is evaluated against four modern architectures on an identical 500-slice BraTS 2023 subset. AQML achieves Dice = 0.910 and AUC = 0.970, outperforming nnU-Net (Dice 0.890), SwinUNet (0.876), TransUNet (0.872), and UNet (0.860). Entropy-aware regularization reduced mean quantum entropy from 0.48 to 0.37, and uncertainty calibration improved by 12% over the classical UNet baseline. IBM Quantum hardware characterization (ibmq_toronto: mean T1 = 91.5 μs, mean CNOT error = 1.08%) provides a concrete noise baseline for near-term feasibility planning. The Entanglement-Aware Regularization (EAR) term and the LLM-governed agentic feedback loop are mathematically specified in full, with Algorithm 1 detailing the complete Adam-SPSA training protocol and three structured refinement actions. The explainability layer and multi-modal fusion across CT and PET modalities are architectural design features pending formal clinical validation, as explicitly acknowledged in the limitations. This framework lays the foundation for scalable, federated, and ethically grounded quantum-intelligent radiology systems.