Q-LCAN: Anatomy-Constrained Quantum-Agentic Systems for MRI-Derived Structural Surgical Digital Twins in Uterine Adenomyosis.

Abstract

Accurate patient-specific structural assessment of adenomyosis is essential for fertility-preserving treatment planning, but pelvic T2-weighted MRI is difficult to parse because the uterus, endometrium, ectopic glands, hemorrhagic foci, and adenomyoma are nested, overlapping, and highly imbalanced. We propose Q-LCAN, an anatomy-constrained Quantum-Agentic System that converts MRI into a structured surgical digital twin. A dense 3D decoder first estimates target-specific probability maps and signed distance fields. Five target agents then exchange visual evidence under predefined anatomical axioms, while quantum-enhanced differentiable action projection (Q-DAP) calibrates restricted expand, shrink, or hold actions before vision-language level-set control. A detached anatomical correction module and an independent audit reduce and report residual violations. On 658 multicenter examinations, Q-LCAN achieved case-level five-target macro-average DSC/ASSD of 0.814/2.23 mm in internal five-fold validation and 0.802/2.33 mm in external testing. Compared with the strongest protocol-aligned baseline, Q-LCAN improved macro DSC by 0.018 and 0.019 and reduced ASSD by 0.29 and 0.31 mm, with gains concentrated in ectopic glands, hemorrhagic foci, adenomyoma burden, lesion-to-endometrium distance, residual lesion ratio, and candidate resection-boundary distance. Q-LCAN provides an auditable quantum-agentic framework for MRI-derived structural twins; its millimeter-scale descriptors are intended as preoperative decision support, not autonomous operative guidance.