Human-in-the-Loop Cybernetic Integration of Coprime-Array AQA Learning and Beam-Adaptive Imaging for Wireless Dental Plaque Quantification

Abstract

Advancing health diagnostics requires systems that coordinate heterogeneous sensing, communication, and inference under real-world uncertainty. This work presents a cyber--physical--social architecture for wireless dental-plaque quantification. The system integrates three core components: (i) machine-learning-based Angle-of-Arrival (AoA) estimation using coprime antenna arrays, (ii) adaptive beamforming for directional reception, and (iii) a UV-based imaging pipeline fed by wirelessly transmitted intraoral images. The AoA subsystem employs a Random Forest regressor trained on narrowband snapshots, enabling the receiver to dynamically steer its beam toward the emitting endoscope---a closed-loop cybernetic mechanism that compensates for multipath, patient movement, and hardware impairments (phase drift, gain imbalance, I/Q mismatch) typical of clinical environments. The estimator meets clinically motivated thresholds (MAE , RMSE , ) even at the low refresh rates (1-10 Hz) of power-constrained medical devices. Improved link reliability from beam alignment directly enhances the quality of received UV dental images, which are processed via CIE-LAB transformation, UV enhancement, and adaptive thresholding for plaque quantification. Clinician-guided contouring introduces a human-in-the-loop regulatory pathway that ensures robustness to occlusion, saliva reflections, and anatomical variability. By coupling ML-driven AoA detection, adaptive beamforming, and clinician-guided analytics, the proposed architecture forms a multi-level feedback system demonstrating how convergent Systems Science can deliver scalable, objective, and preventive oral-health assessment within a Learning Health System framework.