State-Aware Contactless Respiratory Volume Monitoring for Breath-Hold-Guided Radiotherapy.

Abstract

Respiratory motion increases setup uncertainty in thoracic and abdominal tumor radiotherapy and elevates dose exposure to organs at risk (OARs), such as the heart and lungs. Deep inspiration breath-hold (DIBH) can reduce OARs dose, but clinical deployment requires accurate, real-time, low-burden monitoring of breath-hold stability and tidal volume. This study proposes a contactless tidal volume estimation system for DIBH-guided radiotherapy. A depth camera captured three-dimensional thoracoabdominal surface motion to construct a surface-based volumetric surrogate. A state-aware, surrogate-guided time-series regression model incorporating dynamic state gating, steady-state anchor modeling, adaptive baseline correction, and subject-specific calibration was developed for tidal volume estimation. In this prospective model-development and pilot internal-validation study, synchronized depth-camera data and tidal volume signals from an Active Breathing Coordinator (ABC) system were collected from 22 volunteers. ABC-derived tidal volume served as the reference signal, and performance was evaluated using subject-independent leave-one-subject-out cross-validation. Real-time respiratory feedback was provided through a head-mounted visualization interface. Predicted tidal volume strongly correlated with the ABC reference signal (pooled r=0.968; subject-level Fisher z-transformed mean r=0.966, 95% CI: 0.954-0.975, p<0.001), with an overall MAE of 0.098 L. Phase-specific MAE/MAPE were 0.1059 L/17.27% during breath hold and 0.0967 L/16.68% during normal breathing. GPU-side algorithmic latency was 12.55 ms per frame. The proposed contactless framework enables stable tidal-volume estimation and real-time visual feedback during DIBH while minimizing patient encumbrance, supporting low-burden respiratory monitoring for radiotherapy workflows.