Surface-guided breathing signal integration in breathing-adapted intelligent 4D computed tomography: prototype implementation and comparison with an infrared marker-based system.

Background and purpose: Breathing-adapted intelligent four-dimensional computed tomography (i4DCT) reduces motion artifacts during irregular breathing using real-time surrogate signal analysis to control X-ray timing. Current implementations rely on infrared (IR) markers or pressure belts. Surface-guided radiation therapy (SGRT), a widely used markerless technique in radiotherapy, has not been evaluated for direct i4DCT integration. This study tested SGRT for i4DCT acquisition in a phantom setup and benchmarked it against an infrared marker-based system. Materials and methods: Phantom measurements with clinically relevant regular and irregular breathing signals were performed on a prototype CT scanner with direct SGRT control. To improve SGRT-based surrogate accuracy, a table motion correction profile was empirically modeled, and a vendor-specific prediction algorithm was implemented. Surrogate signal accuracy, latency, and motion reconstruction accuracy were compared between SGRT and an established IR system using amplitude- and phase-based reconstructions. Results: SGRT exhibited an absolute latency of ∼ 63  ms, compared to ∼ 23  ms for the IR system. Across regular and irregular breathing signals, SGRT breathing signals showed Root Mean Square Error (RMSE) values up to ∼ 1.5  mm, while correlation with the ideal signal remained high (r = 1). Tumor center-of-mass deviations in amplitude-based reconstructions reached 1.5  mm without prediction at 20 breaths-per-minute and 15  mm amplitude, but reduced to < 0.5  mm with 50  ms prediction. Both amplitude- and phase-based reconstructions showed improved agreement with ideal motion when prediction was applied, with phase-based reconstructions yielding better accuracy. Conclusions: These findings support SGRT as a clinically viable non-contact alternative to IR tracking in i4DCT, especially when combined with motion correction and predictive modeling.