Author: Marian Axente, Mandeep Kaur 👨🔬
Affiliation: Emory University 🌍
Purpose: To validate a low-cost optical imaging system for respiratory monitoring by comparing its accuracy and feasibility against the clinical gold standard in human subjects.
Methods: Following its initial development and phantom-based validation, we validated the proposed system in healthy volunteers. Utilizing machine learning detection algorithms, the system integrates depth and optical imaging from a commercially available camera, capable of independent 3D motion tracking at 13 ± 4 Hz with a latency of 60.73 ± 0.74 ms. VRT and LNG motion is tracked within IEC 61217 fixed DICOM coordinates. To assess its performance for human respiratory monitoring in the absence of ground truth, we benchmarked the system against the clinical gold standard (VARIAN Identify). Four healthy volunteers (supine on a breast board with arms raised) performed normal breathing, coached breathing, and breath-holds across three recording sessions. Both systems simultaneously tracked respiratory motion, and data was analyzed using Bland Altman plots comparing paired measurements. Mean differences reflected systematic bias, while confidence intervals (95% data within limits of agreement) captured variability across and within subjects.
Results: The proposed system demonstrated strong agreement with the clinical gold standard for VRT motion, with a pooled mean difference of -0.01 mm and 95% limits of agreement between -0.96 mm and 0.93 mm. LNG motion showed a slightly larger pooled mean difference of 0.17 mm, with 95% limits of agreement ranging from -2.83 mm to 3.16 mm. Subject-specific analysis indicated minimal systematic bias across both directions, with greater variability observed in LNG motion due to sensitivity to depth signal noise.
Conclusion: The proposed system showed good agreement with the clinical gold standard, while demonstrating realtime tracking capabilities for respiratory motion in human subjects. Expanding testing to larger cohorts will refine the uncertainty estimates.