3196 - Spine IGRT Quality Control Using Population-Based SPC

12:30pm - 01:45pm ET

Poster Hall - Exhibit Hall A

Screen: 10

POSTER

Presenter(s)

Cenji Yu, PhD - Mayo Clinic College of Medicine and Science Rochester, Rochester, MN

C. Yu¹, L. E. Fong de los Santos², D. J. Moseley³, and S. Shiraishi⁴; ¹Mayo Clinic College of Medicine and Science Rochester, Rochester, MN, United States, ²Mayo Clinic, Rochester, MN, United States, ³Mayo Clinic, Rochester, MN, ⁴Department of Radiation Oncology, Mayo Clinic, Rochester, MN

Purpose/Objective(s):

Accurate vertebral level alignment is critical for spine image-guided radiation therapy (IGRT), yet registration review remains largely qualitative and susceptible to inter-observer variability. This study proposes a quantitative framework using normalized cross-correlation (NCC) combined with statistical process control (SPC) to detect vertebral body misalignment in CT–CBCT image registration.

Materials/Methods:

Phantom experiments were performed to characterize NCC sensitivity to vertebral body shifts across multiple CBCT platforms and imaging protocols. Retrospectively, NCCs were calculated for 2,796 CT–CBCT image pairs from spine radiation therapy patients. Population-based SPC control charts were constructed using each patient’s highest NCC per treatment course to establish baseline process behavior. One-level vertebral body mis-registrations were simulated in 20 patients using box-based automatic registration to reflect realistic clinical failure modes. Receiver operating characteristic (ROC) analysis was used to evaluate misalignment detection performance.

Results:

Phantom validation demonstrated that NCC variability across imaging protocols was substantially smaller than NCC reductions associated with vertebral body shift. For the retrospective clinical cohort, the population-based SPC chart achieved a mean NCC of 0.91 (SD = 0.07), with a calculated lower control limit of 0.70. Excluding patients with metal hardware produced a tighter control chart with a lower control limit of 0.74. ROC analysis demonstrated 100% sensitivity for misalignment detection with a specificity of 97%.

Conclusion:

The NCC-based SPC framework provides a robust, quantitative method for population-level monitoring of spine IGRT registration quality. This approach detects vertebral body misalignment while offering safeguards against automation bias. Future work will focus on identifying and analyzing true vertebral level misalignment events from clinical practice to further validate the proposed detection framework.