1229 - Effect of a Novel Quantitative Cone Beam CT Technique on Soft Tissue Boundary Visualization: A Prospective Interobserver Study
Presenter(s)
U. Pyakurel1, R. C. Bliley IV1, F. Bayat1, I. Boor1, J. Hu2, A. Avant1, J. D. Duryea3, R. Sabounchi1, M. Miften1, C. M. Fisher1, R. A. Rabinovitch1, D. C. Binder1, R. M. Lanning1, T. P. Robin1, B. D. Kavanagh1, and C. Altunbas1; 1Department of Radiation Oncology, University of Colorado School of Medicine, Aurora, CO, 2Cancer Biostatistics Core, University of Colorado School of Medicine, Aurora, CO, 3University of Colorado School of Medicine, AURORA, CO
Purpose/Objective(s): Soft tissue visualization in cone-beam CT (CBCT) is inferior to thin-slice planning CT, limiting confidence in target localization and contouring for CBCT-based image guidance and dosimetric evaluation in adaptive radiotherapy (RT). To address this limitation, a quantitative CBCT (qCBCT) method incorporating robust scatter suppression techniques was developed and implemented. A prospective imaging and interobserver study were conducted to assess the effect of qCBCT on soft tissue boundary visualization in comparison to standard of care CBCT.
Materials/Methods: Twenty-nine radiotherapy patients were enrolled into an IRB-approved imaging study and stratified into prostate, pelvis or abdomen, and head and neck (H&N) cohorts. Each participant was scanned once with qCBCT and once with the standard CBCT (sCBCT) protocol on a conventional linac. CBCT scans incorporated an in-house–developed two-dimensional anti-scatter grid prototype and associated correction algorithms. Both scans used identical acquisition parameters and imaging dose. Twenty-one anatomical structures were contoured on the CBCT images and the planning CT sets by four observers in a blinded setting. A consensus contour was generated for each structure by combining the contours of four observers. Similarity between each observer’s contour and the consensus contour was quantified using volumetric and distance-based similarity metrics. The study endpoint was to evaluate the effect of CBCT imaging method on interobserver contour similarity, a surrogate for soft tissue boundary visualization. Differences between imaging methods were analyzed using a linear mixed-effects model.
Results: A total of 1,584 structures were contoured across all image sets by all observers. In general, qCBCT improved interobserver contour agreement across all anatomical regions. Mean Dice coefficients for sCBCT, qCBCT, and planning CT were 0.87 (95% CI: 0.85-0.90), 0.91 (95% CI: 0.88-0.93), and 0.93 (95% CI: 0.91-0.96) respectively (p < 0.0001). Corresponding Hausdorff distances were 4.9 (95% CI: 3.7-6.4) mm, 3.8 (95% CI: 2.9-4.9) mm, and 3.0 (95% CI: 2.3-4.0) mm respectively (p < 0.01). Analyses using mean surface distance and normalized volume difference yielded similar results, with qCBCT showing higher interobserver contour similarity than sCBCT (p<0.01). Improvement in interobserver agreement in qCBCT compared to sCBCT was most pronounced in the H&N cohort.
Conclusion: qCBCT improved soft tissue boundary visualization compared with sCBCT and was particularly beneficial for organs and targets in the H&N region, as demonstrated by higher interobserver similarity of structure contours. While planning CT consistently provided superior soft tissue boundary visualization, qCBCT markedly improved visualization and narrowed the gap between planning CT and CBCT. This study also demonstrated that the qCBCT technique can be implemented with relatively small modifications to the CBCT system.