Robust respiratory signal extraction from cone-beam CT projections using projection enhancement and statistical analysis of breathing variability

Respiratory motion signals extracted from Cone-Beam CT (CBCT) projections can be treated as physiological time-series signals with breathing information embedded within imaging data. However, the robustness of existing projection-based respiratory signal estimation methods under clinically relevant breathing variability and anatomical visibility changes remains insufficiently characterized. This work presents a systematic robustness evaluation of four widely used data-driven respiratory signal estimation algorithms under heterogeneous breathing conditions, including eupnea, bradypnea, tachypnea, ataxic breathing, and diaphragm occlusion. Amsterdam Shroud (AS), Local Principal Component Analysis (LPCA), Intensity Analysis (IA), and Fourier Transform-Phase (FT-P) were evaluated together with three AS image enhancement techniques (foreground extraction, adaptive z-normalization, and composite enhancement), a projection-based Wiener filtering enhancement, and six anatomically justified regions of interest (ROI). Estimation accuracy was assessed using a local correlation metric on five MRI-based digital phantoms and twenty clinically acquired CBCT projection datasets. Statistical significance among method combinations was evaluated using a multiple-comparison Kruskal-Wallis analysis. Composite enhancement improved both AS and LPCA methods, with LPCA + C achieving a correlation value of 0.84. Projection enhancement further improved AS-based methods, with AS + C + WF achieving 0.88, but reduced LPCA performance by 7.2 %, indicating sensitivity to contrast-preserving image characteristics. ROI analysis showed FT-P achieved the highest correlation (0.91) when diaphragm visibility was maintained, although enhanced AS and IA methods formed a closely performing cluster (0.89). Overall, the enhanced AS framework demonstrated the most consistent performance across breathing pattern variability, diaphragm occlusion conditions, and ROI selections, supporting its suitability for robust respiratory signal estimation in CBCT-guided radiotherapy applications.

Actions (login required)

View Item