Tumor phase recognition using cone‐beam computed tomography projections and external surrogate information

• DOI: 10.1002/mp.14298
• Publication Date: 2020-05-28T20:44:18Z
• Authors: Pingfang Tsai; Guanghua Yan; Chihray Liu; Ying‐Chao Hung; Darren L. Kahler; Ji‐Yeon Park; Nick Potter; Jonathan G. Li; Bo Lu
• Keywords: Robustness

Purpose Directly extracting the respiratory phase pattern of tumor using cone‐beam computed tomography (CBCT) projections is challenging due to poor visibility caused by obstruction multiple anatomic structures on beam's eye view. Predicting information external surrogate also has intrinsic difficulties as patterns between surrogates and tumors are not necessary be congruent. In this work, we developed an algorithm accurately recover primary oscillation components motion combined from both CBCT surrogates. Methods The involved two steps. First, a preliminary was acquired applying local principal component analysis (LPCA) cropped Amsterdam Shroud (AS) images. step, only image region used extract in order minimize impact recognition other structures. Second, performing multivariate singular spectrum (MSSA) containing signal original waveform first recovered. For phantom study, QUASAR with removable tumor‐simulator insert employed acquire projection A comparison LPCA our method assessed power analysis. Also, simulated under shift or various amplitude conditions examine robustness method. Finally, scenarios were attaching heart model, PVC tubes, RANDO® slabs phantom, respectively. Each scenario tested five real‐patient breathing mimic real clinical situations. patient eight patients locations selected. performance then evaluated comparing reference extracted for overall discrepancy, expiration peak, valley accuracy. Results tests shifts variations, peak accuracy −0.009 ± 0.18 sec, no time delay found compared reference. anatomical tests, had 1.6 1.2 % −0.12 0.28 sec accuracy, 0.01 0.15 studies, −1.05 3.0 −1.55 1.45% 0.04 0.13 −0.01 waveforms. Conclusions An innovative capable recognizing developed. With aid extra surrogate, improvement prediction traditional statistical methods, obtained. It enables us employ it ground truth 4D‐CBCT reconstruction, gating treatment, clinic implementations that require accurate information.