Author: Ryan Andosca, Peter Boyle, Minji Victoria Kim, Michael Vincent Lauria, Daniel A. Low, Claudia R. Miller, Drew Moghanaki, Louise Naumann, Dylan P. O'Connell, Ricky R Savjani 👨🔬
Affiliation: Department of Radiation Oncology, University of California, Los Angeles, UCLA, University of California, Los Angeles, UCLA Radiation Oncology 🌍
Purpose: To demonstrate the performance of the model-based CT scanning protocol, 5DCT, as an alternative to 4DCT.
Methods: 5DCT imaging results for 242 patients were analyzed. Implementation of the 5DCT protocol included, a) acquiring 25 consecutive fast-helical free-breathing CTs alongside a respiratory bellows signal, b) use of the imaged diaphragm dome to calibrate and correct the breathing surrogate, c) deformable scan registration using deedsBCV (P. Matthias), and d) motion-model fitting. The calibrated surrogate was segmented to analyze individual breaths. The 90th percentile motion-model residuals were analyzed to estimate motion model errors. Finally, for the 123 patients that had tumors in the lungs, the model error analyses were repeated for the tumor regions. Errors in tumors were compared against the errors of the 90th percentile to determine if the latter error method could be used where there was no contoured tumor. Tumor-based motion and model errors were compared to breathing pattern analyses and surrogate calibration to identify if those parameters drove tumor model precision.
Results: Of the 242 patients, 224 were able to be analyzed due to systematic errors such as poor bellows placement. Overall, the bellows calibration root mean squared residuals ranged from 1.52-15.0mm, with a median of 3.98mm. Mean and SD of the breathing amplitudes and periods were 16.4±6.4mm and 4.13±1.37s, respectively. Tumor motion model error was compared against breathing amplitude, amplitude variation, period variation, and surrogate calibration precision. The resulting cross correlations were 0.441, 0.331, 0.174, and 0.077. 93% of the patients had their 90th percentile errors <3mm, with one tumor having errors >3mm.
Conclusion: The 5D model produced tumor-specific motion models with less than 3mm of error for all but one of the cases. However, no positive correlation between model error and other parameters was found. This makes improving the 5DCT workflow more challenging moving forward.