Determination of optimal timepoint for the prediction of head/neck anatomical changes during radiotherapy using principal component analysis.
Tsiamas P, Bagher-Ebadian H, Siddiqui F, Liu C, Hvid C, Brown S, Wen N, Benjamin M, and Chetty I. Determination of optimal timepoint for the prediction of head/neck anatomical changes during radiotherapy using principal component analysis. Med Phys 2018; 45(6):e672.
Purpose: The goal of this study was to identify the optimal time point during fractionated RT for prediction of geometric changes observed on daily CBCT images, using Principal Component Analysis (PCA). Optimal time point reflects the fraction beyond which geometric changes begin to stabilize. Methods: Eighteen H&N cancer patients were treated with to prescribed dose of 60-70 Gy (2 Gy/fx). Seven normal organs (left and right parotid, left and right submandibular gland, mandible, pharyngeal constrictor muscles and spinal cord) were analyzed. PCA models for each organ were developed for individual patients IP (f, PCi) for different fractions (f = 5, 10, 15, 20, 25 and last fraction) and for different Principal Components (PCs) PCi (1 ≤ i ≤ 5). Displacement-vector-field (DVF) histograms as predicted by the IP (f, PCi) model were compared against the actual DVF's magnitudes for each organ (DVF's were computed from deformable registration between pCT's and the final CBCT fraction). Comparison of DVF histograms was performed using the Kolmogorov-Smirnov (KS) test. Optimal time point was defined as the fraction number fo at which computed histograms resulted in similar KS values, with predicted spatial displacements for each organ being described by a minimum number of PCs. Results: For all scenarios, 2-4 PCs predicted spatial displacement at >95% Confidence Level (CL). Differences in percentage predicted spatial displacement between mean IP models for each organ ranged from 2.8% ± 1.8% (1st-PC) to 0.6% ± 0.4% (4th-PC) when f = last fraction, and 1.2% ± 0.9% (1st-PC) to 0.2% ± 0.2% (4th-PC) when f = 5. KS analysis for left parotid gland showed that 13 ± 3 fractions was the optimal fraction number fo beyond which geometric changes begin to plateau. Conclusion: PCA can be used for prediction of organ spatial displacements for patients with head/neck cancers undergoing fractionated RT. Initial results suggest that the optimal time point for prediction of geometric changes is in the range of 13 ± 3 fractions. Additional investigation is warranted.