Principal Component Analysis of Dose Clustering Patterns for Parotid Toxicity Modeling in Head and Neck Radiation Therapy
M Chao1*, I El Naqa2, Y Lo1, J Penagaricano2, (1) Mount Sinai Medical Center, New York, NY, (2) Moffitt Cancer Center, Tampa, FL
Presentations
MO-IePD-TRACK 5-7 (Monday, 7/26/2021) 3:00 PM - 3:30 PM [Eastern Time (GMT-4)]
Purpose: To develop a predictive model using principal component analysis (PCA) and logistic regression (LR) for parotid complication based on 3D dose clustering in head-and-neck cancer irradiation.
Methods: Spatial dose distribution inside the parotid gland was incorporated into the percolation- based cluster model for 155 patients receiving intensity modulated radiation therapy from three institutions. Radiation induced complication was classified into two categories for patients who never developed xerostomia and those who developed xerostomia with grade 1 and above, respectively. For each patient, forty bins of the normalized largest size of dose clusters were formed with cutoff dose ranging from 11-50 Gy in 1 Gy step size. PCA was applied to these 40-column variables from all enrolled patients. The principal components (PCs) were determined and subsequently input into a multinomial logistic regression model for quantitative analysis of toxicity correlation. The returned fitting parameters were examined to establish the correlation between the principal components and parotid gland toxicity. In addition, three cluster connectivity options representing cluster formation were evaluated for the association of optimal morphological clustering patterns with the gland toxicity.
Results: PCA analysis revealed that first three PCs are the most important ones with the first PC explaining over 91% for all three connectivity variances. The scatter plots between the dominating PCs show distinct clustering patterns among the two groups of patients. These first three PCs were input into the LR model and the p values (<0.05) were found for the first and the third principal components. No significant difference was observed among three cluster connectivity choices.
Conclusion: The PCA-LR model can reveal correlations between the percolation-based cluster model and parotid toxicity. Thus, it can be a valuable tool to augment normal tissue complication modeling and facilitate treatment planning decision making in the head and neck radiation therapy.
ePosters
Keywords
NTCP, Radiobiology, Radiation Therapy
Taxonomy
TH- Radiobiology(RBio)/Biology(Bio): RBio- LQ/TCP/NTCP/outcome modeling
Contact Email